Control charts-Statstical Quality control

Control chart

1. better information --> better decisions --> better health1
Statistical Process Control –An Overview
Public Health Intelligence Training Course –March 2011

2. better information --> better decisions --> better health2
Introduction
•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.

3. better information --> better decisions --> better health3
Introduction
•The questions we need to answer are:
–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.

4. better information --> better decisions --> better health4
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
Introduction to Control Charts

5. better information --> better decisions --> better health5
Types of Variation
1. 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.
2. Special cause variation can be exhibited within or outwith control limits i.e trends,
step functions, drift etc.
3. 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.

6. better information --> better decisions --> better health6
Definitions
•A process is said to be ‘incontrol’ if it
exhibits only “common cause” variation.
–This process is completely stable and predictable.
•A process is said to be ‘out ofcontrol’ if it
exhibits “specialcause” variation.
–This process is unstable.

7. better information --> better decisions --> better health7
Basic control chart layout
Centre line
(usually mean
or median)0.00%
2.00%
4.00%
6.00%
8.00%
10.00%
12.00%
14.00%
Apr-
08
May-
08
Jun-
08
Jul-
08
Aug-
08
Sep-
08
Oct-
08
Nov-
08
Dec-
08
Jan-
09
Feb-
09
Mar-
09
Apr-
09
May-
09
Jun-
09
Jul-
09
Aug-
09
Sep-
09
Oct-
09
Nov-
09
Dec-
09
Jan-
10
Feb-
10
Mar-
10
Date
Under Run Hours as a % of Allocated Hours
Zone A
Zone B
Zone C
Zone A
Zone B
Zone C
Upper control
limit
Lower control limit
Warning zones

8. better information --> better decisions --> better health8
Types of control charts
•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

9. better information --> better decisions --> better health9
Types of control charts
1.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.
2.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

10. better information --> better decisions --> better health10
Using control charts and SPC methods
•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

11. better information --> better decisions --> better health11
Examples –Run Charts & Control Charts
Run Charts:
•Display of data points plotted in chronological order
•Ideally 25 data pointsare required
•Centre line (meanor median) is included to identify types of variation
Control Charts:
•A Run chart plus control limitsand warning limits (optional)
•Control limitsare set at 3 standard deviationsabove and below the mean
Warning limitsare set at 2 standard deviations above and below the mean
•These limits provide an additional toolfor detecting special cause variation

12. better information --> better decisions --> better health12
Run chart –Time to work08:24
08:38
08:52
09:07
09:21
09:36
Mon
Tues Wed
Thurs
Fri
Mon
Tues Wed
Thurs
Fri
Mon
Tues Wed
Thurs
Fri
Mon
Tues Wed
Thurs
Fri
Mon
Tues Wed
Thurs
Fri
Time arrived at work

13. better information --> better decisions --> better health13
Run Chart –Out of control08:24
08:38
08:52
09:07
09:21
09:36
09:50
10:04
10:19
10:33
10:48
Mon
Tues Wed
Thurs
Fri
Mon
Tues Wed
Thurs
Fri
Mon
Tues Wed
Thurs
Fri
Mon
Tues Wed
Thurs
Fri
Mon
Tues Wed
Thurs
Fri
Time arrived at work

14. better information --> better decisions --> better health14
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.0.2
0.7
1.2
1.7
2.2
1 2 3 4 5 6 7 8 910111213141516171819202122232425
Blood Samples
Micrograms/ML
SERUM GENTAMICIN LEVELS -TROUGH

15. better information --> better decisions --> better health15
ADVERSE DRUG REACTIONS
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.0
1
2
3
4
5
12345678910111213141516171819202122232425
Week Number
Number of Adverse Drug
Reactions

16. better information --> better decisions --> better health1675
80
85
90
95
100
105
110
115
120
1 2 3 4 5 6 7 8 910111213141516171819202122232425
INDIVIDUAL PATIENT READINGS
MEASUREMENT
DIASTOLIC BLOOD PRESSURE
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.

17. better information --> better decisions --> better health17
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.0
1
2
3
4
5
6
7
123456789
10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
Time
Observations

18. better information --> better decisions --> better health18
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.0
10
20
30
40
50
60
70
1 2 3 4 5 6 7 8 910111213141516171819202122232425
COLPOSCOPY PATIENTS
TIME IN DAYS
Mean = 35
ABNORMAL PAP TEST FOLLOW -UP PROCESS
UCL

19. better information --> better decisions --> better health19
Determining if the process is out of
control –Control Rules
•One or more points falloutside of the control limits
•8 or moreconsecutive points on same sideof centre line
•7successivepoints all going up or down
•14 consecutivepoints going up and down alternately
•2 out of 3consecutive points in zone Aor beyond
•4 out of 5consecutive points in zone Bor beyond
•15consecutive points in zone C(above and below)

20. better information --> better decisions --> better health20
Answers to Handout

21. better information --> better decisions --> better health21
8+ points on same side of centre line

22. better information --> better decisions --> better health22
16 points going up and down

23. better information --> better decisions --> better health23
Common cause

24. better information --> better decisions --> better health24
Common cause

25. better information --> better decisions --> better health25
7 pointsdecreasing

26. better information --> better decisions --> better health26
4 out of 5 points in zone B or beyond0.00%
2.00%
4.00%
6.00%
8.00%
10.00%
12.00%
Apr-0 8 M ay-0 8 J un-08 J ul-08 Aug -08 Sep -08 O ct-0 8 N ov -08 D ec -08 J an-09 Fe b-09 M ar-09 Apr-0 9 M ay-0 9 J un-09 J ul-09 Aug -09 Sep -09 O ct-0 9 N ov -09 D ec -09 J an-10 Fe b-10 M ar-10
Date
Under Run Hours as a % of Allocated Hours

27. better information --> better decisions --> better health27
Acting on Variation
Special or common cause variation?
CommonSpecial
Is the process capable?
Yes No
Search for and
eliminate
differences in causes
between data points Do
nothing
Search for and eliminate
causes common to all
data points

28. better information --> better decisions --> better health28
Management of Variation
Special Cause Variation Common Cause Variation
•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.
•Recognise that the capability will not
change unless the process is changed.
•Work to reduce variation due to
common causes
•Do notreact to individual occurrences
or differences between high and low
numbers.
•Change the system to react to
special causes
•Treat every occurrence as a special
cause
Inappropriate Action Appropriate Action

29. better information --> better decisions --> better health29
Summary
•Understandingthe causesof variationhas reformed
industry
•Application to healthcare has provided important insight
to inform improvement
•Effectively highlightsareas meriting further investigation
throughsimpledata presentation

30. better information --> better decisions --> better health30
Chart Instability
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

31. better information --> better decisions --> better health31
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.

32. better information --> better decisions --> better health32
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).

33. better information --> better decisions --> better health33
Elective admissions to any surgical
specialty: overall mortality at 120 days.00
.50
1.00
1.50
2.00
2.50
3.00
0 2,000 4,000 6,000 8,000 10,000 12,000 14,000 16,000 18,000
Number of Patients
Mortality rate(%) at 120 days

34. better information --> better decisions --> better health34
Transurethral Prostactectomy for
benign disease: overall mortality at
120 days.00
2.00
4.00
6.00
8.00
10.00
12.00
0 50 100 150 200 250 300 350
Number of Patients
Mortality (%) at 120 days

35. better information --> better decisions --> better health35
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.

36. better information --> better decisions --> better health36
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.

37. better information --> better decisions --> better health37
Example 2:
lung cancer mortality rates
by local area

38. better information --> better decisions --> better health38
Further information
http://www.indicators.scot.
nhs.uk/SPC/Main.html
http://www.apho.org.uk/
resource/item.aspx?RID
=39445