Data Mining: Opportunities for Healthcare Quality Improvement & Cost Control Joseph A. Welfeld, FACHE Long Island University 845.359.7200 x 5410 Joe.welfeld@liu.edu March 7, 2005 The Health Information Technology Summit West
Data Mining: Opportunities for Healthcare Quality Improvement & Cost Control Speaker Profile Data Mining Quality Improvement Changing Behavior with Incentives Cost Control Targeting Key Areas Data Mining Software Practical Applications A Case Study Questions
Speaker Profile Joseph A. Welfeld Regional Operations Director: NY - RelayHealth Program Director: Graduate Program in Health Administration: LIU Rockland Graduate Campus 30 years of healthcare experience CEO - Ocean State Physicians Health Plan Regional VP United Healthcare 10 years in strategy consulting for IPAs, PHOs & Hospital Networks MBA Healthcare Administration CUNY/Mt. Sinai School of Medicine
Data Mining: Definition An information extraction activity whose goal is to discover hidden facts contained in databases. True data mining software doesn't just change the presentation, but actually discovers previously unknown relationships among the data.
The Healthcare Database Minefield Hospital claims data billing systems Medical claims data billing systems Pharmacy claims data PBMs Lab data systems Aggregators: Managed Care Organizations Third Part Administrators Medical Groups/IPAs None of the above
Data Mining: Obstacles in Healthcare Organizations Deer in the headlights look Data what? We don t have any more money to buy software We have all the software we need We just spent $ million on a new system Our IT staff can produce anything we want from our in-house data system Our data analysis could not be better
Quality Improvement The Challenge Finding acceptable standards Combining data from multiple sources Limited financial incentives to promote change Until recently, no financial incentives to change Goal physician behavior change
Quality Improvement The Opportunities HEDIS Standards Leapfrog Group Bridges to Excellence MCO Performance Incentives
Sample HEDIS Report Activity: Beta Blocker Treatment After Heart Attack Members age 35 and older who where discharged with an AMI and were prescribed beta-blockers within 7 days of discharge. Numerator: Members who received an ambulatory prescription for a beta-blocker within 7 days of discharge Denominator: Members with an AMI between Jan 1 and Dec 24 of the measurement year Problem Faced: Linking admission/discharge and prescribing data
Beta Blockers Prescribed after MI Diagnosis: ATENOLOL COREG INDERAL LABETOLOL SOTALOL BETAPACE PROPRANOLOL NORMODYNE
Use of Appropriate Medications: People with Asthma Numerator: Members age 5-56 who received a prescription for a long term control asthma medication such as inhaled cortico-steroids Denominator: Members age 5-56 are identified as having asthma using pharmaceuticals and diagnostic data during the year prior to the measurement year Four dispensing events One ER visit with a principle diagnosis of asthma One acute inpatient discharge with a principal diagnosis of asthma At least four outpatient visits with a diagnosis of asthma and two dispensing events
Cost Control The Challenge Payer Provider trust chasm The my patients are sicker debate Combining data from multiple sources into coherent and logical reports
Cost Control The Opportunities The ability to merge medical claims, hospital claims, drug claims, medical records and clinical outcomes data The ability to analyze episodes of care including drug utilization The ability to rapidly create contract models by user-defined resource and provider categories Ability to drill down into individual patient claims Ability to target high cost trends
Cost Control: Targeting High Cost Trends Puts up to 3 datasets side-by-side. Can compare performance against benchmarks. Unlimited number of resource categories and user-defined resource utilization models allowed Tracks in-patient, professional, lab, pharmacy and other cost categories automatically See example:
Cost Control: Drilling Down to Specifics Isolate a resource category and quickly find highest cost by any factor (disease risk group, age, sex, plan, doctor, etc.) Then drill down to get more information on those results Drill down further to see treatment line items for those specific patients Example on following screens shows disease groups with highest lab costs
ACRG2 Metastatic Category: 5 episodes with very high costs
Those 5 Patient Episodes in the ACRG2 Metastatic Group
Cost Control: Age/Sex Analysis Creates unlimited number of age distribution models to apply against data Select specific resource categories to view Cross-tab against specific values of any factor, i.e., disease group, specialty, etc. The following slide shows the utilization of selected resources by Age/Sex for patients in the Asthma-Diabetes-CHF CRG categories:
Cost Control: Physician Profiling Functions designed to monitor physician activity Monitor ICD9 and CPT code utilization patterns Cross-tab against specific values of any factor, i.e., disease group, specialty, etc. Summarizes all costs by provider and compares on one screen.
ER Utilization Costs by PCP: Outliers shown above dotted line on graph Highest outlier on graph highlighted on chart
CPT Codes for Gastroenterologists: Ranked by Frequency
PCP Utilization Cost Summary by Major Resource Category
Detailed 3M CRG (Clinical Risk Groups) Disease/Severity Cost Distribution
Detailed 3M CRG (Clinical Risk Groups) Disease/Severity Cost Distribution
Hudson IPA A Case Study Strategic Question How to deliver real value to managed care organizations? Replace capitated agreement with performancebased model Provide managed care organizations data analysis capabilities they don t really have Assist with HEDIS performance monitoring and communications a key MCO objective
Data Mining Software Bringing Value Gave IPA: Ability to merge medical claims, hospital claims, drug claims, medical records and clinical outcomes data Ability to analyze episodes of care including drug utilization to meet agreed-upon goals Ability to rapidly create contract models by userdefined resource and provider categories Ability to drill down into individual patient claims Ability to analyze HEDIS performance criteria including diabetes and cardiology care
Data Mining Software Characteristics Powerful disease state management and risk contract functionality Data warehouse designed to merge all types of healthcare data. Physician profiling and resource tracking features Drill down into individual patient claims from either financial or clinical perspectives and retrieve both types of information together
Data Mining Software Characteristics SmartCare Developed by VantagePoint Health Information Systems, Inc. Loads claims data at a rate of 100,000 claims/hr Links pharmacy (PBM), hospital & medical claims Automatically creates episodes of care Computes PM/PM ratios in less than five seconds Powerful graphing & statistical tools No programming/data analysis skills/staff needed Open database for addition of other clinical or administrative fields lab, blood pressure, etc.
Data Mining Applications Summary Gives Physician Organizations: Ability to develop quality indicators, performance improvement programs and incentive-based compensation programs. Ability to analyze HEDIS performance criteria including diabetes and cardiology care. Ability to analyze formulary compliance activity. Tool for additional revenue resources including comprehensive market research, clinical outcomes and pharmaco-economic studies. Ability to monitor risk-contract progress.
Data Mining Applications Summary Can Give Managed Care Organizations: A tool to develop true partnership relationships with provider organizations seeking incentive compensation or risk relationships Ability to develop comprehensive HEDIS analysis and performance reports Ability to combine multiple claims data bases into a single data reporting and analysis system at the contracting level Ability to do rapidly model the impact of fee schedule changes on provider costs and contract performance.
Questions??