Screening for the High-Need Population Using Single Institution Versus State-Wide Admissions Discharge Transfer Feed

Access to programs for high-needs patients depending on single-institution electronic health record data (EHR) carries risks of biased sampling. We investigate a statewide admissions, discharge, transfer feed (ADT), in assessing equity in access to these programs.

This is a retrospective cross-sectional study. We included high-need patients at Vanderbilt University Medical Center (VUMC), who were 18 years or older, with minimum three emergency visits (ED) or hospitalizations in Tennessee from January 1 to June 30, 2021, including at least one at VUMC. We used the Tennessee ADT database to identify high-need patients with at least one VUMC ED/hospitalization, then compared this population with high-need patients identified using VUMC's Epic^® EHR database. The primary outcome was the sensitivity of VUMC-only criteria for identifying high-need patient when compared to statewide ADT reference standard.

We identified 2549 patients that had at least one ED/hospitalization and were assessed to be high-need based on the statewide ADT. Of those, 2100 had VUMC-only visits, and 449 had VUMC and non-VUMC visits. VUMC-only visit screening criteria showed high sensitivity (99.1%, 95% CI: 98.7% - 99.5%), indicating that the high-needs patients admitted to VUMC infrequently access alternative systems. Results demonstrated no meaningful difference in sensitivity when stratified by patient's race or insurance.

ADT allows examination for potential selection bias when relying upon single-institution utilization. In VUMC's high-need patients, there's minimal selection bias when relying upon same-site utilization. Further research needs to understand how biases may vary by site, and durability over time.