Home

UNIVERSIDADE DE COIMBRA

1. 135 Figure 5 4 Survival function for hypothetical CA 125 AUC PCIE 137 6 CA 125 AUC robustness as a predictor for a complete response to primary chemotherapy in ovarian cancer a multivariant analysis using ANN Figure 6 1 Multilayer perceptron MLP network model for each member in the ensemble formation 3 inputs connected to 6 input neurons one hidden layer with 3 neurons and 1 output neurons connected to 1 output 152 Figure 6 2 Response graph for ensemble 2 prediction of a complete response to chemotherapy test using CA 125 AUC value as an independent continuous variable 157 7 The Potential for the Rate of CA 125 Increase to Predict Ovarian Cancer Relapse Figure 7 1 Mean CA 125 increase rate in patients with or without relapse aos 171 Figure 7 2 ROC curve CA 125 increase rate to predict patient relapse aso 171 8 CA 125 AUC as a predictor for epithelial ovarian cancer relapse Figure 8 1 Hypothetical CA 125 evolution after primary treatment of patient with ovarian 190 Figure 8 2 Algorithm for implementation of both criteria A and B in clinical practice uiuo boca e Rada 200 29 1 INTRODUCAO Introdu o 1 1 CANCRO EPITELIAL DO OV RIO 1 1 1 Introdu o Os cancros epiteliais do ov rio s o tumores constitu dos por um ou mais
2. 100 Figura 3 11 Gr fico da fun o log stica com a 1 101 Figura 3 12 Gr fico da fun o tangente hiperb lica 102 Figura 3 13 Modelo de um neur nio artificial no qual est representada Dlas uan an 103 Figura 3 14 Representa o gr fica de uma rede neuronal artificial com uma camada de entrada com quatro neur nios artificiais duas camadas ocultas uma com seis e com cinco neur nios e uma camada de sa da com apenas um 105 27 28 4 CA 125 half life breakpoint between a good and poor prognosis in patients with ovarian cancer Figure 4 1 Survival curve Kaplan Meier in agreement with CA 125 half life breakpoint of 16 days 119 5 CA 125 AUC as a new prognostic factor for patients with ovarian cancer Figure 5 1 CA 125 AUC hypothetical determination in a patient with ovarian cancer treated with first line chemotherapy after cytoreductive surgery A patient with an optimal CA 125 serum levels evolution during treatment B patient with an unstable CA 125 serum levels during P 129 Figure 5 2 125 Mean plot according to patients FIGO Tumor SIAC tin Sta senile Ca E ia RUE 135 Figure 5 3 CA 125 Mean plot according to patient response to primary
3. Nome 2 4 B Doente 2 m j O 3 Doente 3 CIMAGO Concentra es de depois do tratamento pri Jas por data Doente 2 An lise Aumento R pido do CA 125 AUC Data 125 BH y 11 10 2006 10 CA 125 AUC2 12 12 2006 15 a 10 12 20081 730 1 25 125 1 0 85 2 CA 125 Caos CA 125 Ja zi 1 3 17 92200718 08 200716 02 200816 08 2008 ge Sair c 2006 A Mano Falc o e C de Oliveira tS Figura 9 1 Ecr principal do CA 125 Analyser O CA 125 Analyser tem por objectivo ajudar a interpretar os aumentos do CA 125 em doentes consideradas como curadas ap s tratamento prim rio do cancro do ov rio A interpreta o do CA 125 feita usando par metros cin ticos como a rea sobre a curva do CA 125 normalizada pelo tempo e a velocidade de aumento do CA 125 A an lise dos valores do CA 125 feita medida que as doentes s o acompanhadas ap s o tratamento prim rio e origina alertas de recidivas baseando se em tr s c lculos independentes a Detec o do aumento brusco da ASC do CA 125 calcula se a ASC at ltima e pen ltima concentra o do CA 125 e verifica se existe um aumento superior em 25 b Detec o da ultrapassagem de um cut off da ASC do CA 125 calcula se a ASC total e verifica se ocorreu um ultrapassar dos cut off s 25 50 75 ou 100 Ul ml 213 Conclus es Finais C C lcu
4. 136 Table 5 4 Cox proportional hazard model results for CA 125 AUC as an independent factor for predicting patient overall survival scc d ei eU redu EQ Uo Cor DO vede DOR 137 6 CA 125 AUC robustness as a predictor for a complete response to primary chemotherapy in ovarian cancer a multivariant analysis using ANN Table 6 1 Input and output variables conversion used in the artificial neural network models 150 Table 6 2 Summary of patient characteristics 154 Table 6 3 Input combinations used for each ensemble 155 Table 6 4 Sensitivity specificity positive predictive value PPV negative predictive value NPV false positive FP rate false negative FN rate and accuracy for each ensemble tested to predict a complete chemotherapy liec 156 Table 6 5 Rank of each variable given by the sensitivity analysis performed on each ensemble 156 7 The Potential for the Rate of CA 125 Increase to Predict Ovarian Cancer Relapse Table 7 1 Sensitivity specificity positive predictive value PPV and accuracy for several CA 125 increase rate cut offs to predict ovarian cancer patient relapse 172 8 CA 125 AUC as a predictor for epithelial ovarian cancer relapse Table 8 1 Increase factors F used in Criterion A
5. 192 Table 8 2 Summary of patient characteristics 193 Table 8 3 Sensitivity specificity positive predictive value PPV negative predictive value NPV false positives FP false negatives FN 96 and accuracy achieved for each O O ener heen tie 194 Table 8 4 Lead time to recurrence using the best accuracies for each criterion ra 196 25 LISTA DE FIGURAS 1 Introduc o Figura 1 1 Estrutura proposta da mol cula de 125 54 3 Material e M todos Figura 3 1 Ecr principal da aplica o Filtra Dados 72 Figura 3 2 Fluxograma da recolha e tratamento dos dados 73 Figura 3 3 Exemplo de uma curva 77 Figura 3 4 Exemplo de uma curva ROC estimada tamb m designada por curva ROC n o 78 Figura 3 5 Exemplo de uma curva de sobreviv ncia obtida pelo m todo de 85 Figura 3 6 Representac o em diagrama do sistema nervoso 95 Figura 3 7 Ilustra o de um neur nio biol gico 96 Figura 3 8 Modelo de um neur nio 98 Figura 3 9 Gr fico da fun o 99 Figura 3 10 Gr fico da fun o tipo rampa com a 2
6. 45 46 Introdu o for Research and Treatment of Cancer EORTC a quimioterapia na doen a inicial capaz de diminuir o risco de recidiva e prolongar a sobreviv ncia global No entanto mulheres no est dio la ou Ib sem tipo histol gico c lulas claras tumores bem diferenciados G1 e sujeitas a uma citorredu o ptima n o apresentam benef cios resultantes da quimioterapia Em todos os casos restantes a quimioterapia adjuvante est indicada com um regime de carboplatina ou a combina o de carboplatina com paclitaxel 33 e a introdu o de 6 ciclos em vez de 3 ciclos de quimioterapia nas doentes de alto risco para o cancro do ov rio mas com doen a inicial pode reduzir a recidiva 37 Apesar da aceita o do tratamento corrente existem v rias quest es pendentes a A adi o de mais agentes citot xicos b Substitui o do paclitaxel por outro taxano como o docetaxel c O n mero de ciclos e a via de administra o d Dura o do tratamento e administra o de terapia de manuten o Nas mulheres mais idosas e com cancro do ov rio avan ado deve ser feito um esfor o na utiliza o de quimioterapia baseada nas combina es de platina 38 A redu o de dose carboplatina paclitaxel pode ser melhor tolerada com igual efic cia e a idade e v rios par metros geri tricos devem ser considerados na dosagem da quimioterapia nas doentes idosas 39 O cancro do ov rio durante a sua evolu o
7. 8 5 References 9 Conclus es Finais 9 1 Refer ncias RESUMO O cancro do ov rio ocupa a posic o cimeira no que diz respeito mortalidade das neoplasias invasivas do aparelho genital feminino O uso de marcadores tumorais de enorme utilidade na pr tica cl nica podendo ser utilizados em diferentes fases da doenga incluindo testes de rastreio diagn stico factores de progn stico e para monitorizar o tratamento e ou detecc o de uma reca da Na pr tica cl nica do cancro do ov rio o marcador tumoral mais utilizado para monitorizar a evolu o da doen a o CA 125 O objectivo prim rio do trabalho proposto para a presente disserta o de doutoramento consistia no desenvolvimento de uma ferramenta inform tica capaz de melhorar a interpreta o do CA 125 nomeadamente no que diz respeito sua capacidade de prever a recidiva em doentes com cancro do ov rio Ao longo do presente trabalho foram ent o testados v rios par metros cin ticos descritores das concentra es do CA 125 tendo sido inclusivamente introduzido pela primeira vez um novo par metro a rea sob a curva do CA 125 normalizada pelo tempo Procurou se numa primeira fase estudar a cin tica do CA 125 durante o tratamento prim rio com o objectivo de inferir sobre o progn stico da doen a e numa fase posterior estudar o comportamento do CA 125 ap s o tratamento prim
8. Access com ambas as tabelas FiltraDados Informa o das Doentes e da Cin tica do CA 125 em Excel Programa de Estat stica ou Redes Neuronais Artificiais Resultados Figura 3 2 Fluxograma da recolha e tratamento dos dados 73 74 Material e M todos 3 2 M TODOS ESTAT STICOS CONVENCIONAIS 3 2 1 Introdu o No decorrer do trabalho para al m da estat stica descritiva normal foram utilizados outros m todos estat sticos Passaremos sua descri o tendo sido usados com o objectivo de analisar as subpopula es em estudo com a finalidade de determinar a rela o entre a evolu o do CA 125 nas diferentes fases de seguimento das doentes com caracter sticas da doen a e sua evolu o 3 2 2 Avalia o de Testes de Classifica o Bin ria An lise de Sensibilidade e Especificidade Na avalia o de testes de classifica o bin ria s o usadas medidas estat sticas como a sensibilidade e a especificidade Na pr tica cl nica estes testes podem ser inclu dos nos testes de rastreio testes de diagn stico ou nos testes de caracteriza o de determinada doen a Observa es cl nicas ou testes laboratoriais podem assim caracterizar uma pessoa como saud vel ou n o ou inclui la ou n o numa dada categoria patol gica No entanto uma pessoa pode ser erradamente classificada pois podem ocorrer erros nas observa es cl nicas ou nos testes laboratoriais To
9. UNIVERSIDADE DE COIMBRA FACULDADE DE FARM CIA AVALIA O DA UTILIDADE DE PAR METROS CIN TICOS DERIVADOS DO CA 125 NO ACOMPANHAMENTO DO CANCRO EPITELIAL DO OV RIO DISSERTA O DE CANDIDATURA AO GRAU DE DOUTOR APRESENTADA FACULDADE DE FARM CIA DA UNIVERSIDADE DE COIMBRA Ant nio Jo o Sales Mano 2012 Trabalho desenvolvido sob orienta o cient fica dos Professores Doutores Am lcar Falc o e Maria Margarida Caramona no Laborat rio de Farmacologia da Faculdade de Farm cia da Universidade de Coimbra Trabalho financiado pela Fundac o para Ci ncia a Tecnologia no mbito do programa PRAXIS XXI SFRH BD 3248 2000 AGRADECIMENTOS Ao Professor Doutor Am lcar Falc o pela cient fica pela sua disponibilidade pelo constante incentivo apoio fundamentalmente pela amizade Professora Doutora Margarida Caramona pela orientac o cient fica por todo o apoio amizade e facilidades concedidas para a realizac o deste trabalho no Laborat rio de Farmacologia da Faculdade de Farm cia da Universidade de Coimbra Ao Professor Doutor Carlos de Oliveira pela ced ncia da informag o cl nica das doentes do Servico de Ginecologia dos Hospitais da Universidade de Coimbra pelos diversos convites feitos para a divulga o deste trabalho pelo apoio e interesse evidenciados e pela estima e amizade demonstradas Ao Laborat rio de Hormonologia e Monitoriza o de F rmacos
10. Weekly Low Dose Carboplatin and Paclitaxel in the Treatment of Recurrent Ovarian and Peritoneal Cancer Gynecol Oncol 2003 88 51 57 Beshara N Fung Kee Fung M Faught W The role of topotecan as second line therapy in patients with recurrent ovarian cancer Eur J Gynaec Oncol 2002 23 4 287 290 Herzog TJ Clinical experience with topotecan in relapsed ovarian cancer Gynecol Oncol 2003 90 S3 S7 Bhoola SM Coleman RL Herzog T Morris R Bryant C Estes JM Alvarez RD Retrospective analysis of weekly topotecan as salvage therapy in relapsed ovarian cancer Gynecol Oncol 2004 95 564 569 Parazzini F Raspagliesi F Guarnerio P Bolis G Role of secondary surgery in relapsed ovarian cancer Crit Rev Oncol Hematol 2001 37 121 125 CA 125 AUC as a predictor for epithelial ovarian cancer relapse 20 21 22 23 24 25 26 27 28 G nh r M Orta F Arvas M K sebay D S nmezer M K se K The role of secondary cytoreductive surgery for recurrent ovarian cancer Gynecol Oncol 2005 97 74 79 Gronlund B Lundvall L Christensen IJ Knudsen JB Hogdall C Surgical cytoreduction in recurrent ovarian carcinoma in patients with complete response to paclitaxel platinum EJSO 2005 31 67 73 Munkarah AR Coleman RL Critical evaluation of secondary cytoreduction in recurrent ovarian cancer Gynecol Oncol 2004 95 273 280 Gadducci A lacconi P Cosio S Fanucchi A Cristofani R Genazzani AR Complete Salv
11. o 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 Rocconi R P et al The timing of normalization of CA 125 levels during primary chemotherapy is predictive of survival in patients with epithelial ovarian cancer Gynecol Oncol 2009 114 2 p 242 5 Canevari S et al Molecular predictors of response and outcome in ovarian cancer Crit Rev Oncol Hematol 2006 60 1 p 19 37 Aggarwal P and S Kehoe Serum tumour markers in gynaecological cancers Maturitas 2010 67 1 p 46 53 Rapkiewicz A V et al Biomarkers of ovarian tumours Eur J Cancer 2004 40 17 p 2604 12 Moore R G S MacLaughlan and R C Bast Jr Current state of biomarker development for clinical application in epithelial ovarian cancer Gynecol Oncol 2010 116 2 p 240 B Gadducci et al Serum tumor markers in the management of ovarian endometrial and cervical cancer Biomed Pharmacother 2004 58 1 p 24 38 Gadducci A et al Serum and tissue biomarkers as predictive and prognostic variables in epithelial ovarian cancer Crit Rev Oncol Hematol 2009 69 1 p 12 27 Kong F et al Using proteomic approaches to identify new biomarkers for detection and monitoring of ovarian cancer Gynecol Oncol 2006 100 2 p 247 53 Bast R C Jr et al Heactivity of a monoclonal antibody with human ovarian carcinoma J Clin Invest 1981 68 5 p 1331 7 Whitehou
12. podemos escrever que para o neur nio k a entrada final u dada por m Up f rmula 3 29 j l em que m o n mero total de entradas 2 Xm SAO OS sinais de entrada wi Wie Wkm S O OS pesos sin pticos do neur nio k A entrada total aplicada uma func o designada por func o de activa o q tamb m conhecida por fun o de transfer ncia Assim matematicamente podemos dizer que o valor da sa da do neur nio dada y Pu gt y f rmula 3 30 JA sa da yx d se tamb m o nome de n vel de activac o do neur nio k De notar que o objectivo da func o de activac o limitar o valor Material e M todos da sa da do neur nio a um intervalo finito normalmente 0 1 ou 1 1 V rios tipos de func es de activac o podem ser aplicadas podendo estas ser discretas ou continuas num dado intervalo Como exemplos de func es de activac o temos 14 a Func o limiar representada na Figura 3 9 e definida por ee amada E Tre ar 0 6 4 2 0 2 4 5 u k Figura 3 9 Gr fico da func o limiar Quando se utiliza esta func o de activac o o modelo descrito na figura anterior o modelo de McCulloch e Pitts b Func o de sinal lu gt 0 4 0 u 20 f rmula 3 32 lu 0 c Fun o de limiar por partes fun o tipo rampa representada da Figura 3 10 e definida por 99 100 Material e M
13. CA 15 3 and CA 72 4 antigens in patients with a pelvic mass Am J Obstet Gynecol 2002 187 2 p 385 92 Andersen M R et al Use of a Symptom Index CA125 and HE4 to predict ovarian cancer Gynecol Oncol 2010 116 3 p 378 83 57 58 Introdu o 20 21 22 23 24 25 26 27 28 29 30 Hays J L et al Proteomics and ovarian cancer integrating proteomics information into clinical care J Proteomics 2010 73 10 p 1864 72 Husseinzadeh N Status of tumor markers in epithelial ovarian cancer has there been any progress A review Gynecol Oncol 2011 120 1 p 152 7 Meani F et al Clinical application of proteomics in ovarian cancer prevention and treatment Mol Diagn Ther 2009 13 5 p 297 311 Liz Y Han R L C Ovarian Cancer Staging Operative Techniques in General Surgery 2007 9 2 p 53 60 Vine M F et al Characterization of prediagnostic symptoms among primary epithelial ovarian cancer cases and controls Gynecol Oncol 2003 90 1 p 75 82 Goff B A et al Frequency of symptoms of ovarian cancer in women presenting to primary care clinics JAMA 2004 291 22 p 2705 12 Angioli R et al Diagnostic open laparoscopy in the management of advanced ovarian cancer Gynecol Oncol 2006 100 3 p 455 61 Nam E J et al Diagnosis and staging of primary ovarian cancer correlation between PET CT Doppler US and CT or MRI Gynecol Oncol
14. ho t representa como o risco varia com o tempo e a fun o g x representa o efeitos dos factores de progn stico A fun o de risco de base baseline hazard function ho t pode ser interpretada como sendo a fun o de risco quando todos os factores de progn stico s o ignorados g x 1 87 Material e M todos O r cio de risco de dois indiv duos com dois conjuntos de factores de progn stico x e x2 dado heln ASQ 80 f rmula 3 24 h tlx MEER g e uma constante independente do tempo O modelo de riscos proporcionais de Cox assume que a func o g x uma func o exponencial dos factores de progn stico ou seja Ys g x e formula 3 25 sendo b b b gt bp o conjunto de coeficientes par metros do modelo dos factores de progn stico Estes coeficientes podem ser estimados a partir dos dados observados e indicam a magnitude do efeito dos factores de progn stico correspondentes Assim a func o de risco pode ser escrita gt p h tlx hte h Dexp 2b x f rmula 3 26 OU In h t1x 0 Xbx f rmula 3 27 j l Demonstra se que Sqlx s D 3 f rmula 3 28 Material e M todos Assim factores de progn stico podem ser incorporados em func es de sobreviv ncia Para estimar o conjunto de coeficientes Cox prop s uma func o de verosimilhan a parcial com o pressuposto de n o haver empates nos tempos de sobreviv ncia obs
15. mbolos e suas propriedades a aproxima o atrav s das redes neuronais adopta a met fora cerebral que sugere que a intelig ncia emerge a partir de um conjunto de elementos de processamento ligados entre si e cada um executando uma tarefa computacional simples Assim a base de desenvolvimento das redes neuronais artificiais n o mais do que a constata o do funcionamento do c rebro biol gico e como este processa a informa o de um modo totalmente diferente de um computador convencional 13 Podemos ent o definir redes neuronais artificiais como sendo sistemas computacionais constitu dos por unidades de processamento simples maci amente distribu das paralelamente que tentam simular algumas funcionalidades do c rebro biol gico como 14 15 91 92 Material e M todos a Aprendizagem na qual o conhecimento adquirido pela rede a partir um conjunto de treino atrav s de um algoritmo de aprendizagem Conhecimento que armazenado sob a forma dos pesos das ligac es entre as unidades de processamento Generaliza o onde o conhecimento armazenado anteriormente pode ser utilizado na presen a de novos casos Com o objectivo de atingir estas tr s funcionalidades as redes neuronais apresentam outras caracter sticas teis e que justificam tamb m a sua aplicabilidade face a outros m todos Entre outras destacam se a Auto organiza o a rede constr i a sua pr pria representa o
16. o da recidiva Mais uma vez a aplica o da ASC do CA 125 apresenta vantagem sobre a velocidade de aumento melhor exactid o e tempo m dio at a recidiva mais consistente e sem depender de nenhuma concentra o individual do CA 125 havendo um efeito atenuante devido variabilidade das concentra es do CA 125 E permitiu apresentar um tempo mediano at recidiva muito superior ao sugerido por Rustin e colaboradores 111 vs 63 dias com valores de exactid o semelhantes 8596 vs 88 o que pode ter implicac es cl nicas de grande alcance Finalmente importa salientar que os par metros cin ticos derivados do CA 125 com destaque para a ASC podem ser utilizados em estudos cl nicos prospectivos na comparac o de novos tratamentos com os utilizados na actualidade ou ent o servir para ajustar o modo como determinada doente deve ser acompanhada 15 ABSTRACT Ovarian cancer comes in first place when we consider mortality related to malignant neoplasms of the female reproductive tract The use of tumour markers is very useful in clinical practice for screening diagnostic and treatment monitoring purposes In clinical practice CA 125 is the most commonly used tumour marker to monitor ovarian cancer progression Therefore the primary aim of this doctoral thesis was the development of software to assist interpretation of CA 125 given that it is a very useful tool to predict recurrence within individuals with
17. ovarian cancer Several kinetic parameters that relate to CA 125 concentrations were examined For the first time time standardised area under the curve AUC of CA 125 was introduced as an additional parameter The primary objective was to analyse the kinetics of CA 125 during the initial stages of treatment to verify disease prognosis Post early treatment CA 125 was then studied in an attempt to predict cancer recurrence The half life of CA 125 is one of the most closely investigated kinetic parameters during the initial stages of treatment The half life is also used to determine prognosis However it was not always possible to determine a precise half life from our population data For this reason AUC was introduced as an additional parameter 17 18 The advantage of CA 125 AUC is that its calculations do not require any assumption of individual CA 125 levels Consequently it allows an accurate numerical classification During early treatment CA 125 AUC may permit a numerical classification of disease response following a course of adjuvant chemotherapy This would prove useful when comparing different treatment regimens particularly in prospective studies The second part of the overall investigation involved study of potential correlations between CA 125 and disease relapse The increasing speed of CA 125 and the use of CA 125 AUC on relapse detection were both considered As before CA 125 AUC provided benefits o
18. stico antecipado pode ser poss vel se houver um reconhecimento da combina o de sintomas aparentemente n o relacionados 19 Sintomas mais graves e frequentes que o esperado devem ser investigados pois podem estar associados a massas ov ricas benignas ou malignas 20 Normalmente o diagn stico do cancro do ov rio requer laparotomia explorat ria Se na base da hist ria e exame cl nico existirem suspeitas de cancro epitelial do ov rio uma ecografia endovaginal normalmente aconselhada Na presen a de quistos complexos no ov rio a biopsia percut nea n o se deve realizar sob risco de derramamento do tumor sendo necess ria uma cirurgia para definir o diagn stico 2 A laparoscopia aberta como ferramenta de diagn stico pode ser til para avaliar a extens o da doen a com est dio avan ado permitindo avaliar as doentes que puderem ser sujeitas a cirurgia de citorredu o e com melhorias na taxa de citorredu o ptima 21 A utiliza o da tomografia por emiss o de positr es PET combinada com a tomografia axial computorizada TAC ou PET CT parece ser superior ecografia p lvica TAC abdomino p lvica e resson ncia magn tica nuclear RMN p lvica no diagn stico do cancro do ov rio 22 Apesar de exames imagiol gicos poderem at certo ponto detectar a extens o intra abdominal da doen a o estadiamento do cancro do ov rio cir rgico e utiliza se a classifica o d
19. 125 cut off h to Cytoreductive First line surgery date chemotherapy end date A A A A CA 125 AUC 2 3 4 Serum Concentration IU ml 1 1 Normal CA 125 cut off 4 emm EE RNA RA N lt O t2 Cytoreductive First line surgery date chemotherapy end date Figure 5 1 CA 125 AUC hypothetical determination in a patient with ovarian cancer treated with first line chemotherapy after cytoreductive surgery A patient with an optimal CA 125 serum levels evolution during treatment B patient with an unstable CA 125 serum levels during treatment 129 130 CA 125 AUC as a new prognostic factor for patients with ovarian cancer A statistical analysis was conducted The Mann Whitney U test for two groups and Kruskal Wallis test for three or more groups were used to compare the CA 125 AUC across subgroups of patients depending on the FIGO stage patient final state tumor grade histological type residual disease 22 cm and response to primary chemotherapy The area under the Receiver Operating Characteristic ROC curve was established for the discrimination by CA 125 AUC in predicting the patient final state overall survival equal or superior to one three and five years and a full response to chemotherapy Sensitivity specificity positive predictive values PPV and overall accuracy were also determined for several CA 125 AUC cut offs In addition a multivariant regression an
20. 1999 que descreve a utiliza o de diferentes par metros cin ticos na interpreta o da evolu o temporal de v rios marcadores tumorais incluindo o CA 125 Assim em vez de utilizar isoladamente a concentra o do CA 125 passar amos a descrever o seu comportamento atrav s do uso de par metros cin ticos derivados Ao longo da presente disserta o foram ent o testados v rios par metros cin ticos descritores das concentra es do CA 125 tendo sido inclusivamente introduzido pela primeira vez um novo par metro a rea sob a curva do CA 125 normalizada pelo tempo 67 3 MATERIAL E M TODOS Material e M todos 3 1 POPULA O A popula o inicial foi recolhida retrospectivamente a partir da base de dados do Servi o de Ginecologia dos Hospitais da Universidade de Coimbra A base de dados estava desenvolvida em FileMaker Pro da FileMaker Inc vers o 5 e partiu da recolha de informa o cl nica das doentes a partir dos respectivos processos cl nicos sob a orienta o da Dra Isabel Godinho A informa o foi exportada para Microsoft Excel resultando num total de 339 doentes diagnosticadas maioritariamente na d cada de 90 do s culo passado Apesar de algumas concentra es do CA 125 estarem dispon veis na informa o recolhida no Servi o de Ginecologia neste trabalho optou se pela recolha integral de todas as concentra es s ricas do CA 125 registadas na base de dados do Laborat rio de
21. 296 had a tumour grade 3 and fifty six 50 5 patients had no tumour grade information Twenty four 21 696 patients had a residual tumour greater than 2 cm after surgery In thirty four 30 696 patients the primary treatment consisted of surgery only in forty seven 42 496 cases the patients were also given adjuvant chemotherapy and in thirty 27 096 the primary treatment included consolidation chemotherapy The mean duration of treatment for patients that only received adjuvant chemotherapy was 4 1 1 4 11 5 S E 0 2 months while for patients that received both types of chemotherapy was 7 7 2 6 39 1 S E 1 3 months The mean follow up time between the first and the last CA 125 level was 54 3 1 4 110 0 S E 23 1 months in which a mean of 12 9 3 40 5 0 7 CA 125 serum levels were obtained per patient with a mean of 141 9 10 5 521 2 5 8 7 days between samples The mean follow up time between the end of primary treatment and the evaluation date was 69 6 months 5 1 229 8 S E 4 4 In that time twenty six 23 4 patients had a recorded relapse seventy seven 69 496 had no recorded relapse and in eight 7 296 this information was unavailable Relapse was recorded in eight 20 796 patients with FIGO stage I Il against seventeen 65 3 patients with FIGO stage III IV the FIGO stage was missing in one patient with relapse The mean time from the conclusion of treatment to relapse detection was 27 0 5 1 89 4 5 3 9
22. 95 5 FIGO tumor stage Figure 5 2 CA 125 Mean plot according to patients FIGO Tumor Stage 450 400 r 350 300 250 200 150 100 0 Mean Mean 0 95 SE RC RP SRPD Figure 5 3 CA 125 Mean plot according to patient response to primary chemotherapy CR Complete Response PR Partial Response WRDP Without Response or Disease Progression 135 CA 125 AUC as a new prognostic factor for patients with ovarian cancer Table 5 3 Sensitivity specificity positive predictive value PPV and accuracy of several CA 125 AUC cut offs for predicting the patient final state overall survival and chemotherapy complete response CA 125 cut off to predict patient final state alive CA 125 ROC AUC lt 25 lt 50 5100 5200 5300 5400 5500 5600 5800 lt 1000 AUC IU mL S E Sensitivity 0 62 0 86 0 86 0 95 0 97 1 00 1 00 1 00 1 00 1 00 Specificity 0 72 0 64 0 64 0 32 0 30 0 20 0 20 0 14 0 08 0 04 077 PPV 0 65 0 67 0 67 0 54 0 54 051 0 51 0 50 0 48 0 47 0 05 Accuracy 0 67 0 74 0 74 0 61 0 61 0 57 0 57 0 53 0 50 0 48 CA 125 cut off to predict patient overall survival 2 1 year Sensitivity 0 48 0 63 0 63 0 81 0 84 0 90 0 90 0 94 0 98 0 99 Specificity 0 83 0 67 0 67 0 25 0 25 0 17 0 17 0 17 0 17 0 08 067 PPV 0 95 0 93 0 93 0 88 0 88 0 88 0 88 0 88 0 89 0 88 0 08 Accuracy 0 52 0 63 0 63 0 74 0 76 0 80 0 80 0 84 0 87 0 87 CA 125 cut off to predict patient
23. AUC as a predictor for epithelial ovarian cancer relapse methods the use of all available CA 125 serum levels during the follow up period decrease the residual error analytical imprecision and intra individual biological fluctuation of CA 125 In fact for each CA 125 AUC normalized in time value a carry over effect can be observed cumulative effect which promote the dilution of marginal errors An important issue concerning the present methodological approach is that the mean lead time might be influenced by the CA 125 sampling scheme more precisely by the interval between any two consecutive samples Therefore the interval between any two consecutive CA 125 samples might influence the way a particular CA 125 criterion signals the relapse To be explicit the negative lead times to relapse must be considered false negatives corrupting the accuracy This is more notorious in the case of Criterion B since the number of patients with a negative lead time to relapse would make the accuracy fall when compared with Criteria A Due to the study design constrains retrospective study without any pattern the interval between CA 125 samples was not standardised between and within patients Therefore in some cases the rapid tumour burst occurred between CA 125 samples if the interval between samples was too extended Nevertheless the false negatives and false positives in Criterion A were below 10 while in Criterion B as the cut off valu
24. Chemotherapy Response Complete Response CR 52 56 5 Partial Response PR 22 23 9 Without Response or Disease Progression WR DP 13 14 1 Missing 5 5 4 Overall Survival gt 1 year 80 86 9 gt 3 years 44 47 896 gt 5 years 21 22 896 133 CA 125 AUC as new prognostic factor for patients with ovarian cancer Table 5 2 CA 125 AUC behaviour among patients according to several covariates Mean CA 125 Median CA 125 AUC AUC adiado Standard error Q25 Q75 iin IU mL IU mL FIGO Stage 18 2 2 4 14 6 12 0 20 1 II 24 6 7 6 14 4 9 8 20 5 I 147 8 30 8 54 2 23 5 199 5 P 0 05 IV 574 6 134 6 676 3 118 6 973 3 Patient Final State Deceased 234 1 44 4 83 2 23 5 319 5 Alive 40 1 10 5 16 5 12 0 34 0 ibt Tumor Grade 1 100 1 45 5 19 8 12 3 146 9 2 158 1 43 0 65 8 16 6 230 7 NS 3 238 8 114 0 44 8 37 3 375 2 Residual Disease gt 2 cm Yes 207 4 48 3 98 1 31 3 237 9 No 97 9 30 0 16 6 12 0 44 8 lt 0 05 Primary Chemotherapy Response Complete Response CR 48 8 15 9 16 8 12 0 34 1 Partial Response 251 7 65 8 98 1 51 9 319 5 lt 0 05 Without Response Disease Progression WR DP 134 316 5 107 5 116 8 54 2 344 3 125 AUC as new prognostic factor for patients with ovarian cancer CA 125 AUC CA 125 AUC mE 100 Sur 0 1 o Mean I II IH IV LL 0
25. Hormonologia e Monitoriza o de F rmacos do Servi o de Patologia dos Hospitais da Universidade de Coimbra A base de dados estava desenvolvida em Microsoft Visual FoxPro vers o 6 Foram exportadas para Microsoft Excel 14818 concentra es s ricas do CA 125 Ambas as tabelas de Excel foram inclu das numa base de dados no Microsoft Access e foi desenvolvida uma aplica o inform tica designado por Filtra Dados na qual se procedeu combina o de ambas as tabelas utilizando querys de SQL Structured Query Language linguagem de programa o usada para trabalhar com base de dados relacionais Figura 3 1 71 72 Material e M todos Filtra Dados CA 125 Cancro Ovario NUMHUC DNASC NUMHUC Data Resultado 085 DUP 12121937 28 10 1991 11 20 01 1922 13 04 1992 87 15 06 1926 29 12 1992 6 24 11 1994 16 14 02 1967 27 03 1995 39 02 07 1949 17 04 1995 72 10 11 1923 08 05 1995 33 16 10 1981 25 04 1943 25 11 1937 12 12 1927 0 5 21 05 1992 21 05 1993 21 05 1994 E Escala yy Logaritmica Imprime Gr fico a Data da Cirurgia ug DATA 1 CIR RGIA DATA DIAGN STICO DIGTADJ DF QTADJ FIM TRAT D REC 1 DATA AVAL Data 29 10 1990 07 11 1990 29 10 1990 01 03 1995 17 05 1995 pata Ayahspsc ij EST DIO RES QT ADJ pe 10 1900 FALECIDA Fasel TIPO HISTOL GICO T EPITELIAL eis Coef
26. Int OCC QOIS d cee ac bate eU 74 3 2 2 Avalia o de Testes de Classifica o An lise de Sensibilidade e Especificidade 74 3 2 3 UVA ADO ia m 76 3 2 4 Teste de 78 3 2 5 Teste de 81 3 2 6 An lise de Sobreviv ncia Curva de Kaplan Mei dace Rupe n UE 83 3 2 7 An lise de Sobreviv ncia Modelo de Risco Proporcional de 87 3 3 Redes Neuronais Artificiais nennen 90 A o ez oreste M ek 90 3 3 2 Defini o de Redes Neuronais Artificiais 91 3 3 3 Hist ria Breve das Redes Neuronais 93 3 3 4 O C rebro Biol gico 94 3 3 5 O Neur nio ANINIClal e e Ce 97 3 3 6 Topologia das 105 3 3 7 Classificac o das Redes Neuronais Areale see 107 3 3 8 Processos de 108 3 3 9 O STATISTICA Neural Networks 111 9 4 Belerenelas sodes ocio bd 113 4 CA 125 half life breakpoint between a good and poor prognosis in patients with ovarian cancer 115 4 1 io scontato dto tees neta cbt oar 117 4 2 Material Method
27. OVARIAN CANCER International Journal of Gynecology and Obstetrics 2005 CA 125 half life breakpoint between a good and poor prognosis in patients with ovarian cancer 4 1 INTRODUCTION Ovarian cancer generally treated with combination first line chemotherapy after cytoreduction surgery 1 has the highest mortality of all invasive cancers of the female reproductive system CA 125 serum concentration is usually adopted to evaluate the clinical situation in ovarian cancer patients 2 An approach to rapid evaluation of clinical response and monitoring instead of using the coarse CA 125 serum concentration is the determination of tumor marker kinetic parameters associated with changes in its concentrations such as half life and doubling time DT 3 The rate of decline in CA 125 during primary chemotherapy has been an important independent prognostic factor in several multivariate analyses 2 Several studies report the greatest difference in progression rate found at a ty of 20 days while the normal CA 125 half life value to be determined varied from 5 to 10 days 3 4 4 2 MATERIAL METHODS AND RESULTS The aim of the present work is the determination of CA 125 half life breakpoint between a good or poor prognosis in our population Retrospective clinical information was obtained from 339 patients with a diagnosis of ovarian cancer at the Gynaecology Service of Coimbra University Hospita
28. addition CA 125 AUC input in the artificial neural models was introduced as two input variable formats as a categorical variable using CA 125 AUC cut off value inferior or equal to 100 IU mL yes no or as a continuous variable using the CA 125 AUC value itself As a consequence of the adopted strategy two sets of inputs were developed In the first one the FIGO tumor stage I 11 or IV and residual disease were combined with CA 125 AUC cut off value lt 100 IU mL categorical variable while in the second the FIGO stage and residual disease were combined with the CA 125 AUC value continuous variable The input and output variable conversion used in the study is shown in Table 6 1 Table 6 1 Input and output variables conversion used in the artificial neural network models Number of Input Variables Conversion Input Neurons L One of N FIGO Stage 1 Il Ill or IV with N 4 4 Residual Disease gt 2cm Yes No Two state 1 CA 125 AUC Value IU mL Scaled linearly 1 CA 125 AUC 100 IU mL Yes No Two state 1 Number of Output Variable Conversion Output Neurons Complete primary chemotherapy response Ta 4 Yes No A restrictive factor for data analysis was the missing information which excluded the use of all the ninety two patients and the use of tumor grade as an input variable 125 AUC robustness as predictor for a complete response to primary chemotherapy in ovarian cancer a multiv
29. apresentam a condic o ou 75 76 Material e M todos seja a proporc o de indiv duos sem a condic o e que t m um teste negativo T d 2 Especifici dade f rmula 3 2 b d Valor preditivo positivo VPP a probabilidade de tendo um teste positivo apresentar a condi o ou seja a propor o de individuos que apresentam a condi o dentro de aqueles que t m um teste positivo a VPP f rmula 3 3 Valor preditivo negativo VPN a probabilidade de tendo um teste negativo n o apresentar a condic o ou seja a proporc o de individuos que n o apresentam a condic o dentro de aqueles que t m um teste negativo VPN f rmula 3 4 Exactid o a probabilidade de obter um resultado verdadeiro com o teste em causa ou seja a propor o de indiv duos que tiveram um resultado verdadeiro Exactidao gt f rmula 3 5 a b c d 3 2 3 Curvas ROC Como referido anteriormente o uso de medidas ou t cnicas que permitam medir a robustez de determinado teste torna se importante Material e M todos de um ponto de vista cl nico No entanto em v rias situac es a aplica o de determinado teste origina um valor quantitativo sendo necess rio transformar esta vari vel cont nua numa vari vel dicot mica bin ria Para se atingir este objectivo temos que utilizar um determinado valor na escala cont nua que discrimine entre duas classes esse
30. before extrapolation of the present results to other datasets Furthermore several well known ovarian cancer prognostic factors such as FIGO stage patient performance status tumor grade DNA ploidy over expression of the HER 2 neu oncogene residual disease after initial cytoreductive surgery and response to primary treatment should be taken into account together with the CA 125 increase rate Multivariant analysis will be necessary to combine all available covariates with the objective of increasing the accuracy of relapse prediction However in clinical practice early relapse detection based on CA 125 increase is only helpful if further studies clearly The Potential for the Rate of CA 125 Increase to Predict Ovarian Cancer Relapse demonstrate that the treatment for the asymptomatic patient is shown to be beneficial for her quality of life or overall survival 175 176 The Potential for the Rate of CA 125 Increase to Predict Ovarian Cancer Relapse 7 5 REFERENCES 1 Jean Michel B Francois T Christine A Jacqueline C Alain D Nelly J Fran oise L Helene V Kinetic of serum marker concentrations and usefulness in clinical monitoring Clin Chem 1999 45 10 1695 1707 2 Bast RC Jr Feeney M Lazarus H Nadler LM Colvin RB Knapp RC Reactivity of monoclonal antibody with human ovarian carcinoma J Clin Invest 1981 68 1331 1337 3 Bast RC Jr Xu FJ Yu YH Barnhill S Zhang Z Mills GB CA 125 The past and th
31. camada quando se trata de uma liga o entre n s da mesma camada b Inter camada quando se trata de uma liga o entre n s de camadas diferentes mas adjacentes c Supra camada quando se trata de uma liga o entre n s de camadas diferentes mas n o adjacentes Quando todos os n s de uma camada est o ligados a todos os n s da camada adjacente diz se que a rede est totalmente ligada conectada no caso inverso diz que est parcialmente ligada conectada Material e M todos Podemos tamb m descrever a estrutura de uma rede neuronal artificial de uma forma compacta que a resuma Assim para uma rede alimentada para a frente com m n s de entrada h n s na primeira camada oculta n s na segunda camada oculta e p n s de sa da referida como uma rede m h hz p Por exemplo podemos dizer que a rede representada na Figura 3 14 uma rede 4 6 5 1 14 Convencionalmente em redes com camadas m ltiplas a camada entrada contabilizada assim rede com n camadas tem n 1 camadas ocultas uma camada e uma camada entrada 13 3 3 7 Classificac o das Redes Neuronais Artificiais Uma possivel classificac o para os diferentes tipos de redes ser a divis o em tr s grupos redes associativas redes de extracc o de caracter sticas e redes n o adaptativas As redes associativas s o o tipo mais comum de redes e s o modelos que necessitam para o seu treino um conjunto d
32. chemotherapy response Additionally the influence of this new prognostic factor to overall survival in patients with ovarian cancer was also studied in our population 5 2 PATIENTS AND METHODS Retrospective clinical information was gathered from 339 patients with a diagnosis of ovarian cancer at the Gynaecology Service of Coimbra University Hospital CUH main database from 1990 to 2000 In addition CA 125 serum levels of these patients were obtained from the Pathology Service Hormonology amp Drug Monitoring Laboratory of CUH Only 92 patients were included in the present analysis due to the restriction of our inclusion criteria patients that underwent primary line chemotherapy within 4 months after submission to cytoreductive surgery and with a minimum of three CA 125 serum concentrations between the time of surgery and the end of chemotherapy For each patient CA 125 AUC was calculated using the following formula sum of all trapezoidareas between C and t t CA 125 AUC 5 1 where C is the first CA 125 serum concentration after the cytoreduction surgery and is the last 125 serum concentration before the end of first line chemotherapy and t and t are the corresponding dates for C and respectively Figure 5 1 CA 125 AUC as new prognostic factor for patients with ovarian cancer A B CA 125 Serum Concentration IU ml A ASA t t CA 125 AUC Normal CA
33. com um cut off a variar entre 0 3 27 De acordo com o Gynecologic Oncology Group GOG considera se uma citorredu o ptima quando as maiores les es tumorais s o inferiores a 1 em V rios ensaios cl nicos demonstraram que a citorredu o ptima tr s benef cios na resposta quimioterapia e sobreviv ncia global 25 26 No cancro do ov rio n o avangado o uso da laparoscopia uma t cnica segura e efectiva com uma sobreviv ncia global aceit vel 28 Introdu o Estudos recentes mostram que procedimentos cir rgicos cujo alvo a doen a localizada na parte superior do abd men aumenta significativamente a taxa de citorredu o ptima sem o aumento do risco de complica es 27 V rios estudos indicam que sendo o objectivo final da cirurgia de citorredu o a elimina o total da doen a residual vis vel ocorre um aumento da sobreviv ncia global 25 A introdu o e a defini o de indicadores de qualidade podem ser usados para monitorizar e melhorar a qualidade da cirurgia no cancro do ov rio 29 A cirurgia prim ria radical pode ser usada em doentes com cancro avan ado com dissemina o extensa mas n o existem estudos que demonstrem maior sobreviv ncia e com uma morbilidade maior e pior qualidade de vida A cirurgia conservadora praticada somente em mulheres f rteis com menos de 30 anos de idade e se a doente desejar preservar a fertilidade n o tiver hist r
34. curva de sobreviv ncia obtida pelo m todo de Kaplan Meier A compara o entre distribui es de sobreviv ncia bastante importante na investiga o biom dica por exemplo a compara o entre a capacidade de dois tratamentos em aumentar a sobreviv ncia As diferen as podem ser ilustradas pela constru o das curvas de sobreviv ncia estimada mas apenas se obt m uma ligeira ideia sobre a diferen a entre as distribui es n o 85 86 Material e M todos demonstrando se estas diferen as s o estatisticamente significativas Assim torna se necess rio a aplica o de um teste estat stico Tomando como exemplo a compara o entre a capacidade de dois tratamentos em aumentar a sobreviv ncia com as fun es de sobreviv ncia S t e S t respectivamente a hip tese nula a considerar Ho Si t Saft ambos os tratamentos s o igualmente eficazes contra as hip teses alternativas H1 S t gt Saft tratamento 1 mais eficaz que tratamento 2 ou H1 S t lt Saft tratamento 2 mais eficaz que tratamento 1 ou H S1 t A Saft ambos os tratamentos s o igualmente n o eficazes Quando n o existem observac es censuradas considerado um caso especial podem ser utilizados testes n o param tricos padr o como o teste de Wilcoxon ou o teste U de Mann Whitney No entanto para conjuntos de dados com observac es censuradas situac o mais comum devem utilizar se os seguintes testes t
35. da informa o medida que o treino decorre Aprendizagem adaptativa a capacidade da rede em resolver problemas a partir de uma experi ncia inicial prever o futuro conhecendo o passado podendo medida que colocada em contacto com informa o recente aprender e ir melhorando a sua capacidade de previs o Toler ncia a falhas caso ocorra uma destrui o parcial da rede a sua capacidade de generaliza o sofre uma degrada o suave e n o brusca devido distribui o da informa o ao longo da rede Computa o em paralelo importante na implementa o em hardware por exemplo em VLSI Very Large Scale Integration capaz de tirar partido deste tipo de computa o para melhorar o desempenho Material e M todos 3 3 3 Hist ria Breve das Redes Neuronais Artificiais Apesar de aparentemente ser uma tecnologia recente o trabalho em redes neuronais comecou na d cada de quarenta ainda antes do desenvolvimento dos computadores Podemos dividir a hist ria das redes neuronais artificiais em tr s pocas Um primeiro per odo de descoberta e entusiasmo seguido de um per odo de frustrac o finalmente um per odo de inovac o e ressurgimento O in cio da computac o neuronal comecou com o trabalho inovador e pioneiro de McCulloch psiquiatra e neurofisiologista e Pitts matem tico em 1943 Estes autores unificaram os estudos de neurofisiologia com a l gica matem tica formulando um modelo de um ne
36. do Servi o de Patologia dos Hospitais da Universidade de Coimbra nas pessoas da Dr F tima Leit o e do Dr Jorge Santos pela ced ncia da informa o do CA 125 pelo apoio e considera o demonstradas Dr Isabel Godinho a t tulo p stumo o meu obrigado Funda o para a Ci ncia e a Tecnologia agrade o o apoio financeiro sob a forma de uma bolsa de doutoramento SFRH BD 3248 2000 NDICE RESUMO Mm 13 17 Lista de Abreviaturas 19 Estado Tablas caida lll 23 lista de FIQUIAS nen na nun ita 27 AUN edo 31 1 1 Cancro Epitelial do 33 SPI PT PT 33 1 1 2 Diagn stico e 36 LTS FAAO e correre Eoo b Eee e o ER EFE REA 39 AA 39 1 1 3 2 Quimioterapia 42 1 1 9 9 Radiole ia 49 1 1 3 4 Outras Terapias 49 1 1 4 Factores de Progn stico 50 1 1 5 Biomarcadores Utilizados no Cancro Epitelial do Ov rio as 51 1 2 CA 125 RD 52 1 3 Biefer nclds cm 57 2 Laca 65 3 Material e M todos aa 69 9 T OQDUIA O mutum e o 71 3 2 M todos Estat sticos Convencionais 74 9 2 Ts
37. inputs connected to input neurons one hidden layer with H neurons and P output neurons connected to O outputs 155 Table 6 4 Sensitivity specificity positive predictive value PPV negative predictive value NPV false positive FP rate false negative FN rate and accuracy for each ensemble tested to predict a complete chemotherapy response Complete Response to Chemotherapy Ensemble T2 FP FN Pr diction Yes No N Sensitivity Specificity PPV NPV Accuracy 96 96 Yes 45 6 1 Complete No 8 28 83 0 88 0 81 0 88 0 81 0 86 7 2 7 2 Response to Y 4 7 2 Chemotherapy Tn E 8 0 90 0 78 0 87 083 086 84 60 Table 6 5 Rank of each variable given by the sensitivity analysis performed on each ensemble Sensitivity Analysis Al Rank Variable 15 125 AUC lt 100 IU mL 1 a FIGO Stage 3 Residual disease 2cm 1 CA 125 AUC value 2 ond FIGO Stage 3 Residual disease gt 2cm 125 AUC robustness as predictor for a complete response to primary chemotherapy in ovarian cancer a multivariant analysis using ANN Complete Response to Chemotherapy Yes No No 0 50 100 150 200 250 300 350 400 450 500 CA 125 AUC IU ml Figure 6 2 Response graph for ensemble 2 prediction of a complete response to chemotherapy test using CA 125 AUC value as an independent continuous variable 6 4 DISCUSSION In a previous study 8 CA 125 normalized in time AUC was checked as a prognostic factor f
38. months Table 8 2 d 192 CA 125 AUC as a predictor for epithelial ovarian cancer relapse In Criterion A the best accuracy 0 85 was achieved for both F values of 1 25 25 and 1 50 50 and start decreasing with F value equal or greater then 1 75 75 For Criterion B the accuracy start to increase above 8096 with a cut off of 25 IU mL 0 82 achieving a maximum with a cut off of 100 IU mL 0 86 Table 3 The mean time difference between t and obtained with Criterion A 1 25 was 5 6 1 1 16 0 S E 0 7 months and the mean lead time to relapse achieved was 181 0 37 to 843 S E 56 3 days The mean lead time to relapse achieved with Criterion B was 131 270 to 644 S E 55 9 days with a cut off of 25 IU mL 111 103 to 530 S E 51 1 days with a cut off of 50 IU mL 63 292 to 507 S E 54 9 days with a cut off of 75 IU mL and 11 377 to 264 S E 44 8 days with a cut off of 100 IU mL cut off Table 8 4 Table 8 1 Increase factors F used in Criterion A Factor 1 10 AUC greater or equal that AUC in more than 10 1 25 AUC greater or equal that AUC in more than 25 1 50 AUC greater or equal that in more than 50 1 75 AUC greater or equal that in more than 75 2 00 AUC greater or equal that AUC in more than 100 2 50 AUC greater or equal that AUC in more than 150 3 00 AUC greater or equal that AUC in more than 200 CA 125 AUC as a predictor for epithelial o
39. nglios regionais ou inguinais IV Met stases dist ncia met stases peritoneais exclu das Metastases na c psula hep tica correspondem ao est dio Ill e met stases par nquima hep tico representam est dio IV Para poder ser considerado como est dio IV o derrame pleural dever ter uma citologia positiva tamb m til a classifica o TNM tamanho do tumor T grau de extens o aos n dulos linf ticos N e presen a de met stases distantes M da Union for International Cancer Control UICC e a sua liga o com a classifica o da FIGO Tabela 1 2 Na suspeita de cancro do ov rio as doentes devem ser referenciadas para centros especializados com a interven o de ginecologistas oncol gicos 24 Considera se doen a inicial um tumor classificado do est dio a lla e doen a avan ada dos est dios llb a IV Introdu o Tabela 1 2 Rela o entre a classifica o FIGO com a classifica o TNM FIGO UICC TNM la T1a NO MO Ib T1b NO MO Ic T1c NO MO lla T2a NO MO 119 T2b NO MO IIc T2c NO MO T3b T3c ou qualquer T N1 MO IV qualquer T qualquer N M1 1 1 3 Tratamento 1 1 3 1 Cirurgia A cirurgia representa o primeiro passo de quase todos os protocolos terap uticos seja realizada com objectivo de diagn stico terap utico ou para remo o tumoral m xima 1 A cirurgia citorredutora prim ria para al m do estadiamento
40. nicas da doen a Um estudo randomizado recente MRC OVO5 EORTC 55955 mostrou n o haver evid ncias de benef cios na sobreviv ncia com a aplica o do tratamento da recidiva baseada apenas no aumento isolado do CA 125 n o estando provada a utilidade da medi o do CA 125 em doentes que obtiveram uma resposta completa no tratamento prim rio Contudo a determina o do CA 125 pode ser um output muito relevante no caso dos ensaios cl nicos 3 Um aspecto da maior relev ncia que a nossa abordagem original atrav s do uso da ASC permitiu apresentar um tempo mediano at recidiva de 111 dias 37 843 muito superior quase o dobro ao sugerido por Rustin e colaboradores de 63 dias 361 77 com valores de exactid o semelhantes 85 vs 88 4 o que pode ter implica es cl nicas de grande alcance Na realidade os par metros cin ticos derivados do CA 125 podem ser utilizados em estudos cl nicos prospectivos na compara o de novos tratamentos com os utilizados na actualidade ou ent o para ajustar o modo como determinada doente deve ser acompanhada ex frequ ncia das consultas ou pedido de exames complementares de diagn stico Conclus es Finais Com base no acumular de conhecimento foi ent o elaborado um programa inform tico o CA 125 Analyser Figura 9 1 capaz de sinalizar recidivas usando transversalmente os resultados obtidos na presente dissertac o 125 Analyser
41. o da transmiss o do impulso nervoso Na sinapse qu mica h a convers o de um sinal el ctrico em qu mico no neur nio pr sin ptico podendo ocorrer a convers o de um sinal qu mico em el ctrico na c lula p s sin ptica 14 17 Existe um est mulo m nimo a partir do qual um neur nio p s sin ptico reage ocorrendo a transmiss o do impulso nervoso a outro neur nio Mas o aumento do est mulo n o provoca o aumento da intensidade do impulso nervoso transmitido pelo neur nio A esta caracter stica d se o nome de princ pio do tudo ou nada no qual o neur nio dispara com a m xima for a ou n o dispara Assim os impulsos nervosos que chegam a um neur nio podem excit lo ou inib lo 3 3 5 O Neur nio Artificial O neur nio artificial tamb m designado por unidade de processamento o elemento b sico que constitui as redes neuronais artificiais A Figura 3 8 mostra um modelo de um neur nio artificial 97 98 Material e M todos X1 Wk1 p u X2 sa da fung o de y activac o Xm Wim Figura 3 8 Modelo de um neur nio artificial A cada unidade de processamento chega um conjunto de entrada input e cada entrada multiplicada de um valor designado por peso wx an logo for a sin ptica Todos as entradas depois de multiplicadas pelos respectivos pesos s o somadas dando origem entrada final ou total final ou total 13 Matematicamente
42. radiotherapy in the treatment of advanced ovarian cancer Radiother Oncol 1999 53 3 p 213 8 Alberto Biete 1 V Angels Rovirosa Blanca Farr s Francesc Casas Carlos Conill Whole abdominal radiotherapy in ovarian cancer Reports of Practical Oncology amp Radiotherapy 2010 15 2 p 27 30 Introdu o 59 60 61 62 63 64 65 66 67 68 69 70 71 72 E et al Effective palliative radiotherapy for symptomatic recurrent or residual ovarian cancer Gynecol Oncol 2006 102 2 p 204 9 See H T et al Targeted therapy for epithelial ovarian cancer current status and future prospects Int J Gynecol Cancer 2003 13 6 p 701 34 Vaughan M et al mproving outcomes in ovarian cancer N Z Med J 2006 119 1242 p U2171 Holschneider C H and 4 9 Berek Ovarian cancer epidemiology biology and prognostic factors Semin Surg Oncol 2000 19 1 p 3 10 Wimberger P et al Prognostic factors for complete debulking in advanced ovarian cancer and its impact on survival An exploratory analysis of a prospectively randomized phase Ill study of the Arbeitsgemeinschaft Gynaekologische Onkologie Ovarian Cancer Study Group AGO OVAR Gynecol Oncol 2007 106 1 p 69 74 Hornung R et al Analysis of potential prognostic factors in 111 patients with ovarian cancer Cancer Lett 2004 206 1 p 97 106 Akahira J l et al Prognostic factors of stage IV epit
43. rio com o objectivo da detec o da recidiva da doen a Durante a fase de tratamento prim rio o tempo de semi vida do CA 125 um dos par metros cin ticos mais estudados sendo um factor de progn stico para o cancro do ov rio Com a nossa popula o 13 14 conclu mos que com a determina o do tempo de semi vida do CA 125 poss vel classificar o progn stico da doenca No entanto a determinac o num rica do tempo de semi vida nem sempre era poss vel Atendendo a esta limitac o foi experimentado um novo par metro cin tico a rea sob a curva do CA 125 normalizada pelo tempo ASC A ASC do CA 125 apresenta vantagens sobre outros par metros cin ticos pois n o assume qualquer condic o sobre o valor s rico individual do CA 125 para o seu c lculo como por exemplo a estabiliza o do CA 125 abaixo do seu valor normal permitindo assim sua determina o num rica sem quaisquer constrangimentos A ASC do CA 125 durante o tratamento prim rio por estar estreitamente relacionada com uma resposta completa quimioterapia adjuvante pode permitir classificar numericamente o resultado do tratamento especialmente til na compara o de diferentes tipos de tratamento nomeadamente em estudos prospectivos Na segunda fase do trabalho foi estudado o comportamento do CA 125 e qual a sua rela o com a recidiva Foram estudados a velocidade de aumento do CA 125 e a aplica o da ASC do CA 125 na detec
44. survival forecast the best CA 125 AUC cut off was 100 IU mL obtained for predicting an overall survival gt 3 years the patient final state alive and the complete response to chemotherapy with an accuracy of 7296 7496 and 8296 respectively In fact it seems clear that lower CA 125 AUC values are associated with a complete response while higher values are associated with a partial response or even a disease progression Therefore CA 125 AUC could be a useful measure of the primary treatment efficacy not only to evaluate the cytoreductive surgery but also the chemotherapy cocktail adopted Regarding the CA 125 AUC kinetic the objective of initial treatment cytoreductive surgery and primary chemotherapy of ovarian cancer is to produce the lowest CA 125 AUC possible In addition the CA 125 AUC kinetic parameter could be useful as an end point in the development of new chemotherapy drugs or to establish new guidelines for the primary treatment of ovarian cancer Finally CA 125 AUC presents some benefits over other kinetic parameters it is easier to calculate and model independent Nevertheless further studies should be carried out in order to 139 125 AUC as a new prognostic factor for patients with ovarian cancer compare CA 125 AUC with other prognostic factors used in the management of ovarian cancer patients and caution should be exercised before extrapolation of the present results to different data sets 140
45. tem como objectivo a m xima redu o tumoral de modo a que as les es residuais sejam t o pequenas quanto poss vel A teoria por detr s da citorreduc o tumoral est na curva Gompertziana do crescimento tumoral curva sigmoide segundo a qual quando os tumores s o pequenos a taxa de crescimento maior o mesmo acontecendo com a taxa de morte celular Em teoria a remoc o m xima do tumor aumenta as hip teses de resposta quimioterapia 25 Tumores volumosos com irriga o sangu nea 39 40 Introdu o insuficiente ou zonas tumorais com crescimento celular lento s o insens veis a f rmacos citot xicos Nos tumores pequenos bem perfundidos e com crescimento celular aumentado a difus o de agentes quimioterap uticos maior resultando num aumento da sua efic cia e com uma necessidade de um menor n mero de ciclos de quimioterapia e com redu o da resist ncia aos f rmacos citot xicos A remo o de tumores volumosos teoricamente tamb m melhora a resposta imune e com uma melhoria sintom tica das doentes 1 25 26 Uma cirurgia citorredutora implica a remo o de mais de 90 do tumor no entanto tal s poss vel em 40 das doentes com doen a peritoneal no est dio III 1 Doen a residual definida como sendo o maior di metro da maior massa tumoral que permanece ap s cirurgia de citorredu o O termo citorredu o ptima tem sofrido altera es ao longo das ltimas d cadas
46. the Rate of CA 125 Increase to Predict Ovarian Cancer Relapse Mann Whitney U 140 0 p lt 0 05 CA 125 Increase Rate IU day o Mean No Yes T 0 95 SE Relapse Figure 7 1 Mean CA 125 increase rate in patients with or without relapse Area under the ROC plot 0 870 5 0 045 Sensitivity 0 0 2 0 4 0 6 0 8 1 0 1 Specificity Figure 7 2 ROC curve CA 125 increase rate to predict patient relapse an Table 7 1 Sensitivity specificity positive predictive value PPV and accuracy for several CA 125 increase rate cut offs to predict ovarian cancer patient relapse CA 125 Increase Rate 2 Cut Off Cut Off IU mL day 0 01 0 02 0 03 0 04 0 05 0 06 0 07 0 08 0 09 0 10 0 20 0 30 0 40 0 50 Sensitivity 96 96 0 92 0 84 0 80 0 76 0 76 0 72 0 68 0 56 0 52 0 44 0 32 0 28 0 20 0 Specificity 96 32 6 60 5 76 7 81 4 83 7 83 7 83 7 86 0 88 4 90 7 95 3 100 0 100 0 100 0 PPV 96 45 3 57 5 67 7 714 731 731 72 0 73 9 73 7 76 5 84 6 100 0 100 0 100 0 Accuracy 96 55 9 72 1 79 4 80 9 80 9 80 9 794 79 4 76 5 76 5 765 75 0 735 70 6 The Potential for the Rate of CA 125 Increase to Predict Ovarian Cancer Relapse 7 4 DISCUSSION Many ovarian cancer patients even with successful first line treatment will eventually suffer relapse The CA 125 level in the majority of the patients will exhibit a rapid increase The aim of the present study was not to calculate the exact value of the rate of CA 125 increase
47. todos l u 2a Q u 4u a lt lt a a IR f rmula 3 33 lu lt a Y 10 s Figura 3 10 Gr fico da fun o tipo rampa com a 2 d Fun es sigmoides O disparo suave de um neur nio utilizando a fun o log stica foi introduzido por Cowan em 1967 As func es sigmoides cujos gr ficos t m a forma de S s o as func es de activac o mais utilizadas na constru o de redes neuronais artificiais S o fun es estritamente crescentes e apresentam uma rela o adequada entre o comportamento linear e n o linear T m tamb m a vantagem de poderem acomodar valores elevados sem satura o enquanto permitem a passagem de valores mais pequenos sem atenua o excessiva Exemplos de fun es sigmoides 1 Fun o log stica definida por Pu f rmula 3 34 l te Material e M todos em que o par metro da inclinac o da func o sigmoide Trata se de uma fun o continua no intervalo 0 1 e quando tende para infinito a func o torna se na func o de limiar Figura 3 11 Uma caracter stica importante desta func o reside na sua derivada ser dada pela express o u qu 1 ag u f rmula 3 35 02 12 10 8 6 4 2 0 2 4 8 10 12 y Figura 3 11 Gr fico da func o log stica com a 1 A equa o 3 25 uma propriedade conveniente desta fun o pois a sua derivada pode ser obtida a partir da pr pria fun o utilizando apena
48. 0 F 2 50 150 0 23 1 00 1 00 0 79 0 81 0 0 19 4 Negative 20 77 Positive 6 0 3 00 200 0 23 1 00 1 00 0 79 0 81 0 0 19 4 Negative 20 77 Table 8 3 Continuation Criterion B Test Relapse E FP FN Sensitivity Specificity PPV NPV Accuracy CA 125 AUC gt C IU mL Yes No Positive 25 25 C 15 IU mL 0 92 0 68 0 49 0 96 0 74 24 3 1 9 Negative 2 52 Positive 19 12 C 25 IU mL 0 73 0 84 0 61 0 90 0 82 11 7 6 8 Negative 7 65 Positive 13 4 C 50 IU mL 0 50 0 95 0 76 0 85 0 83 3 9 12 6 Negative 13 73 Positive 13 2 C 75 IU mL 0 50 0 97 0 87 0 85 0 85 1 9 12 6 Negative 13 75 Positive 13 1 C 100 IU mL 0 50 0 99 0 93 0 85 0 86 1 0 12 6 Negative 13 76 Positive 10 1 C 150 IU mL 0 38 0 99 0 91 0 83 0 83 1 0 15 5 Negative 16 76 Positive 8 1 C 200 IU mL 0 31 0 99 0 89 0 81 0 82 1 0 17 5 Negative 18 76 Positive 5 1 C 400 IU mL 0 19 0 99 0 83 0 78 0 79 1 0 20 4 Negative 21 76 Positive 5 1 C 600 IU mL 0 19 0 99 0 83 0 78 0 79 1 0 20 4 Negative 21 76 Positive 2 1 C 1000 IU mL 0 08 0 99 0 67 0 76 0 76 1 0 23 3 Negative 24 76 196 CA 125 AUC as a predictor for epithelial ovarian cancer relapse Table 8 4 Lead time to recurrence using the best accuracies for each criterion Criterion A Mean lead time to Standard Range Standard Error relapse Deviation Test days days tito relapse days days CA 125 AUC to f gt 1
49. 1 Cross SS Harrison RF Kennedy RL Introduction to neural networks Lancet 1995 346 1075 1079 Bourquin J Schmidli H Van Hoogevest P Leuenberger H Basic concepts of artificial neural networks ANN modeling in the application to pharmaceutical development Pharm Dev Technol 1997 2 2 95 109 Agatonovic Kustrin S Beresford R Basic concepts of artificial neural network ANN modeling and its application in pharmaceutical research J Pharmaceut Biomed Anal 2000 22 717 727 Drew PJ Monson JRT Artificial neural networks Surgery 2000 127 3 11 Lisboa PJG A review of evidence of health benefit from artificial neural networks in medical intervention Neural Networks 2002 15 11 39 Clayton RD Snowden S Weston MJ Mogensen O Eastaugh J LaneG Neural networks in the diagnosis of malignant ovarian tumours Br J Obstet Gynaecol 1999 106 1078 1082 Timmerman D Verrelst H Bourne TH De Moor B Collins WP Vergote 1 Vandewalle J Artificial neural network models for the preoperative discrimination between malignant and benign adnexal masses Ultrasound Obstet Gynecol 1999 13 17 25 Remzi M Djavan B Artificial neural networks in urology 2004 Eur Urol Suppl 2004 3 33 38 Fogel DB Wasson Ill EC Boughton EM Porto VW Evolving artificial neural networks for screening features from mammograms Artif Intell Med 1998 14 317 326 Orr RK Use of an artificial neural network to quantitate risk of malignancy for abnormal
50. 2010 116 3 p 389 94 FIGO Committee on Gynecologic Oncology L D NF Hacker J Gori HW Jones Ill HYS Ngan S Pecorelli Staging Classifications Clinical Practice Guidelines for Gynaecological Cancers H N M NF Hacker MD JL Benedet MD Editor 2000 Elsevier Vernooij F et al The outcomes of ovarian cancer treatment are better when provided by gynecologic oncologists and in specialized hospitals a systematic review Gynecol Oncol 2007 105 3 p 801 12 Shih and D S Chi Maximal cytoreductive effort in epithelial ovarian cancer surgery J Gynecol Oncol 2010 21 2 p 75 80 Trope C and J Kaern Primary surgery for ovarian cancer Eur J Surg Oncol 2006 32 8 p 844 52 Zivanovic O et al Advanced cytoreductive surgery American perspective Gynecol Oncol 2009 114 2 Suppl p S3 9 Tozzi R et al Laparoscopic treatment of early ovarian cancer surgical and survival outcomes Gynecol Oncol 2004 93 1 p 199 203 Verleye L et al EORTC GCG process quality indicators for ovarian cancer surgery Eur J Cancer 2009 45 4 p 517 26 Kehoe S Maximal cytoreductive surgery in advanced ovarian cancer Reviews in Gynaecological Practice 2005 5 4 p 5 Introdu o 31 32 33 34 35 36 37 38 39 40 41 42 43 44 Harries M and M Gore Part I chemotherapy for epithelial ovarian cancer treatment at first diagnosis Lancet Oncol 2002
51. 24 5 18 7 18 7 9 4 35 6 Good x 25 days 32 38 7 21 3 35 3 21 9 54 3 25 p 0 0003 Poor gt 25 days 31 21 0 16 9 17 2 9 0 27 7 Good lt 30 days 40 37 4 21 8 34 7 20 4 54 3 30 p 0 0001 Poor ty2 gt 30 days 23 17 1 11 8 16 9 8 6 21 5 Good s 35 days 42 36 5 21 7 33 5 19 4 54 0 35 p 0 0002 Poor t12 gt 35 days 21 16 9 12 0 16 9 8 6 19 0 Good s 40 days 44 35 9 21 4 31 6 19 4 51 4 40 p 0 0002 Poor gt 40 days 19 16 4 12 5 15 4 6 2 19 0 Between t 2 16 and ty2 19 days statistically significant difference was first found for tye 1 6 days 16 Good tyes 16 days Poor tip gt 16 days 14 49 43 5 24 3 26 1 18 6 34 8 22 3 66 9 19 3 11 4 35 8 All patients with t 2 lt 0 were classified as Poor p 0 0086 CA 125 half life breakpoint between a good and poor prognosis in patients with ovarian cancer 4 4 REFERENCES 1 Baker VV Treatment Options for Ovarian Cancer Clinl Obstetr Gynecol 2001 44 3 522 530 2 Bast Jr RC Xu FJ Yu YH Barnhill S Zhang Z Mills GB CA 125 The past and the future nt J Biol Markers 1998 13 4 179 187 3 Bidart JM Thuillier F Augereau C Chalas J Daver A Jacob N Labrousse F Voitot H Kinetic of Serum Marker Concentrations and Usefulness in Clinical Monitoring Clin Chem 1999 45 10 1695 1707 4 olakovi S Lukic V Mitrovi L Jeli S u nj
52. 25 AUC s 1000 IU mL 7296 CA 125 AUC s 100 IU mL and 6696 CA 125 AUC lt 25 IU mL for a 1 and 5 years overall survival respectively ROC AUC 20 67 0 75 and 0 73 In predicting a complete response to EN 132 CA 125 AUC as a new prognostic factor for patients with ovarian cancer chemotherapy ROC AUC 0 87 the best CA 125 AUC cut off was 100 IU mL with an accuracy of 8296 Table 5 3 We estimated the parameters in the Cox proportional hazard model using overall survival as a dependent variable CA 125 AUC as an independent variable and the patient final state Alive or Deceased as censoring variable FIGO tumor stage residual disease and primary chemotherapy response were stratification variables Table 5 4 Figure 5 4 shows the survival functions for different values of CA 125 AUC produced with the Cox model without any stratification variable 125 AUC as a new prognostic factor for patients with ovarian cancer Table 5 1 Summary of patient characteristics FIGO Stage Patient Count 96 19 20 7 I 11 11 996 HI 51 55 4 IV 9 9 8 Missing 2 2 2 Patient Final State Deceased 50 54 390 Alive 42 45 796 Tumor Grade 1 12 13 096 2 29 31 5 3 9 9 8 Missing 42 45 7 Histological Type Epithelial 82 89 1 Other 8 8 7 Missing 2 2 296 Residual Disease gt 2 cm Yes 36 39 1 No 53 57 696 Missing 3 3 396 Primary
53. 3 9 p 529 36 Ozols R F Optimum chemotherapy for ovarian cancer Int J Gynecol Cancer 2000 10 51 p 33 37 Pignata S et al Chemotherapy in epithelial ovarian cancer Cancer Lett 2011 303 2 p 73 83 Muggia F Platinum compounds 30 years after the introduction of cisplatin implications for the treatment of ovarian cancer Gynecol Oncol 2009 112 1 p 275 81 Panchagnula Pharmaceutical aspects of paclitaxel International Journal of Pharmaceutics 1998 172 1 2 p 1 15 Aravantinos G et Carboplatin and paclitaxel versus cisplatin paclitaxel and doxorubicin for first line chemotherapy of advanced ovarian cancer a Hellenic Cooperative Oncology Group HeCOG study Eur J Cancer 2008 44 15 p 2169 71 Chan J K et al The potential benefit of 6 vs 3 cycles of chemotherapy in subsets of women with early stage high risk epithelial ovarian cancer an exploratory analysis of a Gynecologic Oncology Group study Gynecol Oncol 2010 116 3 p 301 6 Hershman D et al Effectiveness of platinum based chemotherapy among elderly patients with advanced ovarian cancer Gynecol Oncol 2004 94 2 p 540 9 Fader AN et al Improved tolerance of primary chemotherapy with reduced dose carboplatin and paclitaxel in elderly ovarian cancer patients Gynecol Oncol 2008 109 1 p 33 8 Fung Kee Fung M et al ntraperitoneal chemotherapy for patients with advanced ovarian cancer a review of the evid
54. 81 37 to 843 56 3 245 6 1 25 AUC to Criterion B CA 125 AUC to fj2 25 IU mL 131 270 to 644 55 9 243 7 CA 125 AUC to 52 50 IU mL 111 103 to 530 51 1 184 1 CA 125 AUC to 75 IU mL 63 292 to 507 54 9 197 9 CA 125 AUC to t2 100 IU mL 11 377 to 264 44 8 161 6 8 4 DISCUSSION From a clinical point of view earlier relapse detection might have no impact on overall survival since it will be depend on the success of second line treatments Nevertheless it may have an immediate impact on patient follow up scheme number and frequency of the physician s visits and the request for additional diagnostic tools including CA 125 serum levels However it should be recognized that powerful relapse detection tools associated with other advances in therapeutic options could increase survival and or improve the patient s quality of life in the short term CA 125 AUC as a predictor for epithelial ovarian cancer relapse Although currently there is no evidence that earlier chemotherapy is superior to administering chemotherapy at clinical detection of recurrence it should be stressed that a prospective randomised trial on this issue is currently being carried out by the European Organization for Research and Treatment of Cancer EORTC 8 If incoming studies demonstrate an increased overall survival for asymptomatic patients receiving chemotherapy based on a CA 125 tumour marker information the sooner the relapse is detected t
55. 9 411 Partridge E E and M N Barnes Epithelial ovarian cancer prevention diagnosis and treatment CA Cancer J Clin 1999 49 5 p 297 320 Fleming J S et al ncessant ovulation inflammation and epithelial ovarian carcinogenesis revisiting old hypotheses Mol Cell Endocrinol 2006 247 1 2 p 4 21 Fathalla M F ncessant ovulation a factor in ovarian neoplasia Lancet 1971 2 7716 163 Urban N et al Ovarian cancer screening Hematol Oncol Clin North Am 2003 17 4 p 989 1005 ix Menon U et al Sensitivity and specificity of multimodal and ultrasound screening for ovarian cancer and stage distribution of detected cancers results of the prevalence screen of the UK Collaborative Trial of Ovarian Cancer Screening UKCTOCS Lancet Oncol 2009 10 4 p 327 40 Buys S S et al Ovarian cancer screening in the Prostate Lung Colorectal and Ovarian PLCO cancer screening trial findings from the initial screen of a randomized trial Am J Obstet Gynecol 2005 193 5 p 1630 9 Havrilesky L J et al Evaluation of biomarker panels for early stage ovarian cancer detection and monitoring for disease recurrence Gynecol Oncol 2008 110 3 p 374 82 Zhang Z et al Combining multiple serum tumor markers improves detection of stage I epithelial ovarian cancer Gynecol Oncol 2007 107 3 p 526 31 Schutter E M et al The differential diagnostic potential of a panel of tumor markers CA 125
56. CA 125 AUC as new prognostic factor for patients with ovarian cancer 5 5 REFERENCES 1 Vicki VB Treatment options for ovarian cancer Clin Obstet Gynecol 2001 44 3 522 30 2 Eduard EP Mack NB Epithelial ovarian cancer prevention diagnosis and treatment CA Cancer J Clin 1999 49 297 320 3 Bast RC Jr TL St John Jenison Niloff JM Lazarus H A radioimmunoassay using a monoclonal antibody to monitor the course of epithelial ovarian cancer N Engl J Med 1983 309 883 7 4 Bast RC Jr Xu FJ Yu YH Barnhill S Zhang 2 Mills GB CA 125 The past and the future Int J Biol Markers 1998 13 4 179 87 5 Amin PHM Gunnar BK Janne K Ole PB Vera MA Claes GT Prognostic value of pre and postoperative serum CA 125 levels in ovarian cancer new aspects and multivariate analysis Obstet Gynecol 1992 79 1002 10 6 Jean Michel B Francois T Christine A Jacqueline C Alain D Nelly J Fran oise Helene V Kinetic of serum marker concentrations and usefulness in clinical monitoring Clin Chem 1999 45 10 1695 1707 6 Angiolo G Paolo Z Fabio L Tiziano M Enrico S Tiziana B Renza C Serum half life of CA 125 during early chemotherapy as an independent prognostic variable for patients with advanced epithelial ovarian cancer results of a multicentric italian study Gynecol Oncol 1995 58 42 47 T Van Der Burg MEL Lammes FB Van Putten WLJ Stoter G Ovarian cancer the prognostic value of the serum
57. Esta compara o utilizada no processo de aprendizagem para influenciar o modo como os pesos sin pticos s o ajustados No segundo tipo a aprendizagem feita utilizando um conjunto de treino no qual s est o dispon veis os sinais de entrada Na aprendizagem supervisionada tamb m designada por aprendizagem com um professor ocorre ent o a detec o do erro entre a sa da calculada pela rede e a sa da desejada ocorrendo a incorpora o desse erro no ajuste dos pesos sin pticos Na aprendizagem n o supervisionada sem professor j n o ocorre a compara o entre a sa da calculada pela rede e a sa da desejada pois esta n o est dispon vel no conjunto de treino Normalmente a aprendizagem n o supervisionada apresenta uma menor complexidade computacional apesar de ser menos precisa que a supervisionada sendo til em situa es de tempo real onde n o h tempo suficiente para aplicar a aprendizagem supervisionada 13 e na an lise explorat ria permitindo a descoberta de padr es no conjunto de treino e relacionar esses padr es entre si e organiz los em classes 10 109 110 Material e M todos Considerando um conjunto de treino 7 com n elementos podemos descrever o algoritmo de aprendizagem de uma forma generalizada 1 passo iniciar os pesos sin pticos geralmente este passo envolve a randomizac o dos pesos num dada intervalo 2 passo apresentar camada de entrada da rede o elemento de trein
58. P rate false negative FN rate and accuracy for the prediction made by each ensemble The response graph for ensemble 2 was obtained to show the effect on the output variable prediction complete response to chemotherapy by the CA 125 AUC value as an independent continuous variable To test the importance of the adopted inputs a sensitivity analysis was carried out on each ensemble In practice the sensitivity analysis on the inputs indicates which variables are considered more important by that particular neural network ranking of variables Figure 6 1 Multilayer perceptron MLP network model for each member in the ensemble formation 3 inputs connected to 6 input neurons one hidden layer with 3 neurons and 1 output neurons connected to 1 output 125 AUC robustness as predictor for a complete response to primary chemotherapy in ovarian cancer a multivariant analysis using ANN 6 3 RESULTS The mean age at diagnostic time was found to be 55 5 16 1 82 4 S E 1 65 years with a mean overall survival of 39 2 3 5 100 1 S E 2 6 months According to the FIGO tumor stage nineteen 20 796 patients had stage eleven 11 996 had stage Il fifty one 55 4 had stage Ill nine 9 8 had stage IV and in two 2 2 patients this information was missing Eighty two 89 196 patients had epithelial ovarian cancer for histological type Twelve 13 0 patients had a tumor grade 1 twenty nine 31 596 a tumor gra
59. PREDICTOR FOR EPITHELIAL OVARIAN CANCER RELAPSE Cancer Biomarkers 2008 Abstract Purpose The aim of the present work was to evaluate the usefulness of CA 125 normalized in time area under the curve CA 125 AUC to signalise epithelial ovarian cancer relapse Patients and Methods Data from a hundred and eleven patients were submitted to two different approaches based on CA 125 AUC increase values to predict patient relapse In Criterion A total CA 125 AUC normalized in time value AUC was compared with the immediately previous one AUC using the formulae AUC 2 F AUC several values were tested to find the appropriate close related increment associated to patient relapse In Criterion B total CA 125 AUC normalised in time was calculated and several cut off values were correlated with patient relapse prediction capacity Results In Criterion A the best accuracy was achieved with a factor F of 1 25 increment of 2596 from the previous status while in Criterion B the best accuracies were achieved with cut offs of 25 50 75 and 100 IU mL The mean lead time to relapse achieved with Criterion A was 181 days while with Criterion B they were respectively 131 111 63 and 11 days Conclusion Based on our results we believe that conjugation and sequential application of both criteria in patient relapse detection should be highly advisable CA 125 AUC rapid burst in asymptomatic patients should be firstly evaluated using Cr
60. a Federation Internationale de Gyn cologie et d Obst trique FIGO de acordo com 37 38 Introdu o os procedimentos descritos em Staging Classifications and Clinical Practice Guidelines for Gynaecologic Cancers 23 Tabela 1 1 Tabela 1 1 Classifica o FIGO para os cancros do ov rio Est dio Caracter sticas Tumorais Tumor limitado aos ov rios la Tumor limitado a um ov rio c psula intacta aus ncia de tumor na superf cie ov rica Ib Tumor limitado a ambos os ov rios c psula intacta aus ncia de tumor na superf cie ov rica Ic Tumor limitado a um ou a ambos os ov rios com c psula rota e ou tumor na superf cie ov rica e ou c lulas neopl sicas na ascite ou no lavado peritoneal O tumor envolve um ou ambos os ov rios com extens o p lvica Extens o e ou implantes no tero e ou trompa Ilb Extens o a outras estruturas p lvicas Ilc ou Ilb com vegeta es externas nos ov rios uni ou bilateral ou c psula rota ou ascite com c lulas malignas ou lavado peritoneal positivo para c lulas tumorais Hl O tumor interessa um ou ambos os ov rios com met stases peritoneais confirmadas fora da p lvis e ou met stases ganglionares Illa Met stases peritoneais microsc picas para al m da p lvis IIIb Metastases peritoneais macrosc picas fora da p lvis x 2 na sua maior dimensao Met stases peritoneais fora p lvis gt 2 cm na maior dimens o e ou met stases nos g
61. a certain concentration C t was determined using the formula A A 4 A a EA t to CA 125 AUC formula 8 4 where Ajis the area of trapezoid j calculated by Co BE A gt t j 1 2 3 i formula 8 5 CA 125 AUC as a predictor for epithelial ovarian cancer relapse In order to find the best cut off and due to the fact that this was a retrospective study a forward stepwise approach was carried out for each patient dataset In fact for each patient the corresponding CA 125 AUC were calculated and compared with several cut off values 15 25 50 75 100 150 200 400 600 and 1000 IU mL If the CA 125 AUC values were equal or superior to a considered cut off value the test was considered positive Otherwise the test was considered negative Using the example with a patient having ten CA 125 levels obtained after the end of primary treatment the CA 125 AUC was calculated from Co to Cz all patients should have a minimum of three CA 125 levels then from to and so on until reaching CA 125 AUC from Co to Cio The test stopped when the CA 125 AUC to a certain concentration for instance exceeded the considered cut off value positive test or when the overall CA 125 AUC i e Co to in the present example did not exceeded the selected cut off value negative test For both criteria A and B sensitivity specificity the positive predictive value PPV negative predic
62. a rotura continuada do epit lio do ov rio e sua repara o por ocasi o da ovula o seria respons vel pela transforma o maligna b Hip tese de hipergonadotrofismo segundo a qual uma exposi o excessiva do epit lio do ov rio gonadotrofina leva a transforma o maligna c Hip tese hormonal a qual descreve as v rias influ ncias hormonais negativas ou protectoras a que o epit lio do ov rio est sujeito d Hip tese inflamat ria para a qual o processo ovulat rio se assemelha a uma reac o inflamat ria com infiltra o de leuc citos e produ o de mediadores inflamat rios como citocinas associadas remodela o tecidular Sem sintomas cl nicos espec ficos e sem um teste de rastreio o cancro do ov rio na maioria das vezes diagnosticado num estado avan ado originando uma mortalidade elevada Apesar de amplamente utilizado durante o seguimento da doente com cancro do ov rio o marcador CA 125 n o apresenta especificidade e sensibilidade suficiente para ser utilizado isoladamente como teste de rasteio pois apenas 50 das doentes no est dio apresentam um valor s rico de CA 125 elevado Esta eleva o tamb m pode ocorrer em situa es benignas ou em outras patologias 8 V rios estudos cl nicos randomizados est o em curso com destaque para o UK Collaborative Trial of Ovarian Cancer Screening UKCTOCS 9 e o Prostate Lung Colorectal and Ovarian PLCO Cancer Screening Trial 10 os quais co
63. age Surgical Cytoreduction Improves Further Survival of Patients with Late Recurrent Ovarian Cancer Gynecol Oncol 2000 79 344 349 Tinger A Waldron T Peluso N Katin MJ Dosoretz DE Blitzer PH Rubenstein JH Garton GR Nakfoor BA Patrice SJ Chuang L Orr JW Jr Effective palliative radiation therapy in advanced and recurrent ovarian carcinoma Int J Radiation Biol Phys 2001 51 5 1256 1263 Albuquerque KV Singla R Potkul RK Smith DM Creech S Lo S Emami B Impact of tumor volume directed involved field radiation therapy integrated in the management of recurrent ovarian cancer Gynecol Oncol 2005 96 701 704 Donovan KA Greene PG Shuster JL Jr Partridge EE Tucker DC Treatment Preferences in Recurrent Ovarian Cancer Gynecol Oncol 2002 86 200 211 Penson RT Digman Seiden MV Lee H Gallagher CJ Matulonis UA Olson K Gibbens Gore ME Attitudes to chemotherapy in patients with ovarian cancer Gynecol Oncol 2004 94 427 435 Bast RC Jr Klug TL St John E Jenison E Niloff JM Lazarus H et al A radioimmunoassay using a monoclonal antibody to 203 204 CA 125 AUC as a predictor for epithelial ovarian cancer relapse 29 30 31 32 33 34 35 36 37 monitor the course of epithelial ovarian cancer N Engl J Med 1983 309 15 883 887 Bast RC Jr Xu FJ Yu YH Barnhill S Zhang Z Mills GB CA 125 The past and the future Int J Biol Markers 1998 13 4 179 187 Tuxen MK S
64. almente mulheres peri e p s menop usicas 1 V rios factores influenciam o risco de aparecimento de cancro do ov rio 1 4 5 Como factores protectores temos a O uso de contraceptivos orais apresenta um forte efeito protector que se acentua com a maior dura o da sua administra o b Paridade sendo que mulheres m ltiparas apresentam um decr scimo de risco para o cancro do ov rio em compara o com as nul paras c Lacta o com um efeito de protec o menor Como factores que amplificam o risco temos a Hist ria familiar Trata se de um dos factores de risco mais importantes para o aparecimento do cancro do ov rio Aproximadamente 5 10 de todos os cancros do ov rio est o associados a uma predisposi o gen tica Mulheres com uma hist ria familiar de cancro da mama e ou do ov rio especialmente numa idade mais jovem pode indicar v rias muta es gen ticas nos genes supressores tumorais BRCA1 e ou BRCA2 com os produtos de ambos estando envolvidos na repara o do DNA b Idade c Ovula o d Nuliparidade e Obesidade Observa es epidemiol gicas servem de base para diferentes teorias que tentam explicar a etiopatologia do cancro epitelial do Introdu o ov rio Existem quatro hip teses major que em conjunto suportam a explica o da carcinog nese do cancro do ov rio 1 6 a Hip tese da ovula o incessante segundo a qual de acordo com Fathalla 7
65. also significantly more accurate than the TNM staging system when both use the TNM prognostic factors alone and new prognostic factors can be added to the ANN to further increase prognostic accuracy 22 The aim of this work is to confirm the potential of CA 125 AUC to predict a complete response to chemotherapy in our population For this purpose a multivariant analysis was performed with an ANN using several recognized clinical prognostic factors available in our population 6 2 PATIENTS AND METHODS For the purpose of comparison the same population used in the previous univariant study 8 entered this multivariant analysis with an ANN It included 92 patients with a diagnosis of ovarian cancer at the Gynaecology Service of Coimbra University Hospital HUC main database from 1990 to 2000 and CA 125 serum levels were obtained from the Pathology Service Hormonology amp Drug Monitoring Laboratory of HUC These patients underwent primary line chemotherapy within 4 months after submission to cytoreductive surgery and with a minimum of three CA 125 serum concentrations between the time of surgery and the end of chemotherapy 149 150 125 AUC robustness as a predictor for a complete response to primary chemotherapy in ovarian cancer a multivariant analysis using ANN To carry out with the multivariant analysis MLPs were used in which the output of the neural models was a complete response to chemotherapy yes no In
66. alysis based on the Cox proportional hazard model to test the variation of CA 125 AUC in relation to the overall survival time was also carried out A p level lt 0 05 was considered significant 5 3 RESULTS The mean age at diagnostic time was found to be 55 5 16 1 82 4 S E 1 65 years with a mean overall survival of 39 2 3 5 100 1 S E 2 6 months According to the FIGO tumor stage nineteen 20 7 patients had stage eleven 11 9 had stage Il fifty one 55 4 had stage Ill nine 9 8 had stage IV and in two 2 2 patients this information was missing Eighty two 89 196 patients had epithelial ovarian cancer for histological type Twelve 13 0 patients had a tumor grade 1 twenty nine 31 596 a tumor grade 2 nine 9 896 had a tumor grade 3 and forty two 45 796 patients had no tumor grade information Thirty six 39 196 patients had a residual tumor greater than 2 cm after surgery The mean duration of primary chemotherapy was 4 1 0 7 10 2 S E 0 15 months fifty two 56 5 patients had a complete response to primary chemotherapy twenty two 23 996 had a partial response thirteen 14 196 had no 125 AUC as new prognostic factor for patients with ovarian cancer response or a disease progression and five 5 496 patients had missing information for the response to chemotherapy At evaluation date fifty 54 3 patients were deceased while forty two 45 7 were alive Eighty patients 86 996 had at
67. ancro do ov rio mas necess ria investiga o adicional para demonstrar vantagens na sobreviv ncia tendo em conta o impacto na utiliza o de recursos e qualidade de vida 49 Apesar de elevadas taxas de resposta quimioterapia adjuvante a maioria das mulheres sofre recidiva sendo uma grande propor o delas sujeita a nova quimioterapia No entanto a sensibilidade platina com a medi o do intervalo livre de platina n o uma regra 47 48 Introdu o biol gica que segue um padr o temporal exacto Assim podemos considerar v rios grupos 33 50 a Doentes sens veis platina doentes nas quais a recidiva ocorreu num per odo superior a 12 meses ap s a conclus o do tratamento inicial Neste caso a terap utica consiste na administra o em associa o de derivados de platina b Doentes parcialmente sens veis platina quando a reca da ocorre num per odo entre 6 a 12 meses ap s o tratamento inicial Neste caso o tratamento padr o ainda n o definido c Doentes resistentes platina quando a reca da se d num per odo inferior a 6 meses ap s a conclus o do tratamento inicial Designa se por doen a refract ria como aquela onde ocorre progress o da doen a ainda durante o tratamento inicial Nas doentes parcialmente sens veis ou resistentes platina podem ser usados outros agentes citost ticos normalmente em monoterapia Entre outros podemos incluir neste grupo o paclitax
68. ar S Marinkovi J Prognostic Value of CA 125 kinetics and half life in advanced ovarian cancer Int J Biol Markers 2000 15 2 147 152 deb 5 125 AS NEW PROGNOSTIC FACTOR FOR PATIENTS WITH OVARIAN CANCER Gynecologic Oncology 2005 Abstract Objective The aim of the present study was to investigate the usefulness of the CA 125 area under the curve AUC as a new kinetic parameter for predicting overall survival in patients with ovarian cancer In addition the relationship of CA 125 AUC with other prognostic factors of ovarian cancer was evaluated Methods Ninety two patients that underwent primary line chemotherapy within 4 months after submission to cytoreductive surgery were included For each patient CA 125 AUC was calculated and a statistical analysis was conducted to compare CA 125 AUC behaviour among patients according to several covariates Results The mean age at diagnostic time was found to be 55 5 16 1 82 4 years with a mean survival of 39 2 3 5 100 1 S E 2 6 months Across FIGO stage Il Ill and IV patients had a mean CA 125 AUC of 18 2 24 6 147 8 and 574 6 IU mL respectively p lt 0 05 At the evaluation date living patients had a mean CA 125 AUC of 40 1 in contrast to 234 1 IU mL p 0 05 for deceased ones Patients with a complete response to primary chemotherapy had a mean CA 125 AUC of 48 8 while patients with a partial response had a mean of 251 7 IU mL and patients wit
69. ariant analysis using ANN Due to the restrictions regarding the number of patients the complexity of the ANN models is an important issue To keep the models as simple as possible only MLPs with a single hidden layer were chosen The number of hidden units H was also restricted by using the pyramid guideline 23 where for a single hidden layer H is given by 4 Lx where L is the number of input neurons and the number of output neurons After analyzing the various input combinations and the total number of input neurons models with only three hidden neurons were used Figure 6 1 A hundred epochs of back propagation followed by five hundred epochs of conjugate gradient descent as a second order optimization algorithm were used to train the different MLP models To speed up training at the beginning during the back propagation phase the learning rate and the momentum were adjusted in each epoch The learning rate varied from 0 05 to 0 01 and the momentum varies from 0 6 to 0 3 A sum squared error function was used where the error is the sum of the squared differences between the target and actual output values in each output unit For the activation function for the hidden units H and the output unit O the hyperbolic tangent function X was chosen because of its symmetry This function e e tanh results in a sigmoid curve like the logistic function except that output lies in the ran
70. arker doubling time DT are often used as kinetic parameters for the evaluation of clinical response and follow up of patients with ovarian cancer 6 Serum CA 125 half life during early chemotherapy is an independent prognostic factor for both the achievement of a pathologically complete response and the survival of patients with advanced epithelial ovarian cancer 7 and several studies report that the greatest difference in progression rate was found at a t 2 of 20 days 8 11 Nevertheless although CA 125 level before the 39 course of chemotherapy was considered the best prognostic indicator by Fayers et al it was classified inaccurately for clinical use 12 In addition to CA 125 kinetic parameters several other prognostic factors can be used in the management of ovarian cancer the FIGO F d ration Internationale de Gyn cologie et d Obstr trique tumor stage tumor grade tumor biology over expression of the HER 2 neu oncogene residual disease after initial cytoreductive surgery and rate of response to chemotherapy 13 127 128 125 AUC as a new prognostic factor for patients with ovarian cancer CA 125 tumor marker kinetics are more important than the isolated value of CA 125 serum concentration for patient prognosis and in the present work we propose a new kinetic parameter the CA 125 area under the curve AUC and its relation with the FIGO stage patient final state tumor grade residual disease and primary
71. artificiais requer cuidados adicionais Alguns erros da aplica o de redes neuronais em oncologia s o referidos no artigo de revis o de Schwarzer e colaboradores 1 Um dos erros mais comuns s o as estimativas ineficientes com muitos trabalhos a apresentar um tamanho do conjunto de teste muito pequeno No entanto muitos dos trabalhos publicados poderiam ter sido melhorados atrav s da divis o da amostra com a utiliza o de t cnicas de valida o cruzada Segundo Lisboa 2 uma das quest es metodol gicas na elabora o de aplica es cl nicas usando redes neuronais a robustez da avalia o do desempenho S o diversas as fontes de incerteza capazes de reduzir a capacidade de manter o desempenho de um grupo de doentes para outro Podemos incluir variabilidade intra e Conclus es Finais inter individual variabilidade entre os diferentes grupos de doentes e diferen as entre centros cl nicos Assim na criac o de uma ferramenta inform tica baseada em redes neuronais com utiliza o na pr tica cl nica devemos ter em conta a Uma valida o interna na qual um conjunto de treino com a dimens o adequada usado para modelar o problema em causa Novos dados podem servir como conjunto de valida o do treino e para medir a performance da aplica o b Uma valida o temporal com a entrada de dados recentes do mesmo centro cl nico c Uma valida o externa na qual dados obtidos noutros centros cl nicos e
72. atients with an early stage l lla the surgery includes a total abdominal hysterectomy bilateral salpingo oophorectomy infra colic omentectomy multiple biopsys of the abdominal cavity and peritoneal lavage cytology In rare cases pelvic lymphadenectomy was performed For advanced stages Ilb IV cytoreduction surgery was performed with the attempt of resection of lesions smaller than 2 cm During the follow up period beginning at the end of primary treatment data was appropriately recorded in the patient clinical file database Relapse was first determined by physical examination in most cases by imaging and in few cases by biopsy It should be emphasised that no patients were treated for relapse based on CA 125 information alone Due to the fact that CA 125 analysis only became standard practice at the beginning of the 90s not all patients presented CA 125 levels immediately after the end of primary treatment Therefore for data analysis purpose and to ensure the same starting conditions for all of them the first available CA 125 level Co after the end of primary treatment had to be under 35 IU mL baseline Four patients that only had CA 125 levels after relapse detection and one patient with only one CA 125 level between primary treatment conclusion and relapse were withdrawn from the study Thus a hundred and eleven patients were included in the present study where two different methodological approaches were tested in order to ge
73. because for that purpose a more thorough regression analysis should be performed using rich data However this study suggests that monitoring CA 125 can be used to rapidly predict patient relapse with a high level of certainty corroborated by the high area under the ROC plot 0 87 and using only the first CA 125 increase as well as a simple definition of that increase The rate of increase was chosen with the assumption that at the beginning the increase of CA 125 is linear but if larger differences between and C are observed then the CA 125 increase is exponential and a kinetic parameter such as doubling time DT is a more suitable parameter to characterize the CA 125 increase Out of all the patients in this study only one had a relapse without an increase in CA 125 This might be explained by the existence of ovarian tumours that do not secrete CA 125 Using the third CA 125 concentration and with an increase rate superior or equal to 0 06 IU mL day the mean lead time to relapse was 204 2 days However in four patients the third level was taken after clinical detection of relapse giving a negative lead time to relapse This inability to detect CA 125 increase may be due to an unsuitable interval between CA 125 analyses and a rapid tumour burst This explanation is given further credence by the wide range 21 to 312 days between CA 125 samples Thus a long time interval between samples is unsuitable for the use of CA 125 to mon
74. bserva es teriam se n o estivessem empatadas Calcula se para cada amostra a soma Sj dos respectivos n meros de ordens Calcula se o teste estat stico 3 N 1 f rmula 3 16 jan Or 1 Se existem n meros de ordem empatados o valor de deve ser corrigida atrav s do factor de correcc o Ya Esq f rmula 3 17 onde Uu o n mero de empates em cada amostra e m o n mero de grupos de n meros de ordem empatados e o valor corrigido H formula 3 18 Quando lt 5 ou quando existem poucas observa es em cada amostra os valores cr ticos da distribui o do teste Material e M todos estat stico s o apresentados numa tabela No entanto para amostras com um elevados n mero de observac es ou para k gt 5 que este teste H aproxima se de uma distribui o 7 qui quadrado com k 1 graus de liberdade sendo H comparado com os valores tabelados de 7 3 2 6 An lise de Sobreviv ncia Curva de Kaplan Meier A an lise de sobreviv ncia um conjunto de t cnicas estat sticas utilizadas para analisar o tempo at a ocorr ncia de um evento de interesse ao qual chamamos de um modo gen rico de tempo de sobreviv ncia Este evento pode ser em medicina entre outros a ocorr ncia de determinada patologia uma resposta a um tratamento uma reca da ou bito Para al m da rea biom dica a an lise de sobreviv ncia pode ser ap
75. cancer 12 for instance carboplatin alone 13 or together with paclitaxel 14 15 in sensitive patients topotecan 16 18 doxorubicin and gemcitabine among others for resistant patients Past available data do not supply a conclusive role for surgery in the treatment of relapsed ovarian cancer 20 More recent studies suggest that secondary cytoreduction may improve patient survival 21 22 especially in patients that recur at least six months after first line treatment 23 183 184 CA 125 AUC as a predictor for epithelial ovarian cancer relapse In practice radiation therapy is rarely used due to the amount of radiation that abdomen organs can safely receive Nevertheless radiation therapy for advanced ovarian cancer was reported to be more positive than second and third line chemotherapy in terms of response survival and tolerance 24 and therapy with involved field radiation is effective in controlling localized recurrences particularly after optimal debulking 25 Although in many cases the treatment objective is to prolong survival the majority of women with ovarian cancer prefer continuing to treat their cancer aggressively regardless of poor outcomes and the quality of life becomes secondary 26 Approximately half of patients without any symptoms of relapse and the corresponding physicians want early chemotherapy for a rising CA 125 even with the increase of toxicity 27 There are several surveilla
76. cancer relapse Collect CA 125 sample Number of CA 125 concentrations Calculate CA 125 AUC to the last AUC and previous AUC CA 125 level 125 AUC gt with C225 or 50 or 75 or 100 IU ml CA 125 AUC gt 1 25 AUC Patient Relapse Figure 8 2 Algorithm for implementation of both criteria A and B in clinical practice 200 CA 125 AUC as a predictor for epithelial ovarian cancer relapse 8 5 REFERENCES 10 Lewis S Menon U Screening for ovarian cancer Rev Gynaecol Prac 4 156 161 2004 Einhorn N Bast Knapp R Nilsoon B Zurawski V Sj vall K Long term follow up of the Stockholm screening study on ovarian cancer Gynecol Oncol 2000 79 466 470 Bell R Petticrew M Sheldon T The performance of screening test for ovarian cancer results of a systematic review BJOG 1998 105 1136 1147 Kecmanovic DM Pavlov MJ Kovacevic PA Ceranic MS Stamenkovic AB Cytoreductive surgery for ovarian cancer Eur J Surg Oncol 2003 29 315 320 Harries M Gore M Part I Chemotherapy for epithelial ovarian cancer treatment at first diagnosis Lancet Oncol 2002 3 529 536 Thigpen JT Chemotherapy for advanced ovarian cancer overview of randomized trials Semin Oncol 2000 27 3 suppl 7 11 16 Trimbos JB Vergote Bolis Vermorken JB Mangioni C Madronal C Franchi M Tateo S Zanetta G Scarfone G Giurgea L Timmers P Coens C Pecorelli S Impact of A
77. ce of Coimbra University Hospital HUC main database from 1990 to 2000 and CA 125 serum levels were obtained from the Pathology Service Hormonology amp Drug Monitoring Laboratory of HUC Eighty patients with a diagnosis of ovarian cancer were included in the study The inclusion criteria were a Patients who received adjuvant chemotherapy following cytoreductive surgery with or without consolidation chemotherapy as part of the primary treatment b A reasonable follow up time with a minimum of five CA 125 serum concentrations between the end of treatment and the evaluation date C The first CA 125 level after the end of treatment must be below or equal to 35 IU mL For each patient the first CA 125 increase was determined by recording the first three consecutive levels as C C and Ciz with C being the CA 125 level at time i For each patient the CA 125 increase rate was calculated using the first and third concentrations with the formula CA 125 Increase Rate Ci2 Ci tu2 t formula 7 1 The Mann Whitney U test was used to compare the rate of CA 125 increase across the subgroups of patients defined by a positive or negative relapse The area under the Receiver Operating The Potential for the Rate of CA 125 Increase to Predict Ovarian Cancer Relapse Characteristic ROC plot was established as a measure to quantify the test performance 9 for the discrimination by CA 125 increase rate in predicting pat
78. cers of the female gynaecological system in the western world and is generally treated with combination first line chemotherapy after cytoreduction surgery 1 2 Cancer Antigen 125 CA 125 has been showed to be the most reliable serum marker for ovarian carcinoma Due to its poor prognosis ovarian cancer has been the object of a great deal of effort in developing efficient treatments The measure of treatment efficiency in ovarian cancer represents a challenge mainly because many patients have a nonevaluable disease using classical criteria such as RECIST Response Evaluation Criteria in Solid Tumors Patients without macroscopic disease after surgery or patients with widespread diffuse peritoneal disease could not be enrolled in clinical trials because it is difficult to quantify the lesions with radiological imaging such as CT scans CA 125 has the potential to be used to calculate the response rate to therapy making these women eligible for the clinical trials 9 4 Rustin et al 5 produce definitions based on serial CA 125 levels to quantify the response rate to primary chemotherapy in ovarian cancer but instead of using criteria that compared isolated values of the marker the adoption of kinetic parameters for evaluating the clinical response has been proposed as a suitable alternative CA 125 kinetic parameters may be used to evaluate the clinical response to treatment and follow up making CA 125 an important prognostic factor for
79. dade gastrointestinal prolongada 56 No entanto o papel da radioterapia no cancro avan ado deve ser investigado em ensaios cl nicos prospectivos e randomizados 57 Estudos mais recentes defendem a utiliza o da radioterapia abdominal total ap s cirurgia e quimioterapia baseada em platina por permitir um maior controlo da doen a em regi es localizadas com um risco aceit vel de toxicidade aguda 58 Onde a radioterapia se pode destacar no controlo paliativo dos sintomas da doente sintom tica com diminui o da dor e hemorragia 59 1 1 3 4 Outras Terapias O papel do tratamento hormonal indefinido com poucos ensaios cl nicos randomizados Pode no entanto ter um perfil de toxicidade 49 50 Introdu o melhor que a quimioterapia e pode ser usado na doen a terminal para fins paliativos A imunoterapia uma rea em desenvolvimento pois as c lulas cancer genas do ov rio expressam v rias prote nas antig nicas que poder o servir como potenciais alvos terap uticos Terapia gen tica inibidores da transdu o de sinal e inibidores da angiog nese s o tamb m investiga es em curso 4 60 61 1 1 4 Factores de Progn stico O cancro do ov rio caracterizado pelo seu comportamento biol gico diverso desde tumores com um excelente progn stico at tumores com um progn stico muito reservado 1 V rios elementos cl nicos biol gicos patol gicos e cir rgicos podem ser utilizados co
80. de 2 nine 9 896 had a tumor grade 3 and forty two 45 796 patients had no tumor grade information Thirty six 39 196 patients had a residual tumor greater than 2 cm after surgery The mean duration of primary chemotherapy was 4 1 0 7 10 2 S E 0 15 months fifty two 56 5 patients had a complete response to primary chemotherapy twenty two 23 996 had a partial response thirteen 14 196 had no response or a disease progression and five 5 496 patients had missing information for the response to chemotherapy At evaluation date fifty 54 3 patients were deceased while forty two 45 796 were alive Table 6 2 Table 6 3 shows the input combination used in each ensemble and the number and type of members that constitute the ensemble as well as the number of patients used in training validation and test subset for the leave one out cross validation Due to missing information not all the patients were taken into account by the neural models The performance of each ensemble is recorded in table 6 4 Both input sets achieved the same performance with an accuracy of 0 86 153 125 AUC robustness as a predictor for a complete response to primary chemotherapy in ovarian cancer a multivariant analysis using ANN According to the sensitivity analysis the variable with most weight for the prediction was the CA 125 AUC CA 125 lt 100 IU mL and CA 125 AUC value for ensemble 1 and 2 respectively In Figure 6 2 t
81. djuvant Chemotherapy and Surgical Staging in Early Stage Ovarian Carcinoma European Organisation for Research and Treatment of Cancer Adjuvant ChemoTherapy in Ovarian Neoplasm Trial J Natl Cancer Inst 2003 95 2 113 124 Spriggs D Optimal sequencing in the treatment of recurrent ovarian cancer Gynecol Oncol 2003 90 S39 S44 Ozols RF Treatment goals in ovarian cancer Int J Gynecol Cancer 2005 15 suppl 1 3 11 Markman M Bookman MA Second Line Treatment of Ovarian Cancer Oncologist 2000 5 26 35 201 202 CA 125 AUC as a predictor for epithelial ovarian cancer relapse Baker VV Treatment Options for Ovarian Cancer Clin Obstet Gynecol 2001 44 3 522 530 Harries M Gore M Part Il Chemotherapy for epithelial ovarian cancer treatment of recurrent disease Lancet Oncol 2002 3 537 545 Dizon DS Dupont J Anderson S Sabbatini P Hummer A Aghajanian C Spriggs D Treatment of recurrent ovarian cancer a retrospective analysis of women treated with single agent carboplatin originally treated with carboplatin and paclitaxel The Memorial Sloan Kettering Cancer Center experience Gynecol Oncol 2003 91 584 590 Watanabe Y Nakai H Ueda H Hoshiai H Evaluation of weekly low dose paclitaxel and carboplatin treatment for patients with platinum sensitive relapsed ovarian cancer Gynecol Oncol 2005 96 323 329 Havrilesky LJ Alvarez AA Sayer RA Lancaster JM Soper JT Berchuck A Clarke Pearson DL Rodriguez GC Carney ME
82. e values Nevertheless the existence of CA 125 half life breakpoint to allow the discrimination between good or poor prognosis confirms the suitability of this kinetic parameter for an earlier prediction of the patient s overall survival However the CA 125 half life kinetic parameter should not be used alone but in combination with other recognized prognostic factors md tip gt 16 days or t lt 0 r 0 lt t p lt 16 days o oo 2 a a Cumulative Proportion Surviving o e CA o gt o o 10 20 30 40 50 60 70 80 90 100 Overall Survival months Figure 4 1 Survival curve Kaplan Meier in agreement with CA 125 half life breakpoint of 16 days 119 120 CA 125 half life breakpoint between a good and poor prognosis in patients with ovarian cancer Table 4 1 Overall survival for the different prognostic group concerning the breakpoint Median Overall CA 125 Mean Overall Survival Classification N Survival SD p value Qos Qzs days months months Good lt 10 days 4 47 7 33 7 39 7 23 8 71 5 10 p 0 2149 Poor ty2 gt 10 days 59 28 8 19 8 22 6 13 4 44 1 Good tz s 15 days 10 39 5 24 4 31 6 22 3 48 9 15 p 0 1058 Poor gt 15 days 53 28 2 20 2 21 5 11 8 44 1 Good t2 s 20 days 24 38 8 22 1 34 7 21 9 51 4 20 p 0 0069 Poor gt 20 days 39
83. e 1 83 NS CR PR WR DP Cumulative Proportion Surviving CA 125 AUC 200 UI ml CA 125 AUC 400 UI ml CA 125 AUC 600 UI ml 125 AUC 800 UI ml CA 125 AUC 1000 UI ml Survival Time months Figure 5 4 Survival function for hypothetical CA 125 AUC values 137 138 CA 125 AUC as a new prognostic factor for patients with ovarian cancer 5 4 DISCUSSION Many authors have studied the CA 125 kinetic in monitoring the ovarian cancer patient Kinetic parameters prove to be more useful than rough serum concentration alone Van der Burg et al 7 Hawkins et al 8 Verda J Hunter et al 9 S Colakovi et al 10 and others used the CA 125 half life value to evaluate patient survival finding a of 20 days to be a breakpoint between a good and poor prognosis Buller et al showed that the rate of decline of CA 125 in effectively treated ovarian cancer is described by an exponential model and his study suggests that it is possible to predict overall survival which patients have residual disease at reassessment laparotomy who will be free of disease and who will have a recurrence 14 15 In a review article Jean Michel Bidart et a 5 accentuated the value of different serum marker kinetic parameters in the monitoring of patients in several types of cancer To increase the value of CA 125 kinetics as a prognostic factor in ovarian cancer we proposed and studied a new
84. e future nt J Biol Markers 1998 13 4 179 87 4 Van Der Burg MEL Lammes FB Van Putten WLJ Stoter G Ovarian Cancer The Prognostic Value of the Serum Half Life of CA 125 during Induction Chemotherapy Gynecol Oncol 1988 30 307 312 Hawkins RE Roberts Wiltshaw E J Fryatt IJ McCready VR The prognostic significance of the half life of serum CA 125 in patients responding to chemotherapy for epithelial ovarian carcinoma BJOG 1989 96 1395 1399 6 olakovi S Lukic V Mitrovi L S u njar S Marinkovi J Prognostic Value of CA 125 kinetics and half life in advanced ovarian cancer nt J Biol Markers 2000 15 2 147 152 7 Rustin GJS Nelstrop McClean P Brady McGuire WP Hoskins WJ Mitchell H Lambert HE Defining Response of Ovarian Carcinoma to Initial Chemotherapy According to Serum CA 125 J Clin Oncol 1996 14 5 1545 1551 8 Rustin GJS Nelstrop AE Tuxen MK Lambert HE Defining progression of ovarian carcinoma during follow up according to CA 125 A North Thames Ovary Group Study Ann Oncol 1996 7 361 364 The Potential for the Rate of CA 125 Increase to Predict Ovarian Cancer Relapse 10 Zweig MH Campbell G Receiver operating characteristic ROC plots A fundamental evaluation tool in clinical medicine Clin Chem 1993 30 4 561 577 Ozols RF Treatment goals in ovarian cancer nt J Gynecol Cancer 2005 15 suppl 1 3 11 177 8 CA 125 AUC AS A
85. e increase from 25 to 100 IU mL the false positive rate decreased from 11 796 to 1 096 while the false negative rate changed from 6 896 to a stable 12 696 which are encouraging results for clinical practice Anyway as a general rule the interval between samples must be balanced not only for technical but also for economic and psychological reasons many patients reported CA 125 anxiety 12 In accordance with the results obtained with our population the major advantage of Criterion A is that it can predict patient relapse far sooner than Criterion B and even sooner than other known criteria CA 125 AUC as a predictor for epithelial ovarian cancer relapse using the CA 125 levels However the Criterion A is especially useful if patient experience a rapid increase in CA 125 AUC Otherwise Criterion B should be implemented because under some circumstances i e when patients experienced a long term sustained increase of CA 125 tumour marker relapses can occur even when differences between consecutive CA 125 AUC are less than 2596 For this reason we strongly recommend the use of both criteria in conjugation Firstly Criterion A should be applied to evaluate the situation If a negative answer occurs Criterion B should be attempted to confirm the absence of relapse The algorithm of implementation of both criteria in clinical practice is appropriately illustrated in Figure 8 2 199 CA 125 AUC as a predictor for epithelial ovarian
86. e treino no qual est o inclu dos a informa o de entrada e de sa da sendo o treino realizado atrav s de aprendizagem supervisionada Estas redes produzem uma func o interna atrav s dos valores dos pesos sin pticos capaz de realizar tarefas de previs o e de classifica o A capacidade de previs o da rede est no entanto limitada interpola o dentro do espa o de entrada sa da sendo extrapolac o fora deste normalmente errada O modelo mais representativo deste tipo de redes s o os perceptr es de m ltiplas camadas 107 108 Material e M todos Nas redes de extrac o de caracter sticas a aprendizagem normalmente feita utilizando m todos n o supervisionados ou aprendizagem competitiva S o normalmente aplicadas em redu o da dimensionalidade muitas vezes utilizadas em conjunto com as redes associativas originando redes h bridas desempenhando um papel na filtragem de dados Dentro deste grupo podemos incluir as redes hebbianas e as redes de Kohonen As redes n o adaptativas s o normalmente usadas no reconhecimento de padr es como por exemplo imagens Estas redes aprendem padr es tipos das vari veis de entrada reconstruindo os quando apresentada rede um padr o incompleto danificado ou com ru do Est o inclu das neste grupo as redes de Hopfield 3 3 8 Processos de Aprendizagem A constru o de um sistema baseado em redes neuronais artificiais ir geralmente env
87. el caso n o tenha sido usado anteriormente doxorrubicina lipos mica topetecam gemcitabina decetaxel eposido oral trabectedina ifosfamida altretamina hexametilmelamina pemetrexed 33 50 52 A combina o de bevacizumab platina e gemcitabina tamb m foi estudada mas com altos perfis de toxicidade 53 A import ncia do intervalo livre de paclitaxel deve tamb m ser avaliada em estudos futuros 54 Em geral e quando poss vel estas doentes s o inclu das em ensaios cl nicos V rios mecanismos podem estar envolvidos na resist ncia cisplatina ou carboplatina Entre eles podemos incluir excesso de Introdu o um factor de resist ncia satura o muta o ou altera o dos factores requeridos na morte da c lula diminui o do fluxo sangu neo s c lulas tumorais condi es extracelulares redu o da absor o da platina nas c lulas desintoxica o intracelular ex via glutationa diminui o da liga o ao ADN ex o aumento do pH intracelular repara o do ADN diminui o dos erros de repara o de ADN apoptose imperfeita aumento de factores antiapoptose efeitos sobre v rias vias de sinaliza o celular presen a de c lulas quiescentes 55 1 1 3 3 Radioterapia A radioterapia n o est indicada no tratamento adjuvante do cancro do ov rio Ap s vigil ncia de longo prazo doentes com doen a inicial e tratadas com radioterapia abdominal total adjuvante mostrou toxici
88. ence and standards for the delivery of care Gynecol Oncol 2007 105 3 p 747 56 Lesnock J L et al Completion of intraperitoneal chemotherapy in advanced ovarian cancer and catheter related complications Gynecol Oncol 2010 116 3 p 345 50 Landrum L M et al Intraperitoneal chemotherapy for patients with advanced epithelial ovarian cancer a review of complications and completion rates Gynecol Oncol 2008 108 2 p 342 7 Gray H J et al Alternative intraperitoneal chemotherapy regimens for optimally debulked ovarian cancer Gynecol Oncol 2010 116 3 p 340 4 Naumann R W P Sukumvanich and R P Edwards Practice patterns of intraperitoneal chemotherapy in women with ovarian cancer Gynecol Oncol 2009 114 1 p 37 41 59 60 Introdu o 45 46 47 48 49 50 51 52 5 54 55 56 57 58 Berry E et al An outpatient intraperitoneal chemotherapy regimen for advanced ovarian cancer Gynecol Oncol 2009 113 1 p 63 7 Ryu K S et al Effects of intraperitoneal hyperthermic chemotherapy in ovarian cancer Gynecol Oncol 2004 94 2 p 325 32 Bae J H et al Treatment of ovarian cancer with paclitaxel or carboplatin based intraperitoneal hyperthermic chemotherapy during secondary surgery Gynecol Oncol 2007 106 1 p 193 200 Deraco M et al Advanced cytoreduction as surgical standard of care and hyperthermic intraperitoneal chemoth
89. ento Interliga es entre os neur nios Sinapses Movit artificiais Aprendizagem Treino da rede Mem ria Valores dos Pesos sin pticos Fun o de activa o usada no modelo Princ pio do tudo ou nada de McOulloch e Pitts Tabela 3 3 Comparac o entre termos utilizados em estat stica e redes neuronais artificiais Estat stica Redes Neuronais Artificiais Modelo Rede Aprendizagem Regress o Aprendizagem supervisionada Interpolac o Generalizac o Observac es Conjunto de treino Par metros Pesos sin pticos Vari veis independentes Vari veis de entrada input Vari veis dependentes Vari veis de sa da output 104 Material e M todos 3 3 6 Topologia das Redes A topologia de uma rede refere se organizac o e interligac o dos diferentes neur nios artificiais 13 Podemos representar graficamente uma rede neuronal como um conjunto de n s que representam os neur nios artificiais e linhas que representam as liga es sinapses entres eles Figura 3 14 ND o O S E 0A 4 aan Figura 3 14 Representac o gr fica de uma rede neuronal artificial com uma camada de entrada com quatro neur nios artificiais duas camadas ocultas uma com seis e outra com cinco neur nios e uma camada de sa da com apenas um neur nio A organizac o de uma rede frequentemente es
90. erapy as promising treatment in epithelial ovarian cancer Eur J Surg Oncol 2011 37 1 p 4 9 Foster T et al A review of the current evidence for maintenance therapy in ovarian cancer Gynecol Oncol 2009 115 2 p 290 301 Harries M and M Gore Part Il chemotherapy for epithelial ovarian cancer treatment of recurrent disease Lancet Oncol 2002 3 9 p 537 45 Colombo N and M Gore Treatment of recurrent ovarian cancer relapsing 6 12 months post platinum based chemotherapy Crit Rev Oncol Hematol 2007 64 2 p 129 38 Vergote l et al A randomised double blind phase Il study of two doses of pemetrexed in the treatment of platinum resistant epithelial ovarian or primary peritoneal cancer Eur J Cancer 2009 45 8 p 1415 23 Richardson D L et al Combination gemcitabine platinum and bevacizumab for the treatment of recurrent ovarian cancer Gynecol Oncol 2008 111 3 p 461 6 Linch M et al Experience in a UK cancer centre of weekly paclitaxel in the treatment of relapsed ovarian and primary peritoneal cancer Gynecol Oncol 2008 109 1 p 27 32 Stewart D J Mechanisms of resistance to cisplatin and carboplatin Critical Reviews in Oncology Hematology 2007 63 1 p 12 31 Engelen M J et al Long term morbidity of adjuvant whole abdominal radiotherapy WART or chemotherapy for early stage ovarian cancer Eur J Cancer 2009 45 7 p 1193 200 Einhorn N et al s there place for
91. ervados No entanto na pr tica ocorrem empates nos tempos de sobreviv ncia e esta fun o foi modificada para lidar com empates 8 89 90 Material e M todos 3 3 REDES NEURONAIS ARTIFICIAIS 3 3 1 Introdu o Nos ltimos anos a utiliza o de redes neuronais artificiais tem vindo a captar cada vez mais o interesse de diferentes reas cient ficas como a medicina f sica geologia economia engenharia biologia entre outras onde est o a ser aplicadas com um enorme sucesso numa vasta gama de problemas Problemas que envolvam a previs o a classifica o e o controlo podem ser alvo de aplica o de redes neuronais artificiais As raz es deste sucesso podem estar subjacentes a dois aspectos o poder das redes neuronais artificiais como t cnicas sofisticadas de modela o capazes de modelar fun es complexas nomeadamente fun es n o lineares e a facilidade de uso das redes neuronais Este ltimo aspecto prende se com o facto da aprendizagem executada pelas redes neuronais artificiais ser atrav s de exemplos Assim a partir de um conjunto de dados representativos do problema h a aplica o de algoritmos de treino capazes de aprenderem a estrutura do conjunto de dados Um outro aspecto reside no n vel de conhecimento que o utilizador necessita para aplicar as redes num dado problema mais baixo do que o n vel de conhecimento necess rio para por exemplo usar m todos estat sticos tradicionais
92. es da matriz 79 Muitos destes marcadores s o expressos em outros tipos de cancro ou outras patologias tornando dif cil a sua aplica o espec fica e isolada Com o evento da gen mica e prote mica e a possibilidade de estudar qualitativamente e quantitativamente prote nas expressas em c lulas cancer genas de um modo diferencial quando comparada com tecidos saud veis poder permitir a descoberta de novos marcadores tumorais 74 76 80 1 2 CA 125 O antig nio CA 125 Cancer Antigen 125 uma mucina e foi identificada pela primeira vez em 1981 por Bast e colaboradores utilizando o anticorpo monoclonal OC 125 81 Outros anticorpos foram descritos como sendo capazes de reconhecer o CA 125 e ap s v rios estudos imuno histoqu micos podemos dividir estes anticorpos em 2 grupos OC 125 ike ou M11 ike Considera se um terceiro grupo separado relacionado com o anticorpo OV197 A descoberta que a liga o com o M11 n o era inibida pelo OC 125 sugere que estes anticorpos reconhecem epitopos distintos em Introdu o dom nios separados do antig nio CA 125 A interac o antig nio anticorpo possibilitou o desenvolvimento de radioimunoensaios permitindo a quantifica o do CA 125 82 De acordo com O Brien e colaboradores 83 a mol cula do 125 caracterizada por 3 dom nios principais Figura 1 1 a Dominio extracelular amino terminal b Um grande dominio constituido por unidades repetidas c U
93. espectivas proveni ncias b O conjunto de observa es assim constitu do pela jun o das duas amostras ordenado por ordem crescente atribuindo o n mero de ordem 1 observa o menor e o n mero de ordem n n gt observa o maior i 1 2 j 1 ng C No caso de empates ties a cada uma das observa es empatadas atribu do o n mero de ordem m dio que essas observa es teriam se n o estivessem empatadas 79 80 Material e M todos d Calculam se somas dos n meros de ordem das observac es para cada amostra n e 5 S s f rmulas 3 6 e 3 7 j l j l observamos que N N 1 S S f rmula 3 8 e Calculam se as quantidades 1 U NC S f rmula 3 9 n n 1 _ 2 U lt S f rmula 3 10 observamos que U U f rmula 3 11 f Aestat stica do teste U min U U f rmula 3 12 g Atrav s de tabelas dos quantis da distribui o U de Mann Whitney fazemos a estat stica de rejeitar ou n o hip tese nula para determinado valor de p Se e forem iguais ou superiores a 10 observa es demonstra se que os valores de U seguem uma distribuic o normal de m dia Material e M todos _ A ly f rmula 3 13 e desvio padr o aye m D f rmula 3 14 sendo neste caso a estat stica do teste dada por ra f rmula 3 15 3 2 5 Teste de Kruska
94. este de Gehan s teste de Cox Mantel teste de log rank teste de Peto ou o F teste de Cox 8 Material e M todos 3 2 7 An lise de Sobreviv ncia Modelo de Risco Proporcional de Cox Os m todos param tricos de an lise de sobreviv ncia s o importantes na identifica o de factores de progn stico significativos apenas em situac es onde a distribuic o subjacente sobreviv ncia conhecida No entanto na pr tica a forma exacta da distribui o n o conhecida tornando limitada a aplica o dos m todos param tricos O modelo de riscos proporcionais de Cox n o requer o conhecimento da forma da distribui o de sobreviv ncia podendo a fun o de risco neste modelo assumir qualquer forma assumindo se que a fun o de risco para diferentes indiv duos proporcional e independente do tempo O efeito de um ou mais factores de progn stico covari veis sobre o tempo de sobreviv ncia pode ser analisado pelo modelo de riscos proporcionais de Cox cujos resultados s o expressos em estimativas de r cios de risco hazard ratio 8 Considerando x Xy Xz Xp como sendo o conjunto de factores de progn stico podemos escrever a fun o de risco h t x em fun o da fun o de risco subjacente fun o de risco basal ou de base e uma outra fun o seja g x apenas com o conjunto de factores de progn stico ou seja h t1x x x MOE Xx f rmula 3 23 A fun o de risco subjacente
95. f greater calculation accuracy without assumption of individual CA 125 levels This resulted in less data variability It allowed us to calculate an average time to relapse that was higher than that suggested by Rustin et al 111 vs 63 days with a similar degree of accuracy 8596 vs 8890 This may have important implications for clinical practice Finally it is important to say that kinetic parameters which come from CA 125 such as AUC may be used to evaluate novel treatments and ultimately revolutionise patient management ADALINE ADN AFP Al ANNs ANOVA ART ASC AUC beta HCG BP CA CA 125 CR CUH DNA DT EDVAC ENIAC EORTC FIGO FN FP 4 2 LISTA DE ABREVIATURAS Adaptive linear element cido desoxirribonucleico Alpha fetoprote na Artificial intelligence Artificial neural networks Analysis of variance Adaptative resonance theory rea sob a curva Area under the curve Subunidade beta gonadotrofina cori nica humana Back propagation Cancer antigen Cancer antigen 125 Complete response Computed tomography Coimbra University Hospitals Deoxyribonucleic acid Doubling time Electronic discrete variable automatic computer Electronic numerical integrator and computer European Organisation for Research and Treatment of Cancer F d ration Internationale de Gyn cologie et d Obst trique False negative False positive Gynecologic Oncology Group Human epididymis protei
96. fica confinado ao abd men fazendo com que a administra o dos agentes citot xicos por via intraperitoneal seja em teoria atractiva pois permite a chegada de uma maior concentra o do agente s c lulas tumorais 31 40 Introdu o As propriedades farmacocin ticas do paclitaxel e cisplatina variam quando s o administradas por via intraperitoneal com uma concentra o e tempo de semi vida superiores na cavidade peritoneal 40 Esta via de administra o tem evolu do ao longo dos anos com v rios estudos a indicarem melhorias at 30 no intervalo livre de doen a e sobreviv ncia global No entanto existem cr ticas a estes ensaios e esta op o n o tem sido universalmente aceite por v rios motivos 2 40 44 a Maiores efeitos t xicos com pior aceita o das doentes o Problemas t cnicos com o uso e manutenc o do cateter o Infec es Outras complica es como ades es intraperitoneais D Regime ptimo e n meros de ciclos ainda n o estabelecidos A via intraperitoneal para ser mais pratic vel pode requerer medidas de suporte como hidrata o programada e introdu o de factor estimulante das col nias de granul citos 45 Est ainda em investiga o o uso de quimioterapia intraperitoneal hipert rmica com objectivo de atrasar o desenvolvimento da doen a e prolongar a sobreviv ncia 46 48 Quimioterapia de manuten o pode melhorar os resultados cl nicos no c
97. ge 1 1 In addition due to the limited number of patients available a leave one out cross validation was carried out to prevent overfitting a major problem with neural networks in the generalization question accompanied by the difficulty in quantifying with the same performance in the presence of new data In leave one out cross validation a number of experiments equal to the size of the data set are performed In each experiment a single case is placed in the test subset and the rest of the data is used for training In this study nearly 2 3 of the population was used in the train subset and nearly 1 3 in the validation subset At the end of this cross BD 152 125 AUC robustness as predictor for a complete response to primary chemotherapy in ovarian cancer a multivariant analysis using ANN validation process ensemble was formed with the individual network created in each experiment of leave one out cross validation Ensembles are collections of neural networks that cooperate in performing a prediction In this classification problem the ensemble estimates a value for the output by combining the outputs from the individual networks that constitute the ensemble in which the network s predictions are combined in a winner takes all vote The performance for each input combination was measured by calculating the sensitivity specificity positive predictive value PPV negative predictive value NPV false positive F
98. gression a model for epithelial ovarian cancer response Am J Obstet Gynecol 1991 165 360 7 Richard EB Michael LB Jeffery DB Alberto M Philip JDS Serum CA 125 regression in epithelial ovarian cancer correlation with reassessment findings and survival Gynecol Oncol 1992 47 87 92 6 CA 125 AUC ROBUSTNESS AS A PREDICTOR FOR A COMPLETE RESPONSE TO PRIMARY CHEMOTHERAPY IN OVARIAN CANCER A MULTIVARIANT ANALYSIS USING ANN Recent Advances amp Research Updates 2006 Abstract Purpose The aim of the present work is to confirm the potential of CA 125 AUC to predict a complete response to chemotherapy performing a multivariant analysis using Artificial Neural Networks ANNs involving the prognostic factors available in our population Patients and Methods 92 patients with a diagnosis of ovarian cancer were included Multilayer perceptrons MLPs with a single hidden layer and three hidden neurons were used with a complete response to chemotherapy yes no working as output variable Two sets of inputs were tested In the first the FIGO stage and residual disease were combined with CA 125 AUC cut off value lt 100 IU mL categorical variable while in the second they were combined with CA 125 AUC value continuous variable Back propagation followed by conjugate gradient descent as a second order optimization algorithm was used to train the MLPs A sum squared error function and hyperbolic tangent function for activation func
99. h no response or disease progression had a mean of 316 5 IU mL p 0 05 The best CA 125 AUC performance is in predicting patient complete response to chemotherapy with a cut off of 100 IU mL and an accuracy of 82 Conclusions Despite CA 125 AUC high correlation with the FIGO stage residual disease and patient final outcome the main interest of CA 125 AUC calculation is to evaluate the treatment efficacy and to foresee a full chemotherapy response Further studies should be carried out before extrapolating these results to others data sets Key Words CA 125 kinetics ovarian cancer prognostic factors 125 AUC as new prognostic factor for patients with ovarian cancer 5 1 INTRODUCTION Ovarian cancer generally treated with combination first line chemotherapy after cytoreduction surgery 1 2 has the highest mortality rate of all invasive cancers of the gynaecological system Bast et al 1983 first described a radioimmunoassay that could detect CA 125 Cancer Antigen 125 in the serum of ovarian cancer patients 3 CA 125 serum concentration is usually adopted to evaluate the clinical situation in ovarian cancer patients and the rate of decline in CA 125 during primary chemotherapy has been an important prognostic factor in several multivariate analyses 4 The postoperative serum CA 125 level is an independent prognostic factor in patients with invasive ovarian cancer 5 and CA 125 tumor marker half life t12 and tumor m
100. half life of CA 125 during induction chemotherapy Gynecol Oncol 1988 30 307 312 8 Hawkins RE Roberts Wiltshaw E Mundy J Fryatt lJ McCready VR The prognostic significance of the half life of db 125 AUC as a new prognostic factor for patients with ovarian cancer 10 11 12 13 14 142 serum CA 125 in patients responding to chemotherapy for epithelial ovarian carcinoma Br J Obstet Gynaecol 1989 96 1395 9 Verda JH Lee D Michael H John TS Andrew B Daniel LCP Bast RC Jr The prognostic significance of CA 125 half life in patients with ovarian cancer who have received chemotherapy after surgical cytoreduction Am J Obstet Gynecol 1990 163 1164 7 olakovi S Lukic V Mitrovi L Jelic S u njar S Marinkovi J Prognostic value of CA 125 kinetics and half life in advanced ovarian cancer Int J Biol Markers 2000 15 2 147 52 Fayers PM Rusting G Wood R Nelstrop A Leonard RCF Wilkinson P Cruickshank D Mcallister EJ Redman CWE Parker D Scott IV Slevin ML Roulston JE The prognostic value of serum CA 125 in patients with advanced ovarian carcinoma an analysis of 573 patients by the medical research council working party on gynaecological cancer Int J Gynecol Cancer 1993 3 285 292 Christine HH Jonathan SB Ovarian cancer epidemiology biology and prognostic factors Semin Surg Oncol 2000 19 3 10 Richard EB Michael LB Jeffrey DB Alberto M Philip JDS CA 125 re
101. he greater the chances of being able to catch the tumour at an earlier stage of evolution Working with this scenario earlier relapse detection will play a fundamental role in ovarian cancer patient follow up In addition earlier relapse detection could also be relevant in the design of clinical trials for new anti cancer agents If any test wants to be classified as suitable for predicting ovarian cancer relapse the minimal requirement is to accurately signal the relapse in asymptomatic patients As it can be seen from the results obtained in our population both approaches A and allow it with interesting accuracy values above 80 If Criterion A is adopted an increase of 25 between two consecutive CA 125 AUCs suggests with a substantial mean lead time to the event 181 days that a relapse is ongoing accuracy of 8596 Otherwise if Criterion B is adopted cumulative CA 125 AUCs values between 25 IU mL and 100 IU mL indicates the presence of relapse accuracy above 80 Obviously increasing the cut off from 25 IU mL to 100 IU mL imply a small rise of accuracy from 8296 to 8696 but an important decrease regarding the mean lead time to relapse from 131 days with 25 IU mL to 11 days with 100 IU mL Technically speaking Criterion A in comparison with Criterion B presents the advantage of not considering the upper normal limit or even nadir values while in both approaches in comparison with other 197 198 CA 125
102. he response graph illustrates the effect on the output variable prediction complete response to chemotherapy performed by ensemble 2 using the CA 125 AUC value as an independent continuous variable Table 6 2 Summary of patient characteristics FIGO Stage Patient Count 96 19 20 7 I 11 11 9 Il 51 55 4 96 IV 9 9 8 Missing 2 2 2 Patient Final State Deceased 50 54 3 Alive 42 45 7 Tumor Grade 1 12 13 0 2 29 31 5 3 9 9 8 Missing 42 45 7 Histological Type Epithelial 82 89 1 Other 8 8 7 Missing 2 2 2 Residual Disease gt 2 cm Yes 36 39 1 No 53 57 6 Missing 3 3 3 Primary Chemotherapy Response Complete Response CR 52 56 5 Partial Response PR 22 23 9 Without Response or 3 Disease Progression WR DP 13 14 196 Missing 5 5 4 154 125 AUC robustness as predictor for a complete response to primary chemotherapy in ovarian cancer a multivariant analysis using ANN Table 6 3 Input combinations used for each ensemble Leave one out cross Number Type of validation Ensemble Inputs of members Combination members MLP wisis aan N I L H P O Subset Test 2 3 N an Subset CA 125 AUC s 100 IU mL MLP 1 FIGO Stage 83 3 6 3 1 1 54 28 1 Residual disease 2 cm ve CA 125 AUC value MLP 2 FIGO Stage 83 3 6 3 1 1 54 28 1 Residual disease gt 2 cm MLP I L H P O multilayer perceptron with
103. helial ovarian cancer a multicenter retrospective study Gynecol Oncol 2001 81 3 p 398 403 Polverino G et al Survival and prognostic factors of women with advanced ovarian cancer and complete response after a carboplatin paclitaxel chemotherapy Gynecol Oncol 2005 99 2 p 343 7 Maas H A et al The influence of age and co morbidity on treatment and prognosis of ovarian cancer a population based study Gynecol Oncol 2005 97 1 p 104 9 Shinichi Tate Y H Nobuhiro Takeshima Katuhiko Hasumi CA125 regression during neoadjuvant chemotherapy as an independent prognostic factor for survival in patients with advanced ovarian serous adenocarcinoma Gynecologic Oncology 2005 95 p 143 149 Bast Jr et al CA 125 the past and the future Int J Biol Markers 1998 13 4 p 179 87 Colakovic S L V Mitrovic L Jelic S Susnjar S Marinkovic J Prognostic value of CA 125 kinetics and half life in advanced ovarian cancer Int J Biol Markers 2000 15 2 p 147 152 Hoskins P J N Le and R Correa CA 125 normalization with chemotherapy is independently predictive of survival in advanced endometrial cancer Gynecol Oncol 2011 120 1 p 52 5 Gadducci A et al Serum half life of CA 125 during early chemotherapy as an independent prognostic variable for patients with advanced epithelial ovarian cancer results of a multicentric Italian study Gynecol Oncol 1995 58 1 p 42 7 61 62 Introdu
104. ia de cancro heredit rio do ov rio e se a neoplasia se encontrar num est dio la 1 Na recidiva o papel da cirurgia de citorredu o controverso pela aus ncia de estudos prospectivos No entanto v rios estudos retrospectivos mostram uma melhoria na sobreviv ncia global com o uso da cirurgia de citorredu o na recidiva da doen a 25 30 Assim a cirurgia de citorredu o deve ser considerada em caso de recidiva se for poss vel uma citorredu o ptima 41 42 Introdu o A cirurgia de second look realizada por laparotomia est indicada em doentes num contexto de ensaios cl nicos visto que n o contribui para o aumento da sobreviv ncia e tem como objectivo definir o estado da doen a tumoral Nem todos os estudos suportam citorredu o na cirurgia de second look 30 Cirurgia de citorredu o de intervalo cirurgia de intervalo realizada ap s resposta quimioterapia neoadjuvante ou em doentes a meio de quimioterapia adjuvante controversa no que toca melhoria da sobreviv ncia global 26 30 1 1 3 2 Quimioterapia Como a grande maioria dos cancros do ov rio s o epiteliais estes s o relativamente mais sens veis aos f rmacos citot xicos que noutros tumores s lidos A maioria das mulheres requer quimioterapia adjuvante ap s cirurgia de citorredu o e de estadiamento Historicamente durante as ltimas d cadas v rios ensaios cl nicos levaram adop o de u
105. iciente Aumento fase 2 3 4 5 12 Fase3 oe 100 gt cos 200 2 pa Y Concentracoes C Fase4 3 medi o gt 2 fase 2 3 4 C Fase5 C Fase6 Doentes com Ultimo CA125 gt DAVAL e FALECIDA C Fase AUC cut off fase 8 9 11 Numero de Doentes com valores C Fase8 C Fase9 Pesquisa 125 cut off fase 10 C Fase 10 Guarda Folha de Excel Fase 11 C Fase 12 Ver Tabela C Fase13 Inclui Valores Datas Concentra es e ou AUC 3 Ficheiro c Wazio ls Carregado V 2 3 24 11 2005 Sair Figura 3 1 Ecr principal da aplicag o Filtra Dados No cruzamento da informac o foi essencial o uso nos Hospitais Universidade de Coimbra HUC de um n mero de identificac o nico para cada doente que recorra a esta unidade de sa de Na tabela dos valores s ricos do CA 125 os valores foram ordenados numa primeira fase pelo n mero dos HUC seguida por uma ordenac o ascendente data Com esta aplicac o criada especificamente para o efeito foi poss vel executar todos os c lculos relacionados com os par metros cin ticos testados e exportar informac o das doentes e resultados obtidos para tabelas de Excel que iriam ser analisadas em programas de estat stica e ou de redes neuronais artificiais Figura 3 2 Material e M todos Base Dados do CA 125 Base Dados das Doentes Exporta o Exporta o Dados do CA 125 Dados das Doentes em Excel em Excel gS Base de dados em
106. ient relapse Sensitivity specificity positive predictive values PPV and overall accuracy were also determined for several CA 125 increase rate cut off values We also calculated the mean time from f to relapse date A p level lt 0 05 was considered significant 7 3 RESULTS The mean standard error age at diagnostic time was found to be 52 1 1 79 years with a mean overall survival of 74 6 4 81 months Twenty eight 35 096 patients had FIGO stage while ten 12 5 had stage Il thirty seven 46 3 had stage Ill three 3 7 had stage IV and in two 2 596 patients this information was missing Twelve 15 096 patients had a tumor grade 1 nineteen 23 8 a tumor grade 2 six 7 5 had a tumor grade 3 and forty three 53 7 patients had no tumor grade information Seventy one 88 896 patients had epithelial ovarian cancer for histological type and three 3 796 patients had a borderline tumour Twenty four 30 0 patients had a residual tumor greater than 2cm after surgery Only six 7 496 patients received consolidation chemotherapy after adjuvant chemotherapy as part of the primary treatment Using the CA 125 increase criteria in seventy patients it was possible to calculate the rate of increase and in ten patients no CA 125 increase was detected The mean standard error duration of primary treatment was 182 3 9 4 days The mean time interval between CA 125 samples was 117 5 7 74 days ranging from 21 to 312 da
107. inical response to the first line treatment of patients 36 while doubling time DT can be used as a predictive factor for ovarian cancer recurrence 37 Moreover the intrinsic value of the recently so called normalized in time area under the curve CA 125 AUC kinetic parameter as a prognostic factor and as a primary treatment efficacy outcome has already been described 38 The objective of the present work was to evaluate the potential usefulness of CA 125 AUC kinetic parameter to signal ovarian cancer relapse before and anticipating any clinical detection 8 2 PATIENTS AND METHODS Retrospective clinical information was gathered on patients diagnosed in the late 80s and 90s with epithelial ovarian cancer from the Gynaecology Service of Coimbra University Hospital HUC main database CA 125 serum levels of these patients were obtained from the Pathology Service HUC Hormonology amp Drug Monitoring Laboratory database Patients with a minimum of three CA 125 serum concentrations between the end of primary treatment and the evaluation date were included in the study The end of treatment was determined as the date of curative surgery if the patient did not receive any chemotherapy or the date of conclusion of primary 185 186 CA 125 AUC as a predictor for epithelial ovarian cancer relapse adjuvant chemotherapy or the end of consolidation chemotherapy if the patient received it after the primary chemotherapy In p
108. iterion A with a high accuracy 0 85 and with a substantial mean lead time to relapse 181 days If a negative answer was obtained then Criterion B should performed to confirm the absence of relapse Key Words Ovarian cancer CA 125 CA 125 kinetics follow up relapse recurrence CA 125 AUC as a predictor for epithelial ovarian cancer relapse 8 1 INTRODUCTION Ovarian cancer has the highest mortality among all invasive cancers of the female gynaecological system in the western world In addition without a routine generalized screening test the majority of patients with ovarian cancer present an advanced stage of disease at the time of diagnosis 1 3 Despite the developments in first line cytoreductive surgery 4 and chemotherapy 5 6 most women with the disease at an advanced stage suffer recurrence making ovarian cancer a disease with a poor prognosis Approximately 1096 5096 of patients who receive surgery for treatment of early stage ovarian cancer also have a recurrence 7 Since curative therapy for a recurrence of the disease may not be possible attempts to prolong progression free PFS and overall survival relieve symptoms and extend quality of life become the alternative goals 8 9 The treatment options for patients with recurrent ovarian cancer include additional chemotherapy and in some particular cases surgery and radiation therapy 10 11 Several drugs and regimens can be used in relapsed ovarian
109. itor ovarian cancer after treatment On the other hand small sample time 173 174 The Potential for the Rate of CA 125 Increase to Predict Ovarian Cancer Relapse intervals may be too expensive and because many patients report anxiety before the next CA 125 level physicians report this as the CA 125 psychosis Although a standard interval has not been established the optimal time between sample determination should take these two aspects into consideration Other methods for detecting an increase in the level of CA 125 are possible such as the Rustin et al criteria doubling time or other kinetic parameters Although CA 125 increase detection has an immediate impact in patient follow up in particular on the frequency of CA 125 determination and the request for other examination methods it might not have the necessary impact on patient overall survival thus this will depend more on the success of the second line treatment Although currently there is no evidence to suggest that earlier treatment is superior to administering chemotherapy at clinical detection of recurrence this situation is under study in a prospective randomized trial 10 However the use of capable tools to detect relapse earlier and constant treatment evolution could increase survival or improve the patient s quality of life in the future The suggested rate of CA 125 increase is easy to calculate and model independent However caution should be exercised
110. kh Familial ovarian cancer and early ovarian cancer biologic pathologic and clinical features Int J Gynecol Pathol 2001 20 1 p 48 63 Markman M The Hole of CA 125 in the Management of Ovarian Cancer Oncologist 1997 2 1 p 6 9 Gupta D and C G Lis Hole of CA125 in predicting ovarian cancer survival a review of the epidemiological literature J Ovarian Res 2009 2 p 13 63 2 OBJECTIVO Objectivo O objectivo prim rio do trabalho proposto para a presente disserta o de doutoramento consistia no desenvolvimento de uma ferramenta inform tica capaz de melhorar a interpreta o do CA 125 nomeadamente no que diz respeito sua capacidade de prever a recidiva em doentes com cancro do ov rio No entanto para se conseguir desenvolver uma ferramenta inform tica houve a necessidade de combinar a informa o demogr fica e cl nica das doentes incluindo obviamente as concentra es do CA 125 A an lise preliminar da informa o dispon vel relativamente nossa popula o cedo permitiu concluir que havia dificuldades na utiliza o das concentra es s ricas de CA 125 enquanto tal pois o n mero de concentra es variava de doente para doente o mesmo acontecendo com o intervalo de tempo entre concentra es consecutivas informa o retrospectiva n o sujeita a protocolo espec fico A forma encontrada para contornar esta dificuldade baseou se num trabalho de revis o Bidart et al
111. kinetic parameter CA 125 AUC As can be seen in Figure 1A B the CA 125 AUC calculation is independent of the shape presented by the CA 125 serum concentrations making CA 125 AUC a more suitable kinetic parameter than CA 125 half life Figure 1A In addition the CA 125 AUC is less disturbed by peak phenomena especially after surgery and sources of variability i e intra subject and assay variability In the present work the poor correlation between CA 125 AUC and tumour grade is perhaps the consequence of the high number of patients without this information Inversely CA 125 AUC is highly 125 AUC as new prognostic factor for patients with ovarian cancer correlated with the FIGO stage in which lower values are related with stage and Il middle values with stage Ill and higher values with stage IV Patients with stage IV have a mean CA 125 AUC 31 6 times greater than patients with stage CA 125 AUC is also correlated with residual disease for patients with a residual tumor 2 cm after initial cytoreductive surgery mean CA 125 AUC 2 1 times greater CA 125 AUC is also correlated with patient final state in deceased patients having a mean CA 125 AUC 5 8 times greater than living As shown by the Cox proportional hazard model the CA 125 AUC is an independent prognostic factor for patient overall survival and patients with a lower CA 125 AUC have a better overall survival than patients with a higher CA 125 AUC Concerning
112. l tormost G Dombernowsky P Tumor markers in the management of patients with ovarian cancer Cancer Treat Rev 1995 21 215 245 Rustin GJ Bast RC Jr Kelloff GJ Barrett JC Carter SK Nisen PD Sigman CC Parkinson DR Ruddon RW Use of CA 125 in clinical trial evaluation of new therapeutic drugs for ovarian cancer Clin Cancer Res 2004 10 3919 3926 Guppy AE Rustin GJ CA125 Response Can it Replace the Traditional Response Criteria in Ovarian Cancer Oncologist 2002 7 437 443 Rustin GJ Nelstrop AE Tuxen MK Lambert HE Defining progression of ovarian carcinoma during follow up according to CA 125 A North Thames Ovary Group Study Ann Oncol 1996 7 361 364 Tuxen Sol tormos G Rustin GJ Nelstrop Dombernowsky P Biological variation and analytical imprecision of CA 125 in patients with ovarian cancer Scand J Clin Lab Invest 2000 60 713 722 Tuxen MK Sol tormos G Dombernowsky P Serum tumor marker CA 125 for monitoring ovarian cancer during follow up Scand J Clin Lab Invest 2002 62 177 188 Bidart JM Thuillier F Augereau C Chalas J Daver A Jacob N Labrousse F Voitot H Kinetic of Serum Marker Concentrations and Usefulness in Clinical Monitoring Clin Chem 1999 45 10 1695 1707 Riedinger JM Coudert B Barillot 1 Buffenoir G Mayer F Fargeot P Cuienier J Guerrin J Int r t clinique de l estimation CA 125 AUC as a predictor for epithelial ovarian cancer relapse 38 de la vitesse de crois
113. l Wallis O teste de Kruskal Wallis ou an lise de vari ncia pelos n meros de ordem ranks pode ser considerado como a alternativa n o param trica one way ANOVA da tamb m ser designado por ANOVA em ordens de Kruskal Wallis Este teste uma generalizac o do teste de Mann Whitney que permite comparar um conjunto de amostras independentes Pode ent o ser usado para testar se duas ou mais amostras provenientes de uma mesma popula o ou de popula es diferentes ou prov m de popula es com a mesma distribui o 3 6 7 Sejam k as amostras em an lise o teste de hip teses As distribui es das k amostras s o id nticas Ha Existe pelo menos uma amostra onde a distribui o diferente das distribui es das outras amostras em estudo De notar que quando k 2 este id ntico ao teste de Mann Whitney Wilcoxon 81 82 Material e M todos A aplica o deste teste envolve as seguintes etapas Tal como no teste anterior as n n2 n observa es dos k grupos s o combinadas numa serie com dimens o N e distribu das numa ordem de magnitude da mais pequena para a maior mas mantendo a origem observac o De seguida as observa es s o substitu das pelo numero de ordem ranks de 1 atribu do observa o mais pequena at N atribu do observa o maior No caso de empates a cada uma das observa es empatadas atribu do o n mero de ordem m dio que essas o
114. least one year survival forty four 47 896 had at least three years while only twenty one 22 896 had more than five years survival Table 5 1 The CA 125 AUC across groups of patients is shown in Table 5 2 Patients with FIGO stage Il Ill and IV had a mean CA 125 AUC of 18 2 S E 2 4 IU mL 24 6 S E 7 6 IU mL 147 8 S E 30 8 IU mL and 574 6 S E 134 6 IU mL respectively p 0 05 Figure 5 2 Patients with a tumor grade 1 2 and 3 had a mean CA 125 AUC of 100 1 45 5 IU mL 158 1 43 0 IU mL and 238 8 114 0 IU mL respectively p gt 0 05 Patients with residual disease gt 2cm had a mean CA 125 AUC of 207 4 5 48 3 IU mL while patients without residual disease had 97 9 S E 30 0 IU mL p 0 05 At the evaluation date living patients had a mean CA 125 AUC of 40 1 S E 210 5 IU mL in contrast to deceased patients who had a CA 125 AUC of 234 1 S E 44 4 IU mL p lt 0 05 Fifty two patients had a complete response to primary chemotherapy with a mean CA 125 AUC of 48 8 IU mL S E 15 9 while twenty two patients had a partial response and thirteen patients had no response or disease progression with a mean CA 125 AUC of 251 7 IU mL S E 65 8 and a mean of 316 5 IU mL S E 107 5 respectively p 0 05 Figure 3 For predicting the patient final state the best accuracy 7496 was achieved at CA 125 AUC lt 100 IU mL ROC AUC 0 77 To predict patient survival the most accurate was 8796 CA 1
115. licada em reas t o distintas como ci ncias sociais ex dura o do primeiro casamento economia ex dura o de subscri es de produtos financeiros ou outros engenharia ex tempo de vida de sistemas electr nicos A distribui o dos tempos de sobreviv ncia normalmente descrita ou caracterizada por tr s fun es fun o de sobreviv ncia fun o de densidade de probabilidade e a fun o de risco Estas fun es s o matematicamente equivalentes pois basta conhecer uma para se obter as restantes A fun o de sobreviv ncia S t definida como sendo a probabilidade de um individuo sobreviver para alem de t 8 S t P T gt t f rmula 3 19 Tipicamente alguns elementos t m tempos de sobreviv ncia censurados ou truncados ou seja nestes elementos o tempo de 83 84 Material e M todos sobreviv ncia n o foi totalmente observado possu mos informa o acerca do tempo de sobreviv ncia mas n o temos o tempo de sobreviv ncia exacto o elemento em causa durante o estudo n o alcan ou o limite de interesse Este tipo de situa es deve se a v rios factores como o evento de interesse num conjunto de elementos n o ocorre antes do fim do estudo um elemento n o seguido at ao fim do estudo ou esse elemento abandona o estudo mas sabemos o seu tempo de sobreviv ncia at esse instante Um dos m todos mais populares em bioestat stica para estimar a fun o de sobreviv ncia
116. lise dos dados quer recorrendo a m todos estat sticos convencionais quer recorrendo s redes neuronais de uma forma directa dentro do mesmo ambiente de trabalho com a mesma interface gr fica e sem produzir altera es maneira como os dados est o formatados e tabelados Com este software v rios tipos de redes podem ser desenvolvidos ex perceptr es de m ltiplas camadas redes RBF SFOM entre outras V rios algoritmos de aprendizagem podem ser utilizados como a retropropaga o propaga o r pida quick propagation quasi newton Levenberg Marquardt assim como v rias t cnicas de pr e p s processamento como escalonamento e codifica o de vari veis Os dados podem ser automaticamente ou manualmente separados em dados de treino teste e valida o para garantir uma correcta generaliza o por parte da rede podendo tamb m proceder a reamostragens resampling Em muitos casos a construc o de uma rede neuronal envolve varias tentativas nas quais se experimentam v rios tipos de redes diferentes at se encontrar a melhor poss vel para resolver o problema Esta fase pode ser tediosa principalmente se todos os par metros i e o n mero de neur nios ocultos o n mero de camadas a taxa de aprendizagem o momento etc tenham de ser ajustados a cada treino de cada rede 112 Material e M todos Nesta fase o STATISTICA Neural Networks apresenta uma ferramenta muito til Trata se do In
117. lo da velocidade de aumento do CA 125 calcula se a velocidade de aumento do CA 125 entre a antepen ltima e ultima concentrac o do CA 125 com a condic o de que Cio gt Ci e gt 214 Conclus es Finais 9 1 REFER NCIAS 1 Schwarzer G W Vach and M Schumacher On the misuses of artificial neural networks for prognostic and diagnostic classification in oncology Stat Med 2000 19 4 p 541 61 Lisboa P J A review of evidence of health benefit from artificial neural networks in medical intervention Neural Netw 2002 15 1 p 11 39 Rustin G J et al Early versus delayed treatment of relapsed ovarian cancer MRC OVO5 EORTC 55955 a randomised trial Lancet 2010 376 9747 p 1155 63 Rustin G J et al Defining progression of ovarian carcinoma during follow up according to CA 125 a North Thames Ovary Group Study Ann Oncol 1996 7 4 p 361 4 215
118. lores elevados de CA 125 Situa es n o malignas podem apresentar o CA 125 elevado Entre estas incluem se a gravidez menstrua o doen a inflamat ria p lvica endometriose quistos nos ov rios tero fibr ide e ascites derivadas a doen a hep tica Situa es malignas n o ginecol gicas com met stases peritoneais podem apresentar eleva es no CA 125 como acontece nos cancros da mama c lon pancre tico e do pulm o 91 O uso mais comum para a concentra o s rica do CA 125 na monitoriza o da doente com cancro epitelial do ov rio diagnosticado principalmente quando a doen a n o mensur vel por outros meios Com o tratamento padr o uma percentagem das doentes apresentam uma doen a intra abdominal de pequeno volume ou com res duos microsc picos No entanto doentes com tumores com um volume maior ap s cirurgia inicial podem n o apresentar sintomas e a avalia o da quimioterapia pode n o ser poss vel com o exame f sico ou imagiol gicos Os n veis s ricos do CA 125 podem fornecer informa o til sobre a resposta ao tratamento inicial assim como durante tratamentos futuros e na detec o de reca das 92 V rios estudos confirmam a utilidade do CA 125 em monitorizar o progresso das doentes com cancro epitelial do ov rio O aumento do CA 125 precede a detec o cl nica da reca da Durante a monitoriza o da doen a o CA 125 pode ser 55 56 Introdu o utilizado de v rias forma
119. ls CUH main database from 1990 to 2000 Only 63 patients mean age 58 5 years old range 16 3 82 4 years were included in the present analysis due to the restriction of our inclusion criteria patients that underwent primary line chemotherapy within 3 months after submission to cytoreductive surgery Six patients had FIGO stage five had stage Il forty one had stage Ill seven had stage IV and in four patients this information was missing Fifty eight had epithelial ovarian cancer four patients have other histological 117 ne 125 half life breakpoint between a good and poor prognosis in patients with ovarian cancer types and one patient with no information None borderline tumor was include Seven patients had a tumor grade 1 twenty two a tumor grade 2 nine had a tumor grade 3 and twenty five patients had no tumor grade information The mean duration of primary chemotherapy was 4 2 0 7 10 2 S E 0 2 months twenty seven patients had a complete response to primary chemotherapy eighteen had a partial response fourteen had no response or a disease progression and four patients had missing information At evaluation date forty three patients were deceased while twenty three were alive The mean overall survival was 30 0 1 96 94 2 S E 2 7 months For each patient we determined the CA 125 half life using the formula In 2 t 1 1 2 77 C In where Ci is the first 125 serum c
120. m dominio carboxi terminal que inclui uma liga o transmembranar com um dom nio citoplasm tico curto V rios estudos permitiram identificar a mucina MUC16 como sendo o antig nio CA 125 84 85 Mucinas epiteliais s o glicoprote nas complexas altamente O glicosiladas encontradas no muco ou na superf cie de c lulas de v rios epit lios Elas s o respons veis pelas propriedades f sicas dos mucos e est o envolvidas na protecc o das c lulas epiteliais 86 com func es de hidratac o lubrificac o e de barreira da superficie epitelial Por outro lado podem funcionar como mol culas de sinalizac o 87 V rios estudos realcam o poss vel efeito do CA 125 de inibir a resposta imune que direccionada contra o cancro do ov rio 88 O CA 125 modula selectivamente a sensibilidade das c lulas tumorais do ov rio aos agentes genot xicos com o dom nio carboxi terminal a ser aparentemente suficiente para promover a resist ncia cisplatina 89 53 54 Introdu o Repeats H h BiH n BER DIL O o 1 4 BER Ly x La 4 zu Ole E 1a BER 1H ate 1 RU GS ES BR A B gig Of E B mis MO X Q q BEB SB poo gin des NH Glycosylated fF Amino terminal One of 60 t Domain 40 kD Bi 9 Extracellular Domain Potential Cleavage site PO Plasma membrane 4 S T Phosphorylation site Cytoplasmic Domain COOH Figura 1 1 Estrut
121. m regime terap utico padr o baseado na combina o de carboplatina e paclitaxel Esta escolha decorre de v rios ensaios randomizados controlados e da descoberta e s ntese de v rios agentes citot xicos Durante a d cada de 50 ocorre a s ntese dos agentes alquilantes clorambucilo e melfalano com a sua aplica o isolada nos anos 60 Durante as d cadas de 70 e de 80 ocorre o desenvolvimento da cisplatina e da carboplatina respectivamente ocorrendo ensaios que comparam a efic cia destes agentes Na d cada de 90 ocorre a introdu o do paclitaxel e a realiza o de estudos de combina o deste agente com os compostos de platina Tabela 1 3 Apesar do desenvolvimento dos v rios agentes citot xicos e da realiza o de v rios ensaios cl nicos existe ainda controv rsia e debate sobre as doses a administrar e a via de administra o 31 33 Introdu o Tabela 1 3 Agentes citot xicos mais comuns na quimioterapia do cancro epitelial do ov rio D cada 70 Cisplatina Agente Cl NH3 Mecanismo de Ensaios Cl nicos altera es estruturais no ADN Agentes alquilantes vs cisplatina Efeito citot xico causado por altera es na transcri o e replica o induzindo a apoptose Estes mecanismo devido liga o ao ADN com forma o de liga es intra e intercadeias com Cisplatina isolada vs cisplatina combinada D cada Agente Mecanismo de ac o E
122. mammograms Surgery 2001 129 459 466 125 AUC robustness as predictor for a complete response to primary chemotherapy in ovarian cancer a multivariant analysis using ANN 20 21 22 23 Christoyianni Koutras A Dermatas E Kokkinakis G Computer aided diagnosis of breast cancer in digitized mammograms Comput Med Imag Grap 2002 26 309 319 Markopoulos C Kouskos E Koufopoulos K Kyriakou V Gogas J Use of artificial neural networks computer analysis in the diagnosis of microcalcifications on mammography Eur J Radiol 2001 39 60 65 Burke HB Goodman PH Rosen DB Henson DE Weinstein JN Harrel Jr FE Marks JR Winchester DP Bostwick DG Artificial neural networks improve the accuracy of cancer survival prediction Cancer 1997 74 4 857 862 Masters T Practical Neural Networks Recipes in Academic Press Inc 1993 pp 174 177 EGJ 7 THE POTENTIAL FOR THE RATE OF 125 INCREASE TO PREDICT OVARIAN CANCER RELAPSE European Journal of Hospital Pharmacy Science 2006 Abstract Aim To study the applicability of the rate of CA 125 increase in serum as a predictor for ovarian cancer relapse Methods The study involved eighty patients from the Gynaecology Service of Coimbra University Hospital For each patient the first CA 125 increase determined by the first three consecutive levels were Ci 1 C and Ci2 Ci with C being the CA 125 level at time i The Mann Whitney U test
123. mbinam o uso do CA 125 com ecografia com sonda vaginal 35 Introdu o O uso combinado de v rios biomarcadores uma das estrat gias seguidas para melhorar a detec o antecipada da doen a 11 14 Mais recentemente com o desenvolvimento da tecnologia gen mica e prote mica v rios estudos est o em curso para sinalizar e integrar informa o produzida por estas t cnicas nos ensaios cl nicos e em particular no rasteio do cancro do ov rio 15 17 1 1 2 Diagn stico e Estadiamento Aproximadamente 90 das mulheres diagnosticadas com cancro do ov rio apresentam sinais e sintomas meses antes do diagn stico no entanto n o s o espec ficos 1 18 Entre outros estes sintomas podem incluir 2 a Dilata o press o ou desconforto abdominal o Dor abdominal Anorexia perda de apetite perda de peso n o intencional o Dores nas costas Diarreia gt Fatiga 9 Saciedade r pida h Intoler ncia g strica i N useas v mitos flatul ncia j Dor p lvica k Dor rectal Dificuldades respirat rias m Dis ria Nas mulheres p s menop usicas o ov rio torna se n o palp vel pelo que um ov rio palp vel deve ser investigado n o obstante apenas 10 estarem associados ao cancro do ov rio Mulheres mais Introdu o jovens com um ov rio qu stico e com di metro superior a 5cm dever o igualmente ser investigadas 1 O diagn
124. mo factores de progn stico para o cancro do ov rio Um dos mais importantes factores de progn stico o est dio FIGO no diagn stico revelando assim a import ncia da sua correcta aferi o Um outro factor de progn stico independente para a sobreviv ncia a doen a residual ap s cirurgia citorredutora 62 66 O grau histol gico em conjunto com o est dio pode tamb m ser importante para antever a sobreviv ncia O tipo histol gico geralmente n o apresenta signific ncia progn stica excepto nos tumores de c lulas claras e de c lulas pequenas que est o associados a um pior progn stico 62 As caracter sticas das doentes como a idade e o ndex de performance performance status s o tamb m importantes e est o correlacionados com o resultado da doen a 62 enquanto a co Introdu o morbilidade aparentemente n o apresenta um efeito no progn stico 67 V rios par metros cin ticos e a concentra o s rica do marcador tumoral CA 125 apresentam se como factores de progn stico incluindo a taxa de regress o durante a quimioterapia neoadjuvante o tempo de semi vida do CA 125 ou o tempo de normaliza o do CA 125 durante o tratamento prim rio e a concentra o s rica p s operat ria 68 73 Al m do CA 125 outros biomarcadores s ricos tecidulares e gen ticos s o inclu dos em investiga es para aferir do seu potencial na utiliza o cl nica como factores de progn stico para o canc
125. n a Durante a fase de tratamento prim rio o tempo de semi vida do CA 125 um dos par metros cin ticos mais estudados sendo um factor de progn stico para o cancro do ov rio Por esse motivo no primeiro trabalho Cap tulo 4 com a nossa popula o conclu mos que com a determina o do tempo de semi vida do CA 125 poss vel classificar o progn stico da doen a No entanto a determina o num rica do tempo de semi vida nem sempre era poss vel principalmente quando o CA 125 n o estabilizava abaixo do cut off 209 210 Conclus es Finais de 35Ul mL Atendendo a esta limita o foi experimentado um novo par metro cin tico Cap tulo 5 a rea sob a curva do CA 125 normalizada pelo tempo ASC A ASC do CA 125 apresenta vantagens sobre outros par metros cin ticos pois n o assume qualquer condi o sobre o valor s rico individual do CA 125 para o seu c lculo como por exemplo a estabiliza o do CA 125 abaixo do valor normal de 35Ul mL permitindo assim a sua determinac o num rica sem quaisquer constrangimentos A ASC do CA 125 durante o tratamento prim rio por estar estreitamente relacionada com uma resposta completa a quimioterapia adjuvante pode permitir classificar numericamente o resultado do tratamento ou seja poder ser um end point especialmente til na compara o de diferentes tipos de tratamento nomeadamente em estudos prospectivos A aplica o de redes neuronais
126. n o lineares 10 Nas reas da sa de as redes neuronais podem pelas suas caracter sticas serem ferramentas teis na pesquisa de padr es com o objectivo de acelerar o desenvolvimento de t cnicas e meios capazes de tornar o diagn stico e o combate s doen as mais eficaz No caso do cancro s o variados os exemplos da utiliza o de redes neuronais nas publica es m dicas muitas vezes com resultados promissores O n mero de ensaios cl nicos e ensaios randomizados Material e M todos controlados que envolveram a utiliza o de redes neuronais no diagn stico e progn stico aumentou de 1 para 38 desde 1994 a 2006 No entanto dos 396 estudos sobre cancro nos quais as redes neuronais est o presentes apenas 27 eram ensaios cl nicos ou ensaios randomizados controlados nos quais 21 mostraram um aumento no benef cio da prestac o de cuidados e 6 n o Todavia nenhum destes estudos mostrou uma diminuic o no benef cio 11 3 3 2 Defini o de Redes Neuronais Artificiais Desde o in cio do desenvolvimento dos computadores v rios esforcos t m sido feitos com o objectivo de elevar os computadores de meras m quinas calculadores autom ticas a m quinas inteligentes 12 Podemos definir intelig ncia artificial como sendo o estudo do comportamento inteligente e a implementa o de sistemas computacionais que exibam comportamento inteligente 13 Ao contr rio da aproxima o simb lica que utiliza s
127. n 4 Human epidermal growth factor receptor 2 19 20 ICON LASA LMS M1 1 MLP MUC16 NPV OC125 OV 197 OVX1 PE PET PFS PLCO PPV PR RBF RD RECIST RMN ROC SE SQL luo TAC TFG TNM UICC International Collaborative Ovarian Neoplasm Lipid associated sialic acid Least mean square Monoclonal antibodies M1 1 Multilayer perceptron Mucin 16 Negative predictive value Murine monoclonal antibody 125 125 antibody produced against the ovarian cancer cell line Monoclonal antibodies OV 197 Monoclonal antibody OVX1 Processing elements Tomografia por emiss o de positr es Progression free survival Prostate lung colorectal and ovarian Positive predictive value Partial response Radial basis function networks Residual disease Response evaluation criteria in solid tumors Resson ncia magn tica nuclear Receiver operator characteristic Standard error Structured query language Tempo de semi vida Tomografia axial computorizada Taxa de filtrac o glomerular Sistema TNM de Classifica o dos Tumores Malignos Tamanho do tumor T grau de extens o aos n dulos linf ticos N e presenca de met stases distantes M Union for International Cancer Control UKCTOCS UK collaborative trial of ovarian cancer screening VLSI Very large scale integration WFDC2 WAP four disulfide core domain 2 gene WR Without response 21 LISTA DE TABELAS 1 Introduc o Tabela 1 1 Classifica o FIGO para os ca
128. nce options to manage ovarian cancer patients These include second look laparotomy regular physical examination regular imaging and CA 125 measurement 7 Described by Bast et al 1983 28 CA 125 has an important role in analyzing ovarian cancer follow up Numerous studies indicate that a continuous rise in serum 125 precedes the clinical detection of recurrence 29 30 and CA 125 may have an important role to play in the design of clinical trials from prognosis to follow up 31 32 Several authors tested alternative CA 125 progression criteria to predict ovarian cancer recurrence Rustin et al state that a confirmed rise in CA 125 serum concentration to more than twice the normal upper limit during follow up after first line chemotherapy accurately predicts relapse 33 Tuxen et al reported that analytical imprecision and intra individual biological fluctuation considerably influence the CA 125 AUC as a predictor for epithelial ovarian cancer relapse variation in CA 125 results 34 and developed several progression criteria that could provide early information about ovarian cancer recurrence 35 Instead of using the absolute tumour marker concentration itself the use of tumour marker kinetic parameters seems to be a more rational approach Indeed the serum CA 125 kinetic behaviour has an important role in describing the course of the disease Parameters such as the CA 125 half life can be used to evaluate the cl
129. nces Fourth Edition ed 1987 John Wiley amp Sons Inc Rui Guimar es J C Estat stica 1997 McGraw Hill Portugal Elisa Lee J W Statistical Methods for Survival Data Analysis Third edition ed 2003 John Wiley amp Sons Inc P Armitage G B Estad stica para la Investigati n Biom dica Trad J Peret Tercera Edici n ed 1997 Harcourt Brance de Espafia S A StatSoft STATISTICA 6 0 Neural Networks Manual StatSoft Inc 1984 2000 Lisboa P J and A F Taktak The use of artificial neural networks in decision support in cancer a systematic review Neural Netw 2006 19 4 p 408 15 Masters T Practical Neural Network Hecipes in 1993 Academic Press Inc Fu L Neural Networks in Computer Intelligence 1994 McGraw Hill Inc Haykin S Redes Neurais Princ pios e Pr tica Trad Paulo Engel 2nd Edition ed 2001 Porto Alegre Bookman Bourquin J et al Basic concepts of artificial neural networks ANN modeling in the application to pharmaceutical development Pharm Dev Technol 1997 2 2 p 95 109 Drew P J and J R Monson Artificial neural networks Surgery 2000 127 1 p 3 11 Robert Carola J H Charles Noback Human Anatomy amp Physiology Second edition ed 1992 McGraw Hill Inc Bishop C Neural Networks for Pattern Recognition 2000 Oxford University Press Inc 118 4 CA 125 HALF LIFE BREAKPOINT BETWEEN A GOOD AND PROGNOSIS IN PATIENTS WITH
130. ncros do ov rio 38 Tabela 1 2 Rela o entre a classifica o FIGO com a Classifica o NM scii iod ee adiret eec eene 39 Tabela 1 3 Agentes citot xicos mais comuns na quimioterapia do cancro epitelial do ov rio 43 3 Material e M todos Tabela 3 1 Tabela de conting ncia onde est o registados os resultados que relacionam os resultados de um teste com uma dada condi o 75 Tabela 3 2 Comparac o entre o sistema biol gico e artificial 104 Tabela 3 3 entre termos utilizados em estat stica e redes neuronais artificiais 104 4 CA 125 half life breakpoint between a good and poor prognosis in patients with ovarian cancer Table 4 1 Overall survival for the different prognostic group concerning the t1 2 120 5 CA 125 AUC as a new prognostic factor for patients with ovarian cancer Table 5 1 Summary of patient characteristics 133 Table 5 2 CA 125 AUC behaviour among patients according to several covariates eres 134 23 24 Table 5 3 Sensitivity specificity positive predictive value PPV and accuracy of several CA 125 AUC cut offs for predicting the patient final state overall survival and chemotherapy complete response
131. nsaios Cl nicos 80 Carboplatina O mesmo da cisplatina Carboplatina vs cisplatina Carboplatina isolada vs carboplatina combinada Via endovenosa vs via intraperitoneal 43 Introdu o Tabela 1 3 Continua o D cada 90 Paclitaxel Agente O paclitaxel liga se aos microtubulos interrompendo Mecanismo de ac o o ciclo celular na parte final da fase G2 e nafase M inibindo a replica o celular Combina o platina com paclitaxel vs combina es Ensaios Cl nicos de platina sem taxanos Carboplatina com paclitaxel vs cisplatina com paclitaxel 1 34 2 http pubchem ncbi nim nih gov 3 35 4 adaptado de 31 V rios ensaios mostram que a carboplatina apresenta a mesma efic cia que a cisplatina no entanto regimes com carboplatina s o melhores tolerados devido ao seu perfil toxicol gico com redu o da toxicidade no aparelho auditivo rim e sistema nervoso Apresenta tamb m menos n useas e v mitos 1 31 32 Devido carboplatina ser excretada maioritariamente pela urina a exposi o sistem tica depende da fun o renal Assim para o doseamento da carboplatina utiliza se a f rmula de Calvert na qual se tem em conta a taxa de filtra o glomerular e a rea sob a curva concentrac o tempo dose carboplatina mg ASCxTFG 25 Introdu o Na d cada de 90 com a descoberta dos taxanos nomeadamente do paclitaxel o tratamento do cancro do ov
132. nsidered negative In order to build our model and due to the fact that this was a retrospective study a forward stepwise approach was implemented using the following strategy for each patient dataset Exemplifying with a patient having ten CA 125 levels obtained after the end of 187 188 CA 125 AUC as a predictor for epithelial ovarian cancer relapse primary treatment all patients should have a minimum of three CA 125 levels the CA 125 AUC was calculated from to AUC and compared to CA 125 AUC from to subsequently CA 125 AUC was calculated from to C4 AUC and compared to CA 125 AUC from to AUC and so on until CA 125 AUC from Co to Cro AUC was calculated and compared to CA 125 AUC from Co to AUC The stepwise procedure stopped when the CA 125 AUC to a certain concentration time fulfilled the started condition AUC 2 F AUC positive test or when the started condition was not reached at all for the complete dataset negative test Criterion B While with the Criterion A we were looking for rapid increases in CA 125 AUC the Criterion B aimed to look for sustained increases that could potentially be associated to a relapse Therefore in this approach the idea was to find a fixed CA 125 AUC cut off value that once exceeded could signal patient relapse As defined before the total normalised in time CA 125 AUC from the first concentration Co to to
133. o m todo do produto limite para estimar taxas de sobreviv ncia tamb m conhecido pelo m todo de Kaplan Meier 2 9 Sejam t lt t2 lt lt t os tempos de bito distintos observados numa amostra de tamanho retirada de uma popula o homog nea k lt n pois alguns elementos da amostra podem ser censurados e ou terem o evento ao mesmo tempo seja n o numero de elementos com risco num tempo imediatamente antes de t 1 lt lt s o casos na qual a dura o pelo menos t e seja d o n mero de bitos no tempo f ent o a fun o de sobreviv ncia S t estimada por 50 2 f rmula 3 20 t t sendo designada por estimador do produto limite ou estimador de Kaplan Meier e com um intervalo de confianca a 95 dado por SMe f rmula 3 21 com Material e M todos E d 1 f rmula 3 22 8 0 US Notar que a probabilidade estimada do elemento sofrer o n evento no intervalo de tempo entre t a t jc a probabilidade n estimada do elemento de n o sofrer o evento no mesmo intervalo e o produto na f rmula 3 20 a regra do produto para as probabilidades 2 A constru o da curva de Kaplan Meier implica a determina o de S t para cada t com a constru o da tabela com os elementos t S t e respectiva representa o gr fica Figura 3 5 1 0 S t 0 8 0 6 0 4 0 2 0 0 Tempo Figura 3 5 Exemplo de uma
134. o pois uma entrada diferente das restante e o peso sin ptico relacionado com esta nova entrada wko designa se por bias do neur nio bx Podemos traduzir bias como o peso que provoca desequil brio Figura 3 13 Xo bk Wko Wk1 Wk2 Uk 9 X2 s fun o de yk sa da activag o X1 Xm Wkm Figura 3 13 Modelo de um neur nio artificial no qual est representada o bias Assim as equac es 3 29 e 3 30 tomam a forma b f rmula 3 39 j 1 o que equivale a u wx f rmula 3 40 j 0 y Pu y wx f rmula 3 41 j 0 O bias by tem o efeito de aumentar ou diminuir a entrada total aplicada fun o de activa o dependendo se positivo ou negativo 108 Material e M todos respectivamente e podemos consider lo como um par metro externo do neur nio k 14 Estudado o neur nio artificial podemos agora realizar algumas comparac es teis A Tabela 3 2 mostra algumas comparac es entre c rebro biol gico e as redes neuronais artificiais e a Tabela 3 3 mostra a comparac o entre os termos utilizados na estat stica e os termos utilizados nas redes neuronais artificiais Tabela 3 2 Comparac o entre o sistema biol gico e artificial C rebro biol gico Redes Neuronais Artificiais Concentrac o do BR Peso sin ptico neurotransmissor Neur nios artificias ou unidades de Neur nios processam
135. o t pertencente ao conjunto 7 t varia de 1 a n 3 passo calcular a sa da da rede 4 passo alterar os pesos sin pticos de acordo com a regra de aprendizagem estabelecida 5 passo quando o desempenho da rede for o desejado ent o sair caso contr rio incrementar t em 1 e ir para o 2 passo e repetir Se t n faz se 1 e vai se para o 2 passo e repete se ou sai se e anota se o desempenho final da rede Notar que a cada apresentac o de um elemento de treino d se o nome de poca de treino ou de interacc o de treino devendo ser feita de uma forma aleat ria tornando a procura dos valores dos pesos sin pticas de natureza estoc stica V rios algoritmos de aprendizagem est o dispon veis para o treino de redes neuronais artificiais os quais podem usar diversas regras de aprendizagem O objectivo do algoritmo de aprendizagem ent o o ajuste dos pesos sin pticos de modo a minimizar o erro da feita pela rede neuronal V rias regras de aprendizagem podem ser utilizadas entre elas temos a aprendizagem por correcc o de erro competitiva de Boltzman e Hebbiana por correlac o Material e M todos 3 3 9 O STATISTICA Neural Networks No nosso trabalho foi utilizado um software geral de redes neuronais artificiais Foi escolhido o STATISTICA Neural Networks vers o 6 da StatSoft Inc Trata se de um software que apresenta a vantagem de permitir numa mesma aplica o uma an
136. olver os seguintes passos a Selec o e constru o de um modelo de redes neuronais adequado ao problema em causa b Treino da rede neuronal artificial usando um algoritmo de treino adequado c Uso da rede neuronal artificial treinada para resolver o problema em causa A propriedade mais importante de uma rede neural artificial a capacidade de aprender de treinar e de melhorar o seu desempenho ao longo do tempo Uma rede neuronal artificial aprende atrav s de um processo interactivo de ajustes aos pesos sin pticos 14 Baseada na defini o de Mendel e McClaren 1970 14 e na defini o de Hebert Simon 1983 13 podemos definir aprendizagem ou treino como um processo pelo qual os par metros Material e M todos livres da rede s o moldados atrav s da estimulac o do ambiente onde a rede est inserida permitindo rede executar a mesma tarefa a partir do mesmo ambiente mais eficazmente no tempo seguinte O algoritmo de aprendizagem n o mais do que um conjunto de regras para a resoluc o de um problema de aprendizagem existindo v rios exemplos de algoritmos mas basicamente divergindo entre si na forma como o ajuste dos pesos sin pticos ir ser feito 14 A aprendizagem pode ser supervisionada ou supervisionada Na primeira o conjunto de treino para al m dos sinais de entrada tem tamb m incorporado os sinais de sa da que ir o ser comparados com os sinais de sa da originados pela rede
137. oncentration after cytoreduction formula 4 1 surgery must be superior to the 35 IU mL cut off Cz is the first CA 125 serum concentration below 35 IU mL or the last CA 125 concentration after five CA 125 normal half lives 50 days in case CA 125 did not normalize below 35 IU mL and t and t are the corresponding dates for C and gt respectively Regarding the half life value the patients prognosis was classified into two classes good or poor prognosis if lt X or t gt X respectively where X is equal to 10 15 20 25 30 35 40 days All patients with t jp lt s O were classified as poor A statistical analysis was conducted and the Mann Whitney U test was used to compare the overall survival across subgroups of patients depending on CA 125 ti behavior p lt 0 05 was considered statistically significant Table 4 1 Figure 4 1 shows the survival curve Kaplan Meier in agreement with CA 125 half life breakpoint of 16 days CA 125 half life breakpoint between a good and poor prognosis in patients with ovarian cancer 4 3 DISCUSSION M Colakovi et al 4 reported that patients with a CA 125 half life lt 20 days had 1 8 times longer survival times than those with a longer CA 125 half life while in our study the obtained breakpoint was 16 days and patients with CA 125 half life below this value have a mean survival 1 67 times greater than others with longer CA 125 half lif
138. or ovarian cancer Several CA 125 AUC cut off values were tested for predicting the complete response to chemotherapy patient final state and patient overall survival superior or equal to one three and five years The ROC curve AUC was also calculated for each prediction In spite of a consistent relation between CA 125 AUC and patient survival and final state it became clear that CA 125 AUC had a higher performance in predicting the full response to chemotherapy and best accuracy 0 82 was achieved with a CA 125 AUC lt 100 IU mL cut off The relation between CA 125 AUC and the complete response was also shown by the highest obtained ROC curve AUC for predicting a full response to chemotherapy This relation was also shown by the statistical correlation between low values of CA 125 AUC and the complete response against elevated values of CA 125 AUC correlation with a partial response or without response to chemotherapy without 157 158 125 AUC robustness as predictor for a complete response to primary chemotherapy in ovarian cancer a multivariant analysis using ANN statistical difference between the partial response and without response groups In the present work a confirmation of the relation between CA 125 AUC and the full response to chemotherapy using a multivariant approach was expected In the present study the development of neural models increases the ability to predict the complete response when compared wi
139. os trabalhos surgiram pela m o de v rios investigadores levando ao ressurgimento do interesse na rea das redes neuronais artificiais Grossberg em 1980 desenvolveu a teoria da resson ncia adaptativa ART Adaptative Resonance Theory Em 1982 Hopfield desenvolveu redes com realimentac o Tamb m em 1982 Kohonen publicou um artigo sobre mapas auto organiz veis Em 1986 surgiu um marco importante na hist ria das redes neuronais artificiais desenvolvimento do popular algoritmo de do erro backpropagation por Rumelhart Hinton e Williams Apesar de este algoritmo n o representar o processo de aprendizagem no c rebro biol gico ele permite o treino de perceptr es de camadas m ltiplas de um modo eficaz e elegante e a sua introdu o representa um corte entre modela o do c rebro biol gico e o uso de redes neuronais artificiais como analisadores de dados 15 16 3 3 4 O C rebro Biol gico Apesar de n o ser objectivo deste trabalho a descri o minuciosa da anatomia e fisiologia do c rebro biol gico imp e se uma descri o breve de alguns conceitos b sicos Material e M todos O sistema nervoso pode ser visto como um sistema constitu do por tr s elementos Figura 3 6 os receptores a rede neural e os efectivadores A rede neural o c rebro o elemento central que toma as decis es e que recebe e transmite informac o para os receptores e para os efectivado
140. ovarian cancer 6 8 147 148 125 AUC robustness as predictor for a complete response to primary chemotherapy in ovarian cancer a multivariant analysis using ANN Several prognostic factors can be used in ovarian cancer These include the FIGO F d ration Internationale de Gyn cologie et d Obstr trique tumor stage tumor grade DNA ploidy over expression of the HER 2 neu oncogene residual disease RD after initial cytoreductive surgery rate of response to chemotherapy and patient characteristics such as performance status and age 9 Furthermore in a recent work a univariant analysis was performed using CA 125 normalized in time area under the curve CA 125 AUC as a prognostic factor for ovarian cancer 8 Artificial Neural Networks ANNs are a branch of artificial intelligence Al and consist of computer based programs that attempt to simulate some features of the biological brain such as learning generalizing and abstracting from experience ANNs learn i e train from appropriate learning examples gathering knowledge by detecting patterns and relationships in data ANNs consist of several artificial neurons or processing elements PE connected to each other Each connection is associated with a weight Each neuron is capable of executing simple information processing but the power of ANNs resides in the fact that they are connected in a network According to their topology and function there are several
141. overall survival 2 3 years Sensitivity 0 64 0 82 0 82 0 98 0 98 0 98 0 98 0 98 0 98 1 00 Specificity 0 75 0 63 0 63 0 35 0 31 0 19 0 19 0 13 0 06 0 04 075 PPV 0 70 0 67 0 67 0 58 0 57 0 52 0 52 0 51 0 49 0 49 0 05 Accuracy 0 70 0 72 0 72 0 65 0 63 0 57 0 57 0 53 0 50 0 50 CA 125 cut off to predict patient overall survival 2 5 years Sensitivity 0 71 0 90 0 90 1 00 1 00 1 00 1 00 1 00 1 00 1 00 Specificity 0 65 0 51 0 51 0 25 0 23 0 14 0 14 0 10 0 08 0 04 973 PPV 0 38 0 35 0 35 0 28 0 28 0 26 0 26 0 25 0 44 0 43 0 06 Accuracy 0 66 0 60 0 60 0 42 20 40 0 34 0 34 0 30 0 47 0 45 CA 125 cut off to predict patient complete response to chemotherapy Sensitivity 0 65 0 85 0 85 0 94 0 96 0 98 0 98 0 98 1 00 1 00 Specificity 0 89 0 77 0 77 0 37 0 34 0 23 0 23 0 14 0 11 0 06 0 87 PPV 0 89 0 85 0 85 0 69 0 68 0 65 0 65 0 63 0 63 0 61 0 04 Accuracy 0 75 0 82 0 82 0 71 071 0 68 0 68 0 64 0 64 0 62 S E Standard Error 136 125 AUC as new prognostic factor for patients with ovarian cancer Table 5 4 Cox proportional hazard model results for CA 125 AUC as an independent factor for predicting patient overall survival Dependent Overall survival Independent CA 125 AUC Censoring variable Patient Final State Deceased Alive Stratified by Chi Square p value None 18 19 p lt 0 05 FIGO Tumor Stage l ll Ill IV 6 42 lt 0 05 Residual Disease gt 2 cm 14 80 p 0 05 Primary Chemotherapy Respons
142. pecificada como sendo o n mero de camadas apesar de nem todas as redes possuam uma estrutura em camadas e o n mero de n s por camada As camadas podem ser divididas em a Camada de entrada de input a que recebe informac o codificada que est a ser apresentada rede num dado instante As unidades desta camada processam informac o apenas a distribuem para as outras unidades da serem normalmente representadas graficamente de maneira diferente 105 106 Material e M todos b Camada s oculta s n o directamente observ vel da o seu nome a camada que confere s redes caracter sticas de n o linearidade c Camada de sa da de output a camada que codifica o sinal de sa da das redes podendo representar por exemplo um n mero ou uma classe Em rela o ao esquema de interliga es entre os neur nios artificiais as redes podem ser a Redes alimentadas para a frente nas quais as liga es apontam num s sentido da camada de entrada para a de sa da b Redes recorrentes nas quais existem liga es de retro alimenta o feedback ou saltos loops As liga es podem ser sim tricas quando h uma liga o do neur nio para o ent o h tamb m uma liga o do neur nio para o e os pesos associados s liga es s o iguais wj2w Nos restantes casos as liga es s o assim tricas As liga es podem tamb m ser a Intra
143. que sejam independentes dos dados de treino e valida o interna Com um n mero de doentes reduzido que possu amos mesmo com valida o cruzada a cria o de uma ferramenta inform tica utilizando redes neuronais para utiliza o cl nica plena estava assim comprometida No entanto no trabalho apresentado Cap tulo 6 as redes neuronais apesar de n o mostrarem vantagem sobre os m todos estat sticos convencionais confirmaram o resultado obtido da relev ncia e aplicabilidade da ASC CA 125 como factor de progn stico para o cancro do ov rio Na segunda fase do trabalho foi estudado o comportamento do CA 125 e qual a sua rela o com a recidiva Foram estudados a velocidade de aumento do CA 125 e a aplica o da ASC do CA 125 na detec o da recidiva Cap tulos 7 e 8 respectivamente Mais uma vez a aplica o da ASC do CA 125 apresenta vantagem sobre a velocidade de aumento melhor exactid o e tempo m dio at a recidiva mais consistente e sem depender de nenhuma 211 212 Conclus es Finais concentra o individual do CA 125 havendo um efeito atenuante devidos variabilidade das concentra es do CA 125 Sabe se ent o que o aumento do CA 125 caracterizado ou n o por par metros cin ticos est relacionado com a recidiva e durante o acompanhamento das doentes uma das quest es centrais ser a antecipa o ou n o da quimioterapia apenas com a eleva o do CA 125 sem manifesta es cl
144. r for a rea sob a curva ROC 5 3 2 4 Teste de Mann Whitney O teste de Mann Whitney tamb m conhecido pelo teste de Mann Whitney Wilcoxon ou teste MWW um teste equivalente n o param trico ao teste t Student para comparar amostras independentes Pode utilizar se este teste para avaliar a hip tese nula que assume que as medianas populacionais s o id nticas para os dois grupos Este teste n o exige que as popula es tenham a mesma vari ncia Como este teste baseado na classifica o por Material e M todos n mero de ordens ranks das observa es o teste consegue utilizar mais informa o que um teste de mediana 3 6 7 Este teste pressup e que a As duas amostras tenham sido retiradas respectivamente das suas popula es de uma forma independente e aleat ria b As amostras sejam no m nimo ordinais em termos de escala Se as populac es divergem de todo elas apenas divergem em relac o s suas medianas Na hip tese nula admite se que as medianas de ambas as amostras coincidem e na hip tese alternativa postula se que a diferenca entre as medianas n o nula positiva ou negativa Ho m m H n 1 ou H 7 gt 0u ainda 7 lt 7 Sejam n e ne os tamanhos das duas amostras a estatistica de teste U calculada da seguinte forma a As observa es das duas amostras s o combinadas nica vari vel de tamanho N n n gt sendo identificadas as r
145. res Podemos olhar para as setas que apontam para a direita como a transmiss o para a frente da infoomac o e as que apontam para a esquerda como a retro alimentac o do sistema Rede neural Receptores i Efectivadores Resposta c rebro Figura 3 6 Representa o em diagrama do sistema nervoso O c rebro biol gico constitu do por um n mero elevado de c lulas nervosas chamadas de neur nios Por exemplo estima se que no c rebro humano existam mais de 10 bili es de neur nios 14 Os neur nios apresentam duas propriedades importantes a excitabilidade capacidade de responder a est mulos e a condutividade capacidade de conduzir um sinal Os neur nios est o entre as c lulas mais especializadas est o ligados entre si organizados em redes intensamente interligadas e apesar de exibirem diferentes formas e tamanhos apresentam tr s zonas diferenciadas cada uma com uma fun o espec fica o corpo celular dendrites e um ax nio Figura 3 7 17 95 96 Material e M todos Corpo Figura 3 7 Ilustra o de um neur nio biol gico Retirado de 17 Apesar de os neur nios serem cinco a seis ordens de grandeza mais lentos que os circuitos em sil cio o c rebro biol gico uma estrutura extremamente eficiente Esta efici ncia devida n o s ao elevado n mero de neur nios mas tamb m ao elevado n mero de interliga es entre eles estimando se que no c rebro humano existam ap
146. rio sofre um avan o significativo Ensaios de fase Il demonstraram que o paclitaxel era activo na reca da do cancro do ov rio mesmo nos casos de reca das poucos meses ap s a administra o do tratamento com cis ou carboplatina Como resultado de v rios estudos a combina o de carboplatina com paclitaxel a terapia padr o de primeira linha para a quimioterapia adjuvante no cancro do ov rio com o paclitaxel a ser administrado com uma dose de 175mg m durante 3 horas seguido da carboplatina doseada pela f rmula de Calvert com uma ASC entre 5 0 a 7 5 durante 1 hora e normalmente em 6 ciclos de 3 semanas 4 31 A combina o cisplatina paclitaxel e doxorrubicina demonstrou uma melhoria marginal no intervalo livre de doen a mas sem benef cios na sobreviv ncia global quando comparado com o regime carboplatina paclitaxel 36 Apesar do tratamento padr o levar a uma remiss o completa na maioria das doentes estimado que 50 ir o sofrer uma recidiva 32 Como visto anteriormente devido falta de testes de rastreio e sintomas espec ficos a maioria das mulheres com cancro do ov rio diagnosticada num est dio avan ado Na doen a inicial apesar da cura cir rgica ser efectiva na maioria das vezes ainda ocorrem um n mero elevado de recidivas A quimioterapia adjuvante nestes casos ainda controversa De acordo com estudos do International Collaborative Ovarian Neoplasm ICON e da European Organisation
147. rna se ent o necess rio o uso de medidas que permitam medir a robustez de determinado teste Matriz confus o ou tabelas de informa o cruzada o nome dado matriz CxC em que o valor de um item ij representa quantos casos na amostra com diagn stico foram diagnosticados como i Por conven o o diagn stico padr o representado nas colunas Material e M todos Quando C 2 ou seja quando h apenas dois diagn sticos poss veis como sim ou n o a matriz 2x2 chamada de tabela de conting ncia Atendendo seguinte tabela de conting ncia Tabela 3 1 podemos definir varias medidas estat sticas 1 3 Tabela 3 1 Tabela de conting ncia onde est o registados os resultados que relacionam os resultados de um teste com uma dada condic o Condic o Presente n mero de doentes Sim N o Resultado do Positivo a Teste Negativo C d Assim temos a o n mero de verdadeiros positivos b o n mero de falsos positivos c o n mero de falsos negativos d o n mero de verdadeiros negativos A sensibilidade a capacidade que um teste tem em detectar os positivos dentro dos individuos que apresentam a condi o ou seja a propor o de indiv duos com a condi o que apresentam um teste positivo Sensibilidade f rmula 3 1 A especificidade a capacidade que um teste tem em detectar os negativos dentro dos individuos que n o
148. ro epitelial do ov rio 74 1 1 5 Biomarcadores Utilizados no Cancro Epitelial do Ov rio O uso de biomarcadores marcadores tumorais no caso do cancro de enorme utilidade na pr tica cl nica podendo ser utilizados em diferentes fases da doen a incluindo testes de rasteio diagn stico factores de progn stico e para monitorizar o tratamento e ou detec o de uma reca da 75 Na pr tica cl nica do cancro do ov rio o marcador tumoral mais utilizado para monitorizar a evolu o da doen a o CA 125 No entanto muitos outros biomarcadores est o sob investiga o separada ou em conjunto com o CA 125 Entre eles podemos encontrar glicoprote nas como o CA 15 3 CA 72 4 CA 19 9 cidos si licos associados a l pidos LASA e OVX1 Demonstrou se que a amplifica o do gene HE4 WFDC2 ocorre no carcinoma do ov rio 76 estando em curso estudos sobre a sua aplica o na monitoriza o no cancro epitelial do ov rio 77 A alpha fetoprote na 51 52 Introdu o AFP beta HCG e a concentra o s rica de inibinas podem ser igualmente teis em determinados tipos de cancro do ov rio 78 Como marcadores podemos tamb m encontrar citocinas calicre nas humanas factor de crescimento do endot lio vascular D D mero prote nas reguladoras do ciclo celular e da apoptose ciclina D ciclina p53 Cip Kip entre outras telomerase transcriptase reversa receptores tirosina cinase e metaloproteinas
149. roximadamente 60 trili es de interliga es entre neur nios 14 interliga o entre dois neur nios ou entre um neur nio e uma c lula muscular ou glandular d se o nome de sinapse O neur nio que leva o impulso nervoso em direc o sinapse chamado de pr sin ptico e a c lula que se encontra depois da sinapse chamada de p s sin ptica Enquanto a c lula pr sin ptica sempre um neur nio a p s sin ptica pode ser um neur nio uma c lula muscular ou uma c lula glandular A transmiss o de um impulso nervoso numa sinapse pode ser qu mica ou el ctrica Na sinapse el ctrica apesar de rara as duas c lulas comunicantes est o electricamente ligadas por canais intercelulares das membranas plasm ticas adjacentes A este tipo de liga o d se o nome de liga o tipo gap O movimento livre dos Material e M todos i es nestes canais faz com que o impulso nervoso passe da c lula pr sin ptica para a p s sin ptica de um modo quase instant neo No homem este tipo de ligac o entre neur nios pode ser encontrada na retina 17 A sinapse qu mica a mais comum e existe a interven o de uma subst ncia qu mica transmissora o neurotransmissor Depois do impulso nervoso chegar zona terminal do ax nio pr sin ptico ocorre a liberta o do neurotransmissor para o espa o sin ptico entre os neur nios O neurotransmissor actua na c lula p s sin ptica ocorrendo ou n o a continua
150. s Como factor de progn stico para medir a efic cia do tratamento ou para detectar a reca da Tendo por base estas utiliza es foram estudados v rios crit rios utilizando o CA 125 93 a Valor s rico absoluto do CA 125 pr quimioterapia e sobreviv ncia Valor s rico absoluto do CA 125 p s quimioterapia e sobreviv ncia Valor s rico absoluto do CA 125 durante a quimioterapia e sobreviv ncia Valor s rico absoluto do CA 125 pr cirurgia e sobreviv ncia Valor s rico absoluto do CA 125 p s cirurgia sobreviv ncia Tempo de semi vida do CA 125 e sobreviv ncia Valor s rico nadir valor mais baixo do 125 e sobreviv ncia Tempo at ao valor s rico nadir do 125 e sobreviv ncia rea sob a curva normalizada pelo tempo Valores longitudinais aumentos ou diminui es s ricos do CA 125 e sobreviv ncia Introdu o 1 3 REFER NCIAS 10 11 12 18 14 Oliveira F M e C F d Manual de Ginecologia Volume I Cancro Epitelial do Ov rio 2000 Hennessy B T R L Coleman and M Markman Ovarian cancer Lancet 2009 374 9698 p 1371 82 Global Cancer Facts amp Figures 2011 2nd Edition Available from http www cancer org acs groups content epidemiologysurv eilance documents document acspc 027766 pdf Guppy A E P D Nathan and G J Rustin Epithelial ovarian cancer a review of current management Clin Oncol R Coll Radiol 2005 17 6 p 39
151. s Lancet Oncol 2002 3 529 536 3 Rustin GJ Bast RC Jr Kelloff GJ Barrett JC Carter SK Nisen PD Sigman CC Parkinson DR Ruddon RW Use of CA 125 in clinical trial evaluation of new therapeutic drugs for ovarian cancer Clin Cancer Res 2004 10 11 3919 3926 4 Guppy AE Rustin GJ CA125 Response Can it Replace the Traditional Response Criteria in Ovarian Cancer Oncologist 2002 7 437 443 5 Rustin GJ Nelstrop AE McClean P Brady MF McGuire WP Hoskins WJ Mitchell H Lambert HE Defining Response of Ovarian Carcinoma to Initial Chemotherapy According to Serum CA 125 J Clin Oncol 1996 14 1545 1551 6 Bast RC Xu FJ Yu YH Barnhill S Zhang Z Mills GB CA 125 The past and the future Int J Biol Markers 1998 13 4 179 187 7 Bidart JM Thuillier Augereau C Chalas J Daver A Jacob N Labrousse F Voitot H Kinetic of serum marker concentrations and usefulness in clinical monitoring Clin Chem 1999 45 10 1695 1707 8 Mano A Falc o A Godinho 1 Santos J de Oliveira C Caramona M CA 125 AUC as a new prognostic factor for patients with ovarian cancer Gynecol Oncol 2005 97 2 529 534 9 Holschneider CH Berek JS Ovarian cancer epidemiology biology and prognostic factors Semin Surg Oncol 2000 19 3 10 159 160 125 AUC robustness as predictor for a complete response to primary chemotherapy in ovarian cancer a multivariant analysis using ANN 10 1
152. s and 117 2 95 DISCUSSION 119 4 4 A 121 5 CA 125 AUC as a new prognostic factor for patients with ovarian cancer ra aes decies eed acte ite URU xen a Dele 5 2 Patients and Melodia 59s FIBSUIS nannan Bd DISCUSSION E E E E N o ha pula A E 6 CA 125 AUC robustness as a predictor for a complete response to primary chemotherapy in ovarian cancer a multivariant analysis using 6 1 IMMPOGUCTION cc seele 6 2 Patients and 8 De SS EE 6 4 DISCUSSION a 6 5 References rara 7 The Potential for the Rate of CA 125 Increase to Predict Ovarian Cancer Relapse fis nl ig o o 6 CHE 7 2 Objective Patients and Xm LI we D NIU RA ns TA DISCUS SIO m E M 7 5 References arara rara ana 8 CA 125 AUC as a predictor for epithelial ovarian cancer relapse acc act ar Mar oar CA A aha te 8 1 Introduction rara 8 2 Patients and Methods aa 8 3 Results 8 4 Discussion
153. s duas opera es aritm ticas traduzindo se assim numa vantagem computacional 18 2 Fun o tangente hiperb lica definida por o u tanh u f rmula 3 36 esta fun o pode ser tamb m escrita da seguinte forma u e p u f rmula 3 37 Uy 101 102 Material e M todos e a sua derivada dada por Q u 0 tanh u f rmula 3 38 Esta fun o tem a vantagem face fun o descrita pela equa o 3 34 de poder assumir valores negativos Figura 3 12 trazendo vantagens anal ticas 14 Figura 3 12 Gr fico da fun o tangente hiperb lica Muitas outras fun es de activa o podem ser utilizadas dependendo essencialmente do tipo de rede a ser aplicada ao problema em estudo por exemplo a fun o raiz quadrada utilizada em redes de Kohonen A cada liga o ao neur nio k est associada um peso Os pesos sin pticos podem ser n meros inteiros ou reais podendo pertencer um intervalo limitado Antes do treino o esquema de inicializa o dos pesos pode ser espec fico a uma rede particular mas normalmente s o inicialmente randomizados a n meros inteiros pequenos Durante o treino s o ajustados apesar de alguns poderem ser fixados deliberadamente como no caso do bias Quando terminar o treino todos os pesos s o fixados 13 Material e M todos Podemos adicionar uma entrada fixa na qual xo 1 ou 1 designamos por x
154. sance des r cidives premi res par le temps de doublment du CA 125 Bull Cancer 1997 84 9 855 860 Mano A Falc o A Godinho 1 Santos J Leit o de Oliveira C Caramona M CA 125 AUC as a new prognostic factor for patients with ovarian cancer Gynecol Oncol 2005 97 529 534 205 9 CONCLUS ES FINAIS Conclus es Finais Apesar de cada artigo apresentado possuir j as suas pr prias conclus es Cap tulos 4 8 nesta secc o pretende se proceder a uma integrac o capaz de levar obtenc o de conclus es mais abrangentes sobre o presente trabalho Neste estudo retrospectivo o registo inform tico da informa o m dica mostrou se fundamental pois permitiu uma agrega o da informa o de um modo c lere Apesar das bases de dados estarem em plataformas diferentes a codifica o usada pelos HUC foi fundamental para a interliga o das mesmas Este facto associado ao desenvolvimento do programa Filtra Dados acabou por permitir fazer uma an lise multivariante e inclusivamente calcular diversos par metros cin ticos para o CA 125 em diferentes fases de seguimento das doentes Assim o estudo foi dividido em dois componentes principais a Estudar a cin tica do CA 125 durante a fase de tratamento prim rio com o objectivo de inferir sobre o progn stico da doen a b Estudar o comportamento do CA 125 ap s o tratamento prim rio com o objectivo da detec o da recidiva da doe
155. se C and E Solomon Current status of the molecular characterization of the ovarian cancer antigen CA125 and implications for its use in clinical screening Gynecol Oncol 2003 88 1 Pt 2 p 5152 7 O Brien T J et al The CA 125 gene an extracellular superstructure dominated by repeat sequences Tumour Biol 2001 22 6 p 348 66 Yin B W A Dnistrian and Lloyd Ovarian cancer antigen CA125 is encoded by the MUC16 mucin gene Int J Cancer 2002 98 5 p 737 40 Yin B W and K O Lloyd Molecular cloning of the CA125 ovarian cancer antigen identification as a new mucin MUC16 J Biol Chem 2001 276 29 p 27371 5 Desseyn J L D Tetaert and V Gouyer Architecture of the large membrane bound mucins Gene 2008 410 2 p 215 22 Perez B H and I K Gipson Focus on Molecules human mucin MUC 16 Exp Eye Res 2008 87 5 p 400 1 Patankar M S et al Potent suppression of natural killer cell response mediated by the ovarian tumor marker CA125 Gynecol Oncol 2005 99 3 p 704 13 Introdu o 89 90 91 92 93 Boivin M et al CA125 MUC16 tumor antigen selectively modulates the sensitivity of ovarian cancer cells to genotoxic drug induced apoptosis Gynecol Oncol 2009 115 3 p 407 13 Theriault C et al MUC16 CA125 regulates epithelial ovarian cancer cell growth tumorigenesis and metastasis Gynecol Oncol 2011 121 3 p 434 43 Werness B A and G H Eltabba
156. t the best value from the CA 125 tumour marker to predicted epithelial ovarian cancer relapse Criterion A The aim of this approach was to find in what way a rapid increase in CA 125 AUC could signal patient relapse For this purpose the overall CA 125 AUC as a predictor for epithelial ovarian cancer relapse CA 125 AUC for two consecutive CA 125 sampling times were calculated and the increment between them evaluated Mathematically in this criterion the overall CA 125 AUC normalized in time to C t AUC was compared to the previous one i e the total CA 125 AUC normalized in time to 1 AUC 7 AUC and AUC were determined using the following formulae ti to AUC 1 formula 8 1 _ bh AUC formula 8 2 where in both cases Ajis the area of trapezoid calculated by e Therefore calculated AUCs were compared accordingly the following rational AUC gt F AUC with equal to 1 10 AUC gt AUC in more than 10 of AUC 1 25 AUC 2 AUC in more than 25 of AUC 1 1 50 AUC gt AUC in more than 50 of AUC 1 1 75 AUC 2 in more than 75 of AUC 2 00 AUC z AUC in more than 100 of AUC 1 2 50 AUC 2 AUC in more than 150 of AUC or 3 00 AUC 2 AUC in more than 200 of AUC Table 1 If the condition AUC 2 F AUC was true the test was considered positive Otherwise the test was co
157. telligente Problem Solver IPS qual permite testar v rios tipos de redes com diferentes caracter sticas dentro de cada tipo podendo no final resumir as melhores redes encontradas comparando as utilizando uma medida de desempenho e de erro para cada um dos subconjuntos de treino e validac o Este software permite tamb m realizar an lises de sensibilidade teis para obter informa o sobre a import ncia das vari veis de entrada ainda poss vel desenvolver conjuntos de redes neuronais ensembles que cooperam para melhorar a previs o e evitar os poss veis erros cometidos pela rede vari ncia e bias 10 Material e M todos 3 4 REFER NCIAS eo 10 11 12 13 14 15 16 17 18 Ton Cleophas A Z Statistics Applied to Clinical Studies Fifth Edition ed 2012 Springer Science Business Media B V Le C Introductory Biostatistics 2003 John Wiley amp Sons Inc Carrasco J M todo Estadistico en la Envestigati n M dica Sexta Edici n ed 1995 Editorial Ciencia 3 Shultz E K Multivariate receiver operating characteristic curve analysis prostate cancer screening as an example Clin Chem 1995 41 8 Pt 2 p 1248 55 Zweig and G Campbell Receiver operating characteristic ROC plots a fundamental evaluation tool in clinical medicine Clin Chem 1993 39 4 p 561 77 Daniel W Biostatistics A Foundation for Analysis in the Health Scie
158. th the use of CA 125 AUC lt 100 IU mL cut off alone achieving an accuracy of 0 86 in both tested ensembles with false positive and false negative rates clearly below 1096 This gives a good signal for a possible clinical implementation of the procedure The response graph for ensemble 2 Figure 2 clearly shows that when CA 125 AUC is a continuous value the cut off is still 100 IU mL to discriminate yes no a complete response to chemotherapy This result confirms the best CA 125 AUC cut off value obtained in the previously mentioned study 8 and the fact that ensemble 1 achieved the same performance as ensemble 2 The sensitivity analysis clearly shows that in the performed experiments CA 125 AUC was the most important variable for the prediction made by the MLP ensembles Finally the CA 125 AUC kinetic parameter could be potentially used as an efficiency treatment quantifier which represents a significant leap since the comparison between different drugs or different chemotherapy regimens would be more easily and appropriately performed with especial emphasis on clinical trials 125 AUC robustness as predictor for a complete response to primary chemotherapy in ovarian cancer a multivariant analysis using ANN 6 5 REFERENCES 1 Baker VV Treatment options for ovarian cancer Clin Obstet Gynecol 2001 44 3 522 530 2 Harries M Gore M Part I Chemotherapy for epithelial ovarian cancer treatment at first diagnosi
159. tion were used A leave one out cross validation was carried out to prevent overfitting with an ensemble formation one for each input set The performance for each input set was measure by calculating the accuracy of the prediction made by each ensemble The relative importance of each input variable was tested by performing a sensitivity analysis Results Both ensembles achieved the same performance with an accuracy of 0 86 The obtained response graph shows that the breakpoint for CA 125 AUC continuous variable was precisely 100 IU mL to allow the discrimination of the complete response to chemotherapy yes no In addition according to the sensitivity analysis the CA 125 AUC value is the most important variable for prediction purposes CA 125 lt 100 IU mL for ensemble 1 and CA 125 AUC value for ensemble 2 Conclusion The sensitivity analysis shows that the CA 125 AUC variable is the most important input for predicting a complete response to primary chemotherapy Moreover the CA 125 cut off yes no obtained for our population was 100 IU mL 125 AUC robustness as predictor for a complete response to primary chemotherapy in ovarian cancer a multivariant analysis using ANN 6 1 INTRODUCTION A number of biological markers have been found which can help to diagnose a particular tumour monitor the treatment response or detect the first signs of relapse Ovarian cancer has the highest mortality among all invasive can
160. tipos de epit lio e estroma em combina es variadas A maioria destes tumores derivada do epit lio de revestimento do ov rio e representam aproximadamente 90 de todas as neoplasias invasivas primitivas do ov rio Para al m dos tumores epiteliais do ov rio podemos encontrar tumores de c lulas germinais tumores do estroma gon dico cord es sexuais mesenquimatosos gonadoblastoma tumores n o espec ficos do ov rio tumores metast ticos e sarcomas Histologicamente os tumores epiteliais do ov rio classificam se em tumores serosos mucinosos endometri ides meson fricos e carcinomas concomitantes 1 Em termos epidemiol gicos a incid ncia do cancro epitelial do ov rio mais alta na Europa Estados Unidos e Israel e mais baixa no Jap o e pa ses em desenvolvimento 2 De acordo com a American Cancer Society Global Cancer Facts amp Figures 2011 2nd Edition a incid ncia ajustada para a idade nos pa ses desenvolvidos de 9 4 por 100 000 mulheres com uma mortalidade ajustada para a idade de 5 1 por 100 000 mulheres em contraste com os pa ses em desenvolvimento em que a incid ncia ajustada para a idade de 5 0 por 100 000 mulheres com uma mortalidade ajustada para a idade de 3 1 por 100 000 mulheres 3 O cancro do ov rio ocupa a posic o cimeira no que diz respeito mortalidade no grupo das neoplasias invasivas do aparelho genital 33 34 Introdu o feminino atingindo preferenci
161. tive value NPV and best accuracy for predicting relapse was calculated In all true positive tests the lead time to relapse was measure from t to the relapse date t recorded in the patient clinical file Figure 8 1 A positive lead time indicates that the increase criteria occur before the relapse while a negative time indicates that the increase criteria occur after the relapse 189 CA 125 AUC as a predictor for epithelial ovarian cancer relapse Relapse CA 125 Serum Concentration IU ml End of Primary Evaluation Treatment Date Criterion A Search a concentration time C t where 125 AUC to tiz Factor CA 125 AUC to Criterion B Search a concentration time C t where A A A FA A gt Cut Off U ml Lt CA 125 AUC Figure 8 1 Hypothetical CA 125 evolution after primary treatment of patient with ovarian cancer 190 CA 125 AUC as a predictor for epithelial ovarian cancer relapse 8 3 RESULTS The mean age at diagnostic time was found to be 54 2 17 2 83 8 S E 1 4 years At evaluation date the mean overall survival was 6 4 0 6 19 4 S E 0 4 years with thirty 27 096 patients deceased Fifty 45 1 patients had FIGO stage twelve 10 8 had stage ll forty two 37 8 had stage Ill three 2 7 had stage IV and in four 3 6 patients this information was missing Twenty seven 24 3 patients had a tumour grade 1 twenty 18 096 a tumour grade 2 eight 7
162. ts with cancer 1 Ovarian cancer has the highest mortality rate of all invasive cancers of the gynecological system The cancer antigen CA 125 serum concentration is usually adopted to evaluate the clinical situation of ovarian cancer patients during treatment or follow up CA 125 is a membrane glycoprotein expressed by epithelial cells of different origins and it was first identified in 1981 by the monoclonal antibody OC125 2 The rate of decline in CA 125 during primary chemotherapy has been an important prognostic factor and in the majority of studies rising CA 125 values have preceded clinical detection of recurrence 3 Using the CA 125 half life several studies report that the greatest difference in progression rate is found at a ty of 20 days 4 6 Rustin et al developed definitions based on CA 125 levels to measure response in patients receiving first line chemotherapy 7 and in a different study concluded that a confirmed rise of the CA 125 level to more than twice the upper limit of normal during follow up after first line chemotherapy accurately predicts relapse 8 167 168 The Potential for the Rate of CA 125 Increase to Predict Ovarian Cancer Relapse 7 2 OBJECTIVE PATIENTS AND METHODS In this retrospective study we have examined the ability of the rate of CA 125 serum concentration increase to predict ovarian cancer relapse Retrospective clinical information was gathered the Gynaecology Servi
163. types of Multilayer perceptrons MLPs radial basis function networks RBFs and Kohonen networks are examples of ANN types To train an ANN a learning algorithm is applied There are several learning rules including back propagation BP conjugate gradient descent etc The knowledge of the network is stored in the weights of the connections between the artificial neurons The ANN can generalize when the weights are used in the presence of new data 10 13 The interface of medicine and artificial intelligence especially ANNs has been well emphasized by Lisboa 14 and several works exemplify the use of artificial neural networks in cancer research CA 125 AUC robustness as predictor for a complete response to primary chemotherapy in ovarian cancer a multivariant analysis using ANN Many studies make use of artificial neural networks to improve cancer diagnosis ANNs can be used in the diagnosis of malignant ovarian tumors 15 An ANN can be trained to provide clinically accurate information on whether or not an adnexal mass is malignant from the patient s menopausal status serum CA 125 levels and some simple ultrasonographic criteria 16 In prostate cancer many studies have shown a significant improved accuracy in diagnosing staging and predicting post treatment results for ANNs compared to conventional statistical analysis 17 ANN can be used to examine mammograms in screening breast cancer 18 21 ANNs are
164. ur nio que seguia a lei do tudo ou nada O trabalho de McCulloch e Pitts influenciou muitos outros investigadores como von Neumann que desenvolveu o EDVAC a partir do primeiro computador de prop sito geral constru do entre 1943 e 1946 Em 1949 Hebb publicou o livro The Organization of Behaviour no qual formula pela primeira vez uma regra de aprendizagem fisiol gica para a modifica o sin ptica Rosenblatt em 1958 apresenta uma nova abordagem para o estudo do reconhecimento de padr es ao introduzir o perceptr o com a demonstra o do teorema da converg ncia do perceptr o Em 1960 foi desenvolvido por Widrow e Hoff o ADALINE ADAptive Linear Element destacando se do perceptr o no processo de aprendizagem pela implementa o do algoritmo do m nimo quadrado m dio LMS Least Mean Square Depois de um per odo de grande desenvolvimento com o perceptr o com o qual se pensava em resolver qualquer tipo de problema ent o que surge em 1969 o 93 94 Material e M todos livro de Minsky e Papert no qual demonstraram matematicamente as limita es dos perceptr es de camada nica dando in cio a uma poca de descr dito nas redes neuronais artificiais levando ao desinteresse dos investigadores e ao corte de fundos para esta rea Podemos considerar a d cada de setenta como uma poca de in rcia na investigac o em redes neuronais artificiais No in cio da d cada de oitenta v ri
165. ura proposta da mol cula de CA 125 retirado de 83 O CA 125 MUC16 apresenta um papel funcional fundamental nas c lulas tumorais do ov rio por modulac o do crescimento celular mobilidade celular propriedades invasivas metastiza o carcinogenia Aparentemente o dom nio citoplasm tico curto apresenta uma fun o essencial nestes efeitos 90 O CA 125 o marcador tumoral mais utilizado no cancro epitelial do ov rio No entanto o uso isolado da concentra o s rica do CA 125 como teste de rastreio para o cancro do ov rio num est dio inicial n o adequado Alguns requisitos devem ser tidos em conta nos testes de rastreio Devem ter sensibilidade adequada facilmente realiz veis baixo custo e ter um alto valor preditivo positivo sendo que este ltimo factor depende tamb m da preval ncia da doen a A Introdu o doen a a ser rastreada deve ter uma elevada morbilidade ou mortalidade e a dura o da fase pr cl nica deve ser suficientemente longa para ser detectada num teste de rastreio A aplica o de testes de rastreio ao cancro do ov rio tem sido dif cil por desconhecimento da natureza e dura o da fase pr cl nica e pela baixa preval ncia do cancro do ov rio na popula o Na doen a num est dio avan ado o CA 125 est elevado em cerca de 90 dos casos mas apenas em 50 no est dio para al m de que alguns tipos histol gicos como adenocarcinomas mucinosos n o est o associados a va
166. valor d se o nome de ponto de corte cut off point 2 A an lise utilizando curvas ROC Receiver Operator Characteristic foi originalmente desenvolvida no in cio da d cada de 50 no contexto de detec o electr nica de sinais e introduzida como m todo para avaliar as interpreta es de verdadeiros e falsos positivos em sinais de radar Nas d cadas seguintes este tipo de an lise foi introduzida na rea da medicina com um elevado crescimento de trabalhos publicados 4 5 Uma curva ROC a representa o gr fica da sensibilidade ordenadas em fun o de 1 especificidade abcissas resultante da varia o do valor do ponto de corte ao longo dos seus poss veis valores Figura 3 3 3 4 2 5 0 5 Sensibilidade 0 0 5 1 1 especificidade Figura 3 3 Exemplo de uma curva ROC 77 78 Material e M todos 1 0 e e d o sensibilidade e 0 0 0 0 0 2 0 4 0 6 0 8 1 0 1 especificidade Figura 3 4 Exemplo de uma curva ROC estimada tamb m designada por curva ROC n o param trica A curva ROC permite determinar o ponto de corte ptimo sendo este que mais se aproxima do ponto superior esquerdo 0 1 que corresponde a 10096 de sensibilidade e 10096 de especificidade A rea sob a curva ROC permite quantificar a exactid o de um determinado teste Assim as curvas ROC s o tamb m teis na compara o de diferentes testes tendo um teste uma exactid o tanto maior quanto maio
167. varian cancer relapse Table 8 2 Summary of patient characteristics FIGO Stage Patient Count 96 50 45 196 II 12 10 8 III 42 37 8 IV 3 2 7 Missing 4 3 6 Tumor Grade 1 27 24 3 2 20 18 0 3 8 7 2 Missing 56 50 5 Residual Disease gt 2 cm Yes 24 21 6 No 85 76 6 Missing 2 1 8 Primary Treatment Surgery Surgery Adjuvant Chemotherapy Surgery Adjuvant Chemotherapy Consolidation 34 30 6 47 42 4 30 27 0 Chemotherapy Relapse Yes 26 23 4 No 77 69 4 Missing 8 7 2 Patient Final State Deceased 30 27 0 Alive 81 73 0 193 Table 8 3 Sensitivity specificity positive predictive value PPV negative predictive value NPV false positives FP 96 false negatives FN 96 and accuracy achieved for each criterion N2103 due to missing values Criterion A Test CA 125 AUC to f Relapse e FP FN Sensitivity Specificity PPV NPV Accuracy 2 e 99 Yes No F AUC to t Positive 22 26 1 10 10 0 85 0 66 0 46 0 93 0 71 25 2 3 9 Negative 4 51 Positive 19 8 1 25 25 0 73 0 90 0 70 0 91 0 85 7 8 6 8 Negative 7 69 Positive 15 4 lt 1 50 50 0 58 0 95 0 79 0 87 0 85 3 9 10 7 Negative 11 73 Positive 12 2 1 75 75 0 46 0 97 0 86 0 84 0 84 1 9 13 6 Negative 14 75 Positive 10 1 2 00 100 0 38 0 99 0 91 0 83 0 83 1 0 15 5 Negative 16 76 Positive 6
168. was used to compare the rate of increase in CA 125 serum concentration across the subgroup of patients depending on relapse the area under the Receiver Operating Characteristic ROC plot was established and sensitivity specificity positive predictive values and overall accuracy were also determined for several CA 125 increase rate cut off values Results Patients with relapse had a mean standard error CA 125 increase rate of 0 791 0 306 IU mL day in contrast with a mean increase of 0 036 0 009 IU mL day for patients without relapse The best accuracy 80 9 for predicting the relapse was achieved with a CA 125 increase rate cut off gt 0 06 IU mL day The area under the ROC plot obtained was 0 870 0 045 Conclusion A simple CA 125 increase rate definition could quickly predict patient relapse with a high level of certainty Larger datasets should be used to test this method before it can be applied in clinical practice The Potential for the Rate of CA 125 Increase to Predict Ovarian Cancer Relapse 7 1 INTRODUCTION A number of biological markers have been found which can help to diagnose a particular tumour monitor the treatment response or detect the first signs of relapse In some cases instead of using the isolated value of the marker it is more suitable to use kinetic parameters such as tumor marker half life t 2 and tumor marker doubling time DT for the evaluation of clinical response and follow up of patien
169. ys Twenty six 32 5 patients had a confirmed relapse in contrast with fifty two 65 0 169 170 The Potential for the Rate of CA 125 Increase to Predict Ovarian Cancer Relapse patients without a relapse and in two 2 596 patients this information was missing Among all patients with a confirmed relapse two 7 7 had a FIGO stage four 15 496 had a stage Il nineteen 73 had a stage Ill one 3 8 had a stage IV three 11 596 had a tumor grade 1 eight 30 896 had a tumor grade 2 three 11 596 had a tumor grade 3 and in twelve 46 2 this information was missing At the evaluation date fifty three 66 396 patients were alive while twenty seven 33 796 were deceased Patients with relapse had a mean CA 125 increase rate of 0 791 0 306 IU mL day while patients without relapse had a mean increase of 0 036 0 009 IU mL day Figure 1 For predicting the patients with tumour recurrence the best accuracy 80 996 was achieved with a CA 125 increase rate cut off 2 0 06 IU mL day a higher PPV in relation to the 0 04 IU mL day cut off and greater than the 0 05 cut off making it more suitable for separation of the two groups of patients Table 1 The area under the ROC plot was 0 870 0 045 Figure 2 For patients with relapse and a CA 125 increase rate superior or equal to 0 06 IU mL day the mean time between the third level C2 and relapse was 204 2 55 0 days ranging from 55 to 906 days The Potential for

UNIVERSIDADE DE COIMBRA

Contents

Download Pdf Manuals

Related Search

Related Contents