Longitudinal wearable sensor data enhance precision of Long COVID detection

Area of Science:

• Biomedical Engineering
• Computational Biology
• Infectious Diseases

Background:

• Long COVID diagnosis is challenging due to a lack of objective biomarkers.
• This often results in underdiagnosis and misdiagnosis of persistent post-SARS-CoV-2 symptoms.
• Wearable sensor data presents a potential avenue for objective diagnostic markers.

Purpose of the Study:

• To investigate the utility of wearable heart rate (HR) sensor data for Long COVID diagnosis.
• To develop and evaluate machine learning models for predicting Long COVID status.
• To assess the added value of HR features compared to symptom-based data.

Main Methods:

• Longitudinal HR data from 126 individuals post-SARS-CoV-2 infection were analyzed.
• Machine learning models were trained using HR features (derived from six categories) and reported symptoms.
• Model performance was evaluated using Receiver Operating Characteristic (ROC) and Precision-Recall (PR) curves.

Main Results:

• A combined model using HR and symptom features achieved 95.1% ROC-AUC and 85.9% PR-AUC.
• This represents a significant improvement over symptom-only models.
• HR data provided objective features that enhanced diagnostic discrimination.

Conclusions:

• Wearable HR data can serve as an objective biomarker for Long COVID diagnosis.
• Integrating HR data with symptom information improves diagnostic precision.
• This non-invasive approach can reduce misclassification and improve patient outcomes.