Comparative Assessment of Parametric Accelerated Failure Time and Cure Models for Survival Analysis in Clinical Studies

Abstract

This study examined the survival outcomes of 322 patients diagnosed with COVID-19 and admitted to hospitals in Campinas, São Paulo, Brazil. Parametric Accelerated Failure Time (AFT) and cure models were applied to evaluate survival patterns, identify factors influencing patient survival, and compare the suitability of different parametric distributions for modeling COVID-19 survival outcomes. Model performance was assessed using log-likelihood and Akaike Information Criterion (AIC). Among the AFT models, the Weibull AFT model provided the best fit to the data (log-likelihood = -436.93, AIC = 891.85), outperforming the log-normal and exponential AFT models. Similarly, the Weibull cure model demonstrated superior performance among the cure models (log-likelihood = -441.71, AIC = 903.42). Results from the AFT models showed that age, diabetes, and neurological disorders significantly influenced survival time, while age was the only covariate consistently significant in the cure models. Other factors, including sex, asthma, heart disease, and obesity, were not statistically significant. The findings underscore the effectiveness of the Weibull distribution for modelling survival outcomes and cure fractions, while highlighting the value of cure models in providing a more comprehensive understanding of long-term patient survival.