Avaliação de diferentes métodos de predição dinâmica em análise de sobrevivência

Abstract

This study explored the application of dynamic prediction techniques in survival analysis, assessing their effectiveness in two clinical scenarios: the use of neonatal ICU and the risk of death in patients with Chagas cardiomyopathy. Four main methods were employed: Naive 1, Naive 2, Landmark, and Joint Modeling (JM). Within the Joint Modeling framework, three baseline distributions were evaluated (Weibull, Piecewise, and Spline), allowing for a detailed analysis of the advantages and limitations of each approach. The results indicated that, in the analyzed studies, the JM technique demonstrated strong performance in terms of discrimination and calibration, matching or even surpassing other dynamic prediction approaches studied in this work, particularly due to its ability to integrate longitudinal variables. Additionally, it was observed that the aging of time-dependent variables influences model performance, highlighting the importance of continuous updates. This study reinforces the relevance of dynamic methods for complex clinical scenarios, with a direct impact on clinical practice and evidence-based decision-making.