Effect of doctors' ethnicity and country of qualification on prescribing patterns in single handed general practices: linkage of information collected by questionnaire and from routine dataBMJ 1997; 315 doi: https://doi.org/10.1136/bmj.315.7122.1590 (Published 13 December 1997) Cite this as: BMJ 1997;315:1590
- Paramjit S Gill, research tutora (, )
- Anthony Dowell, directora,
- Conrad M Harris, professor of general practiceb
- a Centre for Research in Primary Care, University of Leeds, Leeds LS2 9JT
- b Academic Unit of General Practice, University of Leeds
- Correspondence to: Dr P S Gill Department of General Practice, University of Birmingham, Edgbaston Birmingham B15 2TT
- Accepted 20 November 1997
Objectives: To test whether Asian general practitioners who qualified in the Indian subcontinent prescribe items more often, more expensive items, and fewer generic drugs than their British trained Asian and non-Asian counterparts.
Design: Linkage study using data collected by questionnaire and from routine sources.
Setting: General practices in England.
Subjects: 155 single handed general practitioners: 42 Asian doctors qualified in United Kingdom (group 1), 58 white doctors qualified in United Kingdom (group 2), and 55 Asian doctors qualified in Indian subcontinent (group 3).
Main outcome measures: Prescribing cost (cost per ASTRO-PU), prescribing frequency (number of items per ASTRO-PU), and generic prescribing (percentage of drugs prescribed that are generic).
Results: Doctors in group 1 were significantly younger than those in the other groups and had a higher proportion of patients who were from deprived wards. There was no difference between the groups in the proportion of female doctors and total list size. After adjustment for confounding factors, there were no significant differences between the three groups for prescribing cost (16.58 (95% confidence interval 6.39 to 26.77) for group 1, 17.31 (6.92 to 27.69) for group 2, 17.80 (7.22 to 28.38) for group 3, P=0.55); prescribing frequency (6.58 (4.60 to 8.40), 6.45 (4.70 to 8.30), 7.89 (6.16 to 9.64), P=0.34); and generic prescribing (44.44 (38.95 to 49.93), 47.41 (42.12 to 52.70), 44.04 (38.75 to 49.33), P=0.37).
Conclusions: Asian doctors qualified from the Indian subcontinent did not differ from British trained doctors in their prescribing practice. This study refutes the common belief that Asian doctors are high volume and high cost prescribers.
There is anecdotal evidence that doctors from the Indian subcontinent issue more prescriptions and more expensive items that do non-Asian doctors
We examined this claim by means of a linkage study using data from 155 single handed general practitioners and routine data sources
There was no significant difference between Asian doctors qualified in the Indian subcontinent and British trained Asian and white doctors for prescribing costs, the number of items prescribed, and the percentage prescribing of generic drugs
These results refute the myth that Asian doctors are high volume and high cost prescribers
The issuing of prescriptions by general practitioners is a complex activity that depends on the interplay of many factors, including variables associated with the doctor, the patient, doctor-patient interaction, and sociopsychological variables.1 We know little about the effect of ethnic group and doctor's country of qualification on variations in prescribing patterns.
The evidence that Asian (from the Indian subcontinent) general practitioners prescribe more often and more expensively than non-Asian general practitioners seems to be anecdotal. Although there are studies reporting an association between a doctor's country of qualification and prescribing patterns, to date no study has reported an association between doctors' ethnicity and prescribing patterns. One of the earliest studies found that foreign trained doctors were more likely to be high cost prescribers than their British trained colleagues,2 and two recent studies found that doctors who qualified outside the British Isles were more likely to issue a prescription.3 4 However, none of these studies took account of variables such as the age and sex of the patients.
The aim of this study was to test the hypotheses that single handed general practitioners who are Asian (of Indian, Pakistani, Bangladeshi, or Sinhalese ethnic groups as defined in the 1991 census) and who qualified in the Indian subcontinent prescribe more items, more expensive items, and fewer generic drugs than their British trained Asian and non-Asian counterparts.
Subjects and methods
Subjects Sample frame
We obtained the sample frame of doctors for the study in four stages (see 1) in order to identify the following three groups of general practitioners: Asian doctors who qualified in the United Kingdom, non-Asian doctors who qualified in the United Kingdom, and Asian doctors who qualified in the Indian subcontinent.
We chose single handed general practitioners for study because their prescribing and analysis cost (PACT) data most closely reflect their own prescribing.5 In the first stage, the Department of Health categorised all single handed principals in England (2701) by their country of qualification. Doctors who qualified outside the United Kingdom and the Indian subcontinent were excluded (242, 8.9%).
During stage II, we obtained the doctors' surnames and initials, as well as the medical school of those who qualified in Britain, from the Department of Health. We used the Medical Register6 to obtain the doctors' forenames and to validate their school of qualification, as the Department of Health database does not differentiate between doctors whose primary degree was in the Indian subcontinent and those who had requalified by sitting one of the licentiate exams.6 Furthermore, the Department of Health database does not include ethnic group. We then categorised the general practitioners into Asian (98) and non-Asian (974) groups by their forenames and surnames.7 This method was valid, as we found good agreement (κ=0.95) with doctors' self determined ethnic group when we sent them a questionnaire (see below).
During stage III, six general practitioners were randomly selected from each group for the pilot study. The remaining 92 Asian doctors who qualified in the United Kingdom (group 1) were included in the main study to achieve sufficient sample size as we expected the response to our questionnaire to be low.8 Further random samples of 92 general practitioners who were white and qualified in the United Kingdom (group 2) and 92 doctors who were Asian and qualified in the Indian subcontinent (group 3) were again supplied by the Department of Health (stage IV).
We sent a questionnaire to these 276 general practitioners that asked for their demographic details. These included their ethnic group, using the categories given in the 1991 census; medical school; year of qualification; possession of the MRCGP diploma; length of time in current post and in general practice; employment of assistants or retainers; practice status (single handed, dispensing, training, or fundholding); and total number of consultations during March 1994.
For inclusion in the study, respondents had to be single handed principals; had to have been in their current practice for more than a year (as the PACT data obtained were for a whole year); and had to be Asian and qualified in the United Kingdom (group 1), white and qualified in the United Kingdom (group 2), or Asian and qualified in the Indian subcontinent (group 3). These criteria resulted in the exclusion of 58 general practitioners: 35 from group 1, 13 from group 2, and 10 from group 3.
We calculated that the necessary sample size was 40 doctors per group (120 in total) in order to detect a difference of two items per patient with a power of 80% and significance of 0.05.
We used a linkage design to test the association between the general practitioners' ethnicity and country of qualification and their prescribing patterns. After getting consent, we obtained data from four sources: the questionnaire sent to the single handed general practitioners (see above), data from family health services authorities, data from the 1991 census, and data from the Prescription Pricing Authority.
Data from family health services authorities
List size, to calculate cost based and item based ASTRO-PUs (age, sex, and temporary resident originated prescribing units)9 using a specific computer program written for this study by the FHS Computer Unit, Exeter. This process accounted for all patients registered with a particular general practitioner even if they were residing in other health authorities (further details from PSG)
Number of patients giving rise to deprivation payments on 31 March 1994
Number of temporary residents on the doctor's list between 1 April 1993 and 31 March 1994.
Data from 1991 census
We calculated the Townsend score10 for each practice by, firstly, assigning the post code of the practice to ward and, secondly, retrieving the relevant variables from the files of the 1991 Census Local Base Statistics maintained at the Manchester Computing Centre by the Census Dissemination Unit using the SASPAC 91 computer package.11 For practices with branch surgeries, we used the mean Townsend index. In addition, we calculated the proportion of Asian patients (residents of Indian, Pakistani, Bangladeshi, and other ethnic groups from the Indian subcontinent) and the social class (mainly non-manual versus manual or equal) of all residents in each ward.12
Data from the Prescription Pricing Authority
These were the total number of items, net ingredient cost, and the percentage of drugs that were generic for all categories of drugs listed in the British National Formulary that were dispensed between 1 April 1993 and 31 March 1994.13
Prescribing outcome measures
We used the following prescribing variables as the outcome measures:
We entered data from all four sources into a database and analysed them using SPSS for Windows.14 We compared the groups using χ2 tests, analysis of variance, and Kruskal-Wallis H tests, and we used Bonferroni correction to adjust for multiple comparisons.
To adjust for potential confounding, we developed parsimonious models for each of the outcome variables using analysis of covariance,15 and we examined assumptions underlying the model using standard residual plots and Levene's test for homogeneity of variances.14
All statistical tests were performed with and without outliers, and we found no difference. We identified and included the following variables in the model: doctors' age, sex, country and year of qualification, length of time in current practice and general practice, MRCGP diploma, and number of consultations a year; status of practice (dispensing, fundholding, or training); social class of patients; Townsend index; percentage of patients attracting deprivation payments; and percentage of patients belonging to an Asian ethnic group.
We chose the Townsend score as a proxy for morbidity because it has been shown to perform well when explaining variation in a range of health measures and it adheres closely to the concept of material deprivation.16 We used the proportion of patients attracting deprivation payments as a proxy measure for deprivation: despite criticism of the Jarman score on methodological grounds, this was a pragmatic decision as general practitioners are paid for patients who reside in deprived wards.17
Characteristics of responders and non-responders
Table 1 shows the numbers of general practitioners who responded to our questionnaire. Using two sample t tests, we found no difference between the doctors who responded and the non-responders in terms of age (mean difference −1.7 (95% confidence interval −4.5 to 0.9)), Townsend score for the practice ward (mean difference 0.6 (−0.6 to 1.8)), and sex (χ2 test, df=1, P=0.87). Compared with all general practitioners in England, the responders were older, were more likely to be male, practised in more deprived areas, and had larger list sizes (Table 2).
Characteristics of responders by group
Table 3 shows that Asian doctors who qualified in the United Kingdom (group 1) were significantly younger than doctors in the other groups and had a higher proportion of patients who were from deprived wards. There was no difference between the three groups in the proportion of female doctors and total list size.
Table 4 shows the mean prescribing variables for the three groups before and after adjustment for confounding factors. There were no significant differences between the groups for total cost (cost per ASTRO-PU), frequency of prescribing (number of items per ASTRO-PU), and rate of prescribing of generic drugs.
Doctors' ethnicity and country of qualification
Our study shows that, for all the drugs issued, the Asian doctors who had qualified in the Indian subcontinent did not issue significantly more items, more expensive items, or fewer generic drugs than Asian and white doctors who had qualified in the United Kingdom. This refutes the myth of Asian doctors being high cost prescribers.
It is hardly surprising that, for those Asian doctors who qualified from British medical schools, their ethnicity had no effect on their prescribing. The medical school of qualification, and hence country, would be more likely to affect prescribing through secondary socialisation, which occurs over five to six years.2 18 Socialisation through medical school, specialty, and practising in a group practice changes prescribing behaviour.19
The models we used in this study explained only a small part of the variation in prescribing patterns that we found (23-37%): a large part was due to other, unmeasured factors. This is in line with a study that used routine statistics on general practices in one family health services authority: the authors found that their model explained only 42% of prescribing costs.20 It must be emphasised that, as we did not study appropriateness, the differing levels of prescribing by general practitioners might have been appropriate for their patients.
This is the first study which specifically addresses the effect of doctors' ethnicity and country of qualification on prescribing patterns. Other studies have found that doctors who were foreign trained were high cost and high volume prescribers.2 3 4 These studies did not, however, account for other confounding factors, particularly patient demography.
Limitations of study Chance
We calculated the power of this study from the number of issued items as the outcome measure. After the study, we found that power calculation for all other outcome measures, such as cost and rate of generic prescribing, was above 90%. This indicates that chance is an unlikely explanation for these results.
The results could have arisen from inaccuracies in list sizes—that is, more patients listed on a health authority register than were actually registered with the general practice.21 Such list inflation could reduce cost of prescribing per patient in inner city areas so that comparison with other practices is invalid.5 However, we attempted to account for all patients by using a unique program to extract data from health authority computers (FHS Computer Unit, Exeter). Furthermore, the general practice contract, which needs accurate denominators for payment for achieving certain targets, has probably reduced inaccuracies of health authority registers. The increasing use of computers within general practice22 may also have contributed to increased accuracy of practice registers of patients' age and sex, from which patient information is passed onto the responsible health authority.
Using data from the 1991 census, we estimated the proportion of Asian residents and social class in each ward to minimise bias, but the census data contain errors because of underenumeration.23 Also, because of the selective way in which patients choose practices, errors are present in census derived variables for general practices.24 However, these derived variables are currently the best measures that are available of the socioeconomic characteristics of practice populations.25
Confounding variables might be an alternative explanation for the results. We identified a number of factors from the literature, which, where feasible, we measured and included in the analysis. These adjusted models still explained little of the variation in the prescribing dependent variables (17-32%), with the rest due to other factors. This should be interpreted with caution as it was not the objective of this study to determine which factors explain variation in prescribing outcomes.
Validity and generalisability
The excellent response rate (71%) from the general practitioners is worthy of comment—single handed general practitioners are generally less likely to respond to questionnaires.8 All non-responders were accounted for and did not differ in terms of age, sex, or practice ward deprivation from the other single handed general practitioners. As single handed practitioners are different from those that are in partnerships,26 these results cannot be generalised to other general practitioners.
This study refutes beliefs that Asian doctors are high volume and high cost prescribers. We hope that this lays another myth to rest. It is important to bear in mind that the issuing of a prescription does not occur in a vacuum: it involves a complex interplay of factors, many of which have not been measured in this study. We do not yet know how the prescribing patterns of different groups of doctors vary with different groups of patients in different settings.
We thank our colleagues who gave invaluable help, especially Glen Scrivener, David Lloyd, Dave Clucas, and Sue Bogle. We also thank all the staff at the family health services authorities, Alan Scott of the FHS Computer Unit, and Steve Parker of the NHS Executive for help with data extraction. We are grateful to all the general practitioners who took time to participate in this study.
Funding: Partly funded by the Scientific Foundation Board of the Royal College of General Practitioners.
Conflict of interest: None.