Discovering and Applying Knowledge in Clinical Databases

PROJECT Summary: The long-term goal of our parent project, "Discovering and applying knowledge in clinicaldatabases," is to learn from data in the electronic health record (EHR) and to apply thatknowledge to understand and improve health. Its first two aims are as follows: (1) Taking aknowledge engineering approach, study the effect of preprocessing and analytic choices onreducing health care process bias, and using machine learning techniques, learn more abouthealth care process bias. (2) Taking a more empirical approach, use dynamic latent factormodeling and variation inference to accommodate health care process bias, learning how apatient's health state and health processes affect censoring, exploiting information from manyvariables at once.For this supplement, we plan to focus on COVID-19. The emergence of the virus SARS-CoV-2and its corresponding disease, COVID-19, has led to about 100,000 deaths in the US and greateconomic loss and human suffering. Carrying out randomized clinical trials to assess treatmentis essential but stymied by the difficulty recruiting sufficient patients and the urgency of thequestion. Clinical databases are beginning to fill with COVID-19 patients, but the acuity andseverity of the disease make drawing causal conclusion much more difficult, resulting in aliterature filled with conflicting observational studies.We propose to employ structural causal modeling in the study of COVID-19, engaging expertisein such modeling. We will use the Columbia University Irving Medical Center's clinical datawarehouse with over 6000 testing positive for SARS-CoV-2 and the Observational Health DataScience and Informatics (OHDSI) network, which includes most COVID-19 patients in Korea,Spain, the US Veterans Administration, Stanford, Tufts, and new sites coming on board.