|
Abstract:
|
The stepped-wedge cluster randomized trial (SW-CRT) design is well-suited for evaluation of healthcare delivery interventions. Appealing features of SW-CRTs include having each cluster acting as their own control and not needing to withhold the intervention from any patient. However, the design and analysis of SW-CRT is complex, and methodology are not available for many settings. Machine learning (ML) based prediction models are gaining popularity in healthcare settings due to their ability to process data through semi-automated electronic medical record (EMR)-embedded pipelines and their capacity to achieve desired prediction accuracy through continuous learning. SW-CRTs are appropriate for evaluating ‘clinical decision support systems (CDSSs)’ built using output from ML algorithms. In this round table, I will discuss an ongoing SW-CRTs at Mount-Sinai Health System evaluating a CDSS intervention using ML based mortality predictive data to improve ‘goals of care’ discussions for solid cancer patients at high risk of short-term mortality. I’ll discuss the complexity in the design and implementation process and highlight scenarios where methodology development is needed.
|
