Abstract by Zoe Gibbs
Using Machine Learning to Predict High Cost Patients
A small number of patients incur a disproportionate amount of total healthcare expenditures. To reduce the risk of loss, actuaries would like to predict which policyholders are likely to be high cost (incur over $100,000 in medical expenses in one year). With a dataset from a large insurance company that includes policyholder characteristics and health care costs in 2012 and 2013, we employ machine learning techniques to develop such a model. Through extreme gradient boosting, we found that diagnosis count, risk adjustment score, and total costs in 2012 are excellent predictors of high cost patients.