Specific therapeutic group age-sex related prescribing units (STAR-PUs): weightings for analysing general practices' prescribing in EnglandBMJ 1995; 311 doi: http://dx.doi.org/10.1136/bmj.311.7011.991 (Published 14 October 1995) Cite this as: BMJ 1995;311:991
- Correspondence to: Professor Harris.
- Accepted 19 September 1995
Abstract Objectives: To derive cost comparators for prescribing by English general practitioners in eight specific therapeutic groups, based on age-sex related weightings, and to confirm, from a new dataset, earlier age-sex weightings for overall prescribing (ASTRO-PUs).
Design: Calculations based on one year's prescribing data from selected practices using AAH Meditel software, held on MediPlus by Intercontinental Medical Statistics (IMS, UK and Ireland), and research practices using VAMP software, held on the General Practice Research Database.
Setting: 112 English practices with 739672 patients and 510 British practices with 3126570 patients.
Main outcome measures: Cost based weightings for 18 age-sex groups and for temporary residents for eight leading specific therapeutic groups and for prescribing overall.
Results: The two datasets were similar in age distribution and in the way that prescription numbers were distributed by age-sex band in each therapeutic group. The cost based weightings for specific therapeutic groups showed great variation in the use of these groups for patients in different age-sex groups. When these weightings were applied to the prescribing of practices in two family health services authorities they differed in their power to predict prescribing costs: for cardiovascular and gastrointestinal drugs predictive power was particularly high; for drugs for infections it was particularly low, since these are widely used at all ages and for both sexes. Cost based weightings for overall prescribing derived from the IMS data were similar to those of the ASTRO-PU system even though they were derived by different methods from different datasets.
Conclusions: The weightings (STAR-PUs) offer a sound basis for cost comparisons at the therapeutic group level. Cost-based weightings for overall prescribing derived from the IMS data were reassuringly similar to those of the existing ASTRO-PU system.
Weightings for each specific therapeutic group are needed that allow for the ageand sex structure of each practice's patient population.
The practices in a family health services authority can be more fairly compared with each other or with a local average when costs for each specific therapeutic group have been standardised for age and sex in the population.
A specific therapeutic group age-sex related prescribing unit (STAR-PU) system is likely to be best at accounting for cost differences between practices where the group's weightings show a wide range of values.
Aggregated data from the computer systems of many practices provide a valuable resource for research.
General practitioners' prescribing patterns, as revealed by data reported quarterly by the Prescription Pricing Authority, are largely judged in terms of how they relate to a local average. This is calculated by multiplying the local average cost of one “prescribing unit” by the number of prescribing units in the practice. In the prescribing unit weighting system each patient under the age of 65 counts as one prescribing unit, while older patients count as three. In a previous paper from this unit a more extensive set of weightings (ASTRO-PUs) for overall prescribing was described, incorporating 18 age-sex groups and one temporary resident group.1 These have been accepted as a fairer way of standardising for the effects of demography on a practice's overall prescribing, and they are used in the setting of prescribing budgets.
The effects of demography on prescribing within each specific therapeutic group will differ according to the kinds of condition for which the drugs are likely to be used, but this is currently not taken into consideration. At the moment, a practice's prescribing in every therapeutic group is compared with a local average based on the prescribing unit weighting. Standardising practice populations for individual therapeutic groups, using the 18 age-sex bands of the ASTRO-PU, would allow more valid comparisons to be made. The only published attempt to create comparators at therapeutic group level has been that of Sleator, who used data for one quarter in 1989 for a number of practices in one family practitioner committee area.2
In this paper we report our work to establish sets of weightings, using the demographic bands of the ASTRO-PU system, for eight specific therapeutic groups that together account for about 85% of the volume and cost of general practice prescribing in England. We have used a year's data from a large number of practices throughout England. The weightings are cost based; as the prescribing units are related to specific therapeutic groups and each age-sex band, they are called, acronymically, STAR-PUs.
The data available allowed us also to calculate cost based age-sex related weightings for overall prescribing and to compare them with the ASTRO-PU weightings that had been derived from the prescribing of a different set of doctors, in a different year, by slightly different methods.
The eight therapeutic groups we studied each corresponded to a chapter in the British National Formulary. They were the leading groups in terms of both volume and cost: gastrointestinal system (chapter 1), cardiovascular system (chapter 2), respiratory system (chapter 3), central nervous system (chapter 4), infections (chapter 5), endocrine system (chapter 6), musculoskeletal and joint diseases (chapter 10), and skin (chapter 13).
Data covering one year were obtained from two sources. One was the MediPlus Database assembled by Intercontinental Medical Statistics (IMS, UK and Ireland) from selected practices using AAH Meditel software; a description of this database and its methods of quality assurance is given in the appendix. The other was the General Practice Research Database (formerly the VAMP research database) now held by the Office of Population Censuses and Surveys and fully described elsewhere.3 4 The former covered 112 practices in England with 348952 male and 390720 female patients from April 1993 to March 1994; the latter, 510 practices in the United Kingdom with 1529317 male and 1597253 female patients from April 1992 to March 1993. Figures were obtained separately for each practice for male and female registered patients in the age groups 0-4, 5-14, 15-24, 25-34, 35-44, 45-54, 55-64, 65-74, 75-84, and 85 and over, and also for male and female temporary residents. The General Practice Research Database data provided only the numbers of items prescribed, while those from IMS gave both the numbers of items and their costs. The age-sex distributions of the two samples of registered patients were closely similar to each other and to the 1993 mid-year population estimates for England produced by the Office of Population Censuses and Surveys. The average percentage of temporary residents per practice differed in the two datasets: in the former it was 1.8 for males and 2.5 for females, and in the latter 0.8 for males and 1.0 for females.
Only the IMS data, which covered both items and costs, were suitable for our objectives; but the distribution of items in the data could be pragmatically validated by comparing it with the distribution of items in the larger General Practice Research Database. From each of the two datasets we therefore calculated item based weightings for each therapeutic group and made the appropriate comparisons.
Having derived cost based STAR-PU values for each of the eight groups from the IMS data we calculated their power in accounting for practice prescribing costs in two family health services authorities for which we had detailed data--Bradford and Leeds.
Finally, we used the study data to derive weightings for overall prescribing that would be directly comparable with ASTRO-PUs but calculated on the drug costs of a different group of doctors in a different year. We had the advantage of knowing the costs of all the items prescribed, whereas in the earlier work items had to be costed by extrapolating from the data of two health centres.
The item based weightings derived from the two datasets were closely similar for seven of the eight therapeutic groups, but for the endocrine group there were some differences. Figures 1 and 2 show how they compared for central nervous system drugs (typical of the seven groups) and for endocrine drugs. Even for the latter the two curves are broadly similar, and the divergence occurs only in the older groups.
Analysis of the IMS item-based figures for gastrointestinal drugs, using a linear model, suggested that there was a significant practice effect. We would expect this with other drug groups too, as a result of factors such as differences between practices in deprivation related morbidity and prescription duration. The weightings, therefore, will never be able to explain all of the variation between practices.
In calculating the cost based weightings from the IMS data, we combined the 75-84 year age groups with those for 85 years and over, since the numbers in the latter were small. The values are given in table I. The range was greatest in the gastrointestinal and cardiovascular drug groups, with age as the important factor. Sex difference was particularly pronounced in the endocrine drugs; skin preparations and drugs used for infections, however, did not vary greatly in their level of use across the age-sex groups. These values are the STAR-PU weightings.
We applied them to 73 practices in Bradford and to 123 practices in Leeds. For each drug group we standardised for population size by calculating the cost per patient and the number of STAR-PUs per patient and modelled cost per patient by STAR-PU per patient in a regression equation. Table II shows the r2 values for Bradford and Leeds expressed as a percentage of the variation explained. It looks as though our system is best at accounting for cost differences between practices in the drug groups whose weightings show a wide range of values; these drug groups also happen to be the most expensive.
Table III shows the cost based weightings for prescribing overall and compares them with ASTRO-PU values. There is a reassuring similarity. The two sets are in fact not absolutely comparable, for two reasons. Firstly, the ASTRO-PU data came from VAMP practices, and their coding scheme included some dressings and appliances not included in the IMS data. Secondly, the number of temporary residents in the IMS data was low compared with the national average.
The new sets of weightings allow age-sex standardised costs to be calculated for each practice for eight therapeutic groups. By using these standardised costs practices may be compared with each other or with a local average that is similarly standardised.
The value of any comparator or performance indicator lies in its ability to highlight areas that may require further exploration. The extent of our standardisation produces comparators that do this more sensitively than those based on local averages weighted only by the prescribing unit system. Nevertheless, the origin of our weightings in real prescribing data means that no implications of “goodness” are carried by the comparators derived from them.
Being cost based, our comparators are most appropriately used in looking at costs. We have not given item based weightings because we have previously shown the unreliability and invalidity of using the number of items as a volume measure.5 When they are available, however, the number of defined daily doses could be used for volume weightings.
The weightings for cardiovascular and gastrointestinal drugs show that these two groups are the most demographically sensitive. Using them for comparators will therefore be of particular value in justifying costs to economical practices that have an unusually high proportion of middle-aged and elderly patients--where they may also reveal apparent underspending. In most of the other groups some age or sex effect is present, though to a lesser extent; comparisons with the new local averages are more likely to pick up apparent overspending.
Not surprisingly, the STAR-PU weightings differed greatly for different therapeutic groups; they also differed from the overall weightings provided by either the prescribing unit or the ASTRO-PU system. We believe that they should be used in the quarterly analyses of prescribing in specifictherapeutic groups sent to general practitioners and that local average comparators based on themshould replace those based on the prescribing unit.
Having calculated cost based weightings for overall prescribing from our data, we were pleased to find that they were so similar to ASTRO-PUs, even though they were derived from the prescribingof different doctors, in a different year, by a different method.
Appendix Description of the MediPlus database and methods of quality assurance
The MediPlus database is built up from information collected by Intercontinental Medical Statistics (UK and Ireland) from practices using AAH Meditel's System 5 software. A panel of 140 practices contributes to the database on a long term basis in order to track the drug management of disease over time. The aim is to obtain data from about 500 general practitioners with an identifiable prescribing codenumber, both full time and part time. Data from trainees are included, though theycannot be identified as such because the scripts are written on trainers' prescription forms. The extent to which the panel is representative of UK practices and the methods of assuring quality control are described below.
Information is collected daily via the AT&T Istel network. Practice systems dial in each night to upload their activities, and the data are downloaded early each morning by IntercontinentalMedical Statistics (IMS). At the initial installation an encryption key is generated at random. This is used to encrypt the patient reference number, so that the number by which IMS knows a patient cannot be related to the number used in the practice system.
Data collection conforms to the guidelines of the Royal College of Practitioners and the General Medical Services Committee of the British Medical Association. IMS employs an independent medical adviser to monitor adherence to these guidelines.
All doctors and staff in a panel practice must agree to operate the system, and the national representativeness of the panel is monitored in terms of doctors' age and sex, fundholdhing status, dispensing status, and age structure of the patient population. Table AI shows that, at the time of this study, the percentage distributions of general practitioners by age and sex were very similar in MediPlus and in the United Kingdom as a whole; table AII shows that the age distribution of the MediPlus population was close to that of the United Kingdom in the 1991 census. In MediPlus 34.5% of the practices were fundholders compared with 36% nationally (up to the fourth wave); 27.6% were dispensing, against a national 11.0%.
Practices in the database are regularly scored on 13 quality specifications. The maximum score is 610, and a practice must score at least 500 for its data to be considered suitable for researchpurposes. The specifications are shown in table AIII.
The data obtained for MediPlus are compared with those obtained from a separate UK practice database operated by IMS, the Medical Data Index. This is based on a structured sample of 500 generalpractitioners each calendar quarter: the doctors record their prescriptions and linked diagnoses for one week, with the weeks being spread over the quarter.
In addition, comparisons with information available to the practices on the total number of items dispensed show that MediPlus picks up more than 95% of these.
We thank IMS (UK and Ireland) and the Office of Population Censuses and Surveys for the data used in the study.
Funding The Prescribing Research Unit is funded by the Department of Health.
Conflict of interest None.