Programmatic Theme: Informatics Implementation

Abstract: Predictive models have been proposed for optimizing advanced care planning but are often not evaluated for their clinical utility. We describe a pre-deployment assessment of a mortality prediction model for its potential for improving advanced care planning (ACP) by comparing it to an existing clinical heuristic. The model significantly outperforms a clinical heuristic in detecting patients in high-need for ACP and provides an average of 2 months lead time over the heuristic for advanced care planning.

Learning Objective: Audience will learn methods for estimating incremental utility of a machine learning model over existing clinical heuristics.

Authors:

Sehj Kashyap (Presenter)
Stanford University

Kenneth Jung, Stanford University
Stephanie Harman, Stanford University School of Medicine
Margaret Smith, Stanford University School of Medicine
Nigam Shah, Stanford University
Ron Li, Stanford University School of Medicine