Counting the cost of social disadvantage in primary care: retrospective analysis of patient data
BMJ 1997; 314 doi: https://doi.org/10.1136/bmj.314.7073.38 (Published 04 January 1997) Cite this as: BMJ 1997;314:38
- Adrian Worrall, research officera,
- J Nicolas Rea, general practitionera,
- Yoav Ben-Shlomo, senior lecturer in clinical epidemiologyb
- aKentish Town Health Centre, London NW5 2AJ
- bDepartment of Social Medicine, Bristol University, Bristol BS8 2PR
- Correspondence to: Dr Rea
- Accepted 31 October 1996
Abstract
Objective: To cost the relation between socioeconomic status and various measures of primary care workload and assess the adequacy of current “deprivation” payments in relation to actual costings for patients living in qualifying areas.
Design: Retrospective data on primary care were collected over a 4.5 year period from both computerised and manually filed records. Standardised data on socioeconomic status were obtained by postal questionnaire.
Setting: Inner city group practice with a socioeconomically diverse population.
Subjects: 382 male and female subjects of all ages, with a total of 1296 person years of observation.
Main outcome measures: Primary care costs resulting from consultations with a general practitioner or a practice nurse and both new and repeat prescriptions.
Results: Morbidity, workload, and costs of drug treatment increased with decreasing socioeconomic status. The difference in cost for patients in social classes IV and V combined compared with those in I and II combined was about £150 per person year at risk (£47 for workload and £103 for drugs). Deprivation payments met only half the extra workload cost for patients from qualifying wards.
Conclusions: The greater workload caused by social disadvantage has been previously underestimated by simple consultation rates. The absolute difference in costs for socially disadvantaged patients increases as more detailed measures of workload and drug treatment are included. Current deprivation payments only partially offset the increased expenditure on workload. This shortfall will have to be addressed to attract general practitioners to, or retain them in, deprived areas.
Key messages
The costs of providing primary health care including drug treatment increased with decreasing socioeconomic status from £107 in social classes I and II to £256 in IV and V per person year at risk
Previous studies that used simply consultation rates have underestimated the differences by socioeconomic status compared with the use of more detailed measures of cost
Repeated drug treatment, indicating chronic illness, was the largest component of the total drug and labour cost
The additional workload costs for patients from qualifying electoral wards were only partially offset by deprivation payments
Additional deprivation payments could be used to expand the work of practice nurses, although this needs further evaluation
Introduction
The impact of social disadvantage on morbidity and mortality is now firmly established,1 but its economic effect on health services is less well documented. Recent work has tried to produce a sensitive and valid formula for funding secondary care, taking measures of social deprivation into account.2 In primary care the additional workload associated with deprivation has been acknowledged since 1990 through the provision for general practitioners of additional capitation payments for patients living in the most deprived electoral wards, as assessed by the underprivileged area score.3 These payments have helped to raise the income of inner city general practitioners towards the review body's recommended level,4 but discussion continues about the fairness and accuracy of these payments.5 6 7 This study examines (a) the relation between socioeconomic status and primary care workload, including costs for doctors and practice nurses and drug related costs; and (b) the adequacy of current deprivation payments in meeting additional workload for patients living in qualifying areas.
Methods
Subjects
Subjects were a 5% computer selected random sample of the 12 014 patients registered with a group practice in north London. The age and sex distribution were compared with the practice and local borough populations.
Explanatory variables
We collected standardised data on lifestyle and socioeconomic status with a postal questionnaire, with covering letter from the subject's general practitioner, addressed to a parent or guardian if the subject was aged less than 16 years. We followed up the non-respondents to two reminders by telephone and personal visits, and by flagging records. We recorded the distance of each patient's address from the practice premises.
We took the questions on occupation from the 1991 census and coded the answers using the classification of the Office of Population Censuses and Surveys.8 We classified married or cohabiting women according to their partner's occupational status. We chose social class to allow comparison with other studies and as a comprehensible measure of social disadvantage. We aggregated it into three groups for analysis: I and II, III non-manual and III manual, and IV and V.
Outcome measures
We collected retrospective data on primary care over four and a half years, ending 1 July 1994, the period for which computerised records were available. Measures comprised the number of contacts with general practitioners or practice nurses and diagnostic, prescribing, and referral information; these were chosen if they were funded from family health services authority (primary care) budgets. Fee-for-service items have therefore been included. We examined both computerised and manually filed records. Where a discrepancy was found, we used the larger of two figures. We validated rates of contact with general practitioners by comparison with national expected rates.9
Procedure
A database inquiry application, miquest,10 was used to extract diagnostic Read codes from the practice Meditel system. These codes were then classified as “serious,” “intermediate,” or “minor,” according to the Office of Population Censuses and Surveys' severity categories.9 Only “serious” diagnoses have been reported here–for example, diseases that are possibly life threatening, frequently needing major surgery, or have a high probability of substantial disability.
We calculated the clinical labour cost for each patient by allocating an independently determined cost to each clinical labour event.11 The amounts used for contact with a general practitioner at the surgery, a general practitioner's home visit, and contact with a practice nurse were £16, £46, and £9.23 respectively (each including an element relating to practice expenses, confirmed as reasonable estimates for the practice studied). A “total” primary health care cost per patient was obtained by adding the cost of prescribed items12 and dispensing fees to the clinical labour cost.
Analytic procedure
We determined for each subject the number of clinical episodes and the respective person years at risk, enabling us to calculate a specific rate that could be aggregated by socioeconomic status.
We used analysis of covariance to determine mean rates per year by social class groups, after adjusting for age and sex as covariates. As the actual rate distribution was positively skewed, significance testing used a logarithmic transformation after addition of a constant term (+ 1) to remove zero values. Assumptions of linearity and equal variance were checked by examining the distribution and variance of the rates. Social class was entered in the model both as a set of dummy variables and as a continuous variable (values 1-3) to test for linear trend. Age was included in the regression by using four dummy variables for the following age groups: 0-4, 5-19 (baseline), 20-44, 45-64, and 65 and over.
We compared the rate of contact with a doctor with that observed in the fourth national morbidity survey by using age group and sex specific rates from the latter and calculating an expected rate (indirect standardisation). We converted consultation rates from the national survey to contact rates using the provided ratios.9 In our study population, “ghost” patients–that is, those registered with the practice but who had died, moved away, or re-registered elsewhere–were excluded from the denominator, unlike in the national survey. We therefore applied the national contact rates to our population including ghost patients. Each ghost patient was assumed to have contributed the average length of person years (4.04 years). The expected number of contacts was then divided by the actual number of person years, excluding ghosts, to allow a comparison that takes into account both demographic differences and the exclusion of ghost patients.
To examine whether the pattern of workload across social class groups was overly influenced by heavy users we repeated the analysis using logistic regression and dichotomising each measure into a high and low workload group at the median rate. Odds ratios obtained this way may be misinterpreted as indicative of actual use, rather than high or low use.13 Accordingly, the age and sex adjusted means have also been included. Interactions between sex, age, and social class were examined by the likelihood ratio test.14
The deprivation payment for each subject was ascertained according to the underprivileged area score for his or her electoral ward.
Results
Only 448 of the 572 subjects were resident in the practice area. In all, 382 out of these 448 subjects responded to the questionnaire (response rate 85%). Respondents had a significantly lower general practitioner contact rate than the 64 non-respondents (4.1 v 5.6 contacts per person year, P<0.01).
Table 1 shows that the sample population contained fewer men aged 15-44 years than either the practice or census populations. This is not surprising as this subgroup had the highest list inflation rate and also is less likely to take part in research studies.15 The social class distributions of our sample (social class I and II 40%, III 34%, IV and V 15%, and unclassified 11%) were fairly similar to those in the census data (37%, 40%, 18%, and 4% respectively). Some of these differences reflect the larger proportion of subjects in our study who were not classifiable–for example, students or unemployed people who had never worked.
Age and sex distribution of study population versus practice and local populations. Values are percentages of patients in whole of each population
The population provided 1296 person years of observation that could be classified by social class (569 male, 727 female). The general practice contact rate per person year at risk for our study population was 3.7 compared with an expected rate of 2.9, extrapolating from the national morbidity survey. With adjustment for the “ghost” population, however, the expected rate increased to 3.8, demonstrating that observed rates were similar to those found elsewhere.
Table 2 presents the mean rates and odds ratios for various measures of workload by socioeconomic status.All the measures showed a linear increase by social class, although the pattern for outpatient referrals and recorded use of accident and emergency departments was both weaker and non-significant. Similarly the odds ratios for being a heavy user varied from 2.0 (seeing a general practitioner at surgery) to 3.7 (“serious” illness) between social class IV and V and social class I and II.
Morbidity and workload rates per year by grouped social class, adjusted for age group* and sex
Distance of residence from the practice had no independent effect on any of the main outcome measures–86% of the study population lived within a radius of 1.6 km.
Figure 1 shows the cumulative costs of various types of clinical contact and drug treatment by social class. The cost per year of contact with a general practitioner combined with contact with a nurse rose from £60 in social class I and II to £107 in social class IV and V. The relative differences in clinical labour costs were greatest for home visits. The difference in the “total” primary healthcare cost including drugs between social class IV and V and social class I and II was £150 per person year at risk. The costs of drugs showed the steepest cost gradients, with repeat drugs being the largest component, rising from £30 in social class I and II to £100 in social class IV and V. All cost measures showed significant linear trends in this direction.
The cumulative costs per year with each additional primary care item by social class. Costs for each activity have been added to previous activity to produce a cumulative value
Table 3 summarises the workload costs according to payment bands based on the underprivileged area score derived from the 1991 census. Remuneration (£43 417) was 47% less than the costs of the additional workload of patients from qualifying wards (£82 303); this shortfall would vary between practices.
Underprivileged area scores and workload costs
Discussion
Both primary care use and drug costs increased in a linear fashion with decreasing socioeconomic status to a greater extent than in several other studies.7 9 16 These gradients increased as more detailed data on cost were included (see figure 1). Attendance rates alone underestimate social class differentials. This is analogous to the increased social gradient seen for mortality based on potential years of life lost rather than standardised mortality ratios.17 Differences may be more apparent in a single practice with a shared approach than in a cross section of practices with varied thresholds for prescribing, home visits, follow up consultations, or referrals.
This increased cost reflects greater levels of morbidity, particularly with chronic diseases, as indicated by the large differential costs for repeat prescriptions. The cost of drugs outweighed clinical labour costs, more so for lower socioeconomic status. Similarly, increased use of home visits probably reflects reduced mobility due to ill health as well as less access to a car.
It is unclear whether the increased use of health care by those of lower socioeconomic status adequately reflects their increased need.18 19 Despite increasing consultation rates for serious illness, the pattern for both outpatient referrals and recorded accident and emergency attendance was weaker and did not show a significant linear trend. Our results cannot determine whether this reflects underuse of hospital services by patients from social class IV and V or overuse by those in social class I and II. Other evidence suggests that poorer patients may be less likely to be referred for a specialist opinion20 or to have an operation.21
On the basis of the current measure of deprivation for determining levels of remuneration, we found an almost 50% shortfall between the extra workload costs and the extra remuneration provided. These differences would vary depending on the practice profile: less of a shortfall for more affluent practices and more of a shortfall for practices in more deprived areas.
Limitations of study
Our study has several potential limitations. Firstly, our sample size was relatively small; this enabled us, however, to extract detailed information from both computerised and manual records. It is unclear whether the consultation behaviour of our patients is generalisable to the rest of Britain; the overall consultation rate, however, was closely similar to that observed in a national survey.9 Secondly, data were collected over a retrospective period. This will selectively deplete the number of terminally ill patients, as they could not have been represented in the sample. This will result in underestimating the social class gradient, as these patients are more likely to be of lower socioeconomic status and have higher rates of use. The possibility of “reverse causality” must also be considered, as ill health may result in downward social mobility, exaggerating the socioeconomic relation. This is, however, unlikely to have a large effect.22 Thirdly, other studies have shown that more detailed composite measures–for example, including housing tenure–will further increase the social gradient.23 24 Finally, we could not measure the actual time or complexity of each consultation, nor was adequate information available on the workload of district nurses and health visitors. Some evidence suggests that doctors give more consultation time to educated or articulate patients.25 26 Deprived patients, however, had more consultations for “serious” illness, which is often accompanied by other social problems that might be more demanding for general practitioners. Future prospective studies should try to measure these more subtle aspects of workload.
Role of practice nurse
The practice studied has long experience of working with nurses.27 28 29 Practice nurses were consulted more often by patients of lower socioeconomic status. This may be a useful approach in managing patients with chronic disease,30 the major component of excess morbidity in underprivileged groups. The nurse may be seen as a more approachable figure, more likely to be consulted about family and social problems related to deprivation.30 31
Implications
Current government policy is aimed at increasing the proportion of fundholding practices. Our results showed that current deprivation payments to this practice fell short of the true increased workload costs. The total costs including drug treatment for patients in social class IV and V were around £150 a year more than for their counterparts in social class I and II. The maximum deprivation payment covers less than 10% of this additional expenditure: it therefore does not make economic sense for practices with a high proportion of disadvantaged patients to become fundholders. In addition, because the relation between socioeconomic status and workload is continuous, it is inappropriate to use the current, extreme cut off level in determining payments.32 Many practices that do not receive payments will experience above average workloads because of the social class composition of their population.
Conclusion
Deprivation payments might be used to enable practice nurses to work more effectively in primary care, although this needs more detailed evaluation. Practice nurses are particularly suited to treating chronic illness, the major component of the excess morbidity in underprivileged groups. The greater workload and expense associated with these groups has so far been underestimated and inadequately targeted. This will have to be addressed to attract general practitioners to, and retain them in, deprived areas and increase the range of care that they provide.
Acknowledgments
We thank Mr Martin Shipley for his statistical advice, the partners and staff of the practice, and all the patients who helped by taking part in the study.
Funding: Partly funded by the Research and Development Directorate of the North East Thames Regional Health Authority.
Conflict of interest: None.