Disease-Course Adapting Machine Learning Prognostication Models in Elderly Patients Critically Ill With COVID-19:...

Area of Science:

• Critical care medicine
• Infectious diseases
• Machine learning in healthcare

Background:

• The COVID-19 pandemic poses significant challenges to global healthcare systems.
• Elderly individuals are disproportionately affected by severe COVID-19 and face higher mortality risks.

Purpose of the Study:

• To evaluate machine learning (ML) models for prognostication in critically ill elderly COVID-19 patients.
• To assess the dynamic incorporation of multifaceted clinical information into ML models for disease evolution.

Main Methods:

• A multicenter cohort study (COVIP) involving 151 ICUs across 26 countries.
• Development of baseline ML models (logistic regression, random forest, extreme gradient boosting) using admission variables only.
• Derivation of final ML models incorporating clinical events and time-to-event data, with external validation on a non-European cohort.

Main Results:

• 1432 elderly (≥70 years) COVID-19 ICU patients were analyzed; 56.49% survived 30 days.
• Final ML models integrating dynamic clinical data showed superior performance (AUC 0.81) compared to baseline models and conventional ICU scoring systems (e.g., SOFA score).
• Average precision improved significantly from 0.65 to 0.77 with the inclusion of dynamic clinical information.

Conclusions:

• Integrating clinical events and time-to-event data significantly enhances the accuracy of 30-day mortality prediction in critically ill elderly COVID-19 patients.
• ML models offer valuable supplementary information, potentially aiding complex decision-making for this patient group.