the first 24 hours: understanding new claims
TRANSCRIPT
The First 24 Hours: Understanding New Claims
Gary Anderberg, PhD, Practice Leader, Analytics and Outcomes Bangalore Gunashakar, Senior Technical Consultant, e-Triage Sergo Grigalashvili, VP, Architecture, Analytics and Global Systems Roadmap
Predictive Analytics World, San Francisco, March 5th, 2012
The Business and the Problem
Broadspire is the third largest Third Party Administrator for workers’ compensation.
We get hundreds of new claims daily.
Most of these new claims are puppy dogs, but some of them are Cujo. They often look alike coming in the door.
How do we know which are “one touch” claims and which claims will need intensive resource allocations to resolve successfully?
And don’t forget the creeping cats.
Data Mining– going beyond the usual suspects
Workers’ Comp claim files are data poor
The curse of the text box
Causative factors and influences—
» Medical
» Psychosocial
» Family dynamics
» The work relationship
A vast literature of peer reviewed published scientific studies, many including multi-factor models
Creating an expert database
Asking all the right questions, or why it pays to be nosy
An intelligent guided interview process Capture responses as data points—
» Drop down boxes » Radio buttons » Y/N questions » Defined value questions
Use high gain questions Gather required information and highly predictive responses in the same
process Score the responses and deliver—
» Claim issues » Recommendations » Overall “interesting claim” score
Build the Strategic Plan of Action for the claim
How e-Triage was built
A team of experienced programmers, medical advisors and researchers, everyday users carefully designed the system.
Microsoft .NET Technology platform, Visual Studio 2008, C# .NET, .NET 3.5 Framework, SQL Server Database, SQL Server Reporting Services, Tableau
Modules – DM System, HC System, Expert System
Black Box – Expert System is where all the Magic happens
Contains medical research, references, guidelines, issues, deterrents, recommendations
Interviews are Dynamic and Flows based on the response to questions.
Claimant \ Employer \ Provider are contacted and interviewed within the first 24 hours of the claim being reported
How are answers scored
All responses from the interview & claim data collected are run through the “Black Box” processing
Expert System uses the medical research, custom business rules, derivatives of predictive analytics for processing.
Each answer is evaluated against criteria and when found to be true a score is assigned.
Scores of all answers are summarized and compared against the threshold score
Issues are raised when the answer is found to meet the criteria of one or more research studies.
Issues: are flags the application displays in the DM System to highlight a piece of information that may be helpful in managing the claim.
Deterrents: are Psychosocial factors triggered by responses in the interview. Clusters of like deterrents or cumulative deterrents will trigger specific recommendations that require action.
How scores and issues are flagged
Each answer could raise Issues & Deterrents and/or Scored based on the research criteria that are true
Red, Yellow or Green flags are raised based on the summary & threshold score, as well as the issues raised
Red flagged claims are the complex claims that require high attention and a senior claims adjuster, senior nurse, and/ or physician be assigned
Yellow flagged claims have the potential to turn into complex claims and are closely monitored on the progress of the claim
Green flagged claims are sometimes one-touch claims and they are expected to go through the system very smoothly
Referrals and/or Reviews are generated based on the issues raised and/or upon the summary score exceeding the threshold score
How recommendations are generated
The Issues & Deterrents identified are linked to the published research for causative factors & influences.
The recommendations and guidelines generated are also based on the published research, custom business processes rules, and derivates of predictive analytics
Sample - Points Assigned
Sample Issues & Deterrents to Recovery
Sample Recommended Actions
2011 Analytics – Scores \ Issues
Digging into Data
Characteristics overrepresented in high severity claims
Data set: about 330k WC claims since 2007
Predicting Claim Severity
Dynamic Claim Processing
Score outcomes and optimize each phase of claim processing
Optimize the entire process through dynamic process routing: fast track / straight through / one touch processing
Measure impact from the dynamic processing and fine tune the models
What have we accomplished with e-Triage?
e-Triage was designed to detect and define problem claims.
e-Triage has been in place for six years; we are now running V 2.3.
e-Triage is how we do claims at Broadspire; initial studies indicate savings of about 10% on certain claim costs.
e-Triage has proven the usefulness of predictive modeling in our claim world; we are enhancing e-Triage and exploring new uses for predictive analytics.