Temporal Phenotypes and Risk Models for the Post-COVID Syndrome and its sub-types

点击翻译按钮获取中文摘要

Project Summary/Abstract The fight against the SARS-CoV-2, the coronavirus that causes COVID-19, is ramping up with vaccinations and therapeutics. Yet there is a growing urgency to study and address the other side of the pandemic, a shapeshifting byproduct known as the post-acute sequelae of COVID-19 (PASC), among other names. Even if several millions are successfully vaccinated against the virus, many more are still likely to be infected and for hundreds of thousands (if not millions) of those, recovery from the acute phase of COVID-19 infection will be grueling with a debilitating second act. A collection of persistent physical (e.g., fatigue, dyspnea, chest pain, cough), psychological (e.g., anxiety, depression, post-traumatic stress disorder), and neurocognitive symptoms (e.g., impaired memory and concentration) can appear and last for weeks or months in patients after acute COVID- 19, impeding their ability to function normally and costing the U.S. economy billions of dollars annually in medical bills and lost incomes. However, little is known about the post-acute sequelae of COVID-19, the extent and causes of its lingering health issues, which patients might develop them, and how to address them. We seek to leverage electronic health records (EHRs) data from 7 hospital systems across the U.S. to develop and validate a novel framework for studying evolving temporal phenotypes of the post-acute sequelae of COVID-19. For a period of four years after each patient’s SARS-CoV-2 infection, we will track their clinical data, including clinical notes and laboratory tests recorded in EHR notes to Curate validated cohorts with gold-standard, rule-based, and silver-standard (computationally interpolated) labels for PASC phenotypes (Aim 1). We will utilize these cohorts to develop and validate consistent and interpretable cohort identification and risk models of PASC phenotypes, accounting for temporal ordering and progression of evolving phenotypes over time (Aim 2). Finally, we will evaluate the generalizability the PASC models to develop a framework for modeling evolving temporal phenotypes with EHR data through an objective methodology for evaluating bias in medical AI. (Aim 3). This study will yield new knowledge regarding the phenotypic characteristics of the post-acute effects following known SARS-CoV-2 infection and the underlying drivers that influence their presentation and onset. The novel framework for studying evolving PASC phenotypes will capture and characterize new PASC problems that may present 2-3 years post-acute infection and update risk models. Given uncertainties around the efficacy of current vaccines against future mutations, the proposed learning systems will improve our capacity for adaptive pandemic decision-making. Finally, the framework and underlying methodology developed in this study will lend insights towards understanding persistent sequelae of other known/suspected viral infections and modeling other evolving temporal phenotypes.