Presentation on theme: "HSAG Performance Improvement Projects Using Data to Develop Interventions and Statistical Testing to Evaluate Results Breakout Session #1 Florida EQR."— Presentation transcript:
1HSAG Performance Improvement Projects Using Data to Develop Interventions and Statistical Testing to Evaluate Results Breakout Session #1 Florida EQR Quarterly Meeting June 18, 2008Presented by:Donald Grostic, MSAssociate Director, Research and Analysis TeamYolanda Strozier, MBAEQRO Project Manager
3What does the intervention cycle have to do with CMS PIP Activities? IdentifyPlanImplementEvaluateActivity 7Assess the Improvement Strategy☑Activity 8Review Data Analysis & Interpretation of ResultsActivity 9Reported Improvement is Real?Activity 10Sustained Improvement?
5Data Mining What is data mining? Answer: Data mining is the process of sorting through large amounts of data and picking out relevant information.
6Data Mining (continued) What is data mining used for?Answers:Data mining is the statistical and logical analysis of large sets of data, looking for patterns of care, or service delivery that can aid decision making.To identify and determine areas of non-compliance that will be analyzed during the causal/barrier analysis.
7Data Mining vs. Data Analysis Plan How does data mining differ from a data analysis plan?Answer:A data analysis plan includes calculating and comparing overall indicator rates between measurement periods using statistical testing.Data mining will include analysis that goes beyond just calculating and comparing indicator rates between measurements.
8Data Mining–Example PIP topic (clinical): Indicator: Follow-up after acute care inpatient hospitalization.Indicator:The percentage of members with follow-up within 7 days after acute care discharge for a mental health diagnosis.
9Data Mining Example Step One Group the population or sample.First, group members by county or ZIP code. For our example, the population breaks into three counties: County A, County B, and County C.
10Data Mining Example Step Two Calculate compliance and noncompliance for each county.The percentage compliant and noncompliant by county are presented in the following table.Question: Which county should you data mine further?Percentage CompliantPercentage Non-CompliantCounty A65%35%County BCounty C20%80%
11Data Mining Example Step Three Identify groups where the majority of members are noncompliant.Answer:First we need to know how many members of the population are in each county.Selecting County B will have the greatest effect on the compliance rate because it has the majority of the population and the second lowest compliance rate.Percentage CompliantPercentage Non-CompliantNumber of MembersCounty A65%35%80County B220County C20%80%20
12Data Mining Example Next Steps Now that you have identified County B, what should you do next?Answer:Continue the process of grouping and selecting to find the group that will have the greatest effect on compliance.For County B, you may consider grouping the data by PCP or facility next.
13Data Mining Caution! Words of caution: Grouping and selecting can be taken to a point where the groups selected may be too small to make an impact.Always keep in mind the number of members affected in the selected group relative to the total population.If there is difficulty identifying noncompliant groups or non-compliance is equally distributed among groups, you may be dealing with a systemwide issue.Please keep in mind that data mining is a dynamic, iterative process that takes practice.The more you data mine the better you will become at selecting groups that yield the best effect on rates.
15What is a Causal/Barrier Analysis? A causal/barrier analysis is:A systematic process for identifying the problem.A method for determining what causes the barriers.A way to identify what improvement opportunities are available.Causal/barrier analysis has also been called:Root cause analysis
16How do I perform a causal/barrier analysis? Determine why an event or condition occurs.What is the problem?- Define the problem and explain why it’s a concern.Determine the significance of the problem.- Look at the data and see how the problem impacts your members and/or health plan.
17How do I perform a causal/barrier analysis? (cont.) 3. Identify the causes/barriers.- Conduct analysis of chart review data,surveys, focus groups.- Brainstorming at quality improvementcommittee meetings.- Literature review.4. Develop/implement interventions based on identified barriers.
18Causal/Barrier Methods and Tools Quality improvement committeesDevelop an internal task forceFocus groupsConsensus expert panelsTools:FishboneControl chartFlow chart (process mapping)Barrier/intervention table
21A Physical Health Example What questions could be asked to drill down these causes?What data are needed to identify the most crucial cause?
22A Mental Health Example Discharge planningClientCommunicationTransportationCommunity involvementNo follow-up appointment set at time of dischargeTime lag/claim dataNot client focusedProvider accessCulture changeDemographic informationWhat questions could be asked to drill down these causes?What data are needed to identify the most crucial cause?
23Interventions Checklist Analyze barriers (root causes)Choose and understand target audienceSelect interventions based on cost/benefitImplement interventionsTrack intermediate results (optional)RemeasureModify interventions as needed
26The ‘Implement’ Stage Three tips: Observe and document whether the intervention is implemented as intendedNote any lesson(s) learnedDocument any change(s) that may threaten the results between measurement periodsMethodology (e.g. definition of indicators, sampling)Circumstances (e.g. merger, population, provider)
29Statistical Significance Testing What is statistical testing and why do we use it?Answers:Statistical testing is calculating specific test statistics and associated p values to determine if an observed difference is a true difference and not due to chance alone.The CMS Protocols require that statistical testing be used to prove that any improvement in rates is real.Without statistical testing, a PIP would not meet the CMS Protocols.
30Statistical Significance Testing What type of statistical testing is appropriate for my PIP?Answer:Fisher’s Exact Test or Chi-square test for rates or proportions.T test for means would be the appropriate statistical testing.
31Statistical Significance Testing What type of statistical testing is appropriate for this indicator?Indicator A: The percentage of members with follow-up within 7 days after acute care discharge for a mental health diagnosis.Answer:Fisher’s Exact Test or Chi-square test for rates or proportions.
32Statistical Significance Testing What is the difference between Fisher’s Exact Test and a Chi-square test?Answer:Fisher’s Exact Test will provide the exact p value while the Chi-square test is an approximation of the p value.As your numerators and denominators increase in size, the Chi-square test and Fisher’s Exact Test produce the same p value.If in doubt about which test to use, use Fisher’s Exact Test.
33Statistical Significance Testing What type of statistical testing is appropriate for this indicator?Indicator B: The average response from a member satisfaction survey where answers range from 1=satisfied to 5=dissatisfied.Answer:T test for means would be the appropriate statistical testing.
34Statistical Significance Testing How do I report statistical significance testing results?Answer:When using a Fisher’s Exact Test, Chi-square test or a t test, report the test used, its associated p value along with each indicator, and its numerator and denominator in tabular form.
35Statistical Significance Testing Indicator A: The percentage of members with follow-up within 7 days after acute care discharge for a mental health diagnosis.Time PeriodsMeasurement PeriodsNumeratorDenominatorRate or ResultsIndustry BenchmarkStatistical Testing and SignificanceCY 2003Baseline204148.8%60%N/ACY 2004Remeasurement 1275152.9%Fisher’s Exact TestP value =Chi-square testP value =NOT SIGNIFICANT AT THE 95% CONFIDENCE LEVEL
36Statistical Significance Testing Indicator B: The average response from a member satisfaction survey where answers range from 1=satisfied to 5=dissatisfied.Time PeriodsMeasurement PeriodsNumeratorDenominatorRate or ResultsIndustry BenchmarkStatistical Testing and SignificanceCY 2008Baseline2531002.53N/ASTD DEV = 1.298CY 2009Remeasurement 13711133.28STD DEV = 1.561T-testP value =SIGNIFICANT AT THE 95% CONFIDENCE LEVEL
37Statistical Significance Testing If I use the entire population for my study, do I still have to do statistical significance testing?Answer:Yes. It is appropriate to do statisticaltesting on the entire eligible population.
38Reasons for Statistical Significance Testing on Entire Populations CMS is interested in performance over time.The population will continuously change over time.The members who are studied in one year may or may not appear in the following years.A population that is selected at one point in time is a sample from the true population that contains all members.The entire eligible population for a measure in one year is a sample population drawn from a universe of “all years” or “all populations” that could be selected.CMS has approved statistical testing on populations.
40The ‘Evaluate’ Stage: Linking Intervention to Outcomes NoReviseYesImproved?IdentifyEvaluateStandardizePlanImplement
41The ‘Evaluate’ Stage: Linking Intervention to Outcomes Threats to internal/external validity: Any environmental, organizational, methodological changes between measurement periods?NoYesOutcome: ImprovedIntervention seems to be effectiveConsider standardizing the intervention to subsequent measurement periodsCannot ascertain if the improvement really is due to interventionInvestigate the relationship between the change circumstances and the outcomesOutcome: No Change or WorsensIntervention does not seem to be effectiveConsider revising the intervention to subsequent measurement periods
42The ‘Evaluate’ Stage: Linking Intervention to Outcomes NoReviseEvaluateImproved?YesStandardizeIdentifyQIImplementPlan