Deep Survival Analysis With Clinical Variables for COVID-19.

Area of Science:

• Medical Informatics
• Computational Biology
• Machine Learning

Background:

• Coronavirus disease 2019 (COVID-19) has caused significant global mortality.
• Artificial intelligence (AI) is increasingly vital in patient care, particularly for prognostics.
• Predictive modeling for COVID-19 survival outcomes remains a critical area of research.

Purpose of the Study:

• To introduce a novel predictive model for COVID-19 patient survival using one dimensional convolutional neural networks (1D CNN).
• To leverage clinical variables for accurate short-term and long-term survival predictions.
• To compare the performance of the 1D CNN model against established methods like Random Forest (RF).

Main Methods:

• Survival analysis was conducted using univariate analysis (Log-rank test, Kaplan-Meier estimator).
• A Random Forest (RF) model served as the baseline for comparison.
• A proposed 1D CNN model was developed and evaluated using a comprehensive set of 44 clinical variables.

Main Results:

• Univariate analysis identified nine clinical variables significantly associated with COVID-19 survival (corrected p < 0.05).
• The 1D CNN model demonstrated superior performance metrics compared to the RF baseline and other state-of-the-art techniques.
• The model's effectiveness was validated using clinical variables, showing promising predictive capabilities.

Conclusions:

• The 1D CNN model offers a promising tool for early detection of mortality risk in COVID-19 patients.
• Timely identification of at-risk patients can facilitate the development of effective treatment plans.
• The integration of AI with clinical data holds potential for point-of-care services and rapid healthcare system learning.