Fun With Control Charts Asaph Rolnitsky, MD
FUN WITH CONTROL CHARTS OR: HOW TO IMPRESS YOUR FRIENDS WITH PROCESS CONTROL
Asaph Rolnitsky, MD Sunnybrook Health Sciences Centre NICU, University Of Toronto MSc candidate, Queen’s University VAQS Graduate
DISCLOSURE • No conflicts to disclose
April 5, 2016
1
Fun With Control Charts Asaph Rolnitsky, MD
OBJECTIVES • To understands what SPC is • To understand a run chart • To understand a control chart • To find tools for use • To have fun
CONTENT • • • • • • • •
April 5, 2016
History SPC simplified Basic definitions Run charts examples Control charts examples Analysis by rules Available tools Demonstration?
2
Fun With Control Charts Asaph Rolnitsky, MD
“YOU IDIOTS!!.. WE’LL
You idiots!GET We’ll never DOWN get this thing NEVER THING down the hole! THE HOLE.”
HISTORY • Walter Shewhart (1891 – 1967), a physicist, engineer, and statistician.
• The father of statistical quality control, the control charts and the Shewhart (PDSA) cycle.
• Working in Bell labs, Shewart developed a methods of measuring, analyzing, and presenting changes in product quality.
April 5, 2016
3
Fun With Control Charts Asaph Rolnitsky, MD
WHAT’S WRONG WITH DESCRIPTIVE STATISTICS?
THE PROBLEM WITH “BEFORE AND AFTER” • Before and after can be very impressive and bring your improvement project to its desired goal: presenting favourable data. • And maybe publication.
• In fact, descriptive statistics (before and after) does not show how you got there, and cannot predict where you’re headed.
April 5, 2016
4
Fun With Control Charts Asaph Rolnitsky, MD
PROCESS CONTROL VS DESCRIPTIVE STATISTICS Belly massage to reduce mor tality in hernia repair patients 12.00%
10.00%
Massage Massageq8h q8h 8.00%
6.00%
4.00%
6.3% 2.00%
3.4% 0.00% 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
Average Average Mortality Average Linear (Average) Linear (Average)
TREND LINE VS SPC
April 5, 2016
5
Fun With Control Charts Asaph Rolnitsky, MD
WHAT ARE CONTROL CHARTS? • Control charts present data dispersion over time. • They show the limits of the process, based on standard deviation of the data and where data is within the limits.
CONTROL CHART- WHY? • To distinguish between common cause and special cause variation. • (Out of control/special cause can be good)
• Center line often the mean. • UCL and LCL similar to standard deviation. • Type of analysis depends on distribution of data.
April 5, 2016
6
Fun With Control Charts Asaph Rolnitsky, MD
Visual display of data over time.
CONTROL CHART: REDUCING HYPOTHERMIA AT ADMISSION
DeMauro Pediatrics 2013
April 5, 2016
7
Fun With Control Charts Asaph Rolnitsky, MD
BELL CURVE (NOT TRUE FOR ALL CHARTS)
99.7% OF POINT IN A SAMPLE FALL WITHIN 3SDS.
April 5, 2016
8
Fun With Control Charts Asaph Rolnitsky, MD before
after
12
10
8
6
4
2
0 Jul-09
Jan-10
Aug-10
Feb-11
Sep-11
Apr-12
Oct-12
May-13
Nov-13
Jun-14
Dec-14
Jul-15
NICU i Chart 16 14.670 14
12
UCL
NICU
10
8 4
6.830 4
6 CL
4
2
0
DATE/TIME/PERIOD
April 5, 2016
9
Fun With Control Charts Asaph Rolnitsky, MD
COMPONENTS OF CONTROL CHARTS Data points
Data Control points limits
Central line
SD lines
COMPONENTS OF CONTROL CHARTS
PDSA1: new order set
target
April 5, 2016
10
Fun With Control Charts Asaph Rolnitsky, MD
ZONES Zone A Zone B Zone C Zone B Zone A
RUN CHART • • • •
April 5, 2016
The simplest chart. Simple to see major trends. Easily drawn, no necessary tool. Easy to present on a scoreboard • How many km I ran • How many calories I ate • How long it took to arrive to work • Etc…
11
Fun With Control Charts Asaph Rolnitsky, MD
RUN CHART • A “run” is a point or a cluster of points in any side of the MEDIAN line (and not ON the median).
• Special cause variation indication: • • • •
April 5, 2016
Too many/few runs (available table) A shift: ≥8 points on any side of the median. A trend of ≥6 (or 7?) up or down. “Astronomical point”
12
Fun With Control Charts Asaph Rolnitsky, MD
•
Special cause: • Too many/few runs? • 13/13. • A shift: ≥8 points? • no • A trend of ≥6? • No • Astronomical point? • ?No
Run Charts
Control Charts
Simple & easy to understand and interpret complex, not just one type to consider More Easy to create with paper or in Excel Need a special template or special Less sensitive and can miss some special
software
cause signals
More sensitive and powerful tool—control
No measure of the amount of variation or
limits
‘precision’ in the data
provide additional tests
Minimum 10 data points
Control limits show the precision and more accurately predict future behaviour Minimum 25 data points
April 5, 2016
13
Fun With Control Charts Asaph Rolnitsky, MD
TYPES OF CONTROL CHARTS •
Each chart looks at different characteristics of your data.
• • •
Weight loss
•
Each has different charts.
Medication errors Proportion of ROP
WHAT ARE WE MEASURING?
April 5, 2016
• • •
Sepsis episodes/month (some months have less patients)
•
Days between sepsis
•
Line insertions between infections
Patients with sepsis/month? (one patient can have>1) Sepsis episodes/100 patient days
14
Fun With Control Charts Asaph Rolnitsky, MD
WHAT TYPE? Discrete counts, different distribution
Continuous variables, normally distributed
WHAT TYPE? Counts for EVENTS, or AFFECTED subjects (ie: intubation attempts vs intubated patients Measurement for one observation (ie: Time to admission)
April 5, 2016
15
Fun With Control Charts Asaph Rolnitsky, MD
Type?
continuous
attribute
What are you counting?
Subjects (“defectives”) i.e.: children that had sepsis
XmR (I)
Events (“defects”) i.e.: Sepsis episodes
P
nP if constant sample or sample=1
U
C if constant sample or sample=1
EXAMPLES:
April 5, 2016
KQI
Type of chart Why?
Temperature at admission
X
Measured continuous variable
Number of blood transfusions given in the NICU
C
Counted (“defects”) or “events”
Sepsis workups, intubation attempts
C
Counted “events”
Abx days/100 patient
U
Counted “defects”, corrected/patient
EBM feeds/Total feeds
P
Counted defects, corrected/variable denominator
TPN days/100 admission days
Np
Counted defects, corrected/constant denominator
Days between line sepsis
T
Counted time between event
IVH>2
C
Counted “events”
Complaints/patient
P
Counted defects, corrected/variable denominator
16
Fun With Control Charts Asaph Rolnitsky, MD
WHY IS THE TYPE IMPORTANT? • Each control chart has specific calculation of control limits. • Choosing the wrong chart may falsely show your process is in control or not. • Some expert suggest to use I chart (XmR) as much as possible.
COMMON CAUSE VS SPECIAL CAUSE VARIATION • Common cause variation is the expected, random variability in data signals. • If process is stable but not optimal, may mean need for redesign. • i.e.: Sepsis rate is stable, but high.
April 5, 2016
17
Fun With Control Charts Asaph Rolnitsky, MD
COMMON CAUSE VS SPECIAL CAUSE VARIATION • Special cause is a signal that is statistically significantly deviating from the previous collection of points. • May mean a change to study, or may mean that a change is indeed causing an effect.
RULES OF SPECIAL CAUSE Westgard
AIAG
Montgomery Western Electric
Healthcare
Control Chart Rules 1. Points above UCL or Below LCL
1
1
1
1
1
1
2. Zone A n of n + 1 points above/below 2 sigma
2
2
2
2
2
2
3. Zone B n of n + 1 points above/below 1 sigma
4
4
4
4
4. n points in a row above or below center line
8
9
7
8
8
5. Trends of n points in a row increasing or decreasing
7
6
6
6
6
6. Zone C - n points in a row inside Zone C (hugging)
15
15
15
15
7. n points in a row alternating up and down
14
14
14
8
8
8
8. Zone C - n points in a row outside Zone C 9. Zone B n points above/below 1sigma; 2 points one above, one below 2sigma
April 5, 2016
NelsonJuran
8
4
18
Fun With Control Charts Asaph Rolnitsky, MD
CONTROL CHART RULES
120%
Rule 1: 1 Point above/Below CL
100%
P chart: Feeding Error Rates Rule 6: 15 Points in zone C (HUGGING)
Rule 5: 6 Points trending Rule 4: 8 Points in one side of CL (SHIFT)
80%
60%
40%
20%
0%
April 5, 2016
Rule 2: 2/3 Points above/Below 2SDs
Date/Time/Period
19
Fun With Control Charts Asaph Rolnitsky, MD
RULES TRANSLATING TO SPECIAL CAUSE VARIATION • The rules give description of when there is low probability that your measurements are due to common cause, and more likely represent special cause.
• The rules translate probabilities to points clusters.
EXAMPLES IN PLAIN LANGUAGE: • The chance to have a signal out of the 3SDs is