There are large literatures around how to define, operationalise and measure equity in relation to health care services 123, although equity is generally taken to mean 'fair' or 'just'. Equity has been divided into three domains: equal access to health care for people in equal need; equal treatment for people in equal need; and equal outcomes for people in equal need 1. Whilst this is a simplification of the nature of equity, it is useful in delineating the various domains in which inequities may arise.

The current paper is focussed around the equal prescribing (i.e. equal treatment) for people in equal need. Using the example of the current study, an analysis of equity would assess the differences in prescribing rates provided to the population of one GP practice compared to another GP practice, weighted to take account of the levels of need for CHD drugs in their patient populations. Therefore, it would be equitable to have higher prescribing rates for populations with higher levels of health care need and lower prescribing rates for populations with lower levels of health care need. However, it would be inequitable to have higher prescribing rates for populations with lower levels of health care need and lower prescribing rates for populations with higher levels of health care need. The identification of populations where prescribing was deemed inequitable could then be targeted for further resources aimed at redressing the balance between prescribing rates and health care need.

A recent paper by the authors questioned the equity of GP practice prescribing rates for a range of CHD drugs4 and highlighted the contemporary relevance of the 'inverse care law'5 in the context of GP prescribing. That paper presented the findings of bivariate correlations between prescribing rates and healthcare needs indicators (HCNIs). One of the inherent problems with bivariate analysis is that prescribing rates are likely to be associated with a number of HCNIs. Therefore, the purpose of this paper is to present the findings of multivariate regression analyses between prescribing rates and the HCNIs and ultimately to examine the independent associations between prescribing rates and HCNIs. In doing so, this paper sets a benchmark for future studies aimed at assessing the effectiveness of the National Service Framework for CHD in developing CHD services commensurate with healthcare need 6.

There is a growing body of research which has highlighted large variations in overall prescribing rates between GP practices, which are only partially explained by factors other than health care need 47891011. Statin prescribing has been shown to vary between health authorities and GPs 12131415 and between patients on the basis of gender 13161718, demographics 1319, ethnicity 20 and deprivation 21. Prescribing rates of beta-blockers have also been found to be lower in patients aged over 75 years and in minority ethnic groups 2223.

Studies attempting to 'explain' the variation in statin prescribing rates have been modest, with most studies explaining around 20 per cent of the variation 13152124. In addition, one study explained around 40 per cent of the variation in prescribing rates for all cardiovascular drugs 21. The prevalence of CHD explained 12 per cent of the variation in statin prescribing in men, and 7 per cent in women 13, deprivation explained 14 per cent 15, and a combination of nitrate prescribing rates and population aged between 35 and 74 years explained 18 per cent 24. In addition, the indicative prevalence of CHD was moderately correlated with prescribing rates for statins, and was the most important variable in the multiple regression model 21. However, characteristics of GP practices such as their training status, the number of GPs, or their single-handed status (i.e. whether or not they are 'lone' GPs or work in a multi-partner practice) have been found to have no relationship with prescribing rates for statins 1524. Therefore, the majority of variations in statin prescribing rates in addition to prescribing rates for other CHD drugs remain unexplained.

The planning and provision of health care to local populations in England is now the role of primary care trusts (PCTs). Essentially, PCTs are organisations whose main responsibilities are around developing, commissioning and providing services, which are targeted to the needs of local people, and ultimately to improve the health (and reduce health inequalities) of local people 2526. PCTs have taken over these responsibilities from health authorities, which no longer exist 27 and are responsible for spending 75 per cent of the overall NHS budget in England 28.

This study was undertaken in 4 PCTs in England (referred to as PCT1, PCT2, PCT3, and PCT4 throughout this paper), which included 132 GP practices (PCT1 had 50 GP practices, PCT2 had 24, PCT3 had 31 and PCT4 had 27). In terms of patient populations, we excluded patients aged under 35 years, since prevalence of CHD is particularly low in this age group. In total, there were 353,897 registered patients aged over 35 year across all 4 PCTs.

When an NHS prescription is dispensed in primary care, the prescription form (FP10) is sent to the Prescription Pricing Authority (PPA) for processing. The PPA collates these data and provides them to GP practices and PCTs in the form of Prescribing Analysis and Cost (PACT) data. PACT data are available for all GP practices in England, and allow detailed interrogation in terms of drugs prescribed along with their dosages, pack sizes and formulations. For example, for a specific time period, we can collect data on which statins were prescribed by a GP practice in addition to the dosages and pack sizes. This allows for a complex and timely analysis of PACT data. Useful critiques of PACT data can be found elsewhere 3132.

PACT data were obtained for all GP practices in the 4 PCTs for the 12-month period October 1999 to September 2000. These data were collected for statins, ACE-inhibitors, beta-blockers, aspirin, and bendrofluazide (a full list of drugs obtained are listed in Appendix A). These drug groups were chosen because they represent major drug groups recommended for the prevention (primary and secondary) of coronary heart disease (CHD) in the United Kingdom (UK) 6. Using prescribing rates for aspirin was potentially difficult, since it can also be purchased over the counter (OTC) in community pharmacies and therefore may not represent the totality of aspirin use within populations 3334. It has also been found that non-prescription aspirin use was higher in men aged under 65 years, and also in more affluent areas 35. Therefore, PACT data may underestimate actual aspirin use within the community. Nevertheless, given the importance of aspirin within the management and prevention of CHD 173637, it was postulated that prescribing rates for aspirin may reflect CHD prevalence within GP practice populations, at least as well or even better than prescribing rates for the other CHD drugs.

The numerator in all prescribing rates was based on a measure of prescription volume, as opposed to prescription cost. The validity of using the number of prescription items or total cost as a measure of prescribing volume has been called into question 3839 since it does not specify the quantity of prescription medication (e.g. number and/or dosage of tablets). Therefore, a measure of prescription volume which calculates the total number of grams prescribed is much more useful. The main options available are defined daily doses (DDDs) 4041 and Average Daily Quantities (ADQs) 404243. The Prescribing Support Unit website provides up to date lists of DDDs http://www.psu.co.uk/ddds.html and ADQs http://www.psu.ac.uk/drugs/adqind.html for all drugs for which they have been developed. Within this study, total ADQs were used as the unit of analysis since they represent prescribing practices in the UK, as opposed to DDDs which represent prescribing practices internationally.

The denominator was the total registered (and resident) patient population aged over 35 years. This age group was chosen since the prevalence of CHD is particularly low in people aged less than 35 years 44.

In total, 24 HCNIs were developed for each GP practice in this study, and all of these were entered into multiple regression models (all HCNIs are outlined in Appendix B). These HCNIs reflect patient demographics, ethnicity, socio-economic status, long term limiting illness, CHD mortality and CHD morbidity. However, the only HCNIs discussed here are those included in the final regression models (see Table 1).

In order to assess the equity of prescribing for CHD drugs, we needed to develop a set of variables which measured (or acted as proxies for) the health care need in GP practice populations. In terms of coronary heart disease (CHD), there are a number of identifiable risk-factors, all of which increase the risk of CHD morbidity and/or mortality. However, readily available data are not available for a number of these risk-factors, and therefore could not be explored within this study. Nevertheless, data were available on the following risk-factors: age, gender ethnicity, socio-economic status. Evidence on the importance of these risk-factors for assessing health care need is presented below.

Rates of CHD related morbidity and mortality increase dramatically with age 4546. For example, the prevalence of angina in England was 1 per cent in people aged 16–44 years, 3 per cent in those aged 45–54 years, although this increased to 10 per cent in those aged 55–64 years, 16 per cent in those aged 65–74 years and 18 per cent in those aged over 75 years 47.

Rates of CHD mortality and morbidity in the UK are generally seen to be higher for people from South-Asian groups, than from white-English people, and much lower for people born in the Caribbean. It has been suggested that the mortality rate for South Asians is 50 per cent higher than for the general population 48. In the UK, the standardised mortality rates for CHD in South-Asian men was 146, compared to 100 for all men and just 46 for men born in the Caribbean 49. It is also recognised that there are differences within South Asian groups, especially between Indians, who generally experience better health, and Pakistanis or Bangladeshis, who have generally worse health 50515253. However, reliable data on CHD morbidity or mortality down to these more complex ethnic groupings are not readily available, and any estimates are, at best, imprecise 50. Nevertheless, data on the ethnic minority profiles of GP practice populations may prove very useful in determining the need for CHD drugs.

There is an extensive literature on socio-economic inequalities in CHD mortality and morbidity 54555657. Whilst rates of CHD have been declining in the UK for almost 20 years, they have not been falling as fast as countries such as Australia and the United States 46. Declining CHD mortality rates are only partially explained by reductions in established cardiovascular risk factors 585960 and it is possible that general social and economic improvement over time has contributed to this trend 61. However, it is noteworthy that these benefits have not been observed by all of the socio-economic groups within the UK 62. For example, deaths from CHD in men in the highest social class have halved in the past 20 years, but remain almost unchanged among men in the lowest social class 60. Between 1986 and 1992, people from the highest social classes had a rate of 160 CHD deaths per 100,000 population, whereas the rate for the lowest social classes was 266 per 100,000 49. In addition, a 22-year follow-up study on the relationship between socioeconomic status and CHD in middle-aged men 63 found that irrespective of length of follow-up, lower social classes had a clearly increased risk of fatal CHD – after 8 years they had a 69 per cent increased risk, which dropped to 67 per cent after 15 years and 59 per cent after 22 years. Therefore, socio-economic status is significantly related to CHD mortality and morbidity, and as such, represents an important indicator of health care need for CHD drugs.

Overall, the age, ethnicity and socio-economic status are all important factors in shaping the epidemiology of CHD in the UK, and as such, are important variables for use in developing the HCNIs in this study. The actual development of HCNIs relating to these risk factors are outlined below, in addition to additional HCNIs related more directly to CHD morbidity and mortality.

Some HCNIs were collected directly from GP practice lists (proportion of patients aged over 75 years and proportion of patients aged 55–74 years), some have been calculated for GP practices (Low Income Scheme Index (LISI) score) 64, and others were calculated from data directly attributable to GP practices (data based on hospital episode statistics). However, due to the lack of information available from GP practices on variables such as socio-economic status and ethnicity (in this study, this refers to South Asian groups) of patients, a number of HCNIs were developed which estimated this from data such as the 1991 Census and Local Authority statistics. The method of patient weighted attribution was used to develop these estimates using data at enumeration district level (small geographical areas), whereby the postcodes of patients were linked to enumeration districts. Data on the enumeration districts of all registered patients were then aggregated for each GP practice and divided by the total registered population in order to provide a patient-weighted score. In this way, data from the Census were directly applied to GP practice populations. Further information and critiques of this method can be found elsewhere 656667. Whilst 1991 Census data may be regarded as rather old now (and has since been superseded by 2001 Census data), these were the only data available at the time of the study, since GP practices do not routinely collect data on the ethnicity of patients. Nevertheless, the HCNI relating to ethnicity may be seen with caution in this paper.

Demographic HCNIs were developed directly from GP practice list data, and these relate to the proportion of patients aged 55–74 years, and the proportion aged over 75 years. Both of these demographic groups are indicators of health care need for CHD drugs 44. The Low Income Scheme Index (LISI) was used as a proxy for low income since it represents the proportion of prescriptions which are exempt from prescription charges due to low income 64. The proportion of patients from South Asian groups was estimated for each GP practice using data from the 1991 census.

Data were also obtained from hospital episode statistics (HES) on specific hospital procedures (coronary artery bypass graft (CABG), percutaneous transluminal angioplasty (PTCA), and coronary angiogram) and diagnoses (primary diagnosis of CHD at discharge). Although HES relate to the supply, as opposed to need for health care services 68, it was hypothesised that in the absence of other CHD morbidity data, they may represent a useful proxy of CHD morbidity in GP practice populations. Due to the low numbers of procedures and diagnoses within a GP practice population, data were aggregated for 6 years (1995 to 2000 inclusive). Crude rates (per 1000 patients aged over 35 years) were calculated for CHD HES, which represents the CHD procedures + CHD diagnoses.

As outlined in Appendix B, a number of other variables relating to CHD mortality and morbidity were entered into the regression models, although these were not statistically significant. Indeed, as outlined in a previous paper by the authors 4, bivariate analyses revealed that associations between prescribing rates and standardised mortality rates for CHD were often not statistically significant and in some cases had a negative association. Therefore, prescribing rates and CHD mortality rates do not exhibit a strong relationship, adding to the suggestion of an inequity in prescribing rates.

Multiple linear regression modelling was undertaken for each drug group in each PCT in addition to the combined dataset. The dependent variable in each model was the respective prescribing rate and the independent variables were all 24 HCNIs developed in the study. The forward-stepwise approach was taken but not slavishly adhered to since considerations of model coherence and plausibility were taken into account when seeking the most parsimonious models. The final regression models only contained the HCNIs which had statistically significant (p < 0.05) independent associations with the dependent variable, and each model was checked for collinearity and normality of residuals. Overall, for each multiple regression model, all 24 HCNIs were entered as independent variables, and the final model included only those variables that were statistically significant and added to the fit of the model.

Table 2 provides details of variations between primary care trusts (PCTs) in prescribing rates for each of the study drug groups. PCT1 generally had the highest median prescribing rates, followed by PCT2, with the lowest prescribing rates occurring in PCT3 and PCT4. The median ADQs per patient over 35 years for all study CHD drugs combined were 105 in PCT1, 93 in PCT2, 90 in PCT3, and 84 in PCT4.

PCT1 had the highest median prescribing rates for all drugs except aspirin, where PCT4 had the highest median rate. PCT3 had the lowest median prescribing rates for aspirin, beta-blockers and ACE inhibitors, and PCT4 had the lowest median prescribing rates for statins, bendrofluazide, and all study CHD drugs combined. The difference between the median prescribing rates for PCT1 and PCT4 for bendrofluazide, statins and all CHD drugs was around 4, 5, and 21 ADQs per patient over 35 years. Therefore, an average GP practice in PCT1 with 3000 registered patients over 35 years, prescribed 12000 more ADQs for bendrofluazide, 15000 more ADQs for statins, and 63000 more ADQs for all study CHD drugs. In addition, a similar GP practice in PCT1 also prescribed almost 30000 more ADQs for ACE inhibitors than a comparably sized GP practice in PCT3.

An initial overview of the health care needs of the PCTs suggests that PCT4 had the highest levels of CHD related health care needs within the study, whereas PCT1 had the lowest health care needs. PCT4 was the most deprived of all PCTs, had the highest proportions of patients aged over 75 years, and the highest median score for standardised mortality rates (SMR) for CHD. In contrast, PCT1 may be seen as the 'least needy' of all PCTs on the basis of the HCNIs developed in this study. PCT1 was the least deprived, had the lowest proportions of South Asian groups, had the lowest median SMR and the lowest median rate of CHD hospital procedures.

These overall descriptive findings are indicative of inequitable prescribing rates, whereby PCT1 had high prescribing rates and low comparative health care needs, whereas PCT4 had low prescribing rates and high comparative health care needs. However, these are purely descriptive and do not take into account the multiple risk factors for CHD or the interactions between HCNIs. Therefore, findings from the multiple regression analyses are presented below.

Table 3 presents the percentage of variation explained in each of the models (i.e. R² * 100), Table 1 presents a description of the HCNIs included in the final regression models and Table 4 presents details for the combined dataset in terms of the HCNIs included in each model, their standardised beta coefficients and their contribution (in terms of percentage variation explained) to the total R² of the model.

Regression models generally explained much more variation in prescribing rates in PCT2 than any other PCT. For example, in PCT2, modelling explained around 55 per cent of the variation in prescribing rates for ACE inhibitors and bendrofluazide, almost 60 per cent for statins and almost 90 per cent of variation in aspirin prescribing rates. Multiple regression models explained around 25 to 55 per cent of the variation in prescribing rates in the combined dataset, around 25 to 60 per cent in PCT3, and around 40 to 55 per cent in PCT4. PCT1 had the lowest R² values in general, and no model was derived for ACE inhibitors. Given the reduced explanatory power of the regression models in PCT1, when regression modelling was undertaken for the combined dataset, the R² values were much lower than those for PCT2, and generally lower than those for both PCT3 and PCT4.

In general, regression modelling explained more variation in prescribing rates for aspirin than for the other drug groups. Not including PCT1, regression modelling explained between 50 and 90 per cent of the variation in aspirin prescribing rates. Comparative percentages were between 30 and 60 per cent for statins, 30 and 45 per cent for beta-blockers, and 25 and 55 per cent for ACE inhibitors and bendrofluazide.

Table 4 presents details of the regression models derived for each drug group in the combined dataset. Many of the models in Table 4 are quite similar in terms of the HCNIs included within them, with all models (except for ACE inhibitors) including the rate of CHD HES (positive association) and all models (except for aspirin) including the proportion of patients aged between 55 and 74 years (positive association). In addition, the LISI score was included in the models for aspirin, bendrofluazide and all CHD drugs (negative association).

The models for statins, aspirin and ACE inhibitors also included the proportion of patients aged over 75 years although this indicator had a negative association with prescribing rates for statins and ACE inhibitors, and a positive association with prescribing rates for aspirin. This suggests that GP practices with higher proportions of patients aged over 75 years have lower prescribing rates for statins and ACE inhibitors, and higher prescribing rates for aspirin.

The CHD HES rate varied considerably between models, in terms of the amount of variation explained. For example, it explained 33 per cent of the variation in aspirin prescribing rates, 20 per cent of the variation in prescribing rates for all CHD drugs and beta-blockers, 15 per cent of the variation in statin prescribing rates, and 7 per cent of the variation in prescribing rates for bendrofluazide.

The LISI score explained 23 per cent of the variation in prescribing rates for all CHD drugs, 13 for bendrofluazide and 10 per cent for aspirin. In addition, the proportion of patients aged between 55 and 74 years explained between 2.5 and 7 per cent of the variation for statins, ACE inhibitors, beta-blockers, bendrofluazide, and all CHD drugs.