MultSurv: A Multimodal Deep Learning Model for Hospitalized Patients Survival Analysis in the Context of a Pandemic
Resumo
Respiratory infectious diseases continue to be a significant public health challenge. For example, the COVID-19 pandemic has exposed critical weaknesses in healthcare systems worldwide. The heterogeneity of clinical manifestations in hospitalized patients highlights the need for risk stratification methods to aid in better patient management and resource allocation. In this study, we propose MultSurv MultSurv, a multimodal survival analysis model that integrates clinical, laboratory, and chest X-ray data to capture the temporal dynamics of disease progression. The methodology comprises five main components: pre-processing, feature encoders, temporal attention, CheXReport for image analysis, and multitask neural networks for risk assessment. The model was evaluated using a private dataset, demonstrating superior performance over state-of-the-art methods, with a C-index of 0.723(t = 1) and 0.695(t = 7). The proposed approach can improve patient prioritization in pandemic scenarios and be adapted for broader clinical applications, potentially contributing to better decision-making in healthcare settings.
Referências
Lee et al. (2019). Dynamic-deephit: A deep learning approach for dynamic survival analysis with competing risks based on longitudinal data. IEEE Transactions on Biomedical Engineering.
Rahman et al. (2020). Zoonotic diseases: etiology, impact, and control. Microorganisms.
Zeiser, F. A. et al. (2022). First and second covid-19 waves in brazil: A cross-sectional study of patients’ characteristics related to hospitalization and in-hospital mortality. The Lancet Regional Health-Americas.
Zeiser, F. A. et al. (2024). Chexreport: A transformer-based architecture to generate chest x-ray reports suggestions. Expert Systems with Applications.
