Clinical Prediction With Time-Varying Predictors in Pediatric Hospital Medicine

Clinical prediction models traditionally rely on measurements obtained at a single time point to estimate risk and guide management. However, the clinical status of hospitalized children often evolves during admission, and time-varying measurements obtained over the course of hospitalization may provide more accurate, dynamic estimates of risk that better reflect clinical trajectories. Time-varying prediction models are therefore an increasingly important methodological opportunity for pediatric hospitalists seeking to leverage patient-specific longitudinal data in clinical practice. In this article, we review the development of clinical prediction models using time-varying predictors in pediatric hospital medicine, including appropriate clinical settings and data recording, data structuring, analytic approaches, and advantages and limitations. We illustrate these principles using a published example of a time-varying model developed to predict in-hospital mortality among severely malnourished children, highlighting how incorporating daily clinical signs improved predictive accuracy compared with a single time point model using admission data alone.