Statistical Process Control An Overview Public Health Intelligence

Statistical Process Control – An Overview Public Health Intelligence Training Course – March 2011 1 better information --> better decisions --> better health

Introduction • • • 2 Public health practice commonly makes comparisons between areas, groups or institutions. Methods based on ranking, e. g. league tables, percentiles, have a number of flaws. Ranking makes the assumption that differences between organisations are the results of better or poorer performance. It takes no account of inherent system differences. Just because institutions produce different values for an indicator, and we naturally tend to rank these values, doesn’t mean we are observing variation in performance. All systems within which institutions operate, no matter how stable, will produce variable outcomes due to natural variation. better information --> better decisions --> better health

Introduction • The questions we need to answer are: – – • • 3 Is the observed variation more or less than we would normally expect? Are there genuine outliers? Are there exceptionally good performers? What reasons might there be for excess variation? Alternative methods based on understanding variation may be more appropriate. Statistical process control is one such method and helps to answer these questions through the use of control charts. better information --> better decisions --> better health

Introduction to Control Charts Why use control charts? Control charts are used to monitor, control, and improve system or process performance over time by studying variation and its source. What do control charts do? • Focus attention on detecting and monitoring process variation over time • Distinguishes special from common causes of variation, as a guide to local or management action. • Serves as a tool for ongoing control of a process • Helps improve a process to perform consistently and predictably 4 better information --> better decisions --> better health

Types of Variation 1. 2. 3. 5 Common-cause or process variation is variation that is completely random; special-cause or extra-process variation is non-random i. e. is the result of an event or action. Special cause variation can be exhibited within or outwith control limits i. e trends, step functions, drift etc. In any system variation is to be expected. Using statistical techniques we define the limits of variation (control limits and zones). Interpretation of the data relative to these limits or zones identifies points that are worthy of investigation. better information --> better decisions --> better health

Definitions • A process is said to be ‘in control’ if it exhibits only “common cause” variation. – This process is completely stable and predictable. • A process is said to be ‘out of control’ if it exhibits “special cause” variation. – This process is unstable. 6 better information --> better decisions --> better health

Basic control chart layout Upper control limit Warning zones Zone A Zone B Zone C Zone B Zone A Centre line (usually mean or median) Lower control limit 7 better information --> better decisions --> better health

Types of control charts • • 8 Control charts are plots of the data with lines indicating the target value (mean, median) and control limits superimposed. The common types are based on statistical distributions: – Poisson distribution for counts, rates and ratios; e. g number of violent crimes, number of serious accidents – Binomial distribution for proportions; e. g where the response is a category such as success, failure, response, non-response – Normal distribution for continuous data e. g measures such as height, weight, blood pressure better information --> better decisions --> better health

Types of control charts 1. 2. 9 Conventional control charts (run charts) – The indicator of interest is plotted on the y-axis, against time or the unit of analysis on the x-axis. – Control charts can be plotted with small numbers of data points although their power is increased with more data. Funnel plots – A type of chart where the indicator of interest is plotted against the denominator or sample size. – This gives it the characteristic funnel shape better information --> better decisions --> better health

Using control charts and SPC methods • • 10 Control charts can help us to present and interpret our information more intelligently. They can be used – To detect unusual or outlying patterns, e. g. poor performance, outbreaks or unusual patterns of disease – In health profiling and assessing levels of performance – To decide whether or not targets are being met – In assessing health inequalities better information --> better decisions --> better health

Examples – Run Charts & Control Charts Run Charts: • Display of data points plotted in chronological order • Ideally 25 data points are required • Centre line (mean or median) is included to identify types of variation Control Charts: • A Run chart plus control limits and warning limits (optional) • Control limits are set at 3 standard deviations above and below the mean Warning limits are set at 2 standard deviations above and below the mean • These limits provide an additional tool for detecting special cause variation 11 better information --> better decisions --> better health

Run chart – Time to work 12 better information --> better decisions --> better health

Run Chart – Out of control 13 better information --> better decisions --> better health

Special Cause Rule Number 1: Shifts For detecting shifts in the middle value, look for eight or more consecutive points either above of below the center line. Values on the center line are ignored, they do not break a run, and are not counted as points in the run. SERUM GENTAMICIN LEVELS - TROUGH 14 better information --> better decisions --> better health

Special Cause Rule Number 2: Trends For Detecting trends, look for six lines between seven consecutive points all going up or all going down. If the value of two or more consecutive points is the same, ignore the lines connecting those values when counting. Like values do not make or break a trend. ADVERSE DRUG REACTIONS 15 better information --> better decisions --> better health

Special Cause Rule Number 3: Zig-Zag Patterns Any non-random pattern may be an indication of a special cause variation. A general rule is to investigate where 14 consecutive points go up and down alternately. DIASTOLIC BLOOD PRESSURE 16 better information --> better decisions --> better health

Special Cause Rule Number 4: Cyclical Patterns A non-random cyclical pattern may be an indication of a special cause variation. For example, a seasonal pattern occurring across months or quarters of the year. 17 better information --> better decisions --> better health

Special Cause Rule Number 5: Points Outside Limits A point or points outside control limits is/ are evidence of special cause. Control limits are calculated based on data from the process. ABNORMAL PAP TEST FOLLOW-UP PROCESS Mean = 35 UCL 18 better information --> better decisions --> better health

Determining if the process is out of control – Control Rules • One or more points fall outside of the control limits • 8 or more consecutive points on same side of centre line • 7 successive points all going up or down • 14 consecutive points going up and down alternately • 2 out of 3 consecutive points in zone A or beyond • 4 out of 5 consecutive points in zone B or beyond • 15 consecutive points in zone C (above and below) 19 better information --> better decisions --> better health

Answers to Handout 20 better information --> better decisions --> better health

8+ points on same side of centre line 21 better information --> better decisions --> better health

16 points going up and down 22 better information --> better decisions --> better health

Common cause 23 better information --> better decisions --> better health

Common cause 24 better information --> better decisions --> better health

7 points decreasing 25 better information --> better decisions --> better health

4 out of 5 points in zone B or beyond 26 better information --> better decisions --> better health

Acting on Variation Special or common cause variation? Special Common Is the process capable? Search for and eliminate differences in causes between data points 27 Yes Do nothing No Search for and eliminate causes common to all data points better information --> better decisions --> better health

Management of Variation Inappropriate Action Appropriate Action Special Cause Variation 28 • Identify and study the special cause. • React to special cause - If it is a negative impact, prevent it or minimise impact. -If it is a positive impact, build into process. • Change the system to react to special causes Common Cause Variation • Recognise that the capability will not change unless the process is changed. • Work to reduce variation due to common causes • Do not react to individual occurrences or differences between high and low numbers. • Treat every occurrence as a special cause better information --> better decisions --> better health

Summary • Understanding the causes of variation has reformed industry • Application to healthcare has provided important insight to inform improvement • Effectively highlights areas meriting further investigation through simple data presentation 29 better information --> better decisions --> better health

Chart Instability is defined as: No. of control rule violations Total no. of points entered • Charts can be ranked according to their instability • Good way of prioritising the charts to investigate • Can be used as an ‘Early Warning System’ to identify problem charts before they become a real issue 30 better information --> better decisions --> better health

Funnel plots • Conventional control charts are used for count data, proportions and continuous variables • Funnel plots are used for discrete/count data (e. g. deaths and hospital admissions) – Can be used for proportions, directly standardised rates, indirectly standardised rates and ratios, and rate ratios. 31 better information --> better decisions --> better health

Example 1: rate of mortality at 120 days following admission to a surgical specialty • In this example each data point is a hospital (all hospitals in NHS Board X are shaded blue). • The number of people admitted to a surgical specialty is represented on the horizontal axis, which essentially means that smaller hospitals appear towards the left hand side of the graph and larger hospitals towards the right. • The proportion of people who died within 120 days of admission to hospital is represented on the vertical axis – the higher up the data point, the higher the rate of mortality would appear to be. • The funnel formed by the control limits (and from which the graph gets its name) is wider towards the left hand side. This is simply so the level of activity (in this case, the number of admissions) is taken into account when identifying ‘outliers’ (i. e. the larger the denominator, the most stable the data points are). 32 better information --> better decisions --> better health

Elective admissions to any surgical specialty: overall mortality at 120 days 33 better information --> better decisions --> better health

Transurethral Prostactectomy for benign disease: overall mortality at 120 days 34 better information --> better decisions --> better health

Issues with control charts • In the “any surgical specialty” example, there are many areas which lie outside the control limits • Such a large number of points outside the control limits is known as overdispersion • It arises when there are large numbers of events, and case-mix or other risk factors (e. g. deprivation) are not accounted for • In this example, the overdispersion is probably due to the variation in procedures covered and different uptake of these procedures across the Scottish hospitals. 35 better information --> better decisions --> better health

How to handle overdispersion? • In performance management, we try to identify differences that can be attributed to differences in organisational performance. • In this case it’s usual to adjust the control limits or the data to eliminate potential sources of variation, such as case-mix and demography. • This has the effect of creating a ‘level playing field’. • In public health practice, we are likely to be interested in such sources of variation for their own sake (lung cancer example). • Rather than eliminate them, we want to draw attention to them and understand the reasons behind them. • We tend not to alter control limits, and display the variation as it actually is. 36 better information --> better decisions --> better health

Example 2: lung cancer mortality rates by local area 37 better information --> better decisions --> better health

Further information http: //www. apho. org. uk/ resource/item. aspx? RID =39445 38 http: //www. indicators. scot. nhs. uk/SPC/Main. html better information --> better decisions --> better health