Feasibility of continuous fever monitoring using wearable devices

Fever is a critical biomarker for various illnesses, including COVID-19. Current fever monitoring relies on single-point temperature measurements, which lack sensitivity and fail to capture the dynamic nature of temperature fluctuations. This study aimed to assess the feasibility of using continuous temperature data from wearable devices to improve fever detection. The limitations of single-point measurements, such as their inability to account for individual circadian rhythms, menstrual cycles, and other biological factors, make them less reliable. Continuous temperature monitoring offers a potential solution by providing contextual information and high temporal resolution. This research specifically focused on the use of the Oura ring, a commercially available wearable device, to capture continuous physiological data, including peripheral temperature, and to explore its potential for early illness detection and prediction during the COVID-19 pandemic. The study's success could significantly advance public health efforts by providing a readily available and scalable system for monitoring illness across large populations.

Existing literature highlights the importance of fever as an early symptom in various infections, including COVID-19, emphasizing the need for effective fever monitoring. However, traditional single-point temperature measurements have limitations in detecting early-stage illnesses due to individual variability in body temperature and the lack of contextual information. While some research suggests the difficulties in using single-point temperature measures to detect illness, the potential of wearable devices to overcome these limitations has not been comprehensively explored. Studies on wearable temperature sensors for illness detection are limited. The researchers aimed to fill this gap and provide evidence for the feasibility of utilizing such devices in detecting fever and predicting illness onset.

The TemPredict study enrolled 110 participants who reported experiencing COVID-19 symptoms. After excluding participants due to missing data, incomplete information, or inconsistent reporting, the study included 50 participants with sufficient data for analysis. Participants wore the Oura ring, a commercially available wearable device, which continuously monitored peripheral temperature, heart rate, heart rate variability, and respiration rate. The data was collected for 65 days, including a 45-day baseline period before the onset of reported symptoms and a 20-day period afterward. Data was preprocessed to 1-minute resolution using linear interpolation. Daily minimum and maximum temperature values were calculated, normalized, and used to create digital biomarkers for fever detection. Thresholds were applied to the normalized data to identify “fever-like” days. The study also analyzed changes in heart rate, heart rate variability, and respiration rate in conjunction with temperature to assess the potential for multi-metric illness detection. Wavelet analysis was employed to identify changes in circadian power to assess the potential of fever prediction before symptom onset. Statistical analysis included Wilcoxon rank-sum tests, Kruskal-Wallis tests with Tukey-Kramer post-hoc comparisons, and Pearson correlations.

The study demonstrated that wearable temperature sensors can detect illness-associated temperature elevations that correlate with self-reported fever. Significant increases in mean temperature during the symptom window compared to the baseline period were observed. The analysis of daily minimum and maximum temperatures revealed sharp increases around symptom onset in participants who reported fever. The creation of digital biomarkers using temperature thresholds successfully identified fever-like days in the majority of participants who reported fever. The researchers found that integrating additional physiological variables such as heart rate, heart rate variability, and respiration rate enhanced the accuracy of fever detection. A significant number of participants (38 out of 50) showed temperature anomalies before symptom onset, suggesting the potential for early illness prediction using wearable sensors. Wavelet analysis identified unique circadian power peaks within one week of fever-like episodes, further supporting the feasibility of prediction.

The findings refute concerns about the suitability of distal body temperature for fever detection based on single time-point measurements. The study’s use of continuous data revealed temperature changes associated with illness that would be missed by single measurements. The enhanced physiological differences observed after re-sorting participants by digital biomarkers compared to self-reported fever suggest that wearable sensors may be more sensitive in detecting subtle temperature changes that go unnoticed or unreported. The integration of multiple physiological signals offers potential for improved illness prediction accuracy compared to temperature alone. The observed changes in heart rate, heart rate variability, and respiration rate along with temperature fluctuations indicate a coordinated physiological response to infection. The study's limitations include the lack of serological or ground-truthing measures to confirm illness diagnoses and the use of a relatively small sample size.

This study provides strong evidence for the feasibility of using wearable temperature sensors for fever detection and potential illness prediction. The use of continuous data and the integration of multiple physiological variables significantly enhanced the detection and prediction accuracy. Future research should focus on validating these findings with larger, more diverse populations and incorporate additional physiological data for more robust and specific illness detection. The development of more refined digital biomarkers using machine learning techniques should also be explored.

The study’s relatively small sample size limits the generalizability of the findings. The lack of objective confirmation of COVID-19 diagnoses (e.g., serological testing) is another limitation. The study’s reliance on self-reported symptoms may also introduce biases due to recall error or varying symptom awareness among individuals. Further, the specific digital biomarkers developed in this study might not generalize well to different populations or disease types.