Synthetic estimation of healthy lifestyles indicators: User guide
Contents
1. 8 17 2 Discriminating between small areas 9 173 Supporting 10 ESTIMATES iia 11 2 1 The healthy lifestyle indicators 11 211 sCurretit SMOKING iE a iE erini 11 P E a 11 213 Fruit and vegetable consumption 11 2 14 Fruit and vegetable consumption adults 12 RLD BIN GeCTINKING 12 2D oir ne i aa 12 23 13 3 GUIDE TO THE 14 3L Datas ts E 14 3 11 dataset 14 31 2 The covariate datasets atenei e r aA ERE 15 3 2 Deriving the ward estimates 16 3 3 Deriving the PCO es Gmates tise oases 17 34 Validating MOC els ait hes 17 BCE PER ENC 19 APPENDIX AREA CHARACTERISTICS ASSOCIATED WITH THE HEALTHY LIBES LYLE MEASURES 20 APPENDIX PRODUCING SYNTHETIC ESTIMATES A WORKED2. 24 National Centre for Social Research EXECUTIVE SUMMARY e The National Centre for Social Research NatCen was commissioned by the Department of Health to produce estimates of healthy lifestyle behaviours using Health Survey for England HSfE data e The aim of the project was to respond to the twin requirements of developing small area estimates for publication on the Neighbourhood Statistics NeSS website and of providing key public health information not currently available from any other source e Estimates and 95 confidence intervals covering the period 2000 to 2002 have been produced for wards and Primary Care Organisations PCOs The health behaviours covered are current smoking obesity and binge drinking for adults and fruit and vegetable consumption for children and adults separately e Confidence intervals were produced in order to make the margin of error around the estimates clear We recommend that users view the prevalence for a ward or PCO in light of its confidence interval e Statistical modelling was used to produce the estimates because the sample size of national surveys is too small at ward level to provide reliable estimates e These model based estimates are of a different nature from standard survey estimates They must be used with caution The models estimate the expected prevalence of health behaviours for any ward or PCO given the social and demographic characteristics of its po
3. Longhurst et al 2004 and has been extensively reviewed by academics with expert knowledge of small area estimation A range of checks were used to assess the appropriateness of the five models and to examine whether the models were correctly specified The results of the tests showed that the models were indeed well specified and that the assumptions made were valid This provided confidence in the accuracy of the estimates and the confidence intervals attached to them Having generated the estimates a two stage validation process was undertaken to establish the plausibility of the estimates The first stage involved external validation of the estimates by comparison with other health behaviour data sources These data sources were Camden and Islington Health Authority Survey 1999 National Patient Survey in Primary Care Organisations 2003 Wigan Bolton and Bury Health Authority Surveys 2001 Liverpool Sefton St Helens and Knowsley Lifestyle Surveys 2001 and Health Survey for England 2003 The model based estimates were compared with these data sources both by actual value and by rank Statistical measures of association were also computed to assess the relationship between the model based estimates and those available via the external data sources see Pickering et al 2005 The second stage was a consultation exercise that involved local users academics and health related experts as members of project manag
4. 1 per ward ward respondent and PCO 30872 3231 224 2720 31 Binge drinking p f pao o Fruit and vegetable 23 039 consumption adults Fruit and vegetable 8 438 1 989 4 28 consumption children 3 1 2 The covariate dataset The term covariate describes those area level characteristics e g deprivation scores life expectancy rates rural non rural indicator Government Office Region that were potentially related to health behaviours such as smoking and obesity Because of its universal geographical and population coverage the 2001 Census provided the main source for demographic and social covariate data The full set of Census and administrative datasets that were merged together to provide the area level characteristics that were considered for inclusion in the statistical models are shown in Table 3 2 7 The lack of an exact fit between wards and introduces a further source of error when calculating the confidence intervals for the PCO estimates At present the ONS is carrying out work on this problem The results of this research however will not be available until later in the year As yet therefore NatCen has been unable to take account of this additional error meaning that the margin of error around the PCO estimates may be slightly underestimated 15 National Centre for Social Research Table 3 2 Area level characteristics considered for inclusion in the statistical mod
5. Astley Bridge ward 201 917 represents the Census adult count for the Bolton PCO and 0 2151 represents the expected smoking prevalence for Astley Bridge see Table B 4 The summation over the 20 wards nested within Bolton gives 0 2518 which can be multiplied by 100 to give an overall weighted average of 25 Again users are recommended to interpret this result by adopting a statement such as given the characteristics of its local population we would expect a current smoking prevalence of approximately 25 within the Bolton PCO 31 National Centre for Social Research 32
6. achieved in practice by using the example of Bolton a PCO located in the North West region Using the methodology described in the previous section the expected current smoking prevalence for each ward nested within the Bolton PCO and its Census count of adults aged 16 years or more are shown in Table B 4 13 We have omitted the interaction terms for the same reason 29 National Centre for Social Research Table B 4 Estimates of current smoking prevalence and the total adult population for all wards nested within the Bolton PCO WARDname WARDcode Expected prevalence of Census count of adults current smoking Astley Bridge OOBLFA 0 2151 11 067 Blackrod OOBLFB 0 2199 10 304 Bradshaw OOBLFC 0 2283 10 749 Breightmet OOBLFD 0 3155 10 178 Bromley Cross OOBLFE 0 2075 10 924 Burnden OOBLFF 0 2490 9 597 Central OOBLFG 0 2918 8 070 Daubhill OOBLFH 0 2923 9 084 Deane Cum Heaton OOBLFJ 0 2012 13 263 Derby OOBLFK 0 2498 9 352 Farnworth OOBLFL 0 3124 9 617 Halliwell OOBLFM 0 2564 9 429 Harper Green OOBLFN 0 2936 10 234 Horwich OOBLFP 0 2253 11 378 Hulton Park OOBLFQ 0 2010 13 106 Kearsley OOBLFR 0 2968 10 247 Little Lever OOBLFS 0 2804 9 333 Smithills OOBLFT 0 2378 8 647 Tonge OOBLFU 0 2933 7 918 Westhoughton OOBLFW 0 2338 9 420 Total adult count 201 917 Synthetic estimates for PCOs can be calculated by aggregating the model based estimates for the component wards weighting the contribution of each ward in proportion to its popula
7. also be used to discriminate between wards or PCOs by looking at overlapping confidence intervals When comparing two model based estimates one ward may only be said to have a significantly higher or lower prevalence estimate than another if the confidence intervals for the two wards do not overlap ONS 2004a Table 1 3 shows an example of this where three wards are compared side by side Using Table 1 3 we can say that ward A has a significantly higher current smoking rate than ward since the 95 confidence interval for ward A 32 67 falls entirely outside that for ward B 9 30 Ward C however cannot be said to have a significantly lower estimate than ward A since the confidence interval for ward C 34 54 overlaps with that for ward A 32 67 Table 1 3 Smoking estimates and 95 confidence intervals for three wards 95 confidence intervals for percentage who currently smoke Estimate Lower confidence Upper confidence limit limit As described in Section 1 5 2 the average width of the confidence intervals results in there being little scope for discriminating between wards In the vast majority of cases it would not be possible to state that the prevalence in one ward was higher than that in another with any degree of statistical confidence There is more scope however for comparing PCOs by looking at overlapping confidence intervals 4 Note that the estimate for England is a standard survey estimate obtained
8. by using the health survey data alone National Centre for Social Research 1 7 3 Supporting indicators Users may wish to use the model based estimates of healthy lifestyle behaviours in conjunction with other data sources to build up a profile of wards in their area ONS 2004 Table 1 4 shows an example of this where two wards are compared side by side with respect to healthy lifestyle measures and other externally available indicators Table 1 4 Using supporting indicators to build up a ward profile Indicator England Ward A Ward B Survey based estimate of 26 25 27 Not applicable Not applicable current smoking with 95 confidence interval Model based estimate of current Not applicable 49 32 67 17 8 30 smoking with 95 confidence interval Adults claiming Income Support properties in council tax band 0 6 0 8 H 320 000 Urban rural classification of Not applicable Traditional Suburbs and LRN nanen Sito Index of Multiple Deprivation Not applicable 10 3 ranking 2004 10 bands of equal size with 1 indicating the least deprived wards and 10 the most deprived The first row shows the standard survey national estimate for England This estimate has a narrow confidence interval as it was computed using a large national sample of 30 872 adults The second row lists the model based estimates and 95 confidence intervals for two wards Given the characteristics of ward A for example we wo
9. expectancy at birth number of National years 1999 2000 Statistics smr_10a Compendium Mortality from stroke icd10 i60 i69 of Clinicaland indirectly standardised ratios 2001 Outcome Indicators 2003 Mortality from lung cancer icd10 33 c34 indirectly standardised ratios 23 National Centre for Social Research APPENDIX B PRODUCING SYNTHETIC ESTIMATES A WORKED EXAMPLE 1 Modelling health behaviour data a simple example For health behaviour measures such as smoking i e whether a person currently smokes or not a statistical model can be used to examine how characteristics such as age sex and social class influence the propensity of individuals to smoke We may be interested for example in using Health Survey for England data to examine whether males are more likely to smoke than women In this case therefore current smoking status represents the outcome variable about which comparisons are made and sex denotes a factor which may have an influence on that outcome As current smoking status is a two category binary outcome variable a logistic regression model is the natural one to use in order to examine if sex does influence the propensity of individuals to smoke Using the combined 2000 to 2002 HSfEs we can specify a logistic regression model where current smoking status is specified as the outcome variable 1 current smoker 0 not a current smoker and sex denoted as a factor 0
10. methodology adopted for this project was used previously by the Office for National Statistics ONS to produce ward level income estimates and has been extensively reviewed by academics with expert knowledge of small area estimation A range of checks were used to ensure that the assumptions made by the models were valid The published estimates have also been validated against other health behaviour data sources including the 2003 HSfE National Centre for Social Research 1 BACKGROUND AND GUIDANCE ON USE 1 1 Introduction This document provides a guide to how the synthetic estimates of healthy lifestyle behaviours should be used and the way in which the estimates have been developed This first chapter of the report provides the background to the project and guidance on the use of the estimates The second chapter describes the estimates produced by the project The last chapter provides a non technical overview of the methodology used to produce the estimates 1 2 Estimation for small areas The basic problem with national surveys such as the Health Survey for England HSfE is that they are not designed for efficient estimation for small areas such as electoral wards Heady et al 2003 First prevalence estimates of health behaviours such as current smoking based on the sample data can only be computed for a subset of all wards i e those wards containing respondents to the survey The adult respondents to the 2000 to 2002 health su
11. s heaviest drinking night in the previous week These measures were combined to give the number of units of alcohol consumed on the heaviest drinking day Binge drinking was then defined separately for men and women men were defined as having indulged in binge drinking if they had consumed 8 or more units of alcohol on the heaviest drinking day in the previous seven days for women the cut off was 6 or more units of alcohol Of the 30 440 adults in the 2000 to 2002 HSfEs 5 539 18 were defined to have indulged in binge drinking 2 2 Confidence intervals The five sets of estimates have been produced for 7 958 Census Area Statistics wards and 303 PCOs as at 2003 England As well as producing estimates it is also important to be able to assess the accuracy of the estimates We do this by placing confidence intervals around the estimates As the true prevalence is unknown a range is produced i e a confidence interval within which we are fairly certain that the true value lies On average we would expect the confidence interval to contain the true population value 95 of the time 5 Note that the measure of three or more portions was used rather than the target figure of five or more because the proportion of children in the HSfE eating five or more portions was only 12 It was felt that this was too low a prevalence to obtain reliable synthetic estimates 6 Census Area Statistics CAS wards are used for 2001 Census outputs inc
12. tell us what proportions of those living in the ward smoke by age sex or social class 1 6 Banding ranking of estimates NatCen have made no attempt to rank the wards or assign them to bands e g the highest 10 of wards middle 80 and lowest 10 There are two arguments National Centre for Social Research against ranking wards First the estimates are expected prevalences and do not measure actual prevalence Second given the width of the confidence intervals for the ward estimates reflecting the uncertainty in the modelling process the confidence intervals around the ranks would also be very wide Assigning the wards to bands would still require the uncertainty in the ranking banding to be represented Analysis of the smoking estimates for example has shown that there would not be sufficient evidence to state with confidence that any ward belonged to only one band highest 10 middle 80 and lowest 10 once the uncertainty in the banding had been accounted for Hence a ward belonging to the highest 10 of wards could also be plausibly located within the middle 80 1 7 Examples of data use Given that the model based estimates are subject to a number of important limitations we illustrate in this section some examples of appropriate uses for the estimates 1 7 1 Comparing areas with the national average Users may be interested in identifying those wards which have an underlying prevalence of healthy lifestyle behaviours
13. that are not evenly distributed but that certain areas i e postcode sectors are first selected as Primary Sampling Units PSUs and then households are only selected for interview from these Heady et al 2003 By using the clustering information multilevel modelling provides more accurate standard errors confidence intervals and significance tests and these generally will be more conservative than the traditional estimates obtained by ignoring the presence of clustering in the data Goldstein 2003 Furthermore multilevel models are able to partition the variability in health behaviour measures such as smoking into two core elements one representing variability between areas and the other variability within areas As explained in Pickering Scholes and Bajekal 2004 the variability between areas is used as the basis for assigning precision to the synthetic estimates for small areas such as wards For these reasons multilevel modelling was used in this project to model individual health behaviour data generated from the HSfE The purpose of the modelling was to examine how health behaviours such as smoking obesity and binge drinking were related to characteristics of the area in which people lived 25 National Centre for Social Research B 3 Using multilevel models to generate ward level estimates a worked example The process of generating synthetic estimates of healthy lifestyle behaviours for all 7 958 wards in England i
14. that is significantly higher or lower than England as a whole A ward can only be described as significantly different from the national average if the confidence intervals for those estimates do not overlap Table 1 2 shows an example of this where two wards are compared side by side with the national estimate of current smoking prevalence Using Table 1 2 we can say that ward A has a significantly higher current smoking rate than England as a whole at the 5 significance level since the 95 confidence intervals do not overlap i e the confidence interval for ward A 32 67 falls entirely outside that for the national average 25 27 Ward however cannot be said to have a significantly lower estimate than England as a whole since the confidence intervals overlap the interval for ward B 9 30 overlaps that for the national average 25 27 3 For a technical discussion of ranking see Bird et al 2003 National Centre for Social Research Table 1 2 Smoking estimates and 95 confidence intervals for England and two wards 95 confidence intervals for percentage who currently smoke Estimate Lower Upper confidence confidence limit limit The same line of reasoning can be easily extended to comparing wards to the model based estimate for their PCO Such comparisons may enable PCOs to identify wards within their area with high levels of unhealthy behaviours 1 7 2 Discriminating between small areas The estimates could
15. to be a current smoker obese indulge in binge drinking or consume more than the threshold portions of fruit and vegetables e Within each batch of positive and negative terms the variables have been arranged in decreasing order of statistical significance e The terms containing an asterisk are interaction terms The majority of interaction terms involved one of the Government Office Regions meaning that there was evidence to suggest that the ward characteristics had different relationships with the health behaviour in different regions 21 National Centre for Social Research Definitions of the area characteristics in Tables A 1 and A 2 Table A 3 Variable name hloamnty hovercr icobnuk icouple iethnic illsiall illsiwrk inoqual iolevel ipermsic iprofman isroutin iupdcr50 Table A 4 Ward characteristics Census 2001 data Description Proportion of households without central heating Proportion of households overcrowded occupancy rating minus 1 or less Proportion not born in UK Ireland or European Union Proportion 16 residing as couple Proportion non white Proportion with limiting longstanding illness Proportion of working age with limiting longstanding illness Proportion 16 74 with no educational qualifications Proportion 16 74 with highest qualification NVQ 1 or no qualifications Proportion 16 74 permanently sick disabled Proportion 16 74 professional amp managerial occupations NS SEC 1 am
16. 24 National Centre for Social Research ratio for males ranges from 1 03 to 1 14 we can say that males are significantly more likely than females to be current smokers B 2 Multilevel modelling of health behaviour data Although relatively simple for the purposes of this project this type of model suffers from a number of important limitations Bajekal et al 2004 First it has long been recognised that both individual circumstances and the social and physical environment in which people live influence health behaviours From an individual perspective a person s social class may influence health related behaviour such as whether they smoke or not Equally from an area or ecological perspective smoking prevalence may be influenced by social norms of behaviour In addition the individual and ecological influences can interact to mitigate or increase the risk of being a smoker Twigg et al 2000 Using the techniques of multilevel modelling a model can be applied to survey data that simultaneously accounts for both individual and area level influences on behaviour such as smoking Second by explicitly dealing with hierarchical structures e g individuals within households within regions multilevel models are also well equipped to work with the sampling structure of national surveys such as the HSfE that cluster selected individuals and households within postcode sectors The sampling structure of national surveys results in samples
17. als NatCen has produced confidence intervals to accompany the model based estimates in order to make the margin of error around the estimates clear The interval reflects the range between which the true value is believed to lie at a given level of confidence The confidence intervals therefore represent the uncertainty in the modelling process At the 95 confidence level assuming that the model is a good representation of reality the confidence interval is expected to contain the true value around 95 times out of 100 For example if a ward estimate of current smoking is 49 and the 95 confidence interval is 32 67 we know that 95 of the time the true prevalence estimate for that ward based on its local population characteristics will fall within this range It is important to take into account the margin of error around the estimates when interpreting them We therefore recommend that rather than focus exclusively on the prevalence estimate users view the prevalence for a ward or PCO in light of its confidence interval The average width of the confidence intervals both for wards and PCOs varied across the five healthy lifestyle behaviours Table 1 1 shows that the confidence intervals are widest for children s fruit and vegetable consumption and smallest for obesity for more details on the factors influencing the width of the confidence intervals see Pickering et al 2005 Table 1 1 Average width of the 95 confidence interval
18. attendance allowance claimant rate and being located in the South West region There were three interaction terms in the model each a ward level characteristic with a regional indicator This implies that there was evidence to suggest that those characteristics had different relationships with current smoking in different regions the association between smoking and attendance allowance claimant rates was more strongly negative in the North West region the proportion of non white residents was associated with an increased rate of smoking in the South West region compared 27 National Centre for Social Research with a decrease for the other regions and the association between smoking and being in the 3 most deprived band of wards was stronger in the South West region Stage 2 Using the model to derive ward level estimates Having selected the optimal model for current smoking the model was then used to calculate the underlying expected prevalence estimate of current smoking for all 7 958 wards in England As described in Section 3 1 2 the covariate dataset compiled for this project based on the 2001 Census and other administrative data sources contained the known values of various area level characteristics for all wards Table B 3 shows an extract from this covariate dataset for the Little Lever ward Table B 3 shows for example that the Little Lever ward is located in the North West region and is nested within the Greater Mancheste
19. e 2 Pickering et al 2004 and implementation producing small area estimates based a best method identified in stage 2 for five health behaviours and accompanying reports spreadsheets metadata and user guidance for publication on the Neighbourhood Statistics NeSS website and dissemination to the health community Stage 3 Pickering et al 2005 Model based estimates with 95 confidence intervals have been produced for five healthy lifestyle behaviours covering the period 2000 to 2002 The estimates have been produced at two levels Census Area Statistics CAS ward and Primary Care Organisation PCO The five healthy lifestyle behaviours covered are current smoking for adults aged 16 years or obesity for adults aged 16 years or binge drinking for adults aged 16 years or more consumption of five or more portions of fruit and vegetables per day for adults aged 16 years or more and consumption of three or more portions of fruit and vegetables per day for children aged from 5 to 15 years inclusive The aim of the project was to respond to the twin requirements of developing small area estimates for NeSS and of providing key public health information not currently available from any other source In particular we expect the estimates to assist Primary Care Organisations to identify wards within their area with high levels of unhealthy behaviours and to plan local services accordin
20. edures are presented for each health behaviour in Appendix A The second stage involved applying the results from the model to calculate prevalence estimates using the Census administrative information available for all wards A detailed worked example of how model based estimates can be produced in practice is outlined in Appendix 8 The deprivation scores were aggregated to ward level using a weighted average of the deprivation scores produced for lower level Census Super Output Areas 9 The material provided in Appendices and is more technical and hence users should consider them as optional 16 National Centre for Social Research Complex methods were used to derive confidence intervals for the synthetic estimates For a fuller technical description of the methodology users are referred to the Stage 3 report Pickering et al 2005 3 3 Deriving the PCO estimates Synthetic estimates for 303 Primary Care Organisations in England were calculated by aggregating the model based estimates for the component wards weighting the contribution of each ward in proportion to its population size derived from the Census 2001 counts The corresponding confidence intervals for the PCO estimates were generated using a similar method as for wards see Pickering et al 2005 3 4 Validating the models The methodology used for this project was used previously by the Office for National Statistics to produce ward level income estimates
21. egetable consumed on the previous day These measures were combined to give the total number of portions of fruit and vegetable consumed 11 National Centre for Social Research Note that information about fruit or vegetable consumption was not collected in the HSfE 2000 nor for children under 5 years old in the HSfE 2001 and 2002 Of the 8 438 children aged 5 to 15 years in the 2001 and 2002 HSfEs 3 163 37 had consumed three or more portions of fruit and vegetables The healthy lifestyle measure was whether the child had consumed three or more portions or 2 1 4 Fruit and vegetable consumption adults The healthy lifestyle measure for fruit and vegetable consumption for adults aged 16 years or more was generated from the data collected in the 2001 and 2002 HSfEs about the quantities of different types of fruit and vegetable consumed on the previous day These measures were combined to give the total number of portions of fruit and vegetable consumed Of the 23 039 adults in the 2001 and 2002 HSfEs 5 460 24 had consumed five or more portions of fruit and vegetables The healthy lifestyle measure was whether an adult respondent had consumed five or more portions or not 2 1 5 Binge drinking The healthy lifestyle measure for binge drinking was generated from the data collected in the HSfE about the quantities of all the different types of alcoholic drinks beer wine spirits sherry and alcopops consumed on a respondent
22. els of healthy lifestyles Area level characteristics Source Local Authority District level oooh ee oe eee Compendium of Clinical and Outcome Indicators 2003 Deprivation scores ID 2004 Office of the Deputy Prime Minister 2004 Ward level Key Statistics amp Standard Tables Census 2001 All cause Standardised Mortality Ratios Office for National Statistics Area type classification Office for National Statistics 2004 Deprivation scores ID 2004 derived Office of the Deputy Prime Minister 2004 Department of Work and Pensions 200 Rural non rural indicator Department of the Environment Food and Rural Proportionate distribution of properties in Valuation Office Agency 2001 the council tax bands 3 2 Deriving the ward estimates The process of generating model based estimates of healthy lifestyle behaviours involved two main stages using a Statistical model to represent as well as possible the relationships between health behaviours and area level characteristics and applying that model to calculate prevalence estimates for all wards in England In the case of smoking the first stage involved finding the best model to describe the relationship between whether an adult respondent to the HSfE currently smoked or not and the characteristics of the area in which the person lived Different area level characteristics were associated with different health behaviours The results from the modelling proc
23. ement committees This consultation enabled NatCen to invite users to comment upon the plausibility and 10 The adult population counts were used for the adult health behaviours and the 5 15 year old counts were used for children s fruit and vegetable consumption 17 National Centre for Social Research usefulness of the estimates The comments received informed the approach we have used and generally supported the plausibility of the estimates 18 National Centre for Social Research REFERENCES Bajekal M Scholes S Pickering K and Purdon S 2004 Synthetic estimation of healthy lifestyle indicators Stage 1 report http www natcen ac uk Bird S Cox D Farewell V Goldstein H Holt T and Smith P 2003 Performance Indicators Good Bad and Ugly Royal Statistical Society Working Paper on Performance Monitoring in the Public Service Goldstein H 2003 Multilevel Statistical Models London Arnold Heady P Clarke P and others 2003 Model based small area estimation series No 2 Small Area Estimation Project Report ONS Health Development Agency 2004 The Smoking Epidemic in England Longhurst J Cruddas M Goldring S and Mitchell B 2004 Model Based Estimates of Income for Wards 1998 99 Technical Report ONS ONS 2004 Model Based Estimates of Income for Wards in England and Wales 1998 99 User Guide http neighbourhood statistics gov uk information income_estimates pdf ONS 2004b Guidance for the P
24. female 1 male potentially related to smoking The estimates from this model are shown in Table B 1 Table B 1 A logistic regression model of current smoking using the combined 2000 to 2002 HSfEs Variable Odds ratio 95 confidence P value interval for odds ratio Sex Females 1 00 baseline Males 1 08 1 03 1 14 The estimates from the model give a measure of the effect of sex on current smoking status For ease of interpretation the estimates are presented as odds ratios The odds of an outcome is the ratio of the probability of its occurring to the probability of its not occurring e g if the probability of being a smoker is estimated to be 0 8 then the probability of not being a current smoker is 1 0 0 8 0 2 and so the odds of being a smoker equal 0 8 0 2 4 In this case females are selected as the baseline or reference category with males being compared to them There is no estimate therefore for females and the odds ratio defined for males represents the ratio of the odds of being a current smoker for males to those for females As Table B 1 shows compared to females the odds for a male being a current smoker are estimated to be 8 higher than those for females Table B 1 also shows the 95 confidence interval for the odds ratio In logistic regression a 95 confidence interval which does not include 1 0 indicates the given estimate is statistically significant As the confidence interval attached to the odds
25. ged 85 hovercr South East region hovercr South East region aarate South West region 11 Based on their deprivation score wards were grouped into one of 10 roughly equal sized bands where group 1 imd1 represented the least deprived wards up to group 10 imd10 indicating the most deprived 20 National Centre for Social Research Table A 2 Area characteristics associated with fruit and vegetable consumption adults and children Fruit amp vegetable consumption Fruit amp vegetable consumption adults children icobnuk icobnuk South East region lemale smr_10a ipermsic Yorks amp The Humber region Yorks amp The Humber region 274 most deprived band of wards imd9 West Midlands region Built up areas iolevel Proportion female aged 25 34 Proportion female aged 16 19 iupdcr50 smr_14b ipermsic Yorks amp The Humber region isroutin empscore South West region iupdcr50 icobnuk London region smr_10a South East region isroutin Yorks amp The Humber region Notes to Tables A 1 and A 2 e The terms highlighted in bold had positive coefficients in the model that is they were associated with an increased propensity for a person to be a current smoker obese indulge in binge drinking or consume more than the threshold portions of fruit and vegetables e The terms highlighted in italics had negative coefficients that is they were associated with decreased propensity for a person
26. gly 1 4 Generating synthetic estimates a model based approach A model based approach to produce estimates of healthy lifestyle behaviours was used because the sample size of national surveys is too small at ward level to provide reliable estimates Most national surveys are designed to provide a large enough sample to calculate national or regional estimates To ensure that the national sample is representative of different types of people and areas in the country a relatively small number of areas are selected at random from across the country As a result many small areas such as electoral wards either contain no respondents as they were not covered by the survey or too few respondents to calculate reliable estimates The model based method used to produce the ward level estimates combined two sets of information First the HSfE provided health behaviour data e g whether a respondent currently smoked or not Second the 2001 Census and other 1 Reports can be found on the NatCen website www natcen ac uk along with a project summary National Centre for Social Research administrative data sources provided information about the characteristics of the area in which respondents lived A statistical model was used to examine the relationships between the healthy lifestyle behaviours and area characteristics As part of the modelling process for example we examined whether the propensity for a person to be a current smoker varied s
27. haracteristics in the model not because we considered there to be a direct relationship between them Hence the models should not be interpreted by users as explanatory models of health behaviour Tables A 1 and A 2 show the significant area characteristics associated with the health behaviours The definitions of the area characteristics are shown in Tables A 3 to A 6 In the case of smoking for example being located in the North West region was associated with increased propensity for a person to be a current smoker In contrast being located in the South West region was associated with decreased propensity for a person to smoke for more details see Pickering et al 2005 Table A 1 drinking Current smoking Proportion female aged 25 34 illsiwrk 3 4 most deprived band of wards imd8 11 North West region eduscore imd5 icouple iethnic iprofman aarate South West region aarate North West region iethnic South West region imd8 South West region Obesity iolevel isroutin South West region laidscor East of England region propctxg hloamnty rural isroutin laidscor East of England region Area characteristics associated with smoking obesity and binge Binge drinking North Fast region North West region Yorks amp The Humber region Proportion male aged 45 49 icouple iethnic israte Proportion male aged 75 79 South West region Proportion female aged 50 54 Proportion female a
28. ignificantly between regions or between wards with varying proportions of residents who were living as a couple claiming Income Support had a limiting longstanding illness etc The final model was then used to calculate the prevalence estimate of current smoking for all wards and PCOs in England The model based approach generates estimates that are of a different nature from standard survey estimates because they are dependent upon how well the relationship between healthy lifestyle behaviours for individuals and the Census administrative information about the area in which they live is specified Section 3 of this User Guide provides a brief non technical overview of the methodology used For a fuller technical description of the methodology users are referred to the Stage 3 report Pickering et al 2005 1 5 Limitations of the estimates The estimates resulting from this project must be used with caution Synthetic estimates are difficult to interpret because they are model based Although robust they will almost certainly not mirror precisely any available measures from local studies or surveys although research by NatCen and others have shown that they tend to be related In this section we discuss a number of limitations that users must bear in mind when using the data 1 5 1 Synthetic estimates and performance monitoring Ward or PCO level estimates based exclusively on sample respondents located within the area itself are easy to
29. interpret They represent an estimate of the real prevalence of health behaviours such as current smoking for the area in question Synthetic estimates however are more difficult to interpret This is because the synthetic estimate for a particular ward is a model based estimate and the model that we use estimates the underlying expected value of smoking prevalence for any ward given the social and demographic characteristics of its population They are not therefore estimates of the actual prevalence for wards or PCOs To interpret the estimates it is recommended that users adopt statements such as given the characteristics of the local population we would expect approximately x of adults within ward X to smoke be obese etc Health Development Agency 2004 As the synthetic estimates do not measure actual prevalence within small areas we do not encourage any ranking of wards within their PCO or Strategic Health Authority SHA The large margin of error around any such ranking would also render such an exercise to be meaningless for more details on ranking see Section 1 6 National Centre for Social Research It is important that users note that the estimates do not take account of any additional local factors that may impact on the true prevalence rate e g local initiatives designed to reduce smoking obesity or binge drinking The estimates therefore cannot be used to monitor performance or change over time 1 5 2 Confidence interv
30. les in each year were used for the adult health behaviour measures The boost sample of children in 2002 was included however for children s fruit and vegetable consumption 14 National Centre for Social Research The HSfE data were supplied at the individual level with the postcode of the respondent attached The February 2004 release of the All Fields Postcode Directory was used to allocate these postcodes to 2003 Census Area Statistics CAS ward boundaries Local Authority Districts and Government Office Region Although CAS wards the principal estimation area chosen for this project nest within higher level administrative tiers such as Local Authority Districts and Government Office Regions they do not nest perfectly into larger health areas such as PCOs NatCen was provided a best fit one to one look up table to uniquely attribute whole wards to a PCO Table 3 1 summarises the number of survey observations used to calculate the estimates In the case of current smoking for example 30 872 adult respondents to the 2000 to 2002 health surveys covered 3 231 of the 7 958 CAS wards in England The average number of respondents per ward was 10 although 225 wards only contained 1 respondent Table 3 1 Descriptive statistics for the surveyed HSfE wards and PCOs Health behaviour Number of Number of Number of Average Maximum measure HSfE wards wards number of number in respondents covered containing respondents any sampled only
31. luding those available on the NeSS website They are identical to the 2003 Statistical Wards except that 18 of the smallest wards have been merged into other wards to avoid the confidentiality risks of releasing data for very small areas This has occurred to those wards with fewer than 100 residents or 40 households as at the 2001 Census There are a total of 7 969 CAS wards in England For the purposes of this project we have combined together the nine wards in the City of London and four in the Isles of Scilly into one unit respectively to form 7 958 CAS wards This classification of wards mirrors that used by the Department of Work and Pensions for publishing ward level claimant counts 12 National Centre for Social Research Complex methods were used to derive confidence intervals for the synthetic estimates For a fuller technical description of the methodology users are referred to the Stage 3 report Pickering et al 2005 23 Data files Separate excel workbooks have been produced for each healthy lifestyle measure each workbook containing a separate sheet for wards and PCOs The survey based national estimate for England and its accompanying confidence interval is provided at the top of the sheet for reference The variable names and labels in each worksheet are shown in Table 2 1 Table 2 1 Variable names and labels in the data files Column Column label Lower 95 confidence interval limit Upper 95 confidence inter
32. nvolved using a statistical model to represent as well as possible the relationships between individual health behaviours and area level characteristics and applying that model to calculate prevalence estimates for all wards In this section we illustrate how this works in practice by using the example of current smoking for a ward Little Lever in the North West region of England In the following section we take the process a step further and illustrate how the ward estimates can be used to compute estimates for Primary Care Organisations Stage 1 Fitting the relationship between smoking and area level characteristics Using the combined 2000 to 2002 Health Survey for England data we first identified those area level characteristics most strongly related to whether an individual currently smoked or not As described in Section 3 1 only 3 231 of the 7 958 wards in England were represented in this analysis e g Little Lever was not covered by the HSfE whilst a number of neighbouring wards happened to be The area level characteristics in the optimal model for whether adults aged 16 years or more currently smoked in the HSfE 2000 2002 are shown in Table B 2 26 National Centre for Social Research Table B 2 Estimates for smoking model Area level characteristic Estimate log odds scale Main effects only Proportion 16 residing as couple icouple 2 158 Proportion female aged 25 34 5 108 Proportion non white ie
33. p 2 2 Proportion 16 74 in semi routine amp routine occupations NS SEC 6 amp 7 Proportion unpaid carers caring gt 50 hours per week Ward characteristics Administrative data Variable name Source Description aarate dlarate israte propctxb propctxg DWP benefits Attendance allowance claimant rate data Aug 2001 i Disability living allowance claimant rate Income Support claimant rate Valuation Proportion of dwellings in council tax band B Office Agency data Mar 2001 Proportion of dwellings in council tax 12 National Statistics Socio Economic Classification categories 22 National Centre for Social Research Table A 5 Other ward characteristics Variable name Description Built up areas Classification of wards Office for National Statistics 2004 Derived from Ward located in 5 most deprived Index of Multiple Output Area Deprivation IMD band scores produced by the Office of the Deputy Prime Minister 2004 imd8 Ward located in 3 most deprived Index of Multiple Deprivation IMD band imd9 Ward located in 259 most deprived Index of Multiple Deprivation IMD band eduscore Education skills and training score houscore Barriers to housing and services score Table A 6 Local Authority District characteristics Variable name Source Description laidscor Office of the Index of Multiple Deprivation score Deputy Prime Minister 2004 lemale Office for Life
34. pecific health conditions that are repeated at regular intervals Questions relating to smoking and drinking have appeared in each year of the survey 1994 to 2003 Height and weight measurements have also been taken each year A new module of questions relating to fruit and vegetable consumption was introduced in 2001 and has appeared every year since For the purposes of this study three years of HSfE data 2000 2001 and 2002 were merged together to form a combined survey dataset of health behaviour data The reasons for selecting these particular years were that they included the most up to date HSfE information available and that the years were symmetrically arranged either side of 2001 the year the last Census was carried out Each year the HSfE covers a representative sample of people resident in households and in addition in certain years particular population groups are over sampled or boosted In 2000 a separate sample of older people aged 65 and over resident in care homes was included In 2002 a separate sample of infants and children aged 0 15 young adults aged 16 24 and mothers with infants aged less than 1 was undertaken Typically the annual sample size of the general population is about 16 000 adults aged 16 and over and 4 000 children aged 0 15 In years when special populations are boosted the general population sample is halved to about 8 000 as was the case in 2000 and 2002 Only the general population samp
35. pulation They are not therefore estimates of the actual prevalence for wards or PCOs e The large confidence interval around the estimates meant that wards could not be ranked within PCOs or Strategic Health Authorities SHAs or nationally the margin of error around such rankings would render such an exercise to be meaningless e Itis important that users note that the estimates do not take account of any additional local factors that may impact on the true prevalence rate The estimates therefore cannot be used to monitor performance or change over time e The methodology used does not enable separate estimates for specific population sub groups to be produced within each ward or PCO e The estimates could be used in a number of appropriate ways For example the data could be used to identify those wards or PCOs which had an expected prevalence of health behaviours that was significantly higher or lower than England as a whole Wards having an expected prevalence that was significantly higher or lower than the model based estimate for their PCO could also be identified National Centre for Social Research The large width of the confidence intervals attached to the ward level estimates means that it would not be possible to state that the expected prevalence in one ward was higher than that in another with any degree of statistical confidence There is more scope however for using the estimates to discriminate between PCOs The
36. r SHA and Bolton PCO In addition over 60 of household residents in Little Lever live as part of a couple and 8 5 of residents are female aged between 25 and 34 based on Census 2001 data Almost 17 of residents claimed Attendance Allowance in 2001 based on DWP claimant counts Table B 3 Known area level values for the Little Lever ward Ward level characteristic variable Value for Little Lever WARDcode OOBLFS WARDname Little Lever GORcode B GORname North West SHAcode Q14 SHAname Greater Manchester PCOcode SHQ PCOname Bolton Proportion 16 residing as couple icouple 0 633 Proportion female aged 25 34 0 085 Attendance allowance claimant rate aarate 0 167 Although no respondents to the combined 2000 to 2002 HSfEs resided in the Little Lever ward synthetic estimation works by assuming that the relationship between individual smoking behaviour and the area level characteristics found for the 3 231 surveyed wards applies nationally to all wards Using two sets of information the estimates from the fitted model and the known area level values for all wards the following formula can be used to compute a synthetic estimate of smoking prevalence where Y denotes the expected smoking prevalence for the Little Lever ward given its local population characteristics exp represents the exponential function and X denotes the relevant area characteristics taken from the Census and other 28 National Cent
37. re for Social Research administrative data sources is the estimate of the intercept and B the parameter estimates shown in Table B 2 Note that we have only shown the first icouple and last gor_sw main effect terms from the smoking model for purposes of demonstration s The full formula contains all the terms in Table B 2 From the estimates shown in Table 2 0 082 B 2 158 and icouple B eor_sw 20 137 From the known area level values for the Little Lever ward shown 0 633 and X 0 as this ward is in the North West icouple gor _sw in Table B 3 X region Putting all the terms together the formula becomes 1 0 082 2 158 x 0 633 0 137x0 After inserting the model estimates and the known area level values into this formula we obtain a value of 0 2804 which can be multiplied by 100 to give a model based estimate of 28 Users are recommended to interpret this result by adopting a statement such as given the characteristics of its local population we would expect a current smoking prevalence of approximately 28 within the Little Lever ward Health Development Agency 2004 Using ward level estimates to compute PCO estimates Having computed synthetic estimates for all wards a further output for this project involved combining the ward level estimates to estimate the prevalence of healthy lifestyle behaviours for all 303 PCOs as at 2003 in England We illustrate how this can be
38. resentation of Government Statistics for Health Areas at Regional Health Authority Health Board and Primary Care Levels http www statistics gov uk geography health_areas asp Pickering K Scholes S and Bajekal M 2004 Synthetic estimation of healthy lifestyle indicators Stage 2 report http www natcen ac uk Pickering K Scholes S and Bajekal M 2005 Synthetic estimation of healthy lifestyle indicators Stage 3 report http www natcen ac uk Twigg L Moon G and Jones K 2000 Predicting small area health related behaviour a comparison of smoking and drinking indicators Social Science and Medicine 50 7 8 1109 20 19 National Centre for Social Research APPENDIX A AREA CHARACTERISTICS ASSOCIATED WITH THE HEALTHY LIFESTYLE MEASURES The model based approach we have used to calculate the ward level estimates was based on finding a relationship between individual health behaviour measures and Census administrative information about the areas in which people lived This relationship expressed in a statistical model was then used to calculate the prevalence estimates for all wards in England The ward level estimates were then used to calculate estimates for all PCOs In this section we present the five optimal models of healthy lifestyles used to calculate the estimates Note that each item was retained in the model because it had a significant association with the health behaviour measure allowing for the other area c
39. rovide more details on the derivation of these healthy lifestyle indicators using the Health Survey for England 2 1 1 Current smoking The healthy lifestyle indicator for current smoking was generated from the HSfE measure of current smoking status Adult respondents aged 16 years or more to the HSfE were defined to be current smokers if they reported that they were a current cigarette smoker and not a current smoker if they reported that they had never smoked cigarettes at all used to smoke cigarettes occasionally or used to smoke cigarettes regularly Of the 30 872 adults from the combined HSfEs from 2000 to 2002 7 972 26 reported that they were current smokers 2 1 2 Obesity The healthy lifestyle indicator for obesity was generated from the height and weight of adult respondents aged 16 years or more as measured by the HSfE interviewers The Body Mass Index BMI was derived from the height and weight as the weight in kilograms divided by the square of the height in meters Respondents were defined to be obese if their BMI measure was more than 30 Of the 27 120 adults from the combined HSfEs from 2000 to 2002 5 991 22 were obese 2 1 3 Fruit and vegetable consumption children The healthy lifestyle indicator for fruit and vegetable consumption for children aged from 5 to 15 years inclusive was generated from the data collected in the HSfE about the quantities of different types of fruit and v
40. rveys for example belonged to only 40 of the wards in England Second for those wards containing survey respondents the achieved sample size will usually be small and the estimates will thus have low precision This low precision will be reflected in rather wide confidence intervals for the survey estimates Other more complex techniques are therefore needed to generate precise ward level estimates Synthetic estimation describes the several different ways in which more precise ward level estimates might be constructed The key idea is that to produce prevalence estimates of healthy lifestyle behaviours such as current smoking for a particular ward with an adequate level of precision it is necessary to use a technique that takes advantage of information on smoking from wards other than itself This information is brought into the estimation process through a statistical model 1 3 Healthy lifestyle behaviours The National Centre for Social Research NatCen was commissioned by the Department of Health to produce ward level estimates of five healthy lifestyle behaviours using HSfE data The project involved three main stages National Centre for Social Research scoping and feasibility review of existing approaches to synthetic estimation to assess the various options and identification of the data requirements Stage 1 Bajekal et al 2004 testing and validation of selected alternate methods of synthetic estimation Stag
41. s for the 5 healthy lifestyle behaviours wards and PCOs Health behaviour Average width of the 95 confidence interval ee ara Eu Fruit and vegetable consumption adults Fruit and vegetable consumption children The width of the confidence intervals particularly at ward level represent a further limitation on using the estimates As will be discussed in Section 1 7 2 one potential use of the estimates is to discriminate between wards or PCOs by looking at 2 For a ward whose prevalence estimate of current smoking was 30 a confidence interval of width 11 would correspond to a range of 19 to 41 Similarly for a PCO whose estimate was 30 a confidence interval of width 3 would correspond to a range of 27 to 33 National Centre for Social Research overlapping confidence intervals When comparing two model based estimates one ward may only be said to have a significantly higher or lower prevalence estimate than another if the confidence intervals for the two wards do not overlap The average width of the confidence intervals implies however that there would be little scope for discriminating between wards As an extreme example in the case of current smoking only 16 of the 7 958 wards were significantly different from the ward having the average estimate Therefore it is very important that the ward level estimates are used with great caution by users in the vast majority of cases it would not be pos
42. s gt NatCen National Centre for Social Research Synthetic estimation of healthy lifestyles indicators User guide Shaun Scholes Madhavi Bajekal Kevin Pickering Synthetic estimation of healthy lifestyles indicators User guide Shaun Scholes Madhavi Bajekal Kevin Pickering Prepared for the Department of Health January 2005 Contents EXECUTIVE SUMMAR Yssa 1 1 BACKGROUND AND GUIDANCE ON 3 ecole 3 1 2 __ fOr small 3 1 3 Healthy lifestyle Gehia viOuUrs 3 1 4 Generating synthetic estimates a model based 4 1 5 Limitations OF the 5 5 151 Synthetic estimates and performance 5 15 2 Confidence intervals 6 1 5 3 Geographical 1 7 1 5 4 TMM SSS ssiessescliswisisiesssesivbasissiennedsterastins beevecssethsnlddetecsssveuicsigsecstai dies 7 1 5 5 Estimates for subgroups within small 7 1 6 Banding ranking of 7 1 7 Examples GF dat dees tees 8 171 Comparing areas with the national
43. sible to state that the prevalence in one ward was higher than that in another with any degree of statistical confidence 1 5 3 Geographical boundaries The ward estimates have been produced on 2003 Census Area Statistics CAS ward boundaries the standard set of boundaries used for Neighbourhood Statistics and therefore cannot be translated onto any other boundary system Users must be aware of this when using the estimates in any application or drawing conclusions from the data It is inadvisable for example to aggregate the ward estimates to compute a Local Authority District estimate in the absence of any published confidence intervals for that higher level of geography 1 5 4 Timeliness We stress that recombining estimates to new boundaries as they change over time will not be feasible The estimates are also based on specific years of survey data 2000 to 2002 for smoking obesity and binge drinking and 2001 to 2002 for fruit and vegetable consumption and so are only valid for these time periods 1 5 5 Estimates for subgroups within small areas The methodology used to produce the estimates does not support the production of separate estimates for specific population sub groups within each ward or PCO For example the estimate of current smoking prevalence represents the underlying expected value for the demographic and social mix of adults aged 16 years or more living in a ward at the time of the 2001 Census It cannot therefore
44. thnic 0 914 Proportion professional amp managerial occupations aged 1 324 16 74 iprofman Proportion of working age with limiting longstanding 3 860 illness illsiwrk Attendance allowance claimant rate aarate 1 637 3rd most deprived band of wards imd8 0 138 North West GOR gor_nw 0 533 IMD education skills and training score eduscore 0 006 5th most deprived band of wards imd5 0 119 South West GOR gor_sw 0 137 Interactions aarate gor_nw 3 583 iethnic gor_sw 3 573 imd8 gor_sw 0 376 Intercept 0 082 A number of conclusions can be drawn from this model of smoking The ward level characteristics associated with increased propensity for a person to be a current smoker i e having positive estimates were a higher proportion of females aged 25 34 a higher proportion of residents of working age who had a limiting longstanding illness being in the 3rd or 5 most deprived band of wards out of a possible 10 being located in the North West region and a relatively higher education skills and training deprivation score The ward level characteristics associated with decreased propensity for a person to be a current smoker i e having negative estimates were a higher proportion of household residents over the age of 16 who were living as a couple a higher proportion of non white residents a higher proportion of residents who were classified as being in managerial and professional occupations a relatively higher
45. tion size derived from the Census 2001 counts Hence to compute a weighted average for the Bolton PCO the following formula is applied a Censusadultcount a 1 BoltonPCT gt ies ward Census adult count potion pcr where denotes the expected smoking prevalence for the Bolton PCO and the symbol gt indicates a summation over the 20 wards nested within it Y a represents the expected smoking prevalence for each ward in question Wi Applying this formula for the Bolton PCO results in the following summation over 20 wards 14 The ratio of the Census adult count in the ward to the Census adult count for the PCO as a whole provides the weight for the estimate Such a weight ensures that the larger wards within a PCO provide a larger contribution to the overall PCO estimate than smaller wards 30 National Centre for Social Research pope be e sana gis 201 917 201 917 201 917 201 917 10 924 lo 9075 2227 lo 2490 8 070 0 2918 284 0 2923 201917 201917 201 917 201 917 15263 2752 02498 7017_ baia 222 0 2564 201 917 201 917 201 917 201 917 10234 p 2936 12378 02253 12106 10747 p 2068 201917 201917 201917 201917 2333 p ogo4 8 47 02378 7918 0 2933 2 429 0 2338 201917 201917 201 917 201917 Looking at the first term 11 067 represents the Census adult count for the
46. uld expect approximately 49 of adults to be current smokers The remaining rows give a context to the estimates by providing area level information about these wards taken from the 2001 Census and other administrative data sources Ward A belonged to the 10 of wards having the highest overall deprivation score high scores indicating the most deprived wards Ward B belonged to the lower 20 30 group of wards low scores indicating the least deprived wards Compared to a national average of 5 over 20 of adults in ward A were claiming Income Support in 2001 None of the properties in ward A were in council tax band H properties worth more than 320 000 Finally whilst ward A could be described as an area of traditional manufacturing ward B belonged to the suburbs and small towns category 10 National Centre for Social Research 2 ESTIMATES This chapter describes the estimates produced by the project It is important that users note that the methodology used to produce the five sets of estimates is relatively new and as a result may be subject to consultation modification and further development In view of this ongoing work these estimates are being published as experimental statistics 2 1 The healthy lifestyle indicators The five sets of estimates published by the project are for current smoking obesity and binge drinking for adults and fruit and vegetable consumption for children and adults separately In this section we p
47. val limit In accordance with the Guidance for the Presentation of Government Statistics for Health Areas ONS 2004b the ward estimates are presented in the following nested order wards within reporting PCO within SHA within GOR Wards are listed in CAS ward code order and PCOs and SHAs in alphabetical order within GORs These tables are published on the accompanying electronic files on the NeSS website 13 National Centre for Social Research 3 GUIDE TO THE METHODOLOGY This chapter provides a brief non technical description of the methodology used for producing model based estimates of healthy lifestyle behaviours for all wards and PCOs in England A full description of the methodology can be found in the Synthetic Estimation of Healthy Lifestyles Indicators Stage 3 report Pickering et al 2005 3 1 Datasets used 3 1 1 The survey dataset The Health Survey for England HSfE comprises a series of annual surveys All surveys have covered the adult population aged 16 and over living in private households in England The HSfE series is part of an overall programme of surveys commissioned by the Department of Health and designed to provide regular information on various aspects of the nation s health Each survey in the series consists of core questions and measurements for example anthropometric and blood pressure measurements and analysis of blood and saliva samples which are included each year plus modules of questions on s
Download Pdf Manuals
Related Search