Effective prescribing at practice level should be identified and rewarded
BMJ 1998; 316 doi: https://doi.org/10.1136/bmj.316.7133.750 (Published 07 March 1998) Cite this as: BMJ 1998;316:750
- Trisha Greenhalgh (p.greenhalgh{at}ucl.ac.uk), senior lecturer
- Unit for Evidence Based Practice and Policy, Joint Department of Primary Care and Population Sciences, UCLMS/RFHSM, Whittington Hospital, London N19 5NF
In the administrative corridors of the European Union there is no longer any talk of cows, sheep, or pigs but only of “grain consuming units” or GCUs. A similar level of bureaucratic jargon has come to surround what used to be known as patients in the British NHS, who, for the purposes of administering the funds that pay for their medication, are now known as “prescribing units” (PUs). Just as one cow consumes as much grain as three or four sheep and therefore counts as several GCUs, so a person over the age of 65, who is said to consume, on average, three times as many prescription items as someone aged under 65, is generally counted as three prescribing units.1 Attempts by statisticians and health economists to explain and refine the prescribing unit2–8 have generally been little read and poorly understood by those with most to gain or lose by the formulas produced. Yet the principle behind the jargon is simple, and the implications of an invalid model for capitation based drug budgets far reaching.
General practitioners in England and Wales are expected to keep the total cost of their drug prescribing within specified limits (indicative prescribing budgets9) allocated by their district health authority (previously the family health services authority) and generally calculated on the basis of their previous year's performance plus a small allowance for inflation and real cost increases (the much criticised “historical allocation formula”). Although there is currently no binding sanction against general practitioners who exceed their indicative prescribing budgets, there is, buried within the small print of the latest NHS white paper, the news that, from 1999, primary care groups will have a unified budget for commissioning, prescribing, and practice administration—in other words, general practitioners' prescribing will be cash limited.10 In the interim, many health authorities, encouraged by central office,11 are introducing a variety of financial incentives for practices to remain within particular targets for total prescribing costs.12
Hence, there is considerable interest in developing a robust mathematical model that successfully predicts legitimate variation in prescribing costs and exposes (with a view to modifying or penalising) idiosyncratic variation. Given the number of potential influences on the total cost of a general practitioner's prescribing (see box) and the passion with which general practitioners have traditionally guarded their freedom to prescribe as they choose, it is small wonder that attempts to produce such a model have so far generated more heat than light.13
Factors that potentially affect total prescribing costs
At national or regional level
National morbidity trends
New therapeutic advances
At health authority level
Local morbidity trends
Specific incentives
Confounding commercial factors (for example, all orders for mail order appliances from a manufacturer operating within one health authority area will register as prescribed within that authority)
Hospital initiated prescriptions
At practice level
Fundholding status
Dispensing status
Training or teaching status
Repeat prescribing system
Specific morbidity (such as proportion of drug addicts)
Deprivation
List inflation
Workload
Use of locums or deputies
Use of practice formulary
At individual prescriber level
Policy of active searching for conditions to treat (such as hyperlipidaemia)
Threshold for treating particular conditions
Adherence to guidelines and protocols (including decision support systems)
Policy for dealing with drug company representatives
Postgraduate education and professional development
At individual patient level
Morbidity (such as conditions that are particularly expensive to treat)
Deprivation
Expectations
Health education
Measures of prescribing cost
It is important to understand the principles behind the different types of research that have tried to unravel the complex influences on general practitioners' prescribing costs. If the focus of the research is the impact of doctor or patient factors on the decision to prescribe or the choice of drug, the unit of analysis must be the individual prescriber (and, perhaps, the individual consultation).14 If the focus is organisational factors (such as fundholding status or use of locum doctors), the unit of analysis must be the practice. If, however, the focus is a demographic variable (such as age or sex), aggregated data from a large geographical area (regional or national) must be used so that the effects of differences in local morbidity, practice organisation, and prescribing behaviour are smoothed out.
Demographic variables
A national sample of 90 practices drawn from 80 health authority areas was used to refine the crude prescribing unit (weighted only by a flat factor of 3 for patients aged over 65) to take account of sex, finer gradations of age, and the proportion of temporary residents (a highly mobile population increases prescribing costs, and patients often register as temporary residents simply because they forgot to bring their tablets on holiday with them).3 In another analysis, data from over 500 practices were used to determine average costs by different therapeutic group according to age and sex.7 The resulting ASTRO-PU (age, sex, and temporary resident adjusted prescribing unit, which used nine different age bands),4 the formula of which has recently been updated to take account of changing broad trends in general practitioner prescribing,15 and STAR-PU (specific therapeutic group age-sex related prescribing unit)7 provide more sophisticated weightings for legitimate variations in costs, especially in practices with unusual demographic or epidemiological features.
These and other national or regional analyses have shown, for example, that, while women receive more drugs (and hence account for a disproportionate prescribing volume), men tend to receive more expensive items (and hence reduce sex differences in total prescribing costs). Children aged under 5 receive twice as many drugs as older children or young adults, but their medication tends to be low cost and of short duration so that they also account for high prescribing volume but not, in general, for high costs. Although patients aged over 65 receive about three times as many prescription items, their medicines are generally more costly, accounting for 4.6 times the prescribing costs of younger adults.3
Morton-Jones and Pringle analysed the effect of 24 demographic, morbidity, and practice variables on prescribing costs by means of a multiple regression model. They concluded that 81% of the variation in net ingredient cost at health authority level per patient was explained by just two demographic variables (number of pensioners and the mobility of registered populations measured by list inflation) and two proxy measures of morbidity (standardised mortality ratios and number of prepayment certificates issued).3 Indeed, the surprising aspect of macro-level analyses like this is how much, rather than how little, of the variability can be predicted by how few indicators of need.
Other macro-level studies have analysed the influence of fundholding. 5 16 17 Their findings are conflicting and contested,6 but, overall, there are few consistent differences in prescribing costs between fundholders and non-fundholders that are not explained by underlying sociodemographic variables. Similar regional or national studies which compared dispensing with non-dispensing practices demonstrated a tendency of the former to prescribe more items and more brand name preparations. 18 19 Deprivation—whether defined in terms of standard deprivation indices, unemployment rates, or proportion of practice population receiving low income benefit—has also been shown to have considerable influence on prescribing costs.20
Individual variables
Once the unit of analysis is narrowed to practice level or below, variations in costs are less readily explained—precisely because the effects of individual doctor and patient factors are unmasked. Adjustments for demographic variables with the ASTRO-PU probably account for about 25% of the variation between practices' costs,4 leaving most of the variation to be explained by local morbidity patterns, practice variables, and doctor-patient variables. Given that genuine variations in morbidity at the practice level are difficult to distinguish from variations in ascertainment of morbidity, patients' expectations, and individual doctors' threshold for reaching for the prescription pad, we should not be surprised when preliminary models seem to raise more questions than they answer.
Capitation based models
In an analysis of the prescribing behaviour of 131 general practitioners in a single health authority, Majeed et al recently attempted to derive a capitation based formula from demographic data and practice organisational factors (such as whether the practice was fundholding, computerised, had more than two partners, etc).21 They found poor correlation between most of these variables and net ingredient cost per patient, and found that a crude correction for age together with the generic prescribing rate explained only about a third of variability in costs between practices.
Majeed and Head conclude, probably rightly, that capitation based formulas should not be used as a substitute for reflection or negotiation when setting budgets at the practice level.22 They justifiably ask for particularly expensive drugs to be omitted from assessment of targets, so that vulnerable groups are not perceived (or treated) as a financial liability. And they rightly point out that rigid enforcement of indicative budgets will create a perverse incentive for general practitioners to eschew the assiduous search for unmet need (such as diabetes or hyperlipidaemia) or the prophylactic treatment of particular conditions (such as asthma).13
Improving the formula for allocating prescribing budgets
In their wholesale rejection of a capitation formula, Majeed and Head fail to take account of many things that are known about prescribing at the micro-level as opposed to the macro-level. An independent report from the Audit Commission identified several factors that showed high variability between practices and through which, if the worst performing practices improved to the level of the best, substantial savings could be made and patient care improved (see box).23 Using multiple regression modelling, Whynes et al found that two morbidity variables—proportion of certificates of payment exemption for prescriptions (a proxy for level of chronic illness) and number of night visits (possibly a proxy for deprivation)—and one doctor related variable (proportion of items prescribed generically) explained 42% of variation between practices in costs per ASTRO-PU.8
Factors identified by Audit Commission for improving general practitioners' prescribing practice
Prescription by generic rather than brand name
Use of a preferred list of drugs (formulary) to ensure that the most effective and cost effective medicine is selected for a particular condition
Reduce prescription of drugs with limited therapeutic efficacy
Reduce prescription in areas where overprescribing is known to occur
Neither Majeed et al nor Whynes et al addressed other factors identified by the Audit Commission, but it would be potentially possible to develop proxy measures within existing data systems for non-use of formularies (such as number of different diuretics or non-steroidal anti-inflammatory drugs prescribed) and to identify marker drugs for prescription of products of low therapeutic efficacy (such as peripheral vasodilators or appetite suppressants) or those for which therapeutically equivalent cheaper alternatives exist. Practices that record diagnostic as well as prescribing data electronically would be amenable to scrutiny of their prescribing patterns for particular conditions, such as the frequency of antibiotic use for minor respiratory infections.
The British National Primary Care Research and Development Centre is currently undertaking preliminary research into the development of quality markers such as these for general practitioner prescribing (M Roland, personal communication). Ideally, a marker drug should have a single specific clinical indication and no clinical reason for differences between practices. In a recent region-wide survey, Roberts et al used specific marker drugs for prescribing of brand name drugs, those of low therapeutic efficacy, and those with cheaper therapeutic equivalents to monitor the impact of a regional prescribing incentive scheme.12
Valid standards for these and other hypothetical quality markers24 in general practitioner prescribing must surely be determined externally (for example, by evidence assisted peer review) rather than simply by measuring what some or all general practitioners currently achieve. Only by directing analysis at particular compounds and therapeutic areas, and perhaps only by measuring health outcomes along with prescribing costs, will effective and efficient prescribing be distinguished from simple cost containment.
Given the variability in needs and expectations within and between practice populations, a truly equitable, all encompassing formula for allocating prescribing budgets is probably impossible. But indirect evidence suggests that it is theoretically possible for health authorities to identify an approximate band within which a practice's prescribing costs should remain. The time is surely ripe for a pilot study to test the feasibility of this notion.
Acknowledgments
I thank Paul Wallace, Andy Haines, Mike Pringle, Martin Roland, and James Mason for advice on this article. The opinions expressed are mine alone.