Session 61 L, Applications of Data Analytics in Health Insurance Moderator/Presenter: Henning Chiv, FSA, MAAA
Session 61: Applications of Data Analytics in Health Insurance Henning Chiv, FSA, MAAA June 15 th, 2015
Agenda How is technology changing analytics in both pace and insights? What can actuaries bring to the table? + Overview of the analytics cycle + Analytics examples + Challenges and take-aways + Q&A 1
Poll Who has heard of the following terms? Big data Data science Machine Learning Advanced Analytics Structured vs unstructured data Actuarial control cycle General linear model Eigenvalues 2
Analytics cycle Measure results Outcomes to improve? Recommendations Analytics process
Outcomes to improve Healthcare (insurance) is transforming and faces many challenges: Transparency around pricing and reserving Design and management of provider networks Cost savings and improve health outcomes Retention of members Member satisfaction Provider access Treatment and member out-of-pocket expenses
Analytics cycle Measure results Outcomes to improve? Recommendations Analytics process
Analytics process External events Reporting (What happened?) Prediction (What will happen?) Analysis (Why did it happen?) Prescription (Make it happen!) Data Cleansing 6
Analytics techniques Statistical/Stochastic Loss modeling Regressions Blend of both Markov chains Propensity scoring Algebraic Similarity Matching Support Vector Machines 7
Analytics cycle Measure results Outcomes to improve? Recommendations Analytics process
Recommendations Communicate results of analysis Put analysis into context of business Implicit/explicit assumptions Timing of assumptions Timing for forecast Confidence intervals Recommendations 9
Analytics cycle Measure results Outcomes to improve? Recommendations Analytics process
Measure results Need to own analysis and analytics process from end-to-end Cultivate relationship with stakeholders Automation and technology can help track progress and outcomes Have re-usability in mind 11
Challenges + Data content is not normalized across partners and systems + Healthcare is not intuitive requires the build out of metrics and topologies + External events and changes in the environment are not captured in the data 12
Examples
How is my claims experience? $300.0 $250.0 $200.0 $150.0 $100.0 Total Inpatient vs. Mater nity Inpatient PMPMs $50.0 The curse of big data: WHERE are we going to LOOK? $- 40 35 30 25 20 15 201310 201401 201404 201407 201410 Total Inpatient Maternity Only ER vs. Urgent Care Utilization Summar y Decision trees Pattern recognition 10 5 0 201310 201401 201404 201407 201410 Emergency Room Service Count Urgent Care Service Count 14
Is my wellness program effective? + Alternative to conducting a clinical trial + Use observational data to evaluate efficacy of programs of participating and nonparticipating individuals Propensity scoring + Combination of logistic regression analysis and matching algorithms 15
What should I expect when I am expecting? + What types of visits and procedures should a member expect during a pregnancy? + How does the treatment path for a typical 20 year-old differ from a 40 year-old? + Which factors are good predictors for complications? Similarity Matching 16
Take aways
Take aways + Technology enabled acceleration of the analytics cycle + Advancements in technology enable new analytical techniques + Actuaries bring expertise around healthcare finance, but will have to broaden their tool kits to stay up-todate 18
Q&A Resources: + Healthcare Risk Adjustment and Predictive Modeling (Duncan, Actex 2011) + Doing Data Science (O Neil & Schutt, O Reilly 2014) + Data Science from Scratch (Grus, O Reilly 2015) + http://www.continuum.io/ (Anaconda) + Pandas, numpy, seaborn, scikit-learn 19