OBJECTIVES : For some time now, advanced modeling techniques have been employed by BCBSLA to predict the risk of hospitalization (ROH) among its members. Currently, technology enhancements have taken place that increase model performance and are now being employed on Medicaid members. This research reports the initial findings on this new population of interest.

METHODS : BCBSLA has been routinely employing ROH models for member risk management. Building on this approach, enhanced data collection techniques now make it possible to incorporate both structured and unstructured data into the modeling process; which in turn allows for 100k+ factors engineered to capture and measure complex relationships, uncover interactions and discover unknown sources of correlation. Training algorithms are leveraged to continuously retrain, fit and refit, measure and adjust the model rapidly for maximum performance and prediction of a hospitalization among its members. Now this model has been deployed on the sub-population of Medicaid members.

RESULTS : The ROH for all members, the current PPV in the top percentile at 51% is 1.5x greater than the first baseline model. This model captures 76% of all members having hospitalizations within the top 20% of the risk score. For the risk of Emergency Department visits, the PPV in the top percentile is 90%. Nearly 50% of all members having ED visits are captured within the top 20% of the risk score. When this approach has been applied to the Medicaid population, preliminary models yielded impressive results. The highest 1% of risk scores had a positive predictive value of 48%, wherein half of those identified members would be hospitalized.

CONCLUSIONS : The newest modelling approaches employed by BCBSLA demonstrate how advanced technology can be used to accurately predict Medicaid members risk of hospitalization, and thus, can intervene which in turn reduces healthcare burden for both patients and providers.