This presenter has nothing to disclose.

Using Run Charts to

Establish Special
Cause Variation
Carol Haraden, PhD

March 3, 2017
Framework for Clinical Excellence
Patient Safety

Culture
Psychological Accountability
Safety

Leadership
Teamwork &
Communication

Engagement of
Patients & Family
Transparency
Negotiation

Reliability Continuous
Improvement Learning

Learning System &

Measurement

© IHI and Allan Frankel

Coronary Artery Bypass Graft

Mortality Rate (%)

5.9%

1.1%
Jan 13 Jan 14
Coronary Artery Bypass Graft
7

6 CABG Mortality Rate: Clinic I

5

0
Apr

Aug
Jun

Jul
Jan-13

Jan-14
Feb

Oct
Mar

Nov
Sep

Dec
May
Coronary Artery Bypass Graft
7

5 CABG Mortality Rate: Clinic II

0
Apr

Jun

Aug
Jul

Nov
May
Jan-13

Jan-14
Mar
Feb

Sep
Oct

Dec
Coronary Artery Bypass Graft
7

6 CABG Mortality Rate: Clinic III

5

0
Apr

Aug
Jun

Jul
Jan-13

Jan-14
Feb

Oct
Mar

Nov
Sep

Dec
May
 Understanding Data Variation

There are two ways to view data

Unplanned Returns to Ed w/in 72 Hours

Month M A M J J A S O N D J F M A M J J A S
ED/100 41.78 43.89 39.86 40.03 38.01 43.43 39.21 41.90 41.78 43.00 39.66 40.03 48.21 43.89 39.86 36.21 41.78 43.89 31.45
Returns 17 26 13 16 24 27 19 14 33 20 17 22 29 17 36 19 22 24 22
u chart
1.2

1.0

UCL = 0.88

Rate per 100 ED Patients

0.8

0.6
Mean = 0.54

0.4

0.2 LCL = 0.19

0.0

10

11

12

13

14

15

16

17

18

19
1

9
STATIC VIEW DYNAMIC VIEW
Descriptive Statistics Line Chart
Mean, Median & Mode
Run Chart
Minimum/Maximum/Range
Standard Deviation Control Chart
Bar graphs/Pie charts Statistical Process Control (SPC)

7
Kaiser Permanente Improvement Institute
© 2014 Kaiser Foundation Health Plan, Inc. For internal use only.
 Improvement uses static and dynamic data
Static views are suited Dynamic views are
to assess variation at best for measuring
a point in time changes in data
variation

100%
1000

90%

80%
800

Significance Unusual
70%
Processing Time

600 60%

of Factors 50%
Observations
400 40%

30%

200 20%
Sudden
Shifts
10%

0 0%
ll q M ib a HP
De m
pa IB sh
Co To
System

Trends

8 Permanente Improvement Institute

Kaiser
© 2014 Kaiser Foundation Health Plan, Inc. For internal use only.
 Example: Results of New CHF Protocol
(static)

Best Practice
Spread to entire Region!

New CHF
Protocol
Introduced

Readmission Reduced from 30% to 24%!

Kaiser9Permanente Improvement Institute
© 2014 Kaiser Foundation Health Plan, Inc. For internal use only.
 Understanding Data Variation

Same data … dynamic view

New CHF
Protocol
Introduced

Kaiser Permanente Improvement Institute

10
Kaiser Permanente Improvement Institute
© 2014 Kaiser Foundation Health Plan, Inc. For internal use only.
How will we know that a change is an improvement?
1. By understanding the variation that lives
within your data

2. By making good management decisions on

this variation (i.e. don’t overreact to a
special cause and don’t think that random
movement of your data up and down is a
signal of improvement)
Old Way, New Way
Requirement,
Specification or
Threshold

No
action
Reject Action taken
taken
defectives on all
here
occurrences
Better Quality Worse Better Quality Worse
Old Way New Way
(Quality Assurance) (Quality Improvement)

12
Source: Robert Lloyd, Ph.D.
Tabular Data Display
Frozen Section Turnaround Time
(minutes)
16 15 26
8 17 25
25 25 17
7 7 9
9 23
X=16.8 26
16 9 24
24 8 18
16 16 15
17 17 18
Graphical Data Display
Frozen Section Turnaround Time
Run Chart (minutes)
30
28
26 X=16.8
24
22
20
MINUTES

18
16
14
12
10
8
6
4
2
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26

SEQUENCE
Graphical Data Display
Frozen Section Turnaround Time Histogram
(minutes)
5

0
0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
Four Dimensions of Data

SHAPE

CENTER
SPREAD

SEQUENCE
Types of Variation
Common Cause Variation
• Is not ‘good variation’
• Is stable and predictable
• Due to the design of the
process
• Does not mean that the
variation is acceptable
Special Cause Variation
• Is not ‘bad variation’
• Unstable and
unpredictable
• Due to irregular or
unnatural causes-
intentional or
unintentional
• Does not mean that the
variation is acceptable
 18
Your Drive to Work….
• How much time does it usually take at 7:30 AM on a
Monday morning?

• On Tuesday night at 10:00 PM?

• Is this special or common cause variation?

Common Cause Variation
100
90
80

70

60
50

40
30

20

10
0
08

08

08

08

08
8

8
00

00

00

00

00

00

00

00

00

00
20

20

20

20

20
/2

/2

/2

/2

/2

/2

/2

/2

/2

/2
1/

8/

5/

3/

7/
15

22

29

12

19

26

10

17

24

31
3/

3/

4/

5/

6/
3/

3/

3/

4/

4/

4/

5/

5/

5/

5/
Points equally likely above or below center line
There will be a high data point and a low, but this is expected
No trends or shifts or other patterns
Courtesy of Richard Scoville, PhD, IHI Improvement Advisor
Two Types of Special Causes
Unintentional
When the system
is out of control
and unstable

Holding the Gain: Isolated Femur Fractures

Intentional 1200

Minutes ED to OR per
1000

When we’re trying

800

Patient
600

to change the 400

200

system 0

1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64
Sequential Patients

Courtesy of Richard Scoville, PhD, IHI Improvement Advisor

 Common Cause Variation Special Cause Variation

Holding the Gain: Isolated Femur Fractures

1200

Minutes ED to OR per
1000

800

Patient
600

400

200

1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64
Sequential Patients

Normal Sinus Rhythm (a.k.a. Atrial Flutter Rhythm (a.k.a.

Common Cause Variation) Special Cause Variation)

21
Example of Data for Judgment
(Perfect Care Bundles – all aspects of a bundle must
be met in order to receive credit)
Does this tabular display of data help us understand how to
improve care?
Care Region TYD Average
Bundle Average Q1 Q2 Q3 Q4
AMI 79 79 79 81 80 79
CHF 61 56 58 63 62 60
Pneumonia 46 16 16 20 31 20

SSI 52 41 43 54 49 47
Legend
Better than or equal to the Region

Worse then Region Average

CHF: Special Cause or Common Cause?

70

60

50
Bundle Reliability

40

30

20

10

0
1 2 3 4

Quarters
SSI: Special Cause or Common Cause?

70

60

50
Bundle Reliability

40

30

20

10

0
1 2 3 4

Quarters
 25
What is wrong with this chart?
Comparison is region average- is the color assigned based on
best practice or best performance by region even when not best
practice?
Is there enough data to make any decision?
No goal stated- is the goal green or best practice?
What is rewarded? Special cause or common cause?
 Appropriate Management Response to Common & Special 26

Causes of Variation

Is the process stable?

YES NO

Type of variation Only Common Special + Common

Change the process Investigate the origin of the special
Right Choice cause

Treat normal variation as a special Change the process

Wrong Choice cause (tampering)
Wasted resources!
Consequences of Increased (time, effort, morale,
making the wrong money)
choice variation!

Source: Carey, R. and Lloyd, R. Measuring Quality Improvement in Healthcare: A Guide to Statistical Process Control
Applications. ASQ Press, Milwaukee, WI, 2001, page 153.
Attributes of a Leader Who Understands Variation
Leaders understand the different ways that variation is viewed.

They explain changes in terms of common causes and special causes.

They use graphical methods to learn from data and expect others to consider
variation in their decisions and actions.

They understand the concept of stable and unstable processes and the potential
losses due to tampering.

Capability of a process or system is understood before changes are attempted.

Understanding Variation with
Run Charts
 29
How many data points do I need?
Ideally you should have between
10 – 15 data points before constructing a run chart

10 – 15 patients • If you are just starting to

10 – 15 days measure, plot the dots and
make a line graph.
10 – 15 weeks
10 – 15 months • Once you have 8-10 data points
10 – 15 quarters…? make a run chart.
Elements of a Run Chart
6.00 The centerline (CL) on a
5.75
Run Chart is the Median
5.50

5.25
Measure
Pounds of Red Bag Waste

5.00

4.75
Median=4.610
4.50
~
4.25
X (CL)
4.00

3.75

3.50

3.25
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29

Time Point Number

Four simple run rules are used to determine if special cause variation is present
Normal Distribution with Standard Deviations 31
“What is the variation in one system over time?”
Walter A. Shewhart - early 1920’s, Bell Laboratories

Dynamic View UCL

Static View
time

Every process displays variation: LCL

• Controlled variation
stable, consistent pattern of variation
“chance”, constant causes
• Special cause variation
Static View “assignable”
pattern changes over time
Analysis of Run Charts
Special Cause Rule Number 1: Shifts
eight or more consecutive points either above of below the center line (mean or
median). Values on the center line are ignored, they do not break a run, nor are they
counted as points in the run.
SERUM GENTAMICIN LEVELS - TROUGH

Mean = 2.0
2.2
Micrograms/ML

1.7

1.2

0.7

0.2
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

Blood Samples
Analysis of Run Charts
Special Cause Rule Number 2: Trends
Five or more consecutive points all going up or all going down. If the value of
two or more consecutive points is the same, only count the first point and
ignore the repeating values; like values do not make or break a trend.
ADVERSE DRUG REACTIONS
Number of Adverse Drug

Mean = 3.0
5
4
Reactions

3
2
1
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
Week Number
Analysis of Run Charts
Special Cause Rule Number 3: Patterns
Any non-random pattern may be an indication of a special cause variation. A general rule is to
investigate any non-random pattern that recurs eight or more consecutive times.

DIALOSTIC BLOOD PRESSURE

120 Mean = 94.32

115
MEASUREMENT

110
105
100
95
90
85
80
75
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

INDIVIDUAL PATIENT READINGS

 Analysis of Run Charts
Special Cause Rule Number 4: 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

70

60 UCL
TIME IN DAYS

50

40

30

20

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

COLPOSCOPY PATIENTS
Medication Administration Process

45
SHIFT DOWN
Elapsed Time to Administer Medication in
40 Mean = 22.5
35

30
Minutes

25

20

15

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
Medication Sequence
Abnormal Pap Test Follow-up Process

60
PATTERN
Median = 35
50

40
Time in Days

30

20

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
Colposcopy Patients
Process for Obtaining a Stat Consult

6
SHIFT UP
5
Median = 3.75

4
Time in Hours

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
Consult Patients
Process for Admitting from Outpatient Clinic

6
TREND
PATTERN
5

Median = 3.0
4
Time in Hours

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
Patient: Admissions
 P41

Number of Days Between Falls

SHIFT DOWN TREND

Abnormal Pap Test Follow-up Process

60
RANDOM VARIATION
Median = 35
50

40
Time in Days

30

20

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Colposcopy Patients
Take a moment to reflect
on your own work.
What will you incorporate from
this session into your plans?