Genetic and Survey Data Improves Performance of Machine Learning Model for Long COVID

Area of Science:

• Computational biology
• Epidemiology
• Medical informatics

Background:

• Over 200 million individuals worldwide experience persistent symptoms (long COVID) after SARS-CoV-2 infection.
• The National COVID Cohort Collaborative (N3C) developed a machine learning model using electronic health records (EHRs) to identify long COVID patients.
• The potential for multi-modal data integration to improve long COVID prediction remains largely unexplored.

Purpose of the Study:

• To enhance the predictive accuracy of long COVID identification.
• To evaluate the utility of incorporating health survey, mobile device, and genetic data into existing EHR-based models.
• To identify key predictive factors across diverse data modalities.

Main Methods:

• Utilized a cohort of 17,755 SARS-CoV-2 infected individuals from the All of Us program.
• Applied and expanded the N3C long COVID prediction model.
• Tested machine learning infrastructures, including extreme gradient boosting and convolutional neural networks, for survey/mobile and genetic data, respectively.
• Assessed model performance using metrics like Area Under the Curve (AUC) of the Receiver Operating Characteristic (ROC).

Main Results:

• Extreme gradient boosting and convolutional neural networks showed high performance for survey/mobile and genetic data, respectively.
• Combined multi-modal data (survey, genetic, mobile) significantly improved model specificity and AUC compared to the original N3C EHR-only model.
• Identified key contributing factors to long COVID prediction across the integrated datasets.

Conclusions:

• Integrating diverse data sources, including health surveys, mobile data, and genetics, substantially improves the prediction of long COVID.
• Machine learning models can effectively leverage multi-modal data for more accurate identification of patients with persistent post-COVID conditions.
• This enhanced predictive capability can facilitate targeted interventions and research for long COVID.