Deciding which Control to Use Statistical Process Control In contrast to the there are numerous ways to construct control charts. The decision of which to use hinges on the type of you have collected. ‘A one page book’ There are basically 2 types of data - ‘Count what is countable, measure what is measurable. What is not measurable, make measurable'." Discrete or Attributes data Galileo Galilei Continuous or Variables data Counts of event that can be aggregated This one page book covers some of the basic theory regarding the use of Control Charts This is data that can take on differing into discrete categories - acceptable for improvement. It builds on the topics summarised in the MCA ‘Run Charts’ one values on a continuous scale, over vs. non acceptable, infected vs. not time—e.g. length of stay, total number page book and is best used after those concepts have been practiced and are infected, late vs. on time, Attendance of discharges, attendances etc understood. Run charts are convenient and are easy to construct and understand but vs. non attendance they are not as sensitive in detecting special causes as control charts are. The table summarises some of the differences to consider between Run Charts and Control Charts.

Continuous Data Run Charts vs. Control Charts At its most simplistic level the appropriate chart to use for continuous data is the XmR or Run Charts Control Charts single measurement chart. The previous charts in this book are of this type and this is the one most commonly used in healthcare quality improvement. There are other types of Simple & easy to understand and interpret More complex, not just one type to consider charts for continuous data where each subgroup (unit of time on the x axis) has more than Easy to create with paper or in Excel Need a special template or special software 1 observation called X bar and S Charts. Discrete Data Less sensitive and can miss some special More sensitive and powerful tool— The P-chart (P—Percent or Proportion) is the most easily understood and most often used cause signals provide additional tests control chart for discrete data. It plots the percentage of events over time. On this chart No measure of the amount of variation or Control limits show the precision and more the control limits vary for each data point as the denominator is different for each point. ‘precision’ in the data accurately predict future behaviour

Once again there are Minimum 10—15 data points Minimum 25 data points other types of chart that can be used for discrete The anatomy of a Control Chart data, such a the U Chart which is useful for data best measured as a ratio, Upper Control or a C Chart used for Limit discrete data where the

denominator is fairly constant. The resources below have more on these types of charts and their (y) variable A Lower Control application Limit Resources Further information about using Control Charts — Carey (2003) - Improving Healthcare with Control Charts Carey & Lloyd (1995) - Measuring Quality Improvement in Healthcare Time (x) 1 4 ©2014 Trustees of Dartmouth College, Sheffield Microsystem Coaching Academy Note that the centre line is the Mean, as opposed to the on a Run Chart. Basic Control Chart Theory Basic Tests for Detecting Special Cause Signals on a Control Chart

The statistical principles behind the development of control charts were first developed by 1. One Point outside of the Control limits (either Lower or Upper) Walter Shewhart in the 1920’s. Shewhart realised that some variation (Common cause) was part of the normal function of life and some variation was not normal but due to assignable’ causes (Special cause). He then created the concept of control limits to help differentiate between the two types of variation Control limits are grounded in a simple statistical principle - data exhibiting common cause variation will form a curve as below. Once you know the basic shape of this distribution you can use it a frame of reference for future data. Comparing the variation present in our data to the predictable pattern present in a common cause system is the underlying theoretical principle of control charts.

The (SD) or ‘sigma’ UCL of this data is a measure of the dispersion of the distribution. 3 3sSD The control limits on Mean a chart correspond 3 3sSD to +/- 3 SD from the LCL mean as shown here - Statistically there is a 0.3% of having a data point falling beyond 99.7% will be within 6 SDs these limits. 2. Eight or more successive values on the same side of the centre line When a data point ( Shift) 0.3% will be outside 6 SDs does exceed the in a normal distribution UCL or LCL this is an indication of a special cause. We can also predict that if nothing in the system changes 99.7% of future points will be within the two control limits Mean – 3SD Mean Mean + 3SD 3. Six or more values in a row continually increasing or decreasing 6 SD ( Trend) The standard deviation (SD) or sigma of a control chart is not calculated in exactly the 3 same way as the standard deviation of a distribution. In fact there are different formulae Note that there are some other more detailed tests for detecting special cause signals but for calculating the control limits for the different types of control charts introduced on page for most healthcare processes the 3 tests above will usually be adequate. 4. However specialist computer software such as WinChart deals with this for you. (See Carey & Lloyd 1995 for more detail)