Predicting Survival Status in COVID-19 Patients: Machine Learning Models Development with Ventilator-Related and...

Area of Science:

• Critical Care Medicine
• Pulmonology
• Biomedical Engineering

Background:

• Coronavirus disease 2019 (COVID-19) frequently necessitates intubation and mechanical ventilation due to respiratory failure.
• Extubation failure in COVID-19 patients is associated with increased mortality risk.
• Predictive tools are needed to identify patients at higher risk of extubation failure and mortality.

Purpose of the Study:

• To investigate the feasibility of using machine learning to predict survival in COVID-19 patients.
• To identify key biochemical and ventilator parameters predictive of successful extubation and survival.
• To develop early warning models for respiratory failure management in COVID-19.

Main Methods:

• A retrospective study of COVID-19 patients requiring ventilatory support from May 2021 to May 2022.
• Analysis of sequential biochemical and ventilator data (C-reactive protein, pH, PaCO2, P/F ratio) on days 0-2, 3-5, and 6-7.
• Development of survival prediction models using the Random Forest (RF) machine learning algorithm.

Main Results:

• The surviving group showed significantly lower C-reactive protein (CRP) and higher pH levels on days 0-2 compared to the non-surviving group.
• The RF model utilizing data from days 0-2 demonstrated superior predictive performance over models using later data.
• CRP, PaCO2, pH, and the P/F ratio were identified as primary indicators for survival prediction in the early model.

Conclusions:

• Early assessment of specific biochemical and ventilatory parameters can effectively predict survival in COVID-19 patients.
• CRP, pH, PaCO2, and P/F ratio within the first two days are crucial predictors of successful ventilator weaning.
• These findings support the early integration of these parameters into clinical management strategies for COVID-19-related respiratory failure.