Developing and Validating Multi-Modal Models for Mortality Prediction in COVID-19 Patients: a Multi-center...

Area of Science:

• Computational biology and bioinformatics
• Medical informatics
• Machine learning in healthcare

Background:

• The COVID-19 pandemic necessitated the development of predictive models for patient prognostication and resource management.
• Existing machine learning models for COVID-19 often suffer from methodological limitations and inadequate validation, hindering clinical utility.
• Multi-modal data integration, combining clinical information and medical imaging, shows promise for improving predictive accuracy.

Purpose of the Study:

• To develop and validate multi-modal machine learning models for predicting COVID-19 mortality.
• To assess the performance of these models across diverse, multi-center patient cohorts.
• To provide a methodological framework and share code for building robust clinical prediction models.

Main Methods:

• Development of COVID-19 mortality prediction models using retrospective data from Madrid, Spain (N=2547).
• External validation of models in patient cohorts from New Jersey, USA (N=242) and Seoul, Republic of Korea (N=336).
• Utilized multi-modal data, integrating structured electronic health records and chest X-ray imaging.

Main Results:

• Multi-modal models integrating electronic health records and chest X-ray data demonstrated superior 30-day mortality prediction performance across all validation datasets.
• Achieved areas under the receiver operating characteristic curves of 0.85 (95% CI: 0.83-0.87), 0.76 (95% CI: 0.70-0.82), and 0.95 (95% CI: 0.92-0.98) in the respective cohorts.
• Model performance varied across clinical settings, highlighting the importance of guided machine learning implementation.

Conclusions:

• Validated multi-modal machine learning models offer a promising approach for COVID-19 mortality prediction.
• The integration of clinical and imaging data enhances predictive capabilities in diverse healthcare environments.
• Adherence to best practices in machine learning development and validation is essential for clinical decision support tools.