General Practice

A prescription for improvement? An observational study to identify how general practices vary in their growth in prescribing costs

Commentary: Beware regression to the mean

A prescription for improvement? An observational study to identify how general practices vary in their growth in prescribing costs

Anthony J Avery, senior lecturer ^a, Sarah Rodgers, researcher ^a, Tara Heron, researcher ^a, Robert Crombie, medical adviser ^b, David Whynes, professor of health economics ^c, Mike Pringle, professor of general practice ^a, Darrin Baines, senior lecturer ^d, Roland Petchey, director of health services research ^e. 

^a Division of General Practice, School of Community Health Sciences, University Hospital, Nottingham NG7 2UH, ^b Enigma Medical Systems, Cleethorpes, North East Lincolnshire DN35 0HF, ^c School of Economics, University of Nottingham, University Park, Nottingham NG7 2RD, ^d Health Services Management Centre, University of Birmingham, Park House, Birmingham B15 2RT, ^e Health Management Group, School of Humanities and Social Sciences, City University, London EC1V 0HB

Correspondence to: A Avery tony.avery{at}nottingham.ac.uk

	Abstract

Top Abstract Introduction Method Results Discussion Appendix References

Objective: To identify how some general practices have low growth in prescribing costs relative to other practices.
Design: Observational study.
Setting: Trent region of England.
Participants: 162 general practices: 54 with low growth in prescribing costs, 54 with average increases in costs, and 54 with large increases in costs.
Main outcome measures: Changes in prescribing costs in therapeutic categories in which it has been suggested that savings can be made.
Results: There were significant differences between the three groups of practices in terms of their changes in prescribing costs for almost all the variables studied. For the group of practices with lowest growth in costs the most important factors were reducing numbers of prescription items and costs per item; relatively low growth in the costs of "new and expensive" drugs; increasing generic prescribing; and reducing costs for modified release products. This group of practices did not increase costs as much as the others for lipid lowering drugs (P=0.012) and hormone replacement therapy (P=0.007). The practices with the greatest increases in costs had particularly large increases for proton pump inhibitors, selective serotonin reuptake inhibitors, and modified release products. Compared with the other groups these practices had larger increases in costs for "expensive hospital initiated drugs" (P=0.009).
Conclusion: General practices vary in their growth in prescribing costs in many ways, with growth in costs for "new and expensive" drugs being particularly important.

	Introduction

Top Abstract Introduction Method Results Discussion Appendix References

In 1994 the Audit Commission estimated that the NHS could save up to £425 million a year if general practitioners changed their prescribing habits by controlling the volume of prescribing, increasing rates of generic prescribing, and using expensive products more appropriately.¹ The suggestions were based on extrapolating the prescribing patterns of 50 selected practices to the rest of the country. While the commission might have suggested some potentially useful strategies for cost control,¹ however, there was little evidence that practices actually used such strategies to achieve low growth in their prescribing costs. Also, while the commission commented on high cost prescribing patterns, they did not look at how some practices increased their prescribing costs. Finding out how general practices change their prescribing costs is important for the development of successful cost control strategies.

Previous studies have suggested that general practices can control their prescribing costs by reducing the volume of prescribing^2-5 and the cost per unit of volume ^{2 4} and by increasing their rates of generic prescribing.^2-6 Few studies, however, have looked in detail at the range of cost control strategies suggested by the Audit Commission. ^{7 8}

We examined this issue by comparing three groups of practices characterised by different rates of growth in prescribing costs to identify how some general practices have low rates of growth of prescribing costs relative to others.

	Method

Top Abstract Introduction Method Results Discussion Appendix References

The study was done with data from general practices in the Trent region of England. This region is reasonably representative of the rest of England and Wales in terms of general practice and sociodemographic characteristics.⁹ We interviewed all health authority advisers in the region at the beginning of the study and were satisfied that there were no unusual incentives schemes in operation that might have biased the results.

We did an observational study of changes in prescribing costs between two financial years using anonymised data from all general practices in Trent (n=840). Using prescribing data (PACTLINE), obtained electronically through the Prescription Pricing Authority, we ranked the general practices in terms of their percentage changes in net ingredient costs per prescribing unit (NIC/PU) between the financial years April 1994 to March 1995 and April 1995 to March 1996. We excluded practices with greater than 10% change in list size between the two years and obtained our sample from the 776 remaining general practices.

We sampled practices from the top, middle, and bottom fifths for percentage change in net ingredient costs per prescribing unit between the two years. We calculated that we needed at least 36 practices in each group to detect a 2.5% difference between groups in their change in percentage of items prescribed generically, with a type I error of 0.01 and a power of 0.9. Having established this as a minimum sample size, we decided that our resources were sufficient to allow a 50% margin above this minimum. Accordingly, we took the 54 practices with the lowest percentage growth in net ingredient costs per prescribing unit (group 1). We then found the 54 closest matches for these practices (on the basis of net ingredient costs per prescribing unit in the financial year 1994-5) from practices in the middle fifth (group 2) and from those in the fifth with the greatest percentage increase in costs (group 3).

We analysed changes in overall prescribing variables using PACTLINE data. To do more detailed analysis of prescribing patterns, however, we obtained PACT (Prescribing Analysis and CosT) catalogues for each of the study practices for each year of the study from the 10 health authorities in the region. The catalogues were sent to a company (Enigma Medical Systems) specialising in the analysis of prescribing data for entry on a database.¹⁰ We took this approach because we knew the database software (Optimise) to be extremely flexible in terms of the analysis of changes in prescribing patterns.¹⁰

We concentrated our analysis on drugs and preparations within the chapters of the British National Formulary that have the highest costs in general practice (chapters 1-6 and 10, see box for details).¹¹ Also, we looked at the chapter on drugs used for malignant disease (chapter 8) as we were interested in the costs of expensive drugs that had probably been initiated during hospital treatment. From these chapters we calculated the costs for each practice for each year of the study where the Audit Commission had suggested that savings might be made (see box and Appendix ). The drugs and preparations included in each category were based on information in the British National Formulary and were validated by members of the research team and three independent pharmacists.

Analysis
For the analysis of overall prescribing variables we used prescribing units as the denominator (these give a triple weighting to patients aged 65 years and older and were available with the PACTLINE data).¹ For the analysis of specific drug categories (see box) we used "net ingredient costs per 1000 patients" as the denominator, given that prescribing units and ASTRO-PUs ^{4 12} have not been validated for use across many of these categories.

	Results

Top Abstract Introduction Method Results Discussion Appendix References

Practice characteristics
In 1995-6, in groups 1, 2, and 3 respectively, there were 20 (37%), 21 (38%), and 14 (26%) fundholders⁷ and eight, five, and seven dispensers¹³ (² P=0.31 and P=0.67, respectively). Table 1 shows that there were no significant differences in list size between the groups of practices in 1994-5 or 1995-6. 

Analysis of prescribing variables
Table 1 shows that there were no significant differences between the groups in terms of net ingredient costs per prescribing unit, items per prescribing unit, and cost per item in 1994-5. The practices with lowest growth in costs, however, had a lower generic prescribing rate in this year. When we used analysis of covariance to take account of baseline values there were significant differences between the groups of practices for changes in prescribing variables between the financial years 1994-5 and 1995-6.

	Discussion

Top Abstract Introduction Method Results Discussion Appendix References

Principal findings
This study has shown that general practices with low growth in prescribing costs changed their prescribing patterns in ways that the Audit Commission suggested might bring about savings without detriment to patient care.¹ These practices were also conservative in their uptake of lipid lowering drugs and hormone replacement therapy.

Strengths and weaknesses of the study
This was a large study that looked at changes in prescribing costs in much greater detail than in previous studies. ^{2-4 6} Also, the study was not limited to observing changes in particular groups of practices such as fundholders ^{2 3 5 6 8 14} or dispensers. ^{13 15}

Comparison with other studies
Nevertheless, our findings are similar to those of studies that have shown that low growth in prescribing costs is associated with a reduction in the overall volume of prescribing,^2-8 a reduction in costs per unit of volume, ^{2 4} and an increase in generic prescribing. ^{2-6 8} One study showed that fundholders had lower costs than non-fundholders in one financial year for some of the therapeutic areas outlined by the Audit Commission.⁸ Our study showed that general practices have low growth in costs in these therapeutic areas when their overall growth in prescribing expenditure is low.

Implications of the study
This study provides some support for the types of cost control strategy suggested by the Audit Commission.¹ Those involved in managing prescribing costs in general practice should focus on controlling the volume of prescribing, limiting the uptake of new and expensive drugs, controlling the costs of modified release drugs, and prescribing generically when this will result in savings.

Unanswered questions and future research
Two important questions arise from this research. Firstly, what are the underlying reasons for observed changes in prescribing costs? Secondly, what happens to the quality of prescribing when general practitioners have low growth in their prescribing costs?

	Acknowledgments

We thank the health authority pharmaceutical and medical advisers in the Trent region of England who gave us useful comments on our study design and helped us to obtain PACT data, in particular Dr Peter Fitton, who gave invaluable advice in the early stages of the study. Dr John Wilson provided information on the use of "new and expensive" drugs in the region. Linda Stead and Helen Cornfield from Enigma Medical Systems were responsible for the accurate of data entry on the Optimise database. Steve Davies, Brigitte Nicholls, and Phil Dwyer carefully checked the drugs and preparations that we included in the categories shown in the Appendix . Lindsay Groom, Denise Kendrick, and Michael Dewey gave helpful comments on drafts of the paper.

Contributors: AJA conceived the study, wrote the protocol, managed the study on a day-to-day basis, developed the drug categories used in the study, liaised with the health authorities and Enigma Medical Systems, supervised the analysis, and wrote the paper. SR was the main contract researcher on the project on which this paper is based, and she did the statistical analysis together with TH. RC took the drug categories developed by AJA, incorporated these into the Optimise database, and helped with the data analysis. DW, MP, and RP were involved in the conception of the study; they actively contributed to project board meetings, advised on the analysis, and commented on drafts of the paper. DB made an active contribution to project board meetings, advised on the analysis, and commented on drafts of the paper. AJA is the guarantor.

	Footnotes

Funding: Department of Health (as part of the NHS Prescribing Research Initiative).

Competing interests: None declared.

	Appendix: Detailed definitions of drug categories used in the study

Top Abstract Introduction Method Results Discussion Appendix References

Drugs and preparations from the following drug categories (derived from chapters 1-6, 8, and 10 of issue 31 (1996) of the British National Formulary) were used in our analysis.

Combination products
All preparations containing two or more drugs, excluding:

• Those in which clinically important components cannot be prescribed separately: dopa-decarboxylase inhibitors with dopaminergic drugs used in parkinsonism; clavulanic acid in co-amoxiclav; and sulfamethoxazole in co-trimoxazole
• Those in which components are in a dose that could not be prescribed separately but where equivalents could be bought over the counter at a pharmacyfor example, co-codamol, migraleve (these preparations are included in the "Over the counter" section)
• Lisinopril and quinapril preparations (where the combination products were as cheap as the angiotensin converting enzyme inhibitor prescribed alone at the time of the study)

Modified/sustained release preparations
All modified/sustained release preparations with the exception of:

• Adalat MR preparations (because MR preparations are indicated for the treatment of hypertension)
• Diltiazem and felodipine preparations (because no "short acting" equivalent available)
• Products for which the BNF gives a justification (shown in brackets) for the use of a modified release preparation: Theophylline preparations ("the use of rapid release oral theophylline preparations has declined because of the high incidence of side effects associated with absorption"); lithium preparations ("once daily administration is preferred when plasma concentrations [have been] stabilised"); modified release morphine salts (recognised advantages of these preparations, see page 12 BNF); carbamazepine preparations ("use of modified release tablets (Tegretol Retard) also significantly lessens the incidence of dose related side effects"); dopaminergic drugs used in parkinsonism ("modified release preparations may help with "end of dose" deterioration or nocturnal immobility and rigidity")

Drugs of limited therapeutic value
All drugs that the BNF suggests are of limited clinical value, except those for which similar preparations could be bought over the counter at a pharmacy (these appear in the "Over the counter" section).

Over the counter products
All drugs and preparations for which an equivalent could be bought over the counter at a pharmacy excluding enemas, nitrates, and topical non-steroidal anti-inflammatory drugs (these appear in their own section).

Topical non-steroidal anti-inflammatory drugs
All topical non-steroidal anti-inflammatory drugs listed in section 10.3.2 of the BNF.

New and expensive drugs
This section lists:

• Expensive therapeutic groups that showed an increase in costs of over 20% across the Trent region between the financial years 1994-5 and 1995-6: proton pump inhibitors (BNF section 1.3.5); lipid lowering drugs (section 2.12); selective serotonin reuptake inhibitors (section 4.3.3); oestrogens and hormone replacement therapy (section 6.4.1.1)
• A selection of drugs (not included in the above therapeutic groups) that showed an increase in costs of over 20% across the Trent region between the financial years 1994-5 and 1995-6: long acting ₂ stimulants (salmeterol and eformoterol preparations); fluticasone preparations; sumatriptan preparations

Expensive hospital-initiated drugs
Drugs that the research team and three independent pharmacists considered were probably hospital-initiated:

• Drugs used for malignant disease and immunosuppression (BNF, chapter 8)
• Drugs from BNF chapters 1-6 and 10 that were probably hospital initiated and cost over £30 per week at adult dose (according to the BNF): dornase alfa; granisetron; ondansetron; tropistron; zidovudine; didanosine; zalcitabine; ganciclovir; atovaquone; somatropin

	References

Top Abstract Introduction Method Results Discussion Appendix References

(Accepted 4 May 2000)

Commentary: Beware regression to the mean

T J Cole, professor of medical statistics. 

Department of Paediatric Epidemiology and Biostatistics, Institute of Child Health, 30 Guilford Street, London WC1N 1EH

The study shows that prescribing costs increase over time quite differently in general practices with low and high growth in such costs. But how should we interpret this? To what extent are the differences due to the individual practices and how much to random variation? Or to put it another way, are the low growth practices consistently low growth practices or were they just low growth practices in this particular year?

Under the first interpretation the results for the observed year can be generalised to other years and the differences in expenditure can be considered to reflect policy decisions by individual practices. I will call this the "policy" interpretation. The alternative or "noise" interpretation is that prescribing costs increase over time in a broadly random way and are unaffected by policy. The differences in spending between low and high growth practices would therefore represent no more than random noise.

The truth obviously lies somewhere between these extremes of policy and noise, though it is hard to know exactly where. The authors acknowledge this uncertainty and are careful to avoid any suggestion that the low growth practices "control" their expenditure in any sense.

The key to interpretation is the association between baseline costs and growth in costs. In the presence of random variation the two are inversely related because of regression to the mean. Practices with high costs at the start will tend to show the lowest growth in costs and vice versa. Exactly the opposite pattern is to be expected under the policy scenario, with low cost practices having both a low baseline and low growth. If policy is the driving force, baseline and growth in costs should be positively correlated.

In practice, the noise component is always likely to predominate. To minimise its effect the study matches the practices for their baseline net ingredient costs (NIC) per prescribing unit, though this adjusts only partially for regression to the mean. Table 1 shows that in the low growth practices two other facets of baseline costs are consistently worsethat is, net ingredient cost per item is higher and percentage generic prescribing is lower. This looks more like noise than policy, as the authors acknowledge.

But against this, table 2 shows significant trends in the opposite direction for baseline spending on new and expensive drugs, particularly selective serotonin reuptake inhibitors and proton pump inhibitors. The low growth practices not only spend the least in percentage terms, they also show by far the lowest growth (table 3), and this fits clearly with a policy interpretation.

So on balance it looks as if the main policy difference between the practices lies in their speed to embrace the classes of new and expensive drugs. But these are drugs defined to have increased in cost by more than 20% across the Trent region during the year of study. So the slightly tautologous conclusion is this: that practices spending relatively more on the fasting growing sector of the drugs market show the highest growth in costs.