Generating social class data in primary care

Public Health (1994), 108,279-287

t~) The Society of Public Health, 1994

Generating Social Class Data in Primary Care p. Ward, 1 A. J. Morton-Jones, MA PhD, 2 M. A. L. Pringle, MD FRCGP 3 and C. E. D. Chilvers, MSc HonMFPHM 4

IResearch Assistant, Department of Public Health Medicine and Epidemiology, 2Research Fellow, Department of General Practice, 3professor, Department of General Practice and 4Professor, Department of Pubfic Health Medicine and Epidemiology, University of Nottingham, Medical School, Queen's Medical Centre, Nottingham NG7 2UH

The objective of this study was to compare three methods of collecting social class data in general practice. The setting was a rural dispensing practice on the Nottinghamshire/ Lincolnshire border. The methods examined were: (a) a self-administered questionnaire to 200 patients to determine their social class based on the occupation of the head of household; (b) members of the practice staff were asked to assign a social class to these households based on their local knowledge; and (c) use of small area statistics from the 1991 census data using modal and weighted methods. It was found that the practice staff were unable reliably to assign a social class to the households. The modal method of using small area statistics to assign social class to households through their postcode and its link to the census data was also inaccurate. While a personal questionnaire will remain the only method for assigning a social class to individual patients for clinical care or most research, the weighted method of small area statistics is shown to be a cost-effective and sufficiently accurate method for health needs assessment in general practice.

Introduction The relationship between material deprivation and higher mortality, 1 morbidity, 2 consultation and prescribing rates 3 and health need 4 has been established. These studies indicate that similar trends with socio-economic status run through all aspects of health service needs (perceived or otherwise). Despite this, socio-economic features are rarely quantified by general practitioners, except in so far as whether they qualify for deprivation payments, s and indeed such data are generally unavailable at the practice level. The need for a m o r e explicit understanding of the social and economic status of practice populations is increasing. O n a local level, all practices, but especially fundholders, must be aware of the health needs of their patients to inform the contracting process. T h e y also need to be able to demonstrate these needs in o r d e r to gain extra resources through the business planning cycle. Family health services authorities need to assess routinely the social and demographic characteristics of patients registered with different practices in o r d e r to rationalise their resource allocation and to monitor their contracting functions. Furthermore, there is now a national imperative, in the form of The Health of the Nation, 6 for practices, authorities and primary care as a whole to m o n i t o r outcomes, relating them to social and economic status. T h e r e are several available measures of socio-economic status for populations, including practice populations. F r o m the results of the national censuses (1981 and 1991) single and derived variables can be Correspondence to: Professor M. A. L. Pringle, Department of General Practice, University of Nottingham, Medical School, Queen's Medical Centre, Nottingham NG7 2UH.

280

P. Ward et al.

applied to geographical areas. Census data are related, at their finest level, to enumeration districts. If individual patients can be allocated to enumeration districts, the characteristics of the practice population can be assessed. Factors such as social class distribution, level of car ownership, percentage of persons living alone, age distribution (including the ratio of children and elderly), housing tenure and unemployment can all be looked at individually. These factors may also be used to derive indices which measure social status and deprivation, such as the Jarman Underprivileged Area Score5 and the Townsend index. 7 While such anonymous external data applied to practices on the basis of geography8 can be used to describe the population registered with the practice, it cannot meaningfully be used for the clinical care of small groups or individual patients. If practices are to monitor accurately the social characteristics of, for example, those women without cervical cytology, then they will need to gather data themselves--and only individual-based variables have meaning at this level. Ideally, data for deriving social characteristics which are applicable to the care of individual patients would be easy to gather, and the derived index should be slow to change (to maximise the re-collection interval), be suitable for aggregation for populations, and be established as an acceptable measure (to allow comparisons with other practices and existing population statistics). Although some other measures such as housing tenure 9 and car ownership 1° have their advocates, the Registrar General's social class coding 11 comes closest to matching these criteria. Social class can be expensive and difficult to capture, but it has the theoretical virtues of applying to households (data collected about one individual [the head of household] characterises all the other members of the household), being slow to change (mobility between social classes is much less than between, say, occupations), and being a nationally accepted coding system when aggregated. This study was designed to evaluate three methods of collecting social class data for accuracy, completeness and costs. The first was the conventional questionnaire method. The second used the opinion of members of the primary care team--one of the vaunted strengths of primary care is the continuity of care and the depth of the doctor-patient reiationship. 12 It might therefore be thought feasible to derive an accurate social class distribution, if not accurate assignation of individual patients, through pooling the knowledge of the primary care team. The third applied the 1991 census data to the postcodes of patients to test the accuracy of small area statistics for deriving a social class distribution.

Methods

The research was conducted in a rural dispensing practice on the Nottinghamshire/ Lincolnshire border which has 5,600 patients registered with the two full-time and two part-time general practitioner partners. Three methods for establishing social class were examined.

1. Questionnaire ,for individual patients A questionnaire was designed to elicit a respondent's social class according to the Registrar General's classification. 11 The questionnaire asked for the respondent's occupation, whether they were employed or self-employed, and how many people they supervised. If the respondent did not consider themselves to be the economic head of the household, then they were asked to answer the same questions for the head of the household.

Generating Social Class Data

281

A sample of patients attending the surgery were invited to complete the questionnaire under the supervision of a researcher until 200 completed questionnaires had been achieved. This method of recruitment was regarded as satisfactory because the sample is not claimed to be representative of any larger group of patients. The 200 questionnaire responses were coded for social class according to the guidelines on the application of socio-economic classifications based on occupation. 13 A social class was then allocated to each household and all the members of the household on the basis of the head of household's social class. The social class structure of the households thus derived was regarded as the basis for comparison for the other two methods.

2. The knowledge of the primary health care team A list of the 200 households, with all their members and the address, was then drawn up. The list was accompanied by a commentary which offered guidance on the general characteristics of each social class grouping. All members of the primary care team, including doctors, practice nurses, practice manager, secretaries, dispensers and receptionists were asked to consider whether they knew enough about a household to judge its social class and, if so, what social elass would best characterise that household. 3. The use of 1991 census data In the third method, the postcode for each of the 200 households was obtained from the practice computer system. For each postcode the census enumeration district was identified using the PCCAMM software (Claymore Services Limited). The enumeration district breakdown by social class was obtained using available software (SASPAC, Manchester Computing Centre, University of Manchester). The relevant enumeration district social class distribution was then assigned to each household. From this distribution, two methods were used to obtain an overall social class distribution for the 200 households.

(i) Modal method: This method assigned the most common social class in the enumeration district to each household (i.e. the mode of the enumeration district distribution). If two or more social classes were equally represented, then a proportion of each social class would be assigned to the household. For example, if social classes II, IIIN and IIIM had equal (highest) frequencies, then the household would be assigned as ~II, ½IIIN, ½IIIM. The overall social class distribution was then obtained simply by adding the assigned values for all households. (iO Weight method: This method assigned a proportion of each social class in the enumeration district to each household. The proportion corresponded to a weighting based on the numbers in each social class in the enumeration district. For example, if an enumeration district had 5, 10, 10, 8, 5, 2 residents (40 in all) in social classes I, II, IIINd IIolM, IV, V, respectively, the proportions assigned to the household would be 40, 4o, 4o, 40, 4o, 4o for each of the social classes I - V . These proportions were then aggregated over all the households to obtain the social class distribution. Social class information is only available for a 10% sample of the census population. Furthermore the census data and our questionnaire data differed slightly m the treatment of retired people. The census data did not ascribe a social class to those people who have been retired or economically inactive for the past 10 years, whereas we ascribed a social class to all retired and economically inactive people (that

282

P. Ward et al.

is, their last social class when employed). For each of the three methods, the time and cost involved in gathering the social class data were compared, as were their comparative results. The method of assignment of social class by the health care team was compared with the questionnaire method using cross-tabulations and the weighted kappa statistic 14 for each member ,of the team and for the general practitioners as a group. The social class distribution derived from small area statistics was compared with the questionnaire-derived social class. Results

Questionnaire for individual patients The 200 completed questionnaires related to households comprising 547 patients. From the 200 completed questionnaires it was possible to assign a social class to 193 respondents (96%). The causes of failure to assign a social class were insufficient employment details (1) and being a widow and not having an occupation of the late husband (6). The 193 respondents came from 185 households. The assigned social class of seven of the eight duplicated households was the same. The difference in the social class coding in one household was due to inconsistency of reporting of the occupation of head of household by the two respondents, one occupation gave a social class II and the other social class IIIN. The 185 households with a social class code were used for the remainder of the study and included 517 patients. When the social class of the household was allocated to each individual member of that household, the social class distribution of households and of patients within those households was almost identical (X 2 = 3.96, df = 5, P = 0.55). Knowledge of" the primary health care team When the 15 members of the primary care team were asked to code the 185 households for social class, 12(7%) households were not coded by any team member and 128(69%) households were correctly coded (as compared with the questionnaire) by at least one team member. The relative completeness and accuracy of the team members' coding are given in Table I. One of the doctors attempted 156(84%) of the households with an accuracy of 49% of those attempted, or 41% of the whole sample; other practice members attempted much smaller numbers of households. Most of the kappa statistics fell in the range of 'moderate' agreement (kappa in range 0.41-0.60), 15 but as a group the manager and secretaries did well with 'substantial' agreement (kappa in range 0.61-0.80), 12 albeit on small numbers attempted. The doctors as a group displayed the most complete and accurate knowledge of social class. When the four doctors in the practice were asked to code the patients, 18(10%) were not coded by any doctor; 121(63%) were coded inaccurately or inconsistently and 54(28%) were coded accurately and consistently in comparison with the questionnaire. This latter figure includes 22(12%) households which were only coded by one doctor but were coded correctly. The use of 1991 census data The results from the small area statistics analysis are shown in Figure 1. The social class distribution from the small area statistics (weight method) agrees well with the social class distribution obtained from the questionnaire method. However, the modal method does not agree so well, with the social classes I and V being underrepresented.

Generating Social Class Data Table I

283

The completeness and accuracy of social class coding by the primary team members as compared with the questionnaire-derived social class of 185 codable households Attempted answers (No.) (%)

Practice member Doctors A

56 80 70 156

Answers agreeing with patient questionnaire (No. of total) (%)

Percentage accuracy

Weighted kappa (95% confidence limits)

30.3 43.2 37.8 84.3

25 34 32 76

13,5 18.4 17.3 41.1

44.6 42.5 45.7 48.7

0.561( 0.361-0.760) 0.554( 0.387.0.722) 0,490(0.281-0.699) 0.568(°-451-0.684)

44.3

32

17.3

39.0

0.446(0.276-0.617)

9.2 3.8 15.1 14.1

11 5 19 22

5.9 2,7 10.3 11.9

64.7 71.4 67.9 84.6

0.778 (°-583-°.973) 0,615(0-°63-1.168) 0.747( 0.568-0.927) 0.635( o.167-1.1°2)

Receptionists/dispensers J 18 K 48 L 2 M 61

9.7 26.9 1.1 33.0

7 24 1 28

3.8 13.0 0.5 15.1

38.9 50.0 50.0 45.9

0.559(0.3°8-0.810) 0.236(-0.216-0.688) 0.667(°-2°5-1.129) 0.443( °.221-°.665)

Anonymous N O

16.8 41,1

10 37

5.4 20.0

32.2 48.7

0,347(0.037-0.657) 0,506(0-271-0"741)

B

C D Practice nurse 82 E Manager/secretaries F 17 G 7 H 28 I 26

31 76

100

~180

80 _

=

.~ -d

[ ] Weight method [ ] Modal method [ ] Questionnaire

-]160 -]140 -'t120

-6

60 -

loo

x:

o

80

40 20 (1;

I

Itt

IItN

IIIM

IV

V"

Social class Figure i

Social class distribution for questionnaire households (n = 185)

284

P. Ward et al.

Cost comparison Method 1, the patient questionnaire, proved to be the most expensive at a projected cost of £1653.71 for the whole practice. We estimate that the practice population of 5,600 would include 2,004 households based on the same ratio of household:patients. The cost of determining a social class for the whole practice using method 3 is estimated at £791.79, which is less than half the cost of method 1. Full details of the costings are given in Appendix 1. Discussion

The concept of social class as an indicator of relative deprivation has its critics, Problems with its initial allocations of occupations to class by infant mortality rates, the fact that assignation is made on occupation alone (that is, it is an occupational classification rather than a true social classification), the problems of assignment of social class to non-working women, and the assumption that the divisions between the six ranks are unchanging over time have all been highlighted. 16-20 However, more 'sophisticated' socio-economic measures have failed to provide substantially different or improved results from those using social class as a criterion. Furthermore it has the advantage of being an established, ubiquitous standard. Its use may be scorned on theoretical grounds, but the fact remains that it is still the most widely used measure of socio-economic circumstances, and health service teams are likely to wish to use it for some time to come. The results of this study show that, despite detailed knowledge of many of the practice's patients, members of the primary health care team cannot reliably assign a social class to their patients. The use of small area statistics to obtain a social class distribution proved more fruitful, the weighting method providing good agreement for this size of population in a visual comparison. Clearly, if a practice wishes to know the social class of individual patients then there is no substitute for asking all households to complete a census-type questionnaire. However, this study has shown that there are considerable costs attached to this--£30,000'per 100,000 patients--and it would need to be repeated episodically. If social class were to be required for particular patients at specific points in their care, for example as part of the referral minimum data set, then the data could be gathered relatively cheaply, but only for a restricted group of patients. Some research will require the matching of social class of individuals or small groups of patients to other characteristics that they might have. In that case there is, again, no substitute for direct patient questioning. However, if the object of a study were to compare the social class characteristics of two groups of patients as general groups, then the small area statistics method might be sufficient, assuming that the difference was sufficient to allow for potential inaccuracies in the methodology. The greatest potential for application of the small area statistical methodology lies in the more pragmatic field for health care delivery. Practices are increasingly expected to use business plans when bidding for new or continued resources. These plans should be based on the health needs of the population, both met and unmet, and existing provision. Perhaps the most potent indicator of health need is social class, and those practices in deprived areas have become skilled at using census data in its raw form. The added dimension offered by the methodology in this study is that practices can relate the social characteristics of the locality not just to the local geographic spread of the practice but to the true location of the individual households registered with the practice. This will increase the accuracy of their perception of the socially determined needs of their practice population.

Generating Social Class Data

285

As health needs assessments b e c o m e m o r e refined, practices m a y wish to d e m o n strate that certain groups of patients, those with diabetes or those not attending for cervical cytology, are particularly socially d e p r i v e d - - t h u s justifying extra nursing or health visiting support. Provided the patient numbers are sufficient--and this study was based on a sample of 2 0 0 - - t h e n the small area statistical analysis is cheaper and sufficiently accurate. In terms of accuracy, it is encouraging that the small area statistics analysis picked out the same patterns in social class as obtained from the formal social class questionnaire despite the potential sources of inaccuracies. These include the 10% sampling of the census population and the possible inaccuracies in matching postcodes to enumeration districts. H o w e v e r , support for the intrinsic accuracy of this methodology has c o m e f r o m a n o t h e r s t u d y ? 1 It does, however, rely on decennial data which were most current at the time of this study. In conclusion, the use of small area statistics as a m e t h o d of deriving local distributions of social class for health needs assessment is a useful asset in gaining the best (balancing cost, time and accuracy of results) insight into the socio-economic make-up of groups of patients in general practice.

Acknowledgements 'I~e authors wish to thank Ian Turner for assistance with the small area statistics; Carol Coupland and Jim Pearson for statistical advice; David Whynes for advice on the economic analysis; and the Collingham health care team and patients of Collingham practice for their cooperation in this study. This project was supported by Trent Regional Health Authority.

References 1. Townsend, P. & Davidson, N. (eds) (1982). The Black Report. London: Penguin. 2. OPCS (1987). General Household Survey, 1989. London: HMSO. 3. Morton-Jones, A. J. & Pringle, M. A. L. (1993). The use of unemployment rates to predict prescribing rates in England. Accepted by British Journal of General Practice. 4. Whitehead, M. (1988). The Health Divide. London: Penguin. 5. Jarman, B. (1983). Identification of underprivileged areas. British Medical Journal, 286, 1705-1709. 6. HMSO (1992). The Health of the Nation: A Strategy for Health in England. London: HMSO. 7. Townsend, P., Phillimore, P. & Beattie, A, (1988)o Health and Deprivation. UK: Croom Helm. 8. Curtis, S. E. (1990). Use of survey data and small area statistics to assess the link between individual morbidity and neighbourhood deprivation. Journal of Epidemiology and Community Health, 44, 62-68. 9. Fox, A. J., Jones, D. R. & Goldblatt, P. (1984). Approaches to studying the effect of socio-economic circumstances on geographical differences in mortality in England and Wales. British Medical Bulletin, 40 (4), 309-314. 10. Goldblatt, P. (1990). Mortality and alternative social classifications. In: Longitudinal Study: Mortality and Social Organisation. London: HMSO, 163-192. 11. Leete, R. & Fox, J. (1977). Registrar General's Social Classes: origins and uses. Population Trends, 8 (1). London: HMSO. 12. RCGP (1983). The Future General Practitioner Learning and Teaching (6th edn). London: Royal College of General Practitioners. 13. OPCS (1991). Standard Occupational Classification, Vol. 3. London: HMSO. I4. Cohen, J. (1968). Weighted kappa: nominal scale agreement with provision for scaled disagreement or partial credit. Psychological Bulletin, 70,213-220. 15. Landis, J. R. & Koch, G. G. (1977). The measurement of observer agreement for categorical data. Biometrics, 33, 159-174.

286

P. Ward et al.

16. Goldthorpe, J. H. (1983). Women and class analysis, in defence of the conventional view. Sociology, 17,465-488. 17. Stanworth, M. (1984). Women and class analysis, a reply to John Goldthorpe. Sociology, 18, 159-170. 18. Heath, A. & Britten, N. (1984). Women's jobs do make a difference, a reply to Goldthorpe. Sociology, 18, 475-490. 19. Jones, I. G. & Cameron, D. (1984). Social class analysis--an embarrassment to epidemiology. Community Medicine, 6, 37-46. 20. Morgan, M. (1983). Measuring social inequality: occupational classifications and their alternatives. Community Medicine, 5, 116-124. 21. Reading, R. & Openshaw, S. (1993). Do inaccuracies in small area deprivation analyses matter? Journal of Epidemiology & Community Health, 47, 238-241.

Appendix 1: costs

Method 1 Fixed costs: Three OPCS Manuals, Social Class From Occupation

£20.45

Variable costs:

Costs per questionnaire Second-class postage (two ways) Envelopes and printing Labour time* Total

0.36 0.06 0.395 0.815

Cost for 200 households

Total

163.00

Cost for 2004 households

Total

1633.26

Total inc. fixed cost

£183.45

Total inc. fixed cost £1653.71

*Labour time for envelope addressing, folding etc. based on a clerical rate of £3.50 per hour (two hours). Labour time for processing and entering data into computer based on a research assistant rate of £6.00 per hour (12 hours).

Method 2 The costs involved in this method are labour costs only so there are no fixed costs. The time taken to estimate the social class of 200 households is taken as 20 minutes. The overall cost is dependent on the member of staff selected to assign the social class. Cost for 200 Cost for 2004 households* households* 12.83 128.58 General practitioner rate (£28 per hour) 5.65 56.64 Practice nurse rate (£6.25 per hour) 6.23 62.45 Practice manager rate (£8 per hour) 5.24 52.53 Secretarial rate (£5 per hour) 4.74 47.52 Receptionist rate (£3.50 per hour) *Including data entry based oil 334 households per hour at £6.00 per hour.

Method 3 Tile main cost involved for this method is the cost of the computer software module to convert the postcodes to enumeration districts. We have priced this at a cost of

Generating Social Class Data

287

£195 for one single postcode area and in this particular case two postcode areas were required. There is a labour cost involved in the correction of postcodes, which although thought to be accurate were found to have an 8% error rate. We are assuming that adequate computer facilities are available at the general practice. The link to the Small Area Statistics in Manchester is not chargeable for small or intermittent usage, but this assumes access to university computing facilities. The only added cost could be a cost of £95 for computer software maintenance which would take into account postcode changes over time. Fixed costs: Two Postcode Manuals at £3 each PCCAMM Software, two postcode districts at £195 per district Total fixed costs

6.00 390.00 396.00

Variable costs: Labour time for postcode correction, obtaining computer statistics and interpreting data (six hours based on a research assistant rate and one hour at receptionist rate) Cost for 200 households Cost for 2004 households

39.50

Total inc. fixed costs

£435.50

395.79

Total inc. fixed costs

£791.79