OBJECTIVES : Advanced modeling techniques have been employed by BCBSLA to predict the risk of hospitalization (ROH) among its members. BCBSLA noted a strong association between members who were predicted as “high risk” in their ROH AI Predictive Model and members who actually had a COVID-19 admission. As a result, BCBSLA was asked to create a simplified model that could extend to the entire state of Louisiana knowing that detailed member data would not be available to score Louisiana’s entire population.

METHODS : The current ROH model employed by BCBSLA for its members was simplified for State use by focusing on available health factors that could be easily assessed. The simplified model was tested using 2019 BCBSLA membership to predict all future hospitalizations between Jan-Jun 2020.

RESULTS : The factors chosen for the new simplified model were age, comorbid conditions (diseases of the circulatory system, Hypertension, etc.) and particular events such as hospital or skilled nursing home facility admission within the past 12 months. The simplified model was shown to be highly predictive of future all-cause hospitalizations. The top 20% of predicted high risk members accounted for 70% of all admissions; AUC was very good at 0.89. The model was also predictive of COVID-19 admissions on a naïve population. The top 20% of members captured nearly 50% of COVID-19 admissions, AUC 0.74. The model performance was good considering early COVID-19 admissions were likely misclassified in claims data.

CONCLUSIONS : Using simplified rules, BCBSLA was able to modify their current ROH model so that Louisiana’s health department could identify patients who would be at the highest risk for hospitalized if they contracted COVID-19.