Machine learning for COVID-19 mortality prediction: enhancing cart models with node-specific odds ratios

Area of Science:

• Medical Informatics
• Epidemiology
• Biostatistics

Background:

• COVID-19 poses a significant public health threat, with mortality influenced by various clinical and demographic factors.
• Accurate prediction of COVID-19 mortality is crucial for resource allocation and patient management.

Purpose of the Study:

• To evaluate a classification and regression tree (CART) model integrated with odds ratio (OR) calculations for identifying key predictors of COVID-19 mortality.
• To analyze data from 1,432 hospitalized patients during the first year of the pandemic.

Main Methods:

• A CART model was developed using demographic, clinical, and laboratory data from patient admissions.
• Model performance was assessed using multiple metrics, including F1 score, accuracy, and area under the curve (AUC).
• Odds ratios (ORs) were calculated for each node to quantify mortality risk.

Main Results:

• Mechanical ventilation was identified as the strongest predictor of COVID-19 mortality.
• Patients requiring intubation at admission with shorter hospital stays exhibited the highest mortality (99%, OR = 80).
• The CART model demonstrated excellent predictive performance with an F1 score of 0.918, accuracy of 0.903, and AUC of 0.955.

Conclusions:

• The CART-OR model provides a reliable and interpretable tool for predicting COVID-19 mortality risk.
• The model's reliance on accessible clinical parameters makes it valuable in diverse healthcare settings, including resource-limited areas.
• Findings underscore the critical impact of mechanical ventilation on outcomes and the importance of early intervention for high-risk individuals.