Feasibility of Measuring Physiological Responses to Breakthrough Infections and COVID-19 Vaccine Using a Wearable Ring Sensor

Widespread vaccination programs were enacted globally during the second and third years of the COVID-19 pandemic, with vaccinated individuals at lower risk of developing infection. At the same time, the emergence of highly transmissible variants and changes in public policy increased the absolute number of breakthrough infections in this group. For example, infection rates in Finland reached all-time highs in the first quarter of 2022 [1]. Continuous monitoring of the physiological response to infection via wearable biometric sensors has been used to detect illness [2‒7] and could aid efforts to curb transmission rates of COVID-19 and other illnesses by alerting individuals early to possible infection.

This study reports data from the Oura Ring, one of several commercially available wearable devices equipped with temperature, accelerometer, and photoplethysmography sensors. Sensor data are transmitted from the device to the user’s phone, from where they are uploaded to the cloud. Software running on the phone and in the cloud processes data, detecting and quantifying periods of activity and sleep. A phone app then gives consumers detailed information about their exercise and sleep patterns, as well as physiological measurements directly derived from the sensors. Commercial wearable devices are typically worn around the clock on an ongoing basis: for example, the median daily wear time among Oura users is 23.2 h. This enables the establishment of personalized biometric baselines for each user and allows deviations from those baselines to be detected as anomalous. See [6, 8] for more details on Oura hardware and usage.

Previous work has demonstrated that biometric signals from commercial wearable devices enable early detection of COVID-19 and other flu-like illnesses using biometrics including heart rate during rest [9] or rest and movement [2], or in combination with sleep duration and activity levels [5, 10], and breathing rate and dermal temperature [3, 6, 7]. See also survey on medical- and consumer-grade sensors in [4].

Researchers have also described the physiological response measured by wearable sensors to COVID-19 vaccination. Vaccines are associated with deviations in resting heart rate and heart rate variability (HRV) [11, 12], dermal temperature, and time in deep sleep [11]. The magnitude of temperature deviation appears to correlate with the serum titer of antibodies generated by the vaccines [11], suggesting that wearable technology could provide an approximate estimate of antibody protection.

Here, we characterize the physiological response to infection in vaccinated and unvaccinated individuals using data from a commercial wearable. We additionally describe the physiological response to vaccination itself. Our results give a quantitative description of these responses in five biomarkers and support the possibility of using these biomarkers to develop screening tools but are not in themselves clinically actionable.

All data come from users of the Oura commercial product. Users were included in the infection dataset (n = 838) when they entered “Confirmed COVID-19” using the in-app tagging feature and in the vaccination dataset (n = 20,267) when they selected “COVID-19 Vaccine” (see Table 1). Users tag by entering an unprompted “add tag” interface and selecting from a list of diverse events like “alcohol,” “hot bedroom,” etc. Data ranged from December 2020 to September 2021. Users with insufficient data around the time of the tag (14 of 22 days between D-14 and D + 7 for infection dataset, 6 of 9 days between D-3 and D + 5 for vaccination dataset) or without at least 14 of 28 days of baseline data were excluded (41.0% of infection dataset, 13.7% of vaccination dataset). Within the infection dataset, a subset of breakthrough cases (n = 101) was identified by taking users who had reported at least one vaccination event at least 2 weeks before the reported infection confirmation date. See Table 1 for demographic breakdown. Most users did not report which specific vaccine was administered, but among those who did, the most common were Pfizer or Moderna two-dose vaccines.

Nightly average temperature, heart rate, rMSSD, breathing rate, and sleep efficiency (fraction of time in bed spent asleep) were measured for each user in the month before and after the vaccination/infection event. rMSSD is a measure of HRV. Physiological responses were converted to z-scores using mean and standard deviation from a 28-day baseline period 35 to 7 days before the vaccination or 48 to 21 days before the infection report. An earlier window was used for infections than vaccines as there is some uncertainty in the true date of infection, and we intended to minimize the chance that any post-infection data could affect the calculation of baselines.

False discovery rate (FDR) was controlled using the Benjamini-Hochberg procedure across all tests conducted, with an FDR of 5%. Parametric t-tests were chosen given the large dataset sizes. Tests for differences in each metric at the point of infection/vaccination were prespecified. Cohen’s d is reported as effect size along with the t-statistic, degrees of freedom, and FDR-adjusted p value.

Reported confirmed infection is associated with deviations in skin temperature, breathing rate, rMSSD, and sleep efficiency. Deviations around the point of reported infection are statistically distinguishable from individuals’ baseline measurements taken beforehand in both groups. Note that these metric deviations have not been validated against (and should not be equated with) symptom severity.

We observed clear physiological responses related to reported infections (n = 838) as early as 2.5 days before the user-reported event (Fig. 1a; comparison of time range from 7 days before/after tag to baseline, see Methods; temperature: t (837) = 20.98, p < 0.001, d = 0.72; heart rate: t (837) = 10.85, p < 0.001, d = 0.37; breath rate: t (837) = 18.96, p < 0.001, d = 0.65; rMSSD: t (837) = −7.32, p < 0.001, d = −0.25; sleep efficiency: t (837) = −13.14, p < 0.001, d = −0.45. Deviations in average dermal temperature returned to baseline around 10 days following the reported date. Deviations in average breathing rate, heart rate, and rMSSD extend beyond 28 days, perhaps driven by a subset of individuals with longer symptom duration.

Breakthrough cases (n = 101) are defined here as participants who reported a vaccination at least 2 weeks prior to reported infection. We discovered significantly weaker and shorter duration biometric responses in breakthrough cases compared to users who never reported vaccination (n = 737). All five biometric signals had significantly larger deviations on the night following reported infection in users who did not report vaccination compared to those who did (Fig. 1b; temperature t [836] = 5.89, p < 0.001, d = 0.63; heart rate t [836] = 2.66, p = 0.010, d = 0.28; breath rate t [836] = 4.13, p < 0.001, d = 0.44; rMSSD t [836] = −2.17, p = 0.035, d = −0.23; sleep efficiency t [836] = −1.71, p = 0.098, d = −0.18).

Infections reported between July and September 2021 (n = 380), when the majority of infections in the countries studied were the Delta variant, were compared with infections reported between January and March 2021 (n = 278), when the majority of cases reported were Alpha and other variants. For example, EU data for Finland estimates that 0% of cases were Delta in weeks starting in January to March and at least 97.9% in weeks starting in July to September [1]. In likely Delta infections, we observed larger increases in heart rate and larger decreases in rMSSD on the nights following reported infection compared to earlier variants. Deviations in temperature, breathing rate, and sleep efficiency were not significantly different between variants (Fig. 1c; temperature t [656] = −0.90, p = 0.367, d = −0.07; heart rate t [656] = −2.70, p = 0.010, d = −0.21; breath rate t [656] = −1.02, p = 0.329, d = −0.08; rMSSD t [656] = 2.72, p = 0.010, d = 0.21; sleep efficiency t [656] = 1.00, p = 0.329, d = 0.08).

We also observed physiological responses in response to reported vaccination. All biometric signals were significantly different from their baseline beginning on the night after reported vaccination (Fig. 2a; mean of first 3 days compared to baseline; temperature: t [26,140] = 70.32, p < 0.001, d = 0.43; heart rate: t [26,140] = 40.92, p < 0.001, d = 0.25; breath rate: t [26,140] = 89.05, p < 0.001, d = 0.55; rMSSD: t [26,140] = −22.02, p < 0.001, d = −0.14; sleep efficiency: t [26,140] = −26.77, p < 0.001, d = −0.17). All biometric averages remained divergent from baseline for 4 days following reported vaccination.

We compared responses to reported vaccination in two subsets of these data: a younger cohort of under 35s (n = 6,385) and an older cohort of over 50s (n = 6,564). All biometric z-scores were significantly further from baselines in the younger cohort, amounting to a 40–50% greater deviation (Fig. 2b; temperature t [12,947] = 7.37, p < 0.001, d = 0.13; heart rate t [12,947] = 6.45, p < 0.001, d = 0.11; breath rate t [12,947] = 12.66, p < 0.001, d = 0.22; rMSSD t [12,947] = −5.74, p < 0.001, d = −0.10; sleep efficiency t [12,947] = −5.18, p < 0.001, d = −0.09).

We also compared responses to the vaccine’s first dose (n = 398) versus second (n = 897) for those participants who provided that information. All biometric z-scores were significantly further from baselines for the second dose of the vaccine, with an average of around 80% greater deviation (Fig. 2c; temperature t [1,293] = −6.15, p < 0.001, d = −0.37; heart rate t [1,293] = −2.60, p = 0.012, d = −0.16; breath rate t [1,293] = −6.00, p < 0.001, d = −0.36; rMSSD t [1,293] = 2.35, p = 0.023, d = 0.14; sleep efficiency t [1,293] = 3.00, p = 0.004, d = 0.18).

These results are in line with prior findings that biometric measurements provided by consumer-grade wearable devices are predictive of reported confirmed COVID-19 infection [2, 3, 5‒7, 9, 10]. Both the magnitude and duration of biometric effects of COVID-19 infections, with or without prior vaccination, were significantly larger than responses to the COVID-19 vaccination. However, in users reporting breakthrough COVID-19 infections (i.e., after vaccination), the magnitude of the biometric response was significantly smaller compared to unvaccinated individuals. In individuals who reported vaccination, we observed significantly larger responses to the second dose of the vaccine compared to the first. In those who had COVID-19, with or without vaccination, we also observed larger cardiac responses to the Delta variant, but not fever or respiratory responses, compared to pre-Delta variants. In keeping with recent research, larger deviations were seen in younger individuals and following the second dose [11]. These results suggest it may be possible to approximately measure the immune response and consequent antibody generation following vaccination [11].

Thanks to their ability to perform continuous monitoring, consumer wearables measuring temperature, heart rate, breath rate, and HRV are potential platforms on which to build tools for prospective infection screening. Prospective monitoring has applications both in epidemiology/public health and in screening at the individual level. The potential impact on public health is heightened given the possibility of detecting infection during the window before an individual becomes symptomatic, during which they may still be able to transmit the disease [2, 7, 9]. Critically, continuous passive monitoring of biometrics may result in higher compliance than self- or physician-administered testing programs while also allowing the creation of individual baselines that increase sensitivity to illness-related deviations.

This is a prospective observational study, without a randomly assigned control group. Our outcome variables are self-reported, meaning the vaccination and infection sample may be biased, for example, toward more health-conscious individuals. Certain cultural and socioeconomic groups may be overrepresented among Oura users relative to the general population. The Oura Ring’s photoplethysmography sensors have been validated against reference devices [8] but its temperature sensors have not; however, we note that deviation from a baseline does not require accurate calibration of the sensors.

By accepting Oura’s terms of service, users consent to the use of their anonymized data for research purposes. The study protocol was reviewed and the need for approval was waived by the Internal Ethics Board, Oura Health Ltd. The study was deemed exempt as the analysis used only aggregate (group level) data stored without individual identifiers.

All authors declare competing financial interests in the form of ongoing salaried employment and employee equity at Oura Health Ltd., San Francisco. The authors declare no other competing non-financial interests.

The study was funded by Oura Health Ltd., San Francisco.

Gerald Norman Pho performed the analyses and created the figures. Gerald Norman Pho, Nina Thigpen, Shyamal Patel, and Hal Tily contributed to the design of the research and interpretation of the findings. Hal Tily wrote the manuscript with contributions from Gerald Norman Pho, Nina Thigpen, and Shyamal Patel.

All data were provided by Oura users and as such are governed by Oura’s terms of service agreement. Under this agreement, users consent for their anonymized data to be used in research conducted by Oura, but Oura commits to never share those data with any third party without explicit individual consent. This dataset is therefore not available for download. Further inquiries can be directed to the corresponding author.