Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Predictive Analytics in Healthcare


Published on

Edgewater Technology Executive Event
Unlocking the Value of Your Enterprise Data
Maury DePalo
October 1, 2008
Houston, TX

Predictive Analytics in Healthcare

  1. 1. October 1, 2008 Predictive Analytics in Healthcare
  2. 2. <ul><li>Introduction </li></ul><ul><ul><li>Approaches & Methods </li></ul></ul><ul><li>Supporting the Healthcare Value Chain </li></ul><ul><ul><li>Translational Research </li></ul></ul><ul><ul><li>Hospital Operations </li></ul></ul><ul><ul><li>Risk Management </li></ul></ul><ul><li>Implementation </li></ul><ul><ul><li>System & Data Architecture </li></ul></ul><ul><ul><li>Process Framework for Development & Deployment </li></ul></ul>Predictive Analytics: Overview
  3. 3. Predictive Analytics: Various Approaches CLUSTERING FORECASTING MONITORING & ADVISING SIMULATION & SCENARIO PLANNING DECISION TREE  The use of current and past data, in conjunction with statistical, structural or other analytical models and methods, to determine the likelihood of certain future events Predictive methods cover the spectrum from relatively simple classification and forecasting to more advanced techniques such as simulation and advising As you move up the spectrum, the complexity of the approaches and their implementation increase
  4. 4. Clustering as Classification <ul><li>Hierarchical clustering of 107 genes selected from 12,000 </li></ul><ul><li>Progressive expression relative to normal: </li></ul><ul><ul><li>Increase: Red </li></ul></ul><ul><ul><li>Decrease: Green </li></ul></ul>Yeatman, et al: JNCI, April 2002 Osteopontin
  5. 5. Class Discovery and Class Prediction Golub, et al: SCIENCE, October 1999 <ul><li>Diagnostic classification based on molecular characteristics of disease, i.e. differentially expressed genes </li></ul><ul><li>Automatically discovered the distinction between acute myeloid leukemia (AML) and acute lymphoblastic leukemia (ALL) </li></ul><ul><li>Establishes disease classes, and then uses those class definitions to determine the correct classification for new undiagnosed cases </li></ul><ul><li>Also succeeded in discovering these classes independent of previous biological knowledge </li></ul><ul><li>Enables More Effective, Less Toxic </li></ul><ul><li>Treatment </li></ul>
  6. 6. Class Distinction, Combined with Pathway Model Wei, et al: Cancer Cell, October 2006 GC apoptosis (cell death) rapamycin MCL1
  7. 7. Decision Tree: Progressive Class Distinction <ul><li>340 new tumor markers </li></ul><ul><li>100 tumor progression markers whose expression correlated with progressing tumor stage </li></ul><ul><li>Descend tree using any available differentiating attributes; natural, derived or inferred; separately or in combination </li></ul>
  8. 8. Forecasting: Process Model  Structural Model: Bill of Resources Patient Seen in Emergency Dept Admit Patient: Presumptive Diagnosis: Pneumonia Discharge Monitor Care Delivery Standard Order Sets Equipment Labor Materials Facilities Nursing Orders: Respiratory Therapy: Medication Orders: Resource Demand Day 5 Day 4 Day 3 Day 2 Day 1    
  9. 9. Forecasting: Resource Demand vs. Capacity Standard Order Sets Equipment Labor Materials Facilities Nursing Orders: Respiratory Therapy: Medication Orders: Day 5 Day 4 Day 3 Day 2 Day 1    
  10. 10. Simulation: Resource Demand vs. Capacity  What if … … incidence of disease X increases 2x? … process X increases throughput 1.5x? … market share in geography X (with Y / 1000 cases) increases by Z? … we focus our service lines into centers of excellence, shifting our patient mix across our facilities within the system? Facility A Facility B Facility C
  11. 11. Monitoring & Advising: Risk Management Patient & Case Profiled Against Risk Model Incrementally Accumulate Evidence of Emerging Risk Retain Case Instance & Feedback to Risk Model Notify Risk Mgmt Team of Need for Corrective Action Monitor Care Delivery Isolate Root Causes Track Negative Outcomes Track Key Events Within a Process Track Incidents <ul><li>Track outcomes </li></ul><ul><li>Identify variances from standard of care </li></ul><ul><li>Isolate root causes in case characteristics or provider actions </li></ul><ul><li>Identify case as high cost, high risk; discern patterns using attributes of case or care </li></ul><ul><li>Show evidence that case is aligning with risk pattern; notify parties to intervene, or avoid </li></ul><ul><li>Similar approach for protocol compliance; wellness & outreach; “activated/able” patients </li></ul>Forecast Non- Reimbursement Loss Tie to Claims Data Tie to Clinical Data Track Litigation Improve Quality, Avoid Future Incidents
  12. 12. Implementation: System & Data Architecture Clinical Data Operations Data Financial Data External Data Data Warehouse Data Files Data Sets Parameter Management DATA PRESENTATION/ CONSUMPITON LAYER APPLICATION/ MODEL MANAGEMENT LAYER DATA STORAGE LAYER DATA INTEGRATION LAYER DATA SOURCE LAYER ODS Multi-Dimensional Data Store Integration Cleansing Data Quality Formatting Aggregation Predictive Models Test Management Model/Version Management
  13. 13. Process Framework KPI Definition & Decomposition <ul><li>Define KPIs </li></ul><ul><ul><li>Define the KPIs Necessary to Optimize Business Performance </li></ul></ul><ul><li>Decompose KPIs </li></ul><ul><ul><li>Decompose KPIs into Core Components </li></ul></ul><ul><ul><li>Identify Which Components are “Raw Data” and Which are “Derived” </li></ul></ul><ul><ul><li>Determine Availability, Quality and Completeness of Data </li></ul></ul><ul><li>Model Prototyping & </li></ul><ul><li>Validation </li></ul><ul><ul><li>Conduct Iterative Prototypes of Predictive Analytics Models using Representative Test Data </li></ul></ul><ul><ul><li>Leverage Sub-Tests within a Single Model to Verify Different Scenarios </li></ul></ul><ul><ul><li>Be careful of over-fitting the data </li></ul></ul><ul><li>Model Refinement </li></ul><ul><ul><li>Refine Data Models Based on Prototype Results </li></ul></ul>Prototyping and Refinement <ul><li>Data Sourcing </li></ul><ul><ul><li>Determine Authoritative Data Source </li></ul></ul><ul><li>Data Integration </li></ul><ul><ul><li>Integrate Data Into Format Optimized for Model Consumption </li></ul></ul><ul><li>Model Instantiation </li></ul><ul><ul><li>Formalize Model into Model Repository, Including Version Control </li></ul></ul><ul><li>Incorporate Into Presentation Layer </li></ul><ul><ul><li>Address Usability Needs </li></ul></ul><ul><li>Determine Type & Form of Predictive Analytics to be Performed </li></ul><ul><ul><li>Determine Which Methods to be Used </li></ul></ul><ul><li>Determine and Implement Required Architecture Components </li></ul><ul><ul><li>Define Architecture by Layer(s) </li></ul></ul><ul><ul><li>Implement Required Architecture </li></ul></ul><ul><ul><li>Establish Test/Prototype Environment </li></ul></ul>Establish Architecture Iterative Prototyping & Validation Release to Production Maintain, Improve, Expand <ul><li>Maintain </li></ul><ul><ul><li>Verify Continued Validity of Models and Data </li></ul></ul><ul><li>Continuous Improvement </li></ul><ul><ul><li>Refine/Augment Implemented Models </li></ul></ul><ul><li>Expand </li></ul><ul><ul><li>Incorporate Additional Data </li></ul></ul><ul><ul><li>Develop Additional Predictive Analytics Capabilities </li></ul></ul>
  14. 14. Extending Visibility Into The Enterprise Executive User Functional User Power User Highly Aggregated More Detail Complete Raw Data <ul><ul><ul><li>Graphical Display, Dashboards, Interactive </li></ul></ul></ul><ul><ul><ul><li>Aggregated Data, Model Execution </li></ul></ul></ul><ul><ul><ul><li>Limited Drill Down </li></ul></ul></ul><ul><ul><ul><li>Standard & Ad hoc Reporting </li></ul></ul></ul><ul><ul><ul><ul><li>Parameter-Driven by Users at Run-Time </li></ul></ul></ul></ul><ul><ul><ul><ul><li>Sorting, Selection, Filtering, Drill-Down </li></ul></ul></ul></ul><ul><ul><ul><li>Utilizing Standard Functions and Models </li></ul></ul></ul><ul><ul><ul><li>Direct Access to Detailed Raw Data </li></ul></ul></ul><ul><ul><ul><ul><li>“ Just give me the data in SAS” </li></ul></ul></ul></ul><ul><ul><ul><li>Model Development & Deployment </li></ul></ul></ul><ul><li>Traceability </li></ul><ul><li>Consistency </li></ul>
  15. 15. October 1, 2008 Predictive Analytics in Healthcare