JMIR Publications

ABSTRACT

Background:

Heart rate variability (HRV) is a validated biomarker of autonomic and inflammatory regulation and has been associated with COVID-19 and its sequelae. Changes in HRV patterns have been reported during acute infection and in long COVID. While RT-PCR remains the gold standard for diagnosing acute infection, there is still a lack of accessible, noninvasive, and physiological tools to support continuous monitoring and characterization of both COVID-19 infection and long COVID. With the growing use of wearable devices capable of real-time HRV collection, new opportunities emerge for early detection and health status differentiation through machine learning.

Objective:

This study aimed to identify HRV patterns capable of distinguishing individuals with active COVID-19, long COVID, and healthy controls, using data collected from wearable devices and processed with machine learning models. A secondary objective was to assess the feasibility of a near-real-time health monitoring system based on these patterns using wearable-derived HRV data.

Methods:

HRV indices (SDNN, RMSSD, LF%, HF%) were collected from 61 participants (21 with COVID-19, 20 with long COVID, and 20 healthy controls) using two standardized datasets. Classification models were trained using supervised machine learning (Decision Tree, SVM, k-NN, Neural Networks) and validated via cross-validation. An additional variable, recent COVID-19 infection, was included to enhance prediction accuracy. A prototype near-real-time monitoring system was also developed and tested on 4 independent participants.

Results:

HRV indices (SDNN, RMSSD, LF%, HF%) revealed a distinct physiological signature across groups. Participants with active COVID-19 exhibited significantly lower values (SDNN: 18.4±9.4 ms; RMSSD: 10.4±5.5 ms; LF%: 17.4±10.0%; HF%: 13.5±16.4%) compared to healthy controls (SDNN: 34.7±10.2 ms; RMSSD: 28.0±9.9 ms; LF%: 61.5±14.5%; HF%: 30.2±13.4%) and long COVID participants (SDNN: 29.1±9.4 ms; RMSSD: 22.0±8.9 ms; LF%: 60.9±14.5%; HF%: 29.2±14.8%) with P<.05. Machine learning classifiers trained on these HRV signatures achieved 77% accuracy, improving to 96.7% when prior COVID-19 infection history was included as an additional feature. A near-real-time system tested on four independent participants correctly classified their status, demonstrating initial feasibility of wearable-based implementation.

Conclusions:

HRV patterns collected by wearable devices and processed through machine learning successfully differentiated between active COVID-19, long COVID, and healthy states. A system developed for near-real-time monitoring showed promise by functioning as intended, although it remains a proof of concept requiring further validation. These findings support the potential of HRV as a noninvasive biomarker for monitoring and early detection of COVID-19 and its long-term effects.