Dichotomous outcomes vs. survival regression models for identification of predictors of mortality among patients with severe acute respiratory illness during COVID-19 pandemics

Abstract

IntroductionAs the studies predicting mortality in severe acute respiratory illness (SARI) have inferred associations either from dichotomous outcomes or from time-event models, we identified some clinical-epidemiological characteristics and predictors of mortality by comparing and discussing two multivariate models.MethodsTo identify factors associated with death among all SARI hospitalizations occurred in Botucatu (Brazil)/regardless of the infectious agent, and among the COVID-19 subgroup, from March 2020 to 2022, we used a multivariate Poisson regression model with binomial outcomes and Cox proportional hazards (time-event). The performance metrics of both models were also analyzed.ResultsA total of 3,995 hospitalized subjects were included, of whom 1338 (33%) tested positive for SARS-CoV-2. We identified 866 deaths, of which 371 (43%) were due to the COVID-19. In the total number of SARI cases, using both Poisson and Cox models, the predictors of mortality were the presence of neurological diseases, immunosuppression, obesity, older age, and need for invasive ventilation support. However, the Poisson test also revealed that admission to an intensive care unit and the COVID-19 diagnosis were predictors of mortality, with the female gender having a protective effect against death. Likewise, Poisson proved to be more sensitive and specific, and indeed the most suitable model for analyzing risk factors for death in patients with SARI/COVID-19.ConclusionGiven these results and the acute course of SARI and COVID-19, to compare the associations and their different meanings is essential and, therefore, models with dichotomous outcomes are more appropriate than time-to-event/survival approaches.