Transcript Diagnositcs Presentation

Healthcare Redesign:
Diagnostics
Helen Ganley RN, CM, Cert IV QMA, Adv.Dip.QM, MQIHC
Bounty Brokers Pty Ltd
Disclaimers
All data published in this presentation is fictional
except for published text/figures/tables.
Bounty Brokers P/L accepts no liability for any
information provided or its use by
participants.
"Portions of the input and output contained in
this publication/book are printed with
permission of Minitab Inc. All material
remains the exclusive property and copyright
of Minitab Inc. All rights reserved."
Your Bio
•
•
•
•
Introduction
Novice?
Expert?
Hands Up
–
–
–
–
–
–
Table
Pie Chart
Bar graph
Run chart
Control chart
Control chart + others
Redesign Model: Measurement
Dip. Govt.
PSPGOV50101
Today’s Program
Statistical
Thinking
Statistical
Methods
Improvement and
Problem Solving
Objectives
Following the session, you should be better able to:
• Value the use of dynamic data analysis and display
• Understand that variation exists in all processes
• Monitor a process over time to better understand it
• Determine whether or not processes are ‘in control’
• See the effect of a change in a process
• Provide a more accurate basis for prediction for the
purposes of planning, scheduling, budgeting,
resource allocation, improvement, rewarding, etc…..
Focus of Capability Building
• Apply Deming’s principles
• Monitor and improve core
processes
• Monitor adverse events
– and the systems that produce
them
• Learn to use statistical tools and
techniques
• Turn data into information
• React appropriately to variation
Statistical Thinkers Can…..
Statistical thinkers have skills to:
• Assign limited resources
• Determine if a change (decision) was effective
• Know when and if to ask “What happened”
• Understand the system before targets are set
•
Have confidence in making:
– More accurate predictions
– Decisions to do something
– Decisions to do nothing
5 Problems
1. Limited capacity to appropriately collect,
analyse, interpret, report and act on data.
2. Static data display
3. Focus on the person instead of the process
4. No Common Language
5. Data torturing
Problem 1: Limited capacity to appropriately
collect, analyse, interpret, report and act on data
Type 1 error: Take action or adjust
performance when not
warranted
Risk: Tampering
» Increases variation
within a process
» Wastes resources
» Impacts psychologically
Type 2 error: Take no action when
warranted
Risk: Molehill grows into a
mountain
Risk: Wasting time
Duplicating collecting,
analysing, reporting, reviewing,
communicating and discussing
“new” information that is
already ‘known
Risk: Wasting energy
by looking for explanations of a
perceived trend when nothing
has ‘changed’
Problem 2: Static Data Display
LOS LOS LOS
1
2
3
Audiology
2006 Occasions of Service
Category
Jan-06
Feb-06
Mar-06
Apr-06
May -06
Jun-06
Jul-06
Aug-06
Sep-06
Oct-06
Nov -06
Dec-06
1
1.2
1.6
1.9
2
2.2
2.3
2.5
2.3
2.7
2.9
2.8
2.7
3
2.8
2.9
2.9
3.1
3.6
3.8
3.6
3.4
3.6
4
3.9
4.1
4.1
4.6
4.5
4.8
3.4
3.1
3
3.3
3.2
2.8
2.6
3.2
3.3
3.1
3.4
3
2.8
3.1
2.9
1.9
2.5
2
2.4
2.2
2.6
2.4
2
4.3
3.8
4
3.8
4.2
4.1
3.8
4.5
3
3.6
1.9
3.7
4
3.6
3.5
2.5
4
2.5
3.3
3.9
2.3
3.7
2.6
2.7
4.2
3
1.6
3.3
3.1
3.9
3.3
3.2
2.2
4.2
2.7
2.7
1.1
Problem 3:
Inappropriate People Focus
Focus on the person rather than the
process, by:
• Ranking
• Setting inappropriate goals
• Blaming or giving credit for things over which staff
have little or no control
Problem 4: Data Torturing
Data Torturing: When data analysis goes
beyond reasonable interpretation of the
facts.
Problem 5: No Common
Language
Solution
Statistical Thinking and Methods
Statistical Thinking and Methods
Statistical Thinking and Methods
Statistical Thinking and Methods
Statistical Thinking and Methods
Statistical Thinking and Methods
Statistical Thinking and Methods
Statistical Thinking and Methods
Statistical Thinking and Methods
Statistical Thinking and Methods
Statistical Thinking and Methods
Statistical Thinking and Methods
Statistical Thinking and Methods
Statistical Thinking and Methods
Statistical Thinking and Methods
Statistical Thinking and Methods
Why Statistical Thinking?
Many clinicians and other healthcare leaders
underestimate the great contributions that
better statistical thinking could make toward
reducing costs and improving outcomes.
So convinced am I of the power of this principle
of tracking over time that I would suggest
this: if you follow only one piece of advice
from this lecture when you get home, pick a
measurement you care about and begin to
plot it regularly over time.
You won't be sorry.“
D. Berwick 1995
Common Statistical Traps
•
•
•
•
•
•
•
Average Value approach
Ranking
Poorly Presented Percentages
Trending
Smoothing
Tables
Circling
Average Value Approach
As the average is usually near to a
midpoint set of data, one should expect
to be:
– above average about half of the time
– below average about half of the time
Feel bad half
the time
Feel good half
the time
Comparison to Averages
• Results in a characterization of either “above
average” or “below average”
• Characterizes the world in a binary view:
– “operating okay
– “in trouble”
• Ignores the “dead band” of data on either
side of an average
• Treats every fluctuation as important
2 Bucket Average
What is the difference between
these 2 Patient Groups?
Patient A: 50
Patient C: 140
Patient B: 250
Patient D: 160
Total: 300
Total = 300
Average: 150
Average = 150
Consider if this was waiting to see Dr. in ED?
What is the impact of these results on our patients?
Goals
The Use and Abuse of
Numerical Goals
“I never
use
them”
“ I always
provide one and
let other people
figure out how to
achieve it”
Results in:
•Numerical targets expressed as a single point
•Unfairly holding people (departments) accountable for results they
are incapable of achieving
•Achieving goals at the expense of other parts of the system
•Falsifying numbers
Goal /Target/ Specification
Oriented Approach
Targets (voice of the customer) should be
based on:
•
•
•
•
•
customer expectations
benchmarking
competitive requirements
knowledge of those who will do the work
voice of the process – current system capability
The Deceptiveness of Poorly
Presented Percentage Data
Sample
A 2 patients, 1 died = 50%
B 20 patients, 1 died = 5%
C 200 patients, 1 died = 0.5%
mortality
mortality
mortality
Double the numerator with same sample size
A 2 patients, 2 died
B 20 patients, 2 died
C 200 patients, 2 died
= 100% mortality
= 10% mortality
= 1% mortality
You need to know the denominator
(area of opportunity)
Appropriateness of
Trendlines
• Weight of a baby increases as it gets
older
• Reduction in number of kilometres
driven - predict that, within 12 months,
will be driving minus 350 kilometres
per month.
Trending: 6 Possible
Sequences for 3 Numbers
Upward Trend?
Rebound?
Downturn?
Turnaround?
Setback?
Downward trend?
Statistical Representation
of a Trend
Run Chart: Statistical Representation of a Trend
C7
35
25
15
Index
5
10
15
• A sequence of SEVEN or
more points continuously
increasing or decreasing
(SIX if < 20 observations)
• Omit entirely any points
that repeat the preceding
value:
neither add to the
length of the run nor do they
break it.
Ranking
• Given two numbers if they are not the
same, then one will be bigger.
• Ranking provides managers with a tool
for choosing who goes, and in what
order, as the ship begins to list
Statistical Thinking:
Knowledge of Variation
Statistical Thinking
A philosophy of learning and action
based on the following fundamental
principles:
» all work occurs in a system of interconnected
processes
» variation exists in all processes
» understanding and reducing variation are the
keys to success
4 Approaches to Analysis,
Interpretation & Prediction
• Average Value Approach
• Specification Approach (goal / target)
• Run chart approach
• Shewhart Control Chart Approach
Solution: Our Scientific Method
is Statistical Process Control (SPC)
Graph Title & Date
Individual Value
90
80
70
60
50
40
30
20
10
0
15
East
West
North
1st 2nd 3rd 4th
Qtr Qtr Qtr Qtr
UCL=14.14
10
Mean=5.561
5
0
LCL=-3.016
0
10
20
30
40
Consecutive Observation Number
The control chart is the tool of
choice to appropriately display
variation
Control Charts
• Provides a formal method to detect trends
• Provides credibility and rigour at minimum
costs
• Accepted industry standard with a long
history
• Provides performance objective criteria
• Balances false alarms and failures to detect
– Analogous to circuitry in smoke detectors
Control Chart Approach
3 concepts:
– Variation (special / common cause / structural / offtarget)
– Pattern matching
– Decision Making (optional):
» Do something / Do nothing
» Assign limited resources
» Determine who is/are the “best”
Variation and Pattern
Matching
• Review of process variation when
viewed with the mean can help to spot
patterns of variation which are:
– highly improbable
– non-random
– unnatural
– detectable and therefore assignable
4 Types of Variation
• Off Target
• Common Cause
• Special Cause
• Structural
Variation
Variation is the
constant companion
of any data
If there was no
variation,
we would only need
1 number
Effective Measurement System
For accountability or Improvement:
• Objective, reliable, valid data
• Control for confounding, e.g.. age, casemix
• Use of graphical presentations
• Use of comparative data (over time and between hospitals)
• Applying the scientific method when interpreting results
(hypothesis/experiment/test hypothesis)
• Indication of the magnitude of the expected statistical variation
Recommend/Mandate Control
Charts
Who
What
Australian Council on Healthcare Standards
(ACHS).
Clinical Indicator Users’ Manual 2006
ACHS
Risk Management and Quality Improvement
Handbook. Version 1 2007.
Recommends that control charts be used to display longitudinally,
both the absolute numbers and rates
NSWHealth.
Healthcare Associated Infection: Clinical Indicator
Manual 2008
NSWHealth.
CareSafe Performance Agreement 08/09
NSW to report monthly rates with control charts and EWMA
National Health and Medical Research Council
Pilot Program 2005-2007
Level 3-3 Evidence. Experimental study using a comparative
study without concurrent controls such as an interrupted time
series,
Promotes the use of control charts for 30 indicators
Therapeutic Advisory Group.
Indicators for Quality Use of Medicines in
Australian Hospitals. 2007
Recommends control charts to display data and give information
for decision making.
Incident Mgt. Indicator 13.3: Clinical RIBs reported to NSW Health
“Use control charts as presented in RIRC reports”
Recommend/Mandate
Control Charts
Who
What
Independent Pricing & Regulatory Tribunal.
Framework for Performance Improvement
in Health September 2008
Recommendation 17 ”That NSW Department of Health
investigate models in other health services, such as 's
model of statistical process control charting, and monitor
their impact to see if they are appropriate to adopt in the
future.”
Statistical process control ce-learning course
NSWHealth and Centre for Healthcare
Redesign
SQUIRE.
Standards for Quality Improvement
Reporting Excellence.
Describes analytic methods used to demonstrate effects
of time as a variable (for example, control charts) when
reporting improvement projects, as opposed to
Introduction, Methods, Results, Discussion (IMRAD).
Joint Commission (US)
The healthcare accreditation agency requires that all
organisations submit control charts of their clinical
indicators
Bristol Royal Infirmary (BRI)
Paediatric Cardiac Surgery
In a landmark article The UK Cardiac Surgical Register of mortality
rates for children under one year old was analysed using
control charts. The 1988-90 data showed that Bristol mortality
rate was outside the control limits indicating special cause
variation.
Five years later, data for the period 1991-1995 demonstrated that
BRI was again above the upper control limit. Although external
action to address concerns about paediatric cardiac surgery at
BRI took place in 1998, monitoring using the control
charts could have provided a basis for action some
11 years earlier.
Mohammed A Mohammed et al. Bristol, Shipman and Clinical Governance: Shewhart’s Forgotten Lessons. The
Lancet, vol 357 February 10, 2001
Global Trigger Tool
“Plotting this data on
control charts
will give you useful
information
about trends and special
causes
of variation in harm in
your organisation”.
Griffin FA, Resar RK. IHI Global
Trigger Tool for Measuring Adverse
Events (2nd ed). IHI Innovation Series
white paper, Cambridge,
Massachusetts: IHI
Common Cause Variation
• Is an inherent part of every process:
» chronic - often hidden
• Is random
• Due to regular, natural, or ordinary causes
• Produces processes that are stable or “in
control”
• If only common cause variation exists, we
can make predictions about the process
• Management “plans” for this
There are no lessons to be learned from
comparing high dots to low dots
8 Tests
for
Special
Causes
"Portions of the input
and output contained
in this publication
are printed with
permission of Minitab
Inc. All material
remains the
exclusive property and
copyright of Minitab
Inc. All rights
reserved."
Special Cause Variation
I Chart for C1
1
40
Individual Value
• Due to irregular or
unnatural causes
– acute, often out of the
blue - significant
• Not inherent to the process
• Affects some, but necessarily
all outcomes in the process
• Produces processes that are
unstable or “out of control”
• The process is unpredictable
30
UCL=20.99
20
33
3
10
Mean=7.027
222
0
LCL=-6.936
-10
1
-20
Institute for Clinical Excellence. Blood Transfusion
Improvement Collaborative. Final Report. 2003
0
10
20
30
Observation Number
40
How Much Data?
Standard Deviations from
Baseline
Distance from the Baseline versus
Time needed to Identify major Special Causes (MINITAB)
4
1 Point Outside 3S Control Limits
3
2 of 3 Outside 2 Standard Deviations
4 of 5 Outside 1 Standard Deviations
2
6/7 points in a row up / dow n
1
9 same side
0
0
1
2
3
4
5
Months
6
7
8
9
10
Unlocking the Secrets of
Simple Statistical Methods
    

 
1
x1  x y1  y  x2  x y2  y    xn  x yn  y
r n
sx s y

Tools that Generate
Knowledge for Improvement
Process/system
Improvement
Tools
Collaborative
Work Tools
Planning &
Analysis Tools
Statistical
Thinking Tools
Flowchart
Brainstorming
Affinity
Diagram
Run chart
Control Chart
Cause & Effect
Nominal Group
Technique
Force Field
Analysis
Scatterplot
Pareto chart
Multi-voting
Prioritisation
Matrices
Histogram
Degree of Difficulty
Three Uses of Control Charts
• Evaluate the past
• Evaluate the present
• Predict the range of values likely to see
in the near future (where appropriate)
The Standard Deviation
Mean plus 3s
Average/Mean
Mean minus 3s
6 Sigma
Graph Title & Date
15
Individual Value
UCL
Mean
UCL=14.14
10
Mean=5.561
5
0
LCL=-3.016
0
10
20
30
Consecutive Observation Number
6s
40
Basic Control Chart
Title of Graph
Date
Number being Measured (y axis)
12
UCL=12
10
8
_
X=5.38
6
4
2
0
LCL= minus 1
1
2
3
4
5
6
7
8
9
10 11
Observations in time sequence (x axis)
12
13
A run chart with:
• average (green
horizontal line)
• control limits -three
standard deviations
from the mean (red
horizontal lines):
– upper control limit (UCL) and
lower control limit (LCL),
– or +3 or -3 sigma limits
(+ or - 3.0SL)
Empirical Rule
99-100%
90-98%
60-75%
-3SD -2SD -1SD
Empirical = observed.
Only assumption is that
these data are outputs of a
process
+1SD +2SD +3SD
Xbar
3 sigma limits are not
probability limits - not
based on theory (that
random samples from
an underlying
population would give
this result by chance x
times out of 100)
Control Limits are NOT Confidence
Intervals
The control limits describe the
natural variability of a process
over time and are usually set to
three standard deviations (SDs)
or sigma.
Confidence limits of a
distribution describe the degree
of certainty that a given point is
different from the average score
(populations) – as when
“outlier” performance is
demonstrated using comparison
data.
2 or 3 Standard Deviations?
NEED
ACTION
TAKE ACTION
TAKE NO
ACTION
NEEDS NO
ACTION
I
II
TYPE 1
OVER ADJUST
Take action or adjust
performance when
not warranted
TYPE 11
UNDER ADJUST
No action taken when
action is warranted Torki
Because tampering is such a bad thing, common
control charts have limits set to produce:
•low risk of tampering (type 1 error)
•moderate risk of under-controlling (type 11 error)
What was the Question?
The Choice of Control Chart is
determined by the:
• Research Question:
• Number of falls (x)
• Patients that fell
• Area of Opportunity
Area of Opportunity
MEASURE
AREA OF OPPORTUNITY
Number of bacteria
Agar plate
Number of referrals to Dr.
Day
Number of dents
A car
Miles
Gallon
Number of complaints
Bed days
Annual mortality number
Patients who could die each year
Number of vaginal births
All births
Choosing a Control Chart
Minitab® Statistical Software
Variable /Continuous Data
Normal / non-Normal models
Variable Data - 2 Graphs
• Continuous data has no denominator to
estimate variation - uses own variability:
– Xbar-S Chart
» average subgroup standard deviations
– Xbar-R Chart
» average of subgroup ranges
– I MR Chart
» artificial subgroups created from individual
measurements
Common Cause Variation
Average LOS
Average LOS
U C L=6.6633
6.4
_
X=6.2717
LC L=5.8800
1
2
3
4
5
6
7
O bser vation
8
9
10
11
12
U C L=0.4812
0.45
M oving Range
6.5
6.2
6.0
0.30
UCL=6.6633
6.6
Individual Value
Individual V alue
6.6
6.7
6.4
_
X=6.2717
6.3
6.2
6.1
6.0
0.15
__
M R=0.1473
0.00
LC L=0
1
2
3
4
5
6
7
O bser vation
8
9
10
11
12
Individuals Moving Range (IMR chart)
5.9
5.8
LCL=5.8800
C1
Individuals / X Chart
Individuals Chart:
special cause variation
Time spent waiting
1
1
300
3.0SL=295.5
Minutes
5
200
Mean =142.2
100
0
-3.0SL=-11.23
0
5
10
15
20
25
30
35
Consecutive patients
40
45
Length of Stay
Case study: Is there a
Difference?
Balestracci
Variable
n=
Average
LOS
Standard
Deviation
LOS 1
30
3.027
0.978
LOS 2
30
3.073
0.6680
LOS 3
30
3.127
0.8175
Case study:
Appropriate Analysis?
I Chart of los1
5
1
1
1
Individual Value
4
1
1
1
1
1 1
1
1
UCL=3.595
6
3
2
2
2
2
2
1
1
1
6
2 2
UCL=5.841
LCL=2.458
3
2
I Chart of los3
_
X=3.027
2
1
5
1
1
Individual Value
1
1
1
3
6
9
12
15
18
Observation
21
24
27
30
I Chart of los2
4.5
1
Individual Value
1
5
4.0
5
5
5
UCL=4.119
4
_
X=3.127
3
2
1
LCL=0.412
5
0
3.5
_
X=3.073
3.0
2.5
3
6
6
6
6
5
2.0
1
3
6
9
12
LCL=2.028
1
15
18
Observation
1
21
24
27
30
Balestracci
9
12
15
18
Observation
21
24
27
30
Binomial Model
Defectives
‘p’ & ‘np’
charts
Data Categorised by the
Binomial Model
• Count of occurrences and non-occurrences when the
area of opportunity is known and equal, e.g.:







•
•
Head / tail
Acceptable /not acceptable e.g. audits
Infection/no infection
Full bed/empty bed
Operation / cancellation
Working/broken
Dead/alive
Patient fall/no patient fall
Either/or………Defective/Not………….Fraction
Any Percentage Data=P Chart
Chair-Step Limits
P Chart of Compliance all Elements of Care Bundle
0.6
0.5
Proportion
0.4
0.3
UCL=0.2796
0.2
_
P=0.1613
0.1
LCL=0.0430
0.0
1
3
5
7
9
11
13 15
Sample
Tests performed with unequal sample sizes
17
19
21
23
25
Historical Control Chart
NSW Therapeutic Advisory Group Inc. Indicators
for Quality Use of Medicine in Australian Hospitals.
2007
Where are the Control Limits?
Triage Category 1
Proportion meet DoH Goal
2008-2010
1.50
Sample Count
1.25
__
LCL=1
UCL=1
NP=1
1.00
0.75
0.50
2
4
6
8
10
Sample
12
14
16
18
Poisson Distribution
C charts
U charts
Area of opportunity
may not be
known
Run Chart or Control Chart?
Complaints per Thousand Bed days
0.8
Sample Count Per Unit
0.7
0.6
0.5
0.4
_
U=0.3
0.3
0.2
0.1
0.0
1
6
11
16
21
26
Sample
31
36
41
46
Q:
How high
is too
high?
U Chart
Complaints per Thousand Bed days
0.8
UCL=0.7
Sample Count Per Unit
0.7
0.6
0.5
0.4
_
U=0.3
0.3
0.2
0.1
0.0
LCL=0
1
6
11
16
21
26
Sample
Tests performed with unequal sample sizes
31
36
41
46
Is the Process Capable of
Reaching Target?
Process NOT CAPABLE
of Meeting Goal
Weekly Complications
UCL=60.93
60
Sample Count
50
_
C=41.58
40
Goal
30
LCL=22.24
20
1
3
5
7
9
11
13
Sample
15
17
19
21
23
Process IS
CAPABLE of Meeting Goal
Individual Value
Proportion Patients Would Recommend Happy Hospital
Monthly Sample 50 Patients
90
UCL=90%
85
_
X=83%
80
Goal: 80%
1
75
70
65
22
60
55
5
5
5
2
5
Months
1
LCL=77%
Comparisons: ANOM
Units Transfused by Facility
Sample Count Per Unit
4.0
3.5
3.0
_
U=2.465
2.5
2.0
1.5
1.0
rd
co
n
Co
H
HK
hn
Jo
er
nt
u
H
a
pe
e
N
n
e
ng
a
Or
se
Ba
W
PO
w
Ne
l
ya
Ro
ca
le
st
ts
te
en
itu
t
c
s
in
In
.V
H
t
S
PA
SH
RN
R
Facilities
Clinical Excellence Commission BloodWatch Program
Process Capability
Aim:
Process spread
is smaller
than and
contained
within the
specification
spread
Process Capability of BSL
LSL: 4.0
USL: 8.4 mmol/litre
P rocess Data
LS L
4
Target
*
USL
8.3
S ample M ean
5.74845
S ample N
97
S tDev (Within)
0.476507
S tDev (O v erall) 2.6834
P otential (Within)
C apability
Cp
1.50
C P L 1.22
C P U 1.78
C pk
1.22
O v erall C apability
Pp
PPL
PPU
P pk
0.27
0.22
0.32
0.22
4
8
12
16
20
Harold Shipman
In 2000, Harold Shipman, a general practitioner in
Manchester (U.K.) was convicted of murdering 15 of his
patients and of forging the will of one.
The clinical audit revealed clear evidence of a higher level of
death than would have been expected and not just in the
more recent years. It was concluded that the excess of
death did not appear to be explicable of grounds that
Shipman’s practice served populations with markedly
different demographic or health profiles.
Mohammed A Mohammed et al. Bristol, Shipman and Clinical Governance: Shewhart’s
Forgotten Lessons. The Lancet, vol 357 February 10, 2001
Early Warning of Poor Performance
Harold Shipman Versus Comparative GP Death Rate/1000 Patients
Females aged 75 Years or Above
1973 - 1998
Death per Thousand Patients
300
Comparative GPs
1
Shipman
1
250
UCL=205.5
200
150
_
U=98.3
n=17.5
100
50
n=11
0
1
22 2
LCL=0
1973 1978 1983 1988 1993 1998 1977 1982 1987 1992 1997
Year
Department Health: Harold Shipman's Clinical Practice 1974 - 1998
For women aged 75 years or
over, it was predicted that
Shipman had 177 more deaths
than expected.
A summary of the review of
Shipman’s Clinical practice
found Shipman issued 521
death certificates compared
with the highest number of
any of six comparison
practitioners being 210.
The excess number of deaths
were evidenced from the
first few years of Shipman’s
career as a GP; An excess of
deaths occurred at home or
in his practice premises.
How Will we Know that a
Change is an Improvement?
Q2: How will we know that a
change is an improvement?
Teams use quantitative
measures to determine if a
specific change actually
leads to an improvement.
e.g. Proportion of
reconciled medications.
Nolan et al
3 Ways to Get Better
Numbers
1. Improve the System
2. Distort the System
3. Distort the Figures
» Outliers
» Inliars
» Darn Liars
Strategies for Getting Better
Results
• Dis-aggregate:
– e.g. LOS
• Stratify:
aggregate and chop and
splice
– e.g. Patient falls,
Cancelled OR cases
• Experiment
– e.g. Waiting list
management
• Standardisation
Disaggregate: LOS
• Pre-op
• Intra-op
• Recovery
• Post-op Ward
• Rehabilitation
Stratification: Slicing
FACTORS
EXAMPLES – slice the data by….
WHAT
Type of adverse event, Triage code, Cost
WHEN
Month, Day of week, Time of day
WHERE
Area health service, Facility, Location, e.g. sacrum
WHO
Other GPs
Stratify: Comparing AHS
Complaints Resolved within 35 Days
Comparison of AHS Performance
90
Percent
80
80%
DoH
BMK
70
60
50
A
B
C
D
E
F
G
H
Pareto Principle
80% of the trouble comes from
20% of the problem
Standardisation
Ganley H. Cameron M.
Critical Paths – A
Continuous Quality
Improvement Approach to
Improving Patient Care .
The Quality Magazine.
Australian Quality Council
1996.
Ganley HE, Cameron MJ. Momentum, Australian Quality Council. 2001
Ganley HE, Cameron MJ. Momentum, Australian Quality Council. 2001
Composite Compliance: Proportion of Leg Ulcer Bundle Implemented
Composite Compliance: Proportion Leg Ulcer Bundle Implemented
March 2006 & March 2007
March 2006 & March 2007
1.0
0.9
0.8
0.5
0.4
LCL=35%
Proportion
_
P=57%
0.6
_
P=84%
0.8
UCL=78%
0.7
Proportion
UCL=100%
1
LCL=68%
0.6
0.4
0.2
26%
0.3
1
0.2
Mar-2006
Sample Size 6 & 7
Tests performed with unequal sample sizes
O’Brien M, Lawton J, Conn
C, Ganley H. Best Practice
Wound Care. International
Wound Journal. WileyBlackwell. 7 (4) 2011.
Mar-2007
0.0
Mar-2006
Sample Size 6 & 7
Mar-2007
Causation:
Feeling Challenged?
C Chart of Damage Index
12
1
Sample Count
10
8
6
UCL=5.03
4
_
C=1.43
2
0
LCL=0
1
3
5
7
9
11
13
Sample
15
17
19
21
23
Challenger Data: Relationship between Temperature / Likelihood of Damage
12
10
Damage Index
8
6
4
2
0
50
55
60
65
70
Degrees Fahrenheit
75
80
The Aim is
Improvement
Common cause
variation reduced
Special causes
eliminated
Process improved
Special causes
present
Process under control predictable
Process out of control
- unpredictable
Adapted from R. Lendon
Take Away Messages
Plot the dots - make the variation visible
•
•
•
•
•
•
•
Include sample size information
Interpret smoothed data with caution
Wary of drawing conclusion from few data points
Employ subject matter expertise to understand data
Take care in extrapolating data
Stop tampering
Be willing to think differently
Plot the dots - make the variation visible
Deming’s Common
Principles for Action
A common focus on ....................................................................Quality
A common vision ................................................High Quality Service
achieved by fighting
A common enemy .............................................................Variability
using
A common method ............................................Process Improvement
and communicated through
................................Statistics
A common language