The Caerphilly Health and Social Needs Survey

The survey was granted ethical approval by Gwent Local Research Ethics Committee and informed consent was given by participants. The sampling frame was the adult resident population of 132,613 residents aged 18 and over of Caerphilly county borough as recorded on the former Gwent Health Authority 'Exeter' GP administrative register on 31 May 2001. We excluded patients registered with a Caerphilly borough GP who were resident in a neighbouring borough. We took a random sample, stratified by electoral ward, in order to estimate ward prevalences to within ± 5% with 95% confidence, allowing for the finite population correction factor. This corrects for the potentially sizeable reductions in the standard error of an estimate when the sampling fraction is large and the sampling is without replacement [12].

There were 36 wards defined in Caerphilly borough at the 1991 census which were unchanged at the 1998 boundary revision and these ward boundaries were used in the study. The UK electoral ward is the administrative geographical unit defined to elect politicians to local government. In Caerphilly borough, the mean ward adult population was 3684 at the time of the study. The mean required sample size was 350 in each of the 36 wards in the borough, giving a target of 12,600 responses. We aimed to achieve a 60% response and so the number of people sampled in each ward was increased to give a total of 22,290.

In order to try and maximise response, we timed the survey to link in with the electoral registration process in the borough during which electoral register canvassers call on each property to collect the completed electoral registration forms. We held standardised study briefing sessions for the canvassers who received an information pack containing a Frequently Asked Question sheet, the study sample for their patch and contact record sheet, spare questionnaires and envelopes, and 'sorry you were out' slips. The canvassers visited on up to three separate occasions to collect completed questionnaires in a sealed envelope and they noted on the contact sheet whether the subject had moved away or had withheld consent. The canvassers again visited each subject up to three times if necessary to complete the contact sheet. The third and final wave of the survey was completed using a postal return. The figure shows a flow chart of the survey.

We designed the questionnaire to include questions on: "your lifestyle", including age, gender, smoking, alcohol consumption, diet, physical activity, height and weight; "your health", including limiting long-term illness, chronic diseases, and the Short Form SF-36 version 2 health status questionnaire [13]; "your job" including occupational status, used to derive the eight-category NS-SEC classification [14, 15]; and, "your home", including tenure. Where possible we included validated questions that had previously been used in the Welsh Health Survey 1998 [16]. The SF-36 is a validated and reliable instrument to measure health status and has been used widely in population surveys [13, 17]. We used the SF-36 Version 2 since it has been shown to have greater reliability than SF-36 version 1 [18].

The Council Tax and Benefits division of Caerphilly county borough council supplied an electronic extract of property address and CTVB from the council tax register. The statutory requirement to share information between local authorities and the NHS in Wales, set out in the Health, Social Care and Well-being Strategies (Wales) Regulations 2003 of the NHS Reform and Health Professions Act 2000 [19], led to the adoption of a Gwent Information Sharing Protocol signed up to by all NHS organisations and the five local authorities in Gwent, including Caerphilly county borough council. This study operated within this data sharing framework. Using Microsoft Access we matched the sampling frame dataset to the CTVB extract by address to assign the property CTVB to each resident in the sample frame. Using a Geographical Information System, Mapinfo version 6.5, the postcode of each resident was linked using the postcode to the ward of residence.

Dataset coding

We coded the NS-SEC based on current or most recent occupation. Class was allocated based on responses to questions on occupation, employment status (employer, employee or self-employed), whether a supervisor or a manager and the number of employees in the workplace [14, 15]. The Townsend index of deprivation is a widely used and robust measure of small area social and material deprivation, calculated from four census variables: unemployment, car ownership, owner occupation and overcrowding [20, 21]. We calculated the ward Townsend Index using the standard method [21] and created a five-level categorisation with cut-points at the 20, 40, 60 and 80^th centiles of the distribution of ward scores. Using postcode linkage, we assigned each respondent into one of these five categories of ward Townsend score.

Health and lifestyle outcomes

We defined and recoded a subset of the response dataset to use in the analysis and address the aims of this paper. We defined three dichotomous lifestyle variables: smoking (daily or occasionally versus ex-smoker or never smoked), poor diet (eating fruit and vegetables at most once a week versus eating more than once a week) and obesity (body mass index 30 and over). We defined one morbidity variable: limiting long-term illness (do you have any longstanding illness, health problem or disability, which limits your daily activities or the work you can do?); and two health status variables, the physical health (PCS) and mental health (MCS) component summary scores from the SF-36 version 2. We coded the raw scale scores into MCS and PCS scores using the standard algorithm [13]. Lower scores indicate worse health status.

Statistical analysis

All analyses were carried out using SPSS version 12.0. Since the highest CTVB property value categories E, F, G & H contained small numbers of respondents, we aggregated CTVB categories E to H for the analysis. We calculated descriptive statistics for age, gender, social class and Townsend score for each CTVB category and calculated the proportions of respondents reporting each lifestyle and morbidity outcome for each category of CTVB, NS-SEC, and Townsend fifth. We decided a priori that age and gender could be important confounders and taking CTVB category E-H, the NS-SEC higher managerial and professional category and the least deprived Townsend fifth as the reference categories respectively, we used logistic regression to model crude and adjusted odds ratios for the lifestyle and morbidity outcomes. We calculated mean (standard deviation, SD) MCS and PCS scores for each category of the three variables and then mean MCS and PCS scores adjusted for age as a continuous variable and gender using the general linear model procedure in SPSS.

Main results

Overall, our results suggest that the household CTVB assigned to each individual is a useful proxy for socio-economic status. Residents of lower value properties were more likely to be in the lower NS-SEC categories and live in more deprived wards. We found a clear trend of association between CTVBs and each of the lifestyle and health outcomes that we investigated. The trends were similar for CTVBs and NS-SEC and steeper than for the Townsend score, suggesting the CTVB is a satisfactory classification of socio-economic status in relation to the outcomes assessed in this study.

Classification of socio-economic status

The theory of the classification of socio-economic status has been widely debated [22, 23]. Different domains of socio-economic status, such as income, education and occupation, relate to different pathways between socio-economic status and health and each classification should be theoretically based and capable of being measured with readily available valid and reliable data [23]. The three methods of classifying socio-economic status used in this study all have some drawbacks. Although the new NS-SEC is based on an underlying theory of occupational stratification, the classification is less satisfactory for women than men [24] and it has problems classifying people not in the labour market, such as the permanently sick or disabled [15]. Thus although the classification is intended for people who are economically active, the eighth category of 'never worked and long-term unemployed' may include people who are economically inactive. This may explain the particularly large odds ratio for limiting long-term illness and the low MCS and PCS scores for the never worked and long-term unemployed, as these poor outcomes are associated with economic inactivity [25, 26]. The NS-SEC was not designed as an ordinal scale, although studies investigating NS-SEC and health outcomes have found evidence for significant trends across the classification [24, 25, 27]. Analyses reporting associations between area-based deprivation scores and individual health outcome are prone to the ecological fallacy [6], and the use of area methods is less justifiable where individual measures of socio-economic status are available [28].

CTVBs are defined in a walk-past assessment of the house and represent a relative scale of property value within an area, but the absolute value depends on the estimated amount the property might have been sold for on the open market, making the classification locally context-specific. As a household measure, assigning the CTVB classification to individuals may be prone to the ecological fallacy, but the household unit of aggregation is small and recent studies have demonstrated the importance of the homogeneity at household level in explaining variation in health status [29, 30].

Using CTVBs as a measure of socio-economic status is most likely to misclassify students and elderly people on low incomes who live alone in large old houses, or who move in with relatives in a more affluent household. The elderly living alone qualify for a 25% rebate, which effectively lowers their CTVB by one letter, but the implications of this re-categorisation for studies of health outcome has not yet been investigated. Further research is required to distinguish between the CTVB as a measure of income or wealth, since the value of a property may only be weakly associated with income or wealth, particularly in the rented sector.

A further disadvantage of using CTVBs is the differences in the systems between England, Wales and Scotland, including different time intervals between revaluations, and the changes that might result from the recent political debate on the funding and functions of local government. In England, the government has established an independent Inquiry into local government finance led by Sir Michael Lyons, and no decision on the postponed revaluation exercise in England will be taken until the Inquiry reports in late 2006 [31]. In Wales, however, CTVBs were revalued and rebanded in 2005, within a system of bandwidths which aimed to keep 59% of properties in Caerphilly borough in the same band, 25% to move down, 14% to move up one band and 2% to move up two or more bands [32]. Scotland has no plans to revalue CTVBs in the foreseeable future [33]. Although the valuation bands vary between England, Wales and Scotland, the relative position of properties might remain comparable. This needs to be assessed in further research.

CTVBs have one particular advantage over social class when used as a marker of socio-economic status in population surveys. Since the CTVB is known for both the sampling frame and respondents, this allows an assessment of non-response bias to be made by comparing the CTVB distribution between the sample and responders. Social class is only known for responders. Assessment of non-response can also be made by area deprivation score, but will suffer from a greater degree of ecological bias [6].

Comparison with previous studies

Three previous studies have investigated associations between CTVBs and a population health or lifestyle outcome [5, 9, 10]. Based on an analysis of 856 deaths occurring in one general medical practice population in England over a nine year study period, standardised death rates were shown to be significantly higher than average in bands A and B, and significantly lower than average in bands C to H [9]. We did not find consistent evidence of such a threshold effect, with significant trends in all outcomes investigated in this study across the categories of CTVB. A study of 1390 mothers sampled from the ALSPAC study found a significant trend in the risk of smoking cigarettes in pregnancy and up to eight weeks after delivery, and in breast feeding rates at up to four weeks with nearly twice as many mothers breastfeeding in CTVB D to H (57%) as in CTVB A (31%) [5]. A study to ascertain whether CTVBs are associated with household smoking rates found that 55% of CTVB A households included at least one smoker, compared to 22% in CTVB E and above [10]. Our current study corroborates the findings from this research group and extends the range of outcomes assessed to include other lifestyle factors, limiting long-term illness and physical and mental health status.

Methodological issues

One possible limitation of the comparisons made in this study is that each category of the three different measures was based on different proportions of the underlying population and hence the comparisons were not 'like for like'. However the NS-SEC and CTVB categories were fixed and used pragmatically in the form that they are available, although some merging was required to avoid small numbers in some categories of CTVB. The Townsend score categories were researcher defined, but we found little difference in the results between Townsend categories using centile cut-points and Townsend categories defined by approximately the same distribution of the population as the CTVB categories. Thus we have some confidence that the comparisons between the three measures were meaningful. A further limitation is that the analysis did not distinguish between owner-occupation and the rented sector and further work should assess the impact of housing tenure on the utility of CTVBs as a measure of socio-economic status.

Although CTVBs can be assigned to each resident in the sampling frame using the household address and the public domain website [2], this is clearly a laborious process, and was not possible given the size of this study. Using the household address for matching the council tax register to our sampling frame, we were unable to assign a CTVB to 10.2% of the 12,092 respondents. This proportion of missing CTVBs is similar to the 9.9% of respondents missing NS-SEC data. The trend of a higher degree of non-matching in more affluent areas is probably explained by the greater preponderance of house names in affluent areas, which due to spelling inconsistencies are harder to match than a street address. It is hoped that in future the National Land and Property Gazeteer (NLPG) will improve address matching. This will be achieved by linking disparate datasets using the Unique Property Reference Number. Once NLPG data are of a consistently high standard and in use across local government, it is expected that the anonymised matching of address to assign the CTVB to individuals will approach 100%.

Our study was faced with the usual limitations of postal questionnaire health surveys that result ultimately in non-response bias [34–36]. We made strenuous efforts to maximise the response, including an evidence-based approach to question wording, order and format (consistent question layout and fonts, paper colour, A5 booklet size, contact details on front page), a suitably worded covering letter signed, where possible, by the recipients GP, pre-paid return envelopes, three waves of questionnaires and multiple contacts to collect them by electoral canvassers. In order to raise awareness of the survey in the borough we used publicity posters in GP surgeries, wrote articles in local newspapers and publicised the survey on local radio. We cannot be certain that the use of the canvassers increased the survey response, but because they were local residents it is likely that they were able to facilitate the collection of completed questionnaires through personal contact with recipients. In addition they kept an accurate record of reasons for non-completion, distinguishing between people moved away and people who chose not to complete a questionnaire. This was important in accurately determining the second and third mailing list and compiling an accurate adjusted denominator.

The 62.7% overall response to the survey was satisfactory considering the socio-economic characteristics of the borough [34], and in comparison to the 57.7% response achieved in Caerphilly borough by the Welsh Health Survey in 1998 [16]. Non-response bias resulting from CTVB and NS-SEC item non-response rates of around 10% should be considered further. However, since the purpose of the analyses was to compare trends between CTVBs, NS-SEC and the Townsend score, rather than make inferences about the population, item non-response is unlikely to be a threat to the validity of the study.

As expected from the literature on non-response [34–36], we found the lowest response to the survey was from young males. However we found a similar distribution of CTVB categories in respondents compared to the adjusted denominator, but with a small under-representation of respondents in band A. Overall, we achieved a reasonably representative response by housing value.