POSA393 Online Risk Calculator Using Machine Learning for Prediction of Survival After Hospital Discharge from COVID-19 Infection

ABSTRACT

Objectives: Predicting survival and risk of death after hospital discharge due to COVID-19 can help in screening patients who require special care after hospitalization. This study evaluates the survival curve and associated factors with mortality after COVID-19 admissions. Methods: Retrospective analysis until May 2021 from administrative database of 37,462 people. Analysis included 810 inpatients admitted with COVID-19 followed each month regarding survival after hospital discharge. Survival analysis performed using Cox Ridge Penalized Regression (CRPR), Gradient Boost Survival (GBS) and Random Survival Forest (RFS) from the Python library scikit-survival. Dataset separated into training and test set with the proportions of 75% and 25% respectively. Our predictive variables were patient’s age, sex, if had any comorbidity, cancer, hospitalization longer than 14 days or intensive care unit (ICU) stay. Results: From the 810 patients, 125 had died after hospital discharge, mean time of death 9.28 months. Model performance evaluated through the Concordance Index (C-Index) metric. CRPR had better performance with a C-Index of 0.74, while RFS and GBS had a C-Index of 0.73. Risk of death at any time during the follow-up period was significantly higher when presence of previous comorbidities (p=0.020), age greater than 60 years (p <0.001), ICU stay (p <0.001), and higher average length hospital stay (p=0.001). Conclusions: Several tools have been developed for to calculate absolute risk or chances of needing to go into hospital or dying from Covid-19. The online risk calculator that we developed is unique and suitable to predict which person present high risk of death after COVID-19 hospital discharges and prioritizing individuals to receive special care after leaving the hospital. Models like the one we have developed are only as good as the data they are trained on. We will update the calculator as the amount of data we are able to collect increases.