Don Berwick and SPC for clinicians
- Perhaps the most respected doctor working on the improvement of quality in health care is the paediatrician Don Berwick, a U.S. citizen, who was made an Honorary Knight Commander of the Most Excellent Order of the British Empire in 2005. In 1989 he co-founded the United States IHI (Institute for Healthcare Improvement, ihi.org).
- And perhaps his most outstanding paper is one he wrote in 1991, titled ‘Controlling Variation in Healthcare: A consultation from Walter Shewart’ (Medical Care, December 1991, Vol.29, No.12). In this paper he explains what he learnt from Walter Shewart, why this learning is important for the care of individual patients, for the systems in which clinicians have to work, and for the integration of health care across organisations. But he also tells us why clinicians will be wary of anything that is about the ‘reduction of variation’ if this is not properly explained and understood. As he says, ‘physicians, buffeted by regulation, fear handcuffs that will deny them sensible courses of treatment. Hospitals fear mindless inspections to see if they are in line with others. The word “control” charges the discussion, and reason flees’.
What exactly is SPC?
- Walter Shewhart developed the concept of SPC (Statistical Process Control) while working at Bell Telephone Laboratories in the 1920’s. The application of SPC enables us to make sense of the data generated by a process and to distinguish between the two types of variation in the process: ‘common-cause’ and ‘special –cause.’ This involves plotting the data over time, and then calculating the position of a central line, and the upper and lower ‘control’ limits. The type of control chart we should use is to a large extent dependent on the type of data we are looking at. A paper published in 2008 by Mohammed, Worthington and Woodall, illustrates the selection and construction of four commonly used control charts. See below.
SPC is about being able to predict the future with confidence
- If a process displays only ‘common-cause’ or ‘controlled’ variation, then the process is predictable. So you should be able to go away on holiday content in the belief that nothing will go wrong while you are away – the future is predictable, based on the SPC analysis of the data relating to the past behaviour of your process. But if the process displays ‘special-cause’ or ‘uncontrolled’ variation, then it will be harder to go on holiday and relax, because there are causes of variation that you have not yet got under control!
Don’t confuse the two types of variation or you will waste your time
- It is vital to be able to distinguish between the two types of variation. When the variation is controlled (i.e. common-cause) you will be wasting your time (or the time of others) if you try to determine the cause of individual variations. This doesn’t mean that there is no work needed to reduce the spread of the common-cause variation; it just means that there is no point trying to learn from looking at just one point (i.e. time period) because it happens to look higher (or lower) than the others. The control chart below displays common-cause variation:
- When the variation is uncontrolled (i.e. special-cause), then it will be sensible to try to determine and remove the cause of uncontrolled variability, e.g. by looking at a data point outside the control limits. The control chart below displays special-cause variation:
Why reducing variation is important in delivering healthcare
- The overall aim is to reduce variation in the process so that the future is predictable. As Berwick says: ‘for [Shewhart], statistical control and scientific method were inseparable’…, and ‘without managing care as a system, we can create local excellence and systemic garbage at exactly the same time: locally proud; globally shamed.’
How to decide which type of SPC chart to use
- Here is a summary of the types of SPC chart that are described in the paper by Mohammed and colleagues:
a. The Xmr chart (the individuals chart)
Continuous measurement (also discrete count) data, normally distributed.
E.g. weight or height measured over time for an individual patient; length of stay or time from referral to surgery for successive patients.
Mean, x = Sum(xi)/n
Limits = x +/- 3(mean of moving range/1.128)
= x +/- 2.66(mean of moving range)
b. The u-chart
Events occur one at a time with no multiple events occurring simultaneously or in the same location; events are independent in that the occurrence of an event in one time period or region does not affect the probability of the occurrence of any other event.
E.g. falls per patient day, harm events per patient day.
Mean, u = Sum(events)/ Sum(opportunties)
For the control limits, apply this overall mean to each of the time periods in turn, using the formula
Limits = u+/- 3sqrt(u/ ni), where ni is the size of the group in that time period.
c. The c-chart
Events occur one at a time with no multiple events occurring simultaneously or in the same location; and are independent in that the occurrence of an event in one time period or region does not affect the probability of the occurrence of any other event. (Poisson data, with lower counts).
The overall mean, c = Sum(ci)/ n.
Limits = c +/- 3sqrt(c), when c >10. When c<=10, a different approach is required. See reference.
d. The p-chart
Discrete data: events are binary, e.g. alive/dead or infected/ not infected, and have a constant underlying probability of occurring, and they are independent of each other.
E.g. proportion of fracture of neck of femur patients who died, for successive time periods; DNA rate.
The overall mean, p = Sum(xi) / Sum(ni) (e.g. x = number of deaths in the period and ni = number of admissions in the period).
For the control limits, apply this overall mean to each of the time periods in turn, using the formula:
Limits = p +/- 3sqrt(p(1-p)/ni), where ni is the size of the group in that time period. These two criteria must hold: (1) 0.1<=p<=0.9; (2) ni p(1-p) > 5 for the lowest value of ni.
Rules for detection of a special cause:
• Data point appears outside control limit;
• A run of 8 or more points on one side of the centre line;
• Two out of three consecutive points appear beyond 2sd on the same side of the centre line;
• A run of 8 or more points all trending up or down;
• A run of 14 points alternating up and down.
• The person interpreting the chart should have insight and knowledge about the process.
Reference: Plotting basic control charts: tutorial notes for healthcare practitioners; MA Mohammed, P Worthington and WH Woodall, Qual. Saf. Health Care 2008; 17:137-145.
The views and opinions expressed in this article are personal and are not necessarily those of Northumbria Healthcare NHS FT.
The NHS: observations of an Information Specialist 