Learning and Predicting from Dynamic Models for COVID-19 Patient Monitoring

COVID-19 has challenged health systems to learn how to learn. This paper describes the context and methods for learning at one academic health center. Longitudinal regression models are used to represent the joint distribution of major clinical events including discharge, ventilation and death as well as multivariate biomarker processes that describe a patient's disease trajectory. We focus on dynamic models in which both the predictors and outcomes vary over time. We contrast prospective longitudinal models in common use with retrospective analogues that are complementary in the COVID-19 context. The methods are described and then applied to a cohort of 1,678 patients with COVID-19 who were hospitalized in the first year of the pandemic. Emphasis is placed on graphical tools that inform clinical decision making.