PNNL

Abstract

Developing prediction models for emerging infectious diseases from relatively small numbers of source cases is a critical need for improving the early epidemic response. Using COVID-19 as an exemplar, we demonstrate TRANSMED, a methodology for developing high-performance multi-modal BERT model for electronic healthcare records using existing disease data. Our methodological innovations include the application of transfer learning from more prevalent diseases that share important clinical characteristics with patients severely impacted by COVID-19. Our hierarchical multi-modal model integrates baseline risk factors from the natural language processing of clinical notes at admission, time-series measurements of biomarkers obtained from laboratory tests, and discrete diagnostic, procedure and drug codes. We demonstrate the alignment of TRANSMED’s predictions with well-established clinical knowledge about COVID-19 through univariate and multivariate risk factor driven sub-cohort analysis. TRANSMED’s superior performance over state-of-the-art methods shows that leveraging patient data across modalities and transferring prior knowledge from similar disorders is critical for accurate prediction of patient outcomes.