for the practice change fellows program september 28, 2007 washington, dc dennis a. ehrich, md, facc...
TRANSCRIPT
For the Practice Change Fellows ProgramSeptember 28, 2007
Washington, DC
Dennis A. Ehrich, MD, FACCVice President for Medical Affairs
St. Joseph’s Hospital Health CenterSyracuse, New York
Measurement for Improvement
Agenda for the Morning1-Why Do We Measure in Health care?2-The Model for Improvement3-Selecting one’s measures7-Time ordered statistics and understanding
variation 8-The run chart10-Practical exercises11-The control chart (if time permits)
Why We Measure in Health Care
Measuring for Research
Measuring for Judgment
Measuring for Improvement
Purpose To discover new knowledge
To compare to others, to rank
To bring new knowledge into daily
practice
Tests One large trial Public reporting quarterly or with
12 month running averages
Many sequential, observable tests
Biases Control for as many as possible
Severity or risk adjustment where
available
Stabilize the biases from test to test
Data Gather as much data as possible,
just in case
Reports structure, process or outcomes
Small tests of significant changes, accelerates the rate
of improvement
Duration Can require large numbers of
patients and long periods of time to obtain
results
Ongoing data collection and
quarterly public reporting
Short iterative cycles in a limited number of subjects, followed by
spread
Set aims that are measurable, time-specific, and apply to a defined population
The Model for Improvement
Establish measures to determine if a specific change leads to improvement
Select changes most likely to result in improvement
Test the changes
T. Nolan et al. www.ihi.org
The Use of Iterative PDSA Cycles
Implementing the Changes
“Rapid-cycle CQI”
T. Nolan et al. www.ihi.org
Spreading the Change
1-Planning and set-up 2-Spread within the target population 3-Continuous monitoring and feedback on the spread process
T. Nolan et al. www.ihi.org
Donabedian’s Quality Triangle-It’s Relevance to Process Improvement
-Avedis Donabedian, MD, MPH (1919-2000)
Donabedian’s TriadStructure
OrganizationPeopleEquipment/Technology
ProcessThe actual steps taken in accomplishing the
changeResults must be client-focusedMust deliver results reliably
OutcomesClinicalClient perception or satisfactionFinancial
The Three Domains of MeasurementStructural MeasuresProcess measuresOutcomes Measures
Balancing measures
Donabedian
The Three Domains of MeasurementStructural Measures
Describe the environment. How many?Square footage of a clinical unitNumber of staffStaff qualifications and competenciesPresence or absence of technology and its
characteristicsProcess Measures
Process cycle timeThe percentage of patients for whom the process
achieves its desired result
Donabedian
The Three Domains of MeasurementOutcome Measures
The impact of the change initiative on mortality, readmissions to the hospital, ED visits
The satisfaction scores of clients and staff The cost per case, average LOS, revenue per case
Balancing Measures Unintended outcomes that are consequences
of the new programUnanticipated mortality, morbidity or cost Has the shifting of resources in an
organization compromised other client or patient populations?
Donabedian
ACTION
Aim
Selecting A Measure
Operational Definitions
Data Collection Plan
Data Collection
Data Analysis
The Quality Measurement
Roadmap
Modified from Lloyd, Robert: “Quality Health Care A Guide to Using Indicators”
Selecting a Measure:
-When selecting a measure, have clarity as to whether the measure is one of structure, process or outcome
-And select a balanced panel of indicators that reflect the dimensions of performance being evaluated and the change concept(s) being employed
What Dimension of Performance do You Want to Measure?Appropriateness AvailabilityContinuityEffectivenessEfficiencyRespect and caringFinancial/ViabilitySafetyTime lines
Joint Commission (1996)
What Dimension of Performance do You Want to Measure?SafetyEffectivenessPatient-centerednessTimelinessEfficiencyEquity
IOM: Crossing the Quality Chasm (2001)
What is the “Change Concept”?Eliminate wasteImprove work flowShorten a waiting listChange the work environmentImprove the Provider/Client interfaceManage timeFocus on variationError proofing a processFocusing on product or service
The Improvement Guide by Langley, Nolan, Nolan, Norman and Provost. Jossey-Bass
Relating a Change Concept to a Specific Measure
Concept Potential Indicators for this processPatient scheduling •The average number of days between the call
for an appointment and the actual appointment date•The percentage of appointments made within 3 days of the call for an appointment•The number of appointments scheduled each day
Home care visits •The number of home care visits•The average time spent during a home care visit•The percentage of time spent traveling during each home care visit•The number of visits per home care nurse
CQI Training •The number of participants attending a class•The percentage of cancellations•The percentage of no-shows•The information recall scores at 30 and 60 days
Operational DefinitionsIs clear and unambiguousSpecifies the measurement method,
procedures and equipment when appropriateClinical data (chart reviews) vs. administrative dataClient logs vs. a computer database
Define specific criteria for the data to be collectedDefine all inclusions and exclusionsFor percentages or rates, or ratios, define the criteria
for inclusion in the numerator and denominatorAlways ask “How might somebody be confused
by this definition?”
Lloyd, R. Quality Health Care (2004) Jones and Bartlett
Examples of Unclear DefinitionsTimely completion of the screening processA complete medication listThe readmission rateMedication errorCost impactFrom the acute care hospital
A patient fallSurgical start time
Lloyd, R. Quality Health Care (2004) Jones and Bartlett
Data Analysis What descriptive statistics will be used?
Mean, median, modeMinimum, maximum, range, standard
deviationQuantities, proportions (percentages), rates
How will data be displayed?Bar chart, histogram, line chart, pie chart, Pareto
diagramRun chart, control chart
Data ReportingData reporting plan
Who will receive the resultsHow often will they receive the resultsHow will the data be disseminated?
Printed reportsEmailDashboardSpider diagram
Tools for Displaying Time-ordered DataRun charts
Plot of data over time with the median of the data set plotted as a center line
Control chartsPlot of data over time with the mean as the
center line and with upper and lower control limits
Run ChartsEasily constructed by hand or in available
spreadsheet programsProvides a good idea of improvement in a
change initiativeLess sensitive to significant changes
(special cause variation) than the control chart
Control Charts More sensitive to special cause variation
than a run chart Requires specialized computer software to
create There are 9 types of control charts used in
health care, depending upon whether the data collected is distributed normally, is continuous (numerical) or discreet (attributes) and whether the events measured are frequent or infrequent
Have their own set of rules to identify special cause variation
Understanding VariationAll data, collected over time, variesRandom variation (common cause)
The changes occurring are intrinsic to the process being measured
Non-random variation (special cause)The changes are being imposed on the
system by some external factorMay be unintended and un anticipated or
may be by design
Hand-Drawing a Run Chart
Plot data points as a line graph on x-y axes, where “x” is the increment of time and “y” is the measurement.
Calculate the median value of the data set and draw that line on the chart Sort the data from smallest to largest value Count the data points. That count=N If “N” is an odd number: Median=N+1/2. Begin
counting from smallest to the largest number. When the count reaches N+1/2, that number is the median
If “N” is an even number: Median=The average of N/2 and the next number in the series. Begin counting from smallest to the largest number. When the count reaches N/2, stop and take the average of that number and the next number in the series. That average is the median
Calculating the Median
Odd Number of Data Points 1 N=722371112
Median=N+1/2=7+1/2=4The median is the 4th number in the series, which is 3
Even Number of Data Points1 N=634558
Median=The avg. of the number that is N/2 and the next number in the series. =[4 (the third number in the series) +5 (the next number in the series)] / 2=4.5
Balestracci, D., and Barlow, J, Quality Improvement. 1998 Center for Research in Ambulatory Health Care Administration
Definitions A run is 1 or more consecutive data points on the same
side of the median line A useful observation is one that does not fall on the
median line
•Sixteen of the eighteen observations are useful•There are 10 runs on this run chart
Testing for Special Cause Variation on a Run ChartTest 1. Are any runs longer than expected? If
so, then that run represents a special cause.If there are fewer than 20 useful
observations, then 7 or more data points in a run indicate a special cause.
If there are 20 or more useful observations, then 8 or more data points in a run indicate a special cause.
Testing for Special Cause Variation on a Run ChartTest 2. Is there a trend? A trend is an
excessively long series of consecutive increases or decreases in the data.
Total Number of Data Points on the Chart
Number of ConsecutiveAscending or Descending Points Indicating a Special
Cause
5 to 89 to 20
21 to 100
5 or more6 or more7 or more
Applying Tests 1 and 2
Total number of data points=18 Number of useful observations=16
Test 1-Since there are < 20 useful observations it will take ≥ 7 data points in
a run to cause a run to be “too long” defining special cause variation
Test 2-Is there a trend? For 18 total data points, it will take ≥6 consecutive
ascending or descending data points to define a trend.
Testing for Special Cause Variation on a Run ChartTest 3. Are there too few or too many runs
in the data?Determine the number of useful observations
in your data set. Use the following table to determine whether
the number of runs in your data are within the expected range. If the number of runs is above or below the expected range, the data suggest special cause variation
Useful Observatio
ns
Lower
Limit
Upper
Limit
1516171819202122232425262728
455666778899910
1212131314151516161717181919
UsefulObservatio
ns
Lower
Limit
Upper Limit
29303132333435 3637383940
101111111112131313141415
202021222223232425252626
Expected Number of Runs
Applying Test 3
Are there too many runs? Useful observations= 24. Number of runs= 8. Expected number of runs = 8-17. Therefore there is no evidence for special cause variation.
Testing for Special Cause Variation on a Run ChartTest 4. Fourteen or more points alternating above
and below the median line is a saw tooth pattern indicate a special cause.
When this pattern is seen, it indicates either that two different processes are operating at the same time and have been measured together; in which case stratification of the data would be helpful. Or, it may indicate tampering
KQC=Key Quality Characteristic