Objective: To determine which early-stage variables best predicted the deterioration of coronavirus disease 2019 (COVID-19) among community-isolated people infected with severe acute respiratory syndrome coronavirus 2 and to test the performance of prediction using only inexpensive-to-measure variables.
Methods: Medical records of 3145 people isolated in two Fangcang shelter hospitals (large-scale community isolation centers) from February to March 2020 were accessed. Two complementary methods-machine learning algorithms and competing risk survival analyses-were used to test potential predictors, including age, gender, severity upon admission, symptoms (general symptoms, respiratory symptoms, and gastrointestinal symptoms), computed tomography (CT) signs, and comorbid chronic diseases. All variables were measured upon (or shortly after) admission. The outcome was deterioration versus recovery of COVID-19.
Results: More than a quarter of the 3145 people did not present any symptoms, while one-third ended isolation due to deterioration. Machine learning models identified moderate severity upon admission, old age, and CT ground-glass opacity as the most important predictors of deterioration. Removing CT signs did not degrade the performance of models. Competing risk models identified age ≥ 35 years, male gender, moderate severity upon admission, cough, expectoration, CT patchy opacity, CT consolidation, comorbid diabetes, and comorbid cardiovascular or cerebrovascular diseases as significant predictors of deterioration, while a stuffy or runny nose as a predictor of recovery.
Conclusions: Early-stage prediction of COVID-19 deterioration can be made with inexpensive-to-measure variables, such as demographic characteristics, severity upon admission, observable symptoms, and self-reported comorbid diseases, among asymptomatic people and mildly to moderately symptomatic patients.
Keywords: COVID-19; SARS-CoV-2; competing risk survival analysis; machine learning algorithm; progression.
© 2023 The Authors. Journal of Evidence-Based Medicine published by Chinese Cochrane Center, West China Hospital of Sichuan University and John Wiley & Sons Australia, Ltd.