External Medical Records Help Identify USD 4.5 MM In Annual Savings

Case Study

External Medical Records Help Identify USD 4.5 MM In Annual Savings

Business Objective

Our client, a Fortune 100 life insurance company, wanted to create a mortality risk score for prospects using external medical information like prescriptions, procedures, diagnoses, and surgeries. The medical data would be purchased from an external vendor with an additional expense of close to a million dollars every year. The client thus wanted to determine if this initiative would result in better mortality prediction and save on insurance payouts, to justify the additional cost.

The specific objective was to:

• Quantify the expected value from the mortality risk score created using external medical history of an applicant

Challenges

  • Limited access to information underwriters use in making the current underwriting decision
  • Sparse mortality rate (<1%)

Solution Methodology 

  • Data preparation and feature engineering:
    • Created features using seven years of medical history of the individual
    • Some of the features include – distinct occurrence of prescription drug fills, number and types of medical procedures and surgeries, progression of a disease, adherence to medication.
  • Solution design:
    • Created three models with mortality as the target variable
      • Benchmark Model, which represented the effectiveness of existing underwriting process, and was built using policy features and risk class ascertained by underwriters
      • Risk Score Model, which measured the effectiveness of external medical data in predicting mortality
      • Incremental Model, which measured the lift (gain) over the benchmark model
    • The Risk Score model was later provided to the external data vendor to generate a risk score on an ongoing basis that could be fed into the Incremental model to ascertain incoming applicant’s risk score.
  • Model development and validation
    • Settled for logistic regression for the Benchmark and Incremental models to simplify business adoption
    • Built tree-based (XG-Boost and random forest) models to predict mortality risk score, to be able to capture the non-linear relationship in the data effectively

Business Impact

  • Estimated USD 4.5 MM in annual savings driven by improvement in mortality prediction using external medical history of an applicant
Download Case Study

©2017 Tiger Analytics. All rights reserved.

Log in with your credentials

Forgot your details?