Machine learning models predict long COVID outcomes based on baseline clinical and immunologic factors

Area of Science:

• Computational biology
• Infectious disease epidemiology
• Clinical informatics

Background:

• Post-acute sequelae of SARS-CoV-2 (PASC), or long COVID, is a significant health concern with unclear mechanisms and prediction challenges.
• Lack of biomarkers and defined sub-phenotypes hinders accurate prediction of long COVID development.
• Machine learning (ML) offers a potential solution for enhancing diagnostic precision using clinical data.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting long COVID risk.
• To identify key clinical predictors of long COVID development from hospital admission data.
• To leverage ML for early identification of individuals at high risk for long COVID.

Main Methods:

• Utilized clinical data, including antibody titers and viral load at hospital admission, to train ML models.
• Assessed model predictive performance using metrics like AUROC and AUPRC.
• Conducted feature importance analysis to determine significant predictors of long COVID.

Main Results:

• ML models demonstrated predictive capabilities with median AUROC of 0.64-0.66 and AUPRC of 0.51-0.54.
• Low antibody titers and high viral loads at admission were the strongest predictors of long COVID.
• Comorbidities (respiratory, cardiac, neurologic) and female sex were also significant risk factors.

Conclusions:

• Machine learning models effectively identify patients at risk for long COVID based on initial clinical data.
• These models can facilitate early interventions, potentially improving patient outcomes.
• Predictive modeling may help mitigate the long-term public health impact of SARS-CoV-2.