For these relatively deprived populations, prescribing for anxiety and depression are best predicted not by health status, material deprivation or ethnicity but by the proportion of the population with English as a first language and number of GPs per 10 000 population. These analyses suggest that the wide variation in prescribing rates observed in previous studies may be due not only to differences in health status or the cultural acceptability of treatment, but also to differences in access to primary care for those with mental health related symptoms, both in terms of the actual services available and an individual's ability to access them. It is worth noting that the number of GPs per 10 000 population was not correlated with area deprivation or self-reported health, variables that might predict increased need for primary care provision. As we did not have information about the prescribing practices, or any other characteristics, of GPs in the NDC areas, we cannot examine whether these associations are confounded by individual GP prescribing practices. For example, areas with larger Asian populations might also have more overseas trained doctors and this has previously been shown to be associated with lower prescribing rates independent of patient ethnicity[3]. Studies which explore the "supply side" factors that influence prescribing rates in more depth are needed to unravel these issues. Since the numbers of GPs in this data set was based on routine data which may not include accurate information on part-time GPs or GPs in training, some areas may have better provision than routine data suggests.
The lack of association with deprivation may well be due to all the areas identified as NDC communities being significantly deprived. A recent study from east London, that focused on practice characteristics, also found a correlation between anti-depressant prescribing and list size[4]. Similarly a lack of association with age and sex may be explained by the observation that these did not vary significantly between NDC areas.
The finding that the proportion of the population not having English as a first language was a significant predictor of prescribing rates, while proportion with "Asian" ethnicity was not, suggests a more complex relationship between culture, communication and prescribing, than can be measured by self-defined "Asian" ethnicity. It is plausible that identifying a language other than English as a first language reflects both language skills and self-perceived integration into the English-speaking community, both factors that may influence use of primary care services. The lack of independence of factors relating to ethnicity (particularly Asian ethnicity and English not as first language) means that it is difficult to be sure to what extent the underlying issue is language rather than ethnicity, but the multivariant analysis demonstrates that language is a better predictor than ethnicity and therefore potentially the underlying factor, reflected by ethnicity.
The major strength of this study was the availability of large sample survey data from the NDC MORI/NOP household survey. This gave us a validated measure of mental health status and allowed us to include mental health status as a potential explanatory variable. It also gave us a measure of perceived access as reported by respondents, although no information was collected on frequency of use of services. We had expected that at least some of the variation in prescribing between populations would be explained by underlying variation in the SF-36 mental health scores. The results suggest that for these communities, population need does not explain the variation. The remaining variation is likely to be largely due to random effects, local cultural attitudes to symptoms of anxiety and depression that are difficult to quantify and variation in the prescribing practice of individual GPs and other wider determinants of access to drug treatment.
Interpretation of prescribing rates need to bear in mind that these categories of drugs will be prescribed in a minority of cases for non-mental health conditions (for example neuropathic pain or enuresis), although it is likely that this would account for a relatively small proportion of variation. The clinical significance of variation in prescribing of anxiolytic and anti-depressant drugs may be different as there is some evidence from local audits that the former are over-prescribed and the latter under-prescribed relative to best practice. We found a high positive correlation between prescribing in these two categories, suggesting that in this analysis the prescribing patterns do not represent a proxy for "quality of care" (which might be predicted to show a negative correlation in that case) but purely for "access" irrespective of quality.
The main limitation in terms of the generalisability of our findings is that the data set only included relatively deprived urban populations that might be expected to have poorer than average access to GPs. It is possible that in more affluent populations with better access to primary care, population characteristics including mental health needs may explain some variation in prescribing rates. It seems likely, however, that if access is an issue, it will have an impact on prescribing.
These results do have important policy implications, particularly for those responsible for ensuring equitable and adequate access to primary care services. For example, when interpreting prescribing rates it may be worth considering that lower prescribing may not reflect better health or more judicious prescribing but instead reflect poorer access to care. Conversely, increasing prescribing rates may in fact be a reflection of improving access to care rather than deteriorating health status in a local population.
