Public HealthBristol, Shipman, and clinical governance: Shewhart's forgotten lessons
Section snippets
Common-cause and special-cause variation
Consider a process such as writing a signature. Five of MAM's signatures are shown in the left of figure 1. Although these signatures were produced under the same conditions and by the same process, they are not identical. However, although they show variation, the variation is controlled within limits. They are all recognisably the same signature. This kind of variation suggests that a stable process produced the signatures.
In the UK National Health Service, three basic approaches are used to
Variation cannot be eliminated
Shewhart illustrated his concepts by applying them to the best data available to him at the time. This was a data set obtained from an experiment in which almost everything possible was done to obtain perfect results (ie, no variation)—Millikan's Nobel-Prize-winning measurements of the charge of an electron.3 Despite Millikan's best efforts, there was substantial variation in his measurements of the charge of an electron. However, as the control chart (figure 2) of Millikan's data shows, all
Case study 1: Bristol cardiac surgery
A control chart based on data from the UK Cardiac Surgical Register of the mortality rates for children younger than 1 year old during three epochs9 is shown in figure 3. The chart for epoch 1 will be used to explain the interpretation of a control chart.
In epoch 1, the mortality rates for nine hospitals lie within the control limits: common-cause variation. Action to reduce this variation must focus on the underlying process of care common to these nine hospitals. However, two hospitals
Case study 2: Harold Shipman
A control chart (figure 4) of mortality rates for women aged 65 years and older in Thameside and Glossop, UK, during 1992–98 10 shows that in 1992 and 1994, Harold Shipman's mortality rates were within common-cause variation. However, during 1993 and 1995–98, his mortality rates indicated special-cause variation. To reduce special-cause variation, the special cause must be found and removed. Subsequent legal proceedings identified that special cause as being Shipman himself.
Commentators have
Case study 3: IVF treatment
Marshall and Spiegelhalter12 analysed the case-mixadjusted livebirth rate at 52 IVF clinics in the UK (n=24 739 treatment cycles, range of livebirth rate 5–24%). They concluded that league tables were unreliable. No action point emerged from their analysis. In contrast, a control chart (figure 5) with the upper and lower control limits divides the clinics into three groups with guidance for action:
- •
Group A—performance above the upper control limit. Find out why their results are better than
Case study 4: neonatal deaths
Parry and colleagues13 compared mortality for nine neonatal units (n=2671 infants, mortality range 15–28%), concluding that league tables were unreliable indicators of performance. In contrast, a control chart of the neonatal data (figure 6) shows only common-cause variation, suggesting that future improvement is best sought from a fundamental change to the underlying process of care. There are no grounds for taking action on individual neonatal units.
Case study 5: prevalence of coronary heart disease in primary care
The point prevalence of coronary heart disease in a primary group consisting of 16 general practices in Birmingham, UK, was reported (private communication, Birmingham Health Authority, 1999) as 9·67% (2999/3102), with wide variation (1–38%) between practices. A control chart of these data (figure 7) identifies 12 practices within control limits indicating common-cause variation. These practices should be left alone. However, five practices are outside the control limits, indicating
Case Study 6: mortality after fractured hips
Todd and colleagues14 compared differences in mortality after fractured hip in eight hospitals in East Anglia, UK (n=560, mortality range 5–24%). A control chart (figure 8) shows seven hospitals within common-cause variation. Improvement at these seven hospitals can only come from changing the underlying process of care for patients with fractured hip. One hospital had a very low mortality outside the limits of common cause-variation: this mortality rate is therefore likely to have a special
Discussion
These case studies illustrate an important role for Shewhart's approach to understanding and reducing variation. They demonstrate the simplicity and power of control charts at guiding their users towards appropriate action for improvement.
Actions based on Shewhart's approach are subject to two types of mistake.14 Mistake 1 is to treat an outcome resulting from a common cause as if it were a special cause. Mistake 2 is to treat an outcome resulting from a special cause as if it were a common
References (24)
Advanced topics in statistical process control. The power of Shewhart's charts
(1995)- et al.
The NHS's 50 anniversary. Clinical governance and the drive for quality improvement in the new NHS in England
BMJ
(1998) Economic control of quality of manufactured product
(1931)- et al.
Applications of statistical quality control to cardiac surgery
Ann Thorac Surg
(1996) Out of the crisis
(1986)Sample design in business research
(1990)- et al.
The uses and usefulness of probability paper
J Am Stat Assoc
(1949) - et al.
Overview of statistical evidence presented to the Bristol Royal Infirmary Inquiry concerning the nature and outcomes of paediatric cardiac surgical services at Bristol relative to other specialist centres from 1984 to 1995
- et al.
GP probably killed more than 100
The Times
(Feb 1, 2000)
Shipman was a criminal and not a poorly performing doctor
BMJ
Reliability of league tables of in vitro fertilisation clinics: retrospective analysis of live birth rates
BMJ
Cited by (193)
Predictors of hospitalizations for diabetes in Germany: an ecological study on a small-area scale
2019, Public HealthCitation Excerpt :To identify counties with remarkably high crude hospitalization rates, we constructed a funnel plot. Funnel plots are an application of statistical process control and provide two forms of variation: variation due to chance (common cause variation) and variation related to extrinsic factors (special cause variation).26 Based on an a priori specified probability distribution, a funnel plot contains so-called control limits, which are plotted around a target outcome and show the limits of common cause variation.
Early detection of variants of concern via funnel plots of regional reproduction numbers
2023, Scientific ReportsReducing unnecessary diagnostic phlebotomy in intensive care: a prospective quality improvement intervention
2023, BMJ Quality and SafetyVariation in hospital morbidities in an Australian neonatal intensive care unit network
2023, Archives of Disease in Childhood: Fetal and Neonatal EditionPerformance assessment of surgical tracking systems based on statistical process control and longitudinal QA
2023, Computer Assisted SurgeryEconomic analysis of surgical outcome monitoring using control charts: The SHEWHART cluster randomised trial
2023, BMJ Quality and Safety