Abstract:
|
Streaming data are becoming more common in a variety of fields. One common data stream in clinical medicine are electronic health records (EHRs) which have been used to develop risk prediction models. Our motivating application considers the risk of patient deterioration - defined as in-hospital mortality or transfer to the ICU. Duke University Hospital recently implemented an alert risk score for acute care wards: the National Early Warning Score (NEWS). However, it is questionable whether it optimally uses the data and accounts for sparsity in the features for each individual (feature-poor data). We introduce a simple recalibration of the NEWS, and then consider a general cox proportional hazard model that can be tailored to any patient population. We propose a joint longitudinal-Cox model, where the features adaptively update for each individual in real-time. Finally, we introduce a novel set of rich, flexible, and robust semiparametric Bayesian models that allow the features to update adaptively over time to produce dynamic risk evaluations as more data are collected. We apply these methods to the NEWS, and make comparisons on risk and cost of implementation in an EHR setting.
|