Family Practice Vol. 20, No. 1, 69-73
© Oxford University Press 2003
Clinical Research |
Understanding variation in quality improvement: the treatment of sore throats in primary care
Department of Public Health and Epidemiology, University of Birmingham, Edgbaston, Birmingham B15 2TT, UK.
Correspondence to T Marshall; E-mail: T.P.Marshall{at}bham.ac.uk
Marshall T and Mohammed MA. Understanding variation in quality improvement: the treatment of sore throats in primary care. Family Practice 2003; 20: 69–70.
Received 5 March 2002; Revised 19 June 2002; Accepted 9 September 2002.
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionConclusionReferences |
|---|
Background. In 1988, two practices attempted to improve the prescribing of antibiotics for sore throat. The initiative produced only modest improvements in prescribing practice, a finding the authors found difficult to explain. This paper reanalyses the data from an audit of antibiotic prescribing for sore throat in general practice.
Objective. Our aim was to demonstrate the use of Shewhart control charts and to obtain fresh insight into the variations in clinical practice revealed in clinical audit data.
Methods. We use Shewhart control charts to explore variation in antibiotic prescribing between GPs and to suggest the action most likely to result in improvement.
Results. Using control charts, it is possible to distinguish two categories of GPs: low prescribers of antibiotics and high prescribers of antibiotics. Low prescribers of antibiotics show common cause variation, indicating that their prescribing is a stable process. Among low prescribers, improvement can best be achieved by changing the common underlying process. One high prescriber of antibiotics is affected by special cause variation. Among high prescribers, improvement can best be achieved by investigating the special causes affecting this GP and learning lessons from the findings.
Conclusion. The original improvement effort took the same action on all GPs in both practices. Our analysis suggests that such an approach was unlikely to be successful and that different actions were needed for high and low prescribers. The control charts provide fresh insights on the original data and guide improvement efforts.
Keywords. Antibiotics, clinical audit, general practice, quality improvement, sore throat.
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
Sore throat is a common reason for consulting a GP.1 In clinical practice, throat infections may be managed either with or without antibiotic treatment. There are three reasons why it is regarded as good practice to treat few sore throats with antibiotics: (i) the absolute benefits of antibiotic prescribing in sore throat are modest;2 (ii) antibiotic prescription contributes to the development of resistance;3 and (iii) antibiotic prescribing increases the tendency for patients to consult in the future.4 GPs vary in the proportion of patients to whom they prescribe antibiotics. To improve the clinical management of sore throat, we first must understand the information contained in this variation.
In the 1920s, a physicist called Shewhart was charged with improving the quality of telephones manufactured by Bell Laboratories. He realized that the most efficient action to guide improvement must be guided by an understanding of the causes of variation. To help guide improvement action, he developed a simple graphical method—the control chart—to distinguish between variation with no assignable cause and variation with an assignable cause.5 These were later called common cause (non-assignable) and special cause (assignable) variation. The control chart is a means of visualizing data. It has three lines: the upper and lower lines represent the limits of common cause variation. Data points within the control limits indicate common cause variation. Data points outside of the control limits or unusual patterns in the data indicate special cause variation. Control charts have been widely used in industry,6,7 in laboratory settings8 and in control of communicable diseases.9,10 It has been suggested recently that control charts might find applications in clinical governance.11
In 1988, two practices attempted to improve prescribing for sore throat.12 The intervention they chose was a postgraduate education meeting with a focus on the management of sore throat. At this, prescribers were presented results of local audit and were given individual feedback of their own prescribing. Evidence of the effectiveness of antibiotics for sore throat was reviewed and a prescribing policy was suggested. In addition, prescribers were issued with warnings about medicolegal risk with antibiotics. The intervention was evaluated by analysing the GPs’ prescribing rates. This paper uses control charts to reanalyse and interpret the reported variations in prescribing between GPs.
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
Prescribing data for each GP in the two participating practices were obtained from the original paper.12 Since these are binomial data, they are displayed as control charts in the square root domain, using a method which has been described previously.11 On the horizontal axis, we plot the square root of the number of sore throat consultations where antibiotics were not prescribed. On the vertical axis, we plot the square root of the number of sore throat consultations where antibiotics were prescribed. The mean ratio of consultations where antibiotics are prescribed to those where they are not is a straight line through the origin. For a binomial distribution plotted in this way, the control limits are parallel straight lines above and below the mean.13 Data from before (1987) and after the intervention (1988) are presented separately.
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
The data on which the results are based are shown in Table 1
.
|
Before the intervention—1987
The control charts for 1987 are shown in Figure 1
. It is evident from control chart A that prescribing of antibiotics is affected by a number of special causes. However, the data also show a pattern. Visual inspection of control chart A suggests that there are two clusters of data points: GPs who prescribe more antibiotics than average (high prescribers) and GPs who prescribe fewer (low prescribers). There are high and low prescribers in each practice.
|
In control chart B (Fig. 1
), GPs have been divided into two groups: those who prescribe more antibiotics than the average (high prescribers) and those who prescribe fewer (low prescribers). Control limits have been calculated separately for high and low prescribers (Fig. 1). Visual inspection of control chart B indicates that rates of antibiotic prescribing among the low prescribers (GPs C, D, E, K and N) show common cause variation. Variation between low prescribers is consistent with a stable process. It follows that improvement in this group is best achieved by acting on the underlying process. Control charts are also an aid to prediction. If the underlying process remains stable, we would expect the low prescribers to continue to vary within the same control limits. Visual inspection of control chart B indicates that rates of antibiotic prescribing among the high prescribers (GPs A, B, F, I, J, L, M and O) also show common cause variation. Variation between high prescribers is consistent with a stable process. If the underlying process remains stable, we would expect high prescribers to continue to vary within the same control limits.
After the intervention—1988
The control charts for 1988 are shown in Figure 2. It is evident from control chart A that GP prescribing of antibiotics continues to be affected by a number of special causes. However, do the data continue to show the same pattern? Control chart B (Fig. 2) shows the 1988 data with control limits for high prescribers and for low prescribers calculated from the 1987 data.
|
Visual inspection indicates that the GPs identified as low prescribers in 1987 (C, D, E, K and N) continue to be low prescribers. They continue to show common cause variation within the same control limits. Of the three new GPs, two (G and H) are clearly low prescribers and one (P) could be either a low or high prescriber; there are insufficient data to be sure.
Visual inspection of control chart B indicates that all of the GPs identified as high prescribers in 1987, two (F and O) have left their practice and six (A, B, I, J, L and M) continue to be high prescribers. Five continue to show common cause variation within the same control limits. However, A is now affected by a special cause. Variation between high prescribers is not consistent with a stable process. Improvement among this group can best be achieved by investigation to find out what special causes affect GP A’s prescribing.
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
Stable processes produce common cause variation. Shewhart’s insight was to realize the significance of this for quality improvement. If the variation we see is common cause variation, it is reasonable to believe that we are dealing with a stable process. This stable process will continue to produce the same variation. To improve a stable process, we must make fundamental changes to the process itself.
The variation produced by stable processes is not unlimited. It is described by a mathematical theorem called Tchebycheff’s inequality. Tchebycheff’s inequality defines the probability that the next data point from a stable process will lie within t sigma limits of the mean. Shewhart used Tchebycheff’s inequality to design a practical tool to set a practical and economic limit on common cause variation. By this, he meant a limit at which it was reasonable to look for external (special) causes. This is the control chart. He concluded that the most efficient choice of control limits was 3 sigma limits on either side of the mean. Sometimes common cause variation can produce data points outside the control limits. Sometimes special cause variation can produce data points within the control limits. However, data points outside the control limits are sufficient evidence of special cause variation to take action: to look for external factors acting on the process. Additionally, data points within the control limits are sufficient evidence that we are dealing with a stable process.
Special cause variation is important for two reasons. First, these data points are not part of the same process as the rest of the system. They need to be dealt with differently. Secondly, and more importantly, they are signals. Common cause variation is the product of a myriad of interacting factors. How can we determine which factor is critical for improving quality? How can we learn how to improve the process? Special cause variation gives us a clue. Whatever factors are different for these data points are clearly of some importance. They are our clue to improving the system. Do these insights hold true in a clinical setting?
Patients clearly vary in the frequency with which they develop bacterial or viral sore throats. The threshold at which different patients decide whether to consult also varies. There is intrinsic variation in the consultation process by which GPs elicit patients’ symptoms, concerns and expectations. There is also variation in the diagnostic process: the way in which GPs interpret these factors. We therefore expect to observe variation between GPs’ rates of antibiotic prescribing. This variation is produced by the interaction of many factors. We are unlikely to disentangle the influences of these factors successfully. However, there is also a degree of consistency in the consultation process. When deciding to consult, we would expect patients to be influenced by similar factors, even if they accord them different importance. When diagnosing, we would expect GPs to take account of similar symptoms and to use similar diagnostic categories, even if they attach different weights to each symptom. When recommending treatment, we would expect GPs to consider similar strategies, even if they have different thresholds for using each treatment.
What are the implications of this? The treatment decisions of clinicians using similar consultation, diagnostic and treatment decision-making processes will show variation. We would expect this variation to be consistent with the variation found in a stable process. This is common cause variation. If we observe variation greater than is consistent with a stable process, it suggests that something else is affecting the process: this is special cause variation.
Why does this distinction matter? As we observed, the most efficient action to improve quality depends on whether we are dealing with a stable process or one affected by external factors. A health care process showing common cause variation is intrinsically stable. As long as the health care process remains fundamentally the same, it will continue to show the same common cause variation. To improve a process showing common cause variation, we must fundamentally change the health care process itself. A health care process showing special cause variation is affected by external factors—it is not intrinsically stable. Because they cause special cause variation, these external factors clearly exert an important effect on outcomes. To improve a health care process, the most efficient action is to investigate and find the special causes, to learn lessons and to take appropriate action.
The low prescribers in this example are part of a stable process. The high prescribers are affected by a special cause in the second year. Because of these differences, distinctly different improvement efforts should probably be taken for each group. Without this insight, it is likely that the same actions will be taken for all GPs. However, it is unlikely that the same actions will effect changes in both groups. A single approach is therefore unlikely to produce the desired improvement. The intervention reported was a combination of feedback of prescribing data and an educational meeting. This resulted in a reduction in antibiotic prescribing among the high prescribers (except A), but no change among the low prescribers.12
|
Conclusion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
Control charts are an action-oriented approach to data analysis. They distinguish between common cause and special cause variation because these different causes of variation require different actions. The approach is simple to apply and intuitive. In this example, it provides important insights into the prescribing practices of GPs and provides a guide for action to improve their quality of prescribing. It has important implications for clinical governance and for improving quality of care.
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionReferences |
|---|
1 Office of Population Censuses and Surveys. Morbidity Statistics from General Practice. Fourth National Study 1991. London: HMSO, 1994.
2 Del Mar CB, Glasziou PP, Spinks AB. Antibiotics for sore throat (Cochrane Review). The Cochrane Library, Issue 3, 2001. Oxford: Update Software.
3 UK Antimicrobial Resistance Strategy and Action Plan. London: Department of Health, 2000.
4 Little P, Gould C, Williamson I, Warner G, Gantley M, Kinmouth AL. Reattendance and complications in a randomised trial of prescribing strategies for sore throat: the medicalising effect of prescribing antibiotics. Br Med J 1997; 315: 350–352.
5 Shewhart WA. Economic Control of Quality of Manufactured Product, 1931. Republished 1980 by the American Society for Quality Control Quality Press, Michigan, USA.
6 Deming WE. Out of the Crisis. Cambridge (MA): Massachusetts Institute of Technology, 1986.
7 Deming WE. The New Economics. Cambridge (MA): Massachusetts Institute of Technology, 1994.
8 Chesher D, Burnett L. Using Shewhart p control charts of exter-nal quality-assurance program data to monitor analytical performance of a clinical chemistry laboratory. Clin Chem 1996; 42: 1478–1482.
9 Benneyan JC. Statistical quality control methods in infection control and hospital epidemiology. Part 1: introduction and basic theory. Infect Control Hosp Epidemiol 1998; 19: 194–214.[Web of Science][Medline]
10 Benneyan JC. Statistical quality control methods in infection control and hospital. Part 2: chart use, statistical properties, and research issues. Infect Control Hosp Epidemiol 1998; 19: 265–277.[Web of Science][Medline]
11 Mohammed MA, Cheng KK, Rouse A, Marshall T. Bristol Shipman, and clinical governance: Shewhart’s forgotten lessons. Lancet 2001; 357: 463–467.[CrossRef][Web of Science][Medline]
12 Pitts J, Vincent S. What influences doctors’ prescribing? Sore throats revisited. J R Coll Gen Pract 1989; 39: 65–66.[Web of Science][Medline]
13 Mostellor F, Tukey J. The uses and usefulness of probability paper. J Am Stat Assoc 1949; 44: 174–212.[Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  J. Thor, J. Lundberg, J. Ask, J. Olsson, C. Carli, K. P. Harenstam, and M. Brommels Application of statistical process control in healthcare improvement: systematic review Qual. Saf. Health Care, October 1, 2007; 16(5): 387 - 399. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||