OBJECTIVES:

The emergency department (ED) plays an important role in treating millions of patients each year. Some patients live with acute and chronic illness or experience accidents or injuries that require immediate care. However, the ED is expensive and inefficient for those visits where it is avoidable. Predictive modeling can improve the efficiency of ED use by identifying patients at high risk of visiting. This study evaluates a Blue Cross and Blue Shield of Louisiana predictive model that identifies members at high risk of readmitting to the emergency department.

METHODS:

The traditional claims-based risk models meant to target these outcomes rely on completed claims data, which can take months to arrive. An alternative approach is to rely on admission-discharge-transfer (ADT) data. This contains important patient medical information that is created at the start of a medical encounter. For this model, ADT records representing office visits, transfers, ED visits or hospital admissions serve as the index visit. These records are joined to internal sources containing data on a member’s demographics, social determinants of health vulnerabilities and prior health care utilization. Models are trained to predict a visit to the ED within 30 days.

RESULTS:

The final model has an area under the curve statistic of 0.71 in the training and 0.72 in the testing sets. The precision or positive predictive value in the top 1%, 10% and 25% were 70%, 38%, and 28%, while the recall or sensitivity was 15%, 25% and 45%.

CONCLUSIONS:

Using this predictive model and relying on incoming ADTs as a critical source of real-time data, Blue Cross can quickly identify patients at risk for a readmission or ED visit. This speed of identification allows its Care Management team to offer timely intervention leading to improved health outcomes for members.