A Multilevel Mediation Model for Categorical Mediator and Outcome: Assessing the Impact of the HEART Care Pathway on Physician Inpatient Admitting Rates and Patient Admission Risk

• Activity Number: 66 - Novel Bayesian Methodology with Health Applications
• Type: Contributed
• Date/Time: Monday, August 3, 2020 : 10:00 AM to 2:00 PM
• Sponsor: WNAR
• Author(s): Ernest Shen* and Aileen Baecker and Adam Sharp and Yi-Lin Wu and Benjamin Sun
• Companies: Kaiser Permanente and Kaiser Permanente and Kaiser Permanente and Kaiser Permanente and University of Pennsylvania
• Keywords: Bayesian Statistics; Electronic Health Records; Healthcare Utilization; Mediation Analysis

Abstract

In May 2016, Kaiser Permanente Southern California implemented the HEART care pathway in order to “reduce hospitalization and noninvasive stress testing in 13 community EDs in Southern California.” One of the primary outcomes was admission to inpatient/observation care, and the aim was to assess whether admission rates changed post-HEART implementation at the patient level. However, it is also of interest to know whether differences in Emergency Physicians’ behavior mediated the association between HEART implementation and patient IP admission risk. Using the published data, we constructed physician rates in each period (per 100 visits), and linked them with patient outcomes in each period, in order obtain the correct time-ordering of the mediator and outcome variables. We then used a Bayesian multilevel mediation model, with a log-Poisson specification for both the mediator and outcome models, and used the product of rate ratios to quantify the indirect effect of physician-level admitting rates. Our preliminary data indicate a small indirect effect, suggesting that physician admitting rates partially mediated the effect of HEART implementation on patient admission risk.