More information about text formats
Prescribing in dispensing practices: A response to Goldacre et. al.
Goldacre et. al. is a welcome contribution to the ongoing debate about the causes of high cost prescribing by primary care doctors in the NHS. Their paper claims that doctors in “dispensing” practices (that is, rural practices with an dispensary) have a potential financial conflict of interest around their prescribing decisions because they have the opportunity to earn additional revenues by using more expensive medicines. Using the NHS England Primary Care Prescribing dataset for January 2015 only, they examine data on statins, proton pump inhibitors (PPIs), ACE inhibitors (ACEis) and angiotensin receptor blockers (ARBs). These are the four most commonly prescribed classes of drugs where there is clear guidance that higher cost items are no more effective than lower cost products. Within each of these four classes, they identified options with high acquisition costs that are considered to be as effective as those with lower costs. Following statistical analysis, the authors found that dispensing practices are significantly more likely to prescribe drugs from this list of higher cost products across all four categories.
Following a gap in published research on this topic of over twenty years, Goldacre et. al. have been successful in raising again the issue of whether NHS dispensing doctors are more expensive. Although their results seem conclusive, on methodological grounds,...
Following a gap in published research on this topic of over twenty years, Goldacre et. al. have been successful in raising again the issue of whether NHS dispensing doctors are more expensive. Although their results seem conclusive, on methodological grounds, this reply disputes the validity of their claim that this group of physicians directly responds to a financial conflict of interest. In doing so, it is not denied that such behaviour may occur in practice. Instead, it is suggested that Goldacre et. al. failed to perform robust analysis, so their conclusion of financially motivated behaviour is unsupported. In particular, they have not generated sufficient evidence to prove that an underlying economic mechanism (that is, the incentive structure for dispensing doctors) is casual in affecting a measurable financial outcome (that is, differences in prescribing costs). Here, this response explains why this claim is not supported by their work, especially as they openly admit that (because of their use of cross-sectional data) their analysis does not demonstrate causality.
Evidence suggests that dispensing doctors are a heterogenous group of NHS prescribers. In 1957, the Hinchcliffe Committee reported that dispensing doctors, as a group, had lower drug costs because they tended to receive a capitation fee (that is, a drug budget) rather than fee-for-service payments (like pharmacists). With capitation budgets being abolished in the mid-1980s, drug costs in dispensing practices rose rapidly. A seminal BMJ paper by Morton-Jones and Pringle, which highlights higher costs amongst dispensing doctors in a sample of 108 Lincolnshire practices, was widely criticised at the time. Later work by Baines et. al. suggests that Morton-Jones and Pringle’s analysis of costs per patient rather than costs per Age, Sex, and Temporary Resident Originated Prescribing Unit (ASTRO-PU) probably over-estimated the influence of dispensing status. In the following years, Baines et. al. produced a series of papers that demonstrates that prescribing behaviour by dispensing doctors is a bit more complicated than it first seems. For instance, dispensing doctors may prescribe more over-the-counter items because they operate where no pharmacies are located. As investigators are extremely limited in the types of analysis they can perform using readily-available NHS drug data, there has been little research on dispensing doctor costs since the 1990s. Therefore, the recent contribution by Goldacre et. al. should be welcomed because it offers an innovative methodological approach to re-examining this dorment topic. Despite the promise, here it is argued that their novel methods are not robust enough to support their claims of financially motivated prescribing in dispensing practices.
Goldacre et. al. claim that dispensing practices generate additional income by negotiating lower prices on higher cost drugs while being reimbursed at a standard rate. Therefore, dispensing doctors - as individuals who choose medicines for their patients - have a potential financial conflict of interest around prescribing choices. This represents an important hypothesis for testing. As dispensing status is determined exogenously by location, practices cannot be randomly assigned to the scheme in order to scientifically test the above hypothesis. Therefore, the preferred procedure for testing this hypothesis of performing a Randomised Controlled Trial (RCT) cannot be followed. The authors respond by choosing to perform an observational study. This choice may be sufficient, if they only wish to report observed differences in spending patterns. However, their observational study is not sufficiently robust to scientifically test their proposed hypothesis. The reasons for questioning their methods are explained below.
One of the surprising features of the Goldacre et. al. paper is the use of only one month’s data only: an important fact not mentioned in the abstract. Unfortunately, this limited source of information creates several problems for the validity of their analysis. First, the month chosen (January) is unlikely to be representative of the whole year due to seasonal patterns in prescribing. Next, the year chosen (2015) is unlikely to represent the long-term situation due to fluctuations in drug prices. Third, “regression to the mean” suggests that the observed high costs may fall as actual spends fluctuate around trend. If the analysis is re-run on more recent data, the results are unlikely to be the same. These three reasons suggest that, using one month’s data alone, the authors cannot say much about the ongoing situation with dispensing doctors. Therefore, they ought to limit their claims and not suggest that they have found a stable pattern observable at all times. For example, their extrapolation of estimated cost savings from their one month’s data (£628,875) to a hypothetical year (£7,546,497) is methodologically questionable because costs in each month are likely to differ significantly. Unfortunately, their reporting of estimated annual total cost savings suggests that the NHS could make these financial gains on a repeated basis, which is something the authors cannot tell from their limited data source. Therefore, this headline is both misleading for NHS decision-makers and the public, whilst vilifying doctors working in dispensing practices.
Goldacre et. al. use “standard” primary care practices as their basic unit of analysis. With their practice-level data, the inner workings of the surgeries that they sample cannot be directly observed. Each practice is a “black box” and the actions of the incumbent doctors are hidden from view. For instance, a dispensary manager may control the range of drugs purchased for supply, whilst incumbent dispensing doctors - who may prescribe all items using their generic names - may have little influence over the products ultimately distributed. With their chosen data, Goldacre et. al. are unable to identify such occurrences and can only surmise what behaviours occur within the practices they sample. Indeed, the authors cannot tell whether dispensing doctors have a financial conflict of interest and whether this makes them prescribe more expensive products. In response, the title of their paper and their claim that they are directly studying the behaviour of dispensing doctors are misleading.
Need for a model
Within economics, the lack of opportunities to perform RCTs has led to the convention of building theoretical models to support empirical analysis. In claiming that dispensing doctors respond to a financial conflict of interest, Goldacre et. al. fail to provide a conceptual framework (and accompanying model) that specifies the relationship between the economic incentives facing dispensing doctors and their prescribing behaviour. As a result, the patterns of costs they observe may or may not be linked to an identifiable incentive structure. Without a formal model, the economic relationships at play are difficult to identify and there is no map for concluding whether incentives (or something else) encourage dispensing doctors to behave differently. As Goldacre et. al. fail to provide a coherent description of the relationship between incentives and behaviour, their empirical work cannot prove or disprove their study hypothesis. In response, they should not claim that they are examining the effects of a “financial conflict of interest” because they fail to outline the theoretical links between the variables they study.
The method of undertaking primary care research used by Goldacre et. al. and others may be called the “characteristics approach”. Researchers often undertake this type of analysis because it is easier to perform not necessarily because it is the most appropriate. Rarely, do practices have one characteristic. In their analysis, Goldacre et. al. take the “standard” primary care practice as their basic unit of study and assign the characteristic of dispensing to the relevant cases. They dropped all non-standard practices (including out of hours services, community health services, prisons and homeless clinics) from the dataset, leaving only data from standard GP settings. This is an interesting censoring of the data because dispensing practices could also be considered non-standard. Therefore, comparing dispensers with standard practices could skew the results because they are not a good analytical match.
Clearly, Goldacre et al. ignored a range of other characteristics that could have had more explanatory power than dispensing status. For instance, Whynes et. al. found that 42% of the variation in drug costs per ASTRO-PU (which, by itself, accounts for 25% of the variation in NHS drug costs) amongst dispensing and non-dispensing practices in Lincolnshire was explained by factors such as generic prescribing, night visiting, formulary use, exemption certificates and fundholding status. In contrast, the Odds Ratio models constructed by Goldacre et. al. uses a deprivation index, list-size and number of doctors as confounding variables. As primary care data is usually highly correlated, it is hard to tell whether their analysis captures a real relationship between costs and dispensing status (and the confounders) or whether the chosen characteristics are just a form of nominalism made to seem significant by well executed statistical modeling. In the balance of probabilities, it is likely that alternative models with a different range of characteristics could also explain the variation in the dependent variable to the same degree.
At the core of the characteristics approach is the assumption that all practices with the same characteristic are homogenous or similar in one or more respects, although (in everyday operation) they may differ greatly in many ways. For instance, Goldacre et. al. report that dispensing practices vary in terms of the number of patients they dispense to. The authors represent this heterogeneity using the the following groups (percentage of dispensing patients in brackets): low (zero to 35%), medium (35% to 70%) and high (70% to 100%). Using logistic regression modelling, their analysis found that the likelihood of using higher cost drugs increased with the proportion of dispensing patients on a practice’s list. Because of the nature of characteristics research, the inner workings of the practices in these sub-groups are unknown. For instance, Goldacre et. al. cannot answer the question: is the relationship between these categories and costs related to geographic determinants rather than physician decision-making? For instance, practices with fewer dispensing patients may be located in markets towns with local pharmacies, which reduces costs due to increased over-the-counter sales. Similarly, high proportions of dispensing patients may reflect locations in more isolated areas, which may increase the volume of high cost medicines supplied. Evidence suggests that geography affects the operation of dispensing status.[10,11,12] Without the existence of a robust theoretical model to help us understand the importance of heterogeneity amongst practices labelled as dispensing, it is simply conjecture to conclude that having proportionally more dispensing patients increases the temptation for individual doctors to prescribe for profit.
Goldacre et. al. report that dispensing doctors succumb to a morally unacceptable conflict of interest. In making this judgement, the authors are raising moral concerns about the behaviour of a group of physicians that they have not directly interviewed or observed. As a result, they have no direct evidence that the sampled doctors had the intention to behave unethically. Without intention, Goldacre et. al. cannot prove whether these physicians decided to bring about the morally unacceptable consequence of using high cost products when cheaper alternatives are available. Aware of a similar problem in their prescribing cost research, Baines et. al. sampled 105 practices in Lincolnshire, approximately 50 percent of which were dispensing. Using a postal questionnaire, the authors asked respondents to agree or disagree with the following statement: “A GP should provide a patient with a more expensive prescribed drug if the patient demands it, even if that drug is no better than a cheaper alternative.” Of the replies, 47% strongly disagreed, 39% disagreed, 6% were neutral, 7% agreed and 1% strongly agreed. Clearly, this data shows that most dispensing and non-dispensing doctors believed that using an expensive drug when a cheaper drug is available is unethical. This result suggests that either Goldacre et. al. have misunderstood the moral stance of dispensing doctors or GP opinion has changed radically in the last 20 years.
Goldacre et. al. claim to examine whether dispensing practices are more likely to prescribe high-cost options. Following their analysis, they conclude that doctors in dispensing practices are more likely to prescribe higher cost drugs. By focusing on prescribing, the authors can report that any high costs observed were the results of doctor drug choice decisions. Despite their stated aim, Goldacre et. al. did not use data on doctor prescribing. Instead, they analysed data from the Health and Social Care Information Centre (HSCIC) NHS England Primary Care Prescribing dataset for prescriptions dispensed by pharmacies and dispensing practices. Data on prescribing decisions are only available from practice information systems and are not reported to the health service. Information on items dispensed is only available because it is generated from NHS remuneration claims made by chemists and dispensing practices. To be factually correct, the authors should not report that they are studying doctor prescribing, but they should state that they are analysing patterns of items dispensed. In most studies of NHS drug costs, this distinction may be irrelevant. Because Goldacre et. al. focus on dispensing practices, this distinction is important because not all items prescribed are actually collected by patients. In response, the authors should have referred to dispensing - not prescribing - in their title, intentions, results and conclusions.
“Primary non-compliance” (also called “primary medication nonadherence”) occurs when new medications are prescribed, but patients do not collect the products ordered for them. Evidence suggests that the average primary non-compliance rate is about 5-6% of all primary care prescriptions issued, with a variance between 0.5% and 57.1% depending upon time and place.[15.16] According to the literature, the primary reasons for non-fulfillment are patient concerns about the recommended medicines, lack of perceived need for the prescription and affordability issues. When providing NHS services, dispensing doctors will have much lower rates of primary non-compliance than non-dispensing doctors. This is because they tend to give actual medicines directly to patients before they leave their surgeries rather than issuing prescription forms that may (or may not) be cashed later at a chemist shop elsewhere. Given the volume of medicines prescribed but not collected, the difference in dispensing volumes observed by Goldacre et. al. could simply be primary non-compliance in action. In response, the authors should have adjusted their work to account for non-adherence rates before attempting to calculate the degree of inappropriate prescribing behaviour by doctors.
Dispensing doctors may respond to financial incentives and inflate their prescribing costs for personal gain. However, using flawed research methods and limited data, Goldacre et. al. have not provided robust, reliable evidence to prove that this is case. Because they use only one month’s cross-sectional data, the authors cannot tell us something universal about dispensing doctors that extrapolates across time and place. As a result of problems with their methodological approach, little faith can be put in their conclusion that doctors in dispensing practices respond to a financial conflict of interest. On the contrary, the results they observed may be caused by a range of other factors that they could have modelled but chose to ignore. In particular, the procedure of dispensing practices supplying drugs instead of prescription forms to patients may be the real, untold story that their analysis suggests, but that Goldacre et. al. did not expose.
1. Goldacre, B., Reynolds, C., Powell-Smith, A., Walker, A. J., Yates, T. A., Croker, R., & Smeeth, L. (2019). Do doctors in dispensing practices with a financial conflict of interest prescribe more expensive drugs? A cross-sectional analysis of English primary care prescribing data. BMJ Open, 9(2), e026886.
2. Hinchliffe, H. (1959). Final Report of the Committee on the Cost of Prescribing.
3. Morton-Jones, T. J., & Pringle, M. A. (1993). Prescribing costs in dispensing practices. BMJ, 306(6887), 1244-1246.
4. Baines, D. L., Tolley, K. H., & Whynes, D. K. (1996). The costs of prescribing in dispensing practices. Journal of clinical pharmacy and therapeutics, 21(5), 343-348.
5. Baines, D. L., & Whynes, D. K. (1997). Over-the-counter drugs and prescribing in general practice. Br J Gen Pract, 47(417), 221-224.
6. Avery, A. J., Cole, T. J., Rodgers, S., Heron, T., Crombie, R., Whynes, D., ... & Petchey, R. (2000). A prescription for improvement? An observational study to identify how general practices vary in their growth in prescribing costsCommentary: Beware regression to the mean. Bmj, 321(7256), 276-281.
7. Gibbard, A., & Varian, H. R. (1978). Economic models. The Journal of Philosophy, 75(11), 664-677.
8. Baines, D. L. (2001). A healthy disposition? The use and limitations of the characteristics approach to general practice research. Br J Gen Pract, 51(470), 749-752.
9. Whynes, D. K., Baines, D. L., & Tolley, K. H. (1996). Explaining variations in general practice prescribing costs per ASTRO-PU (age, sex, and temporary resident originated prescribing unit). BMJ, 312(7029), 488-489.
10. Baines, D. L., Brigham, P., Phillips, D. R., Tolley, K. H., & Whynes, D. K. (1997). GP fundholding and prescribing in UK general practice: evidence from two rural, English Family Health Services Authorities. Public Health, 111(5), 321-325.
11. WILCOCK, M. (2001). Dispensing doctors and non‐dispensing doctors—a comparison of their prescribing costs. International Journal of Pharmacy Practice, 9(3), 197-202.
12. Lim, D., Emery, J., Lewis, J., & Sunderland, V. B. (2009). A systematic review of the literature comparing the practices of dispensing and non-dispensing doctors. Health policy, 92(1), 1-9.
13. Baines, D. L., Tolley, K. H., & Whynes, D. K. (1998). The ethics of resource allocation: the views of general practitioners in Lincolnshire, UK. Social science & medicine, 47(10), 1555-1564.
14. Beardon, P. H., McGilchrist, M. M., McKendrick, A. D., McDevitt, D. G., & MacDonald, T. M. (1993). Primary non-compliance with prescribed medication in primary care. Bmj, 307(6908), 846-848.
15. Gadkari, A. S., & McHorney, C. A. (2010). Medication nonfulfillment rates and reasons: narrative systematic review. Current medical research and opinion, 26(3), 683-705.
16. Schafheutle, E. I., Seston, E. M., Hassell, K., Noyce, P. R., & Nicolson, M. (2001). Non‐dispensing of NHS prescriptions in community pharmacies. International Journal of Pharmacy Practice, 9(S1), 14-14.
Whilst a systematic review of dispensing doctors found increased prescription and less often in generics [Health Policy. 2009 Sep;92(1):1-9] and concurred with findings from this study, the Australian census data on its dispensing doctors reportedly contradict the international trends [Med J Aust. 2011 Aug 15;195(4):172-5]. In Australia, doctors are only allowed to dispense under its National Health Act where there is no convenient and efficiency access to pharmacists. Community pharmacy owners in Australia are largely represented by the Pharmacy Guild of Australia which is a major lobbyist and advocate with the Australian Government, see for instance the successive Community Pharmacy Agreement. There are other factors identified in the mixed methods study which may explain the better quality of prescribing in Australia as compared to the English findings. Is there cross-jurisdictional lesson to be learned?