Continuous physiological monitoring from consumer wearables offers a potential route to both individual illness detection and population health surveillance. Prior research has often focused on COVID-19 and commonly pooled all non-illness days into a single healthy baseline. This study tested whether individualized wearable signatures of illness can be identified and whether those signatures reproduce at the population scale in a university setting.
The analysis used daily data from Oura Ring devices collected in the LEMURS cohort. The full enrollment included 584 first-year college students across two semesters (October 2022 to May 2023). The primary, paired within-participant analysis included 260 students and covered 3,218 person-weeks. Self-reported weekly illness status and self-tagged day-level illness events were used as reference labels.
Each student’s daily wearable-derived signals were matched to their own weekly self-report of illness, enabling paired comparisons of sick versus non-sick weeks within individuals. Statistical significance was assessed with Benjamini-Hochberg false discovery rate control (q<0.05). Reported effect sizes for paired comparisons were expressed as Cohen’s d.
Five wearable signals reliably differed between a student’s sick and non-sick weeks at FDR q<0.05:
These findings indicate a consistent within-person illness phenotype captured by multiple physiological and behavioral signals from a consumer ring.
The individual-level signature was tested at population scale by tracking the weekly fraction of students with elevated temperature and comparing that fraction to survey-reported illness rates. The correlation between elevated-temperature prevalence and reported illness was Pearson r = 0.66 (95% CI [0.20, 0.92], N = 11 weeks), showing that aggregate temperature monitoring from the cohort reflected concurrent self-reported illness trends.
A complementary day-level analysis used self-tagged illness events (n = 17 events, 27 days). This analysis recovered four of the five primary signals with substantially larger effect sizes (up to Hedges’ g = 3.8), indicating stronger acute deviations around self-tagged illness days. The day-level physiological pattern was evaluated against non-illness states: it was distinct from signatures associated with alcohol/hangover (d = +0.69), luteal-phase physiology (d = +1.45), and self-reported stress (Fisher-z r = +0.01), supporting discriminant validity of the illness signature.
The authors tested an eight-signal composite illness classifier and compared it to temperature alone. In a leave-one-participant-out evaluation the composite achieved an AUC of 0.74 versus 0.71 for temperature alone. The in-sample difference was not statistically significant (p = 0.54), suggesting that simple aggregate temperature monitoring performed comparably to a multivariate composite in this dataset.
These results suggest that a personalized wearable illness signature is robust at the individual level and reproducible when aggregated across a campus cohort. Because temperature alone closely tracked population illness and the composite did not significantly outperform temperature, the authors highlight that simple aggregate temperature monitoring from consumer rings may be sufficient for campus-scale surveillance applications.
The study received IRB approval from the University of Vermont (protocol #2126). The authors state that data processing scripts are available on request or will be released in tandem with a future peer-reviewed journal article. The preprint notes that competing interests were declared as none. The manuscript provides specific sample sizes and reported effect sizes; additional methodological details or full code release are pending the later journal publication.
In a university cohort of first-year students monitored with consumer rings, elevated skin temperature, higher resting heart rate, increased respiratory variation, reduced steps, and lower nightly HRV distinguished self-reported sick weeks from non-sick weeks. The temperature signal reproduced at population scale and a simple temperature-based approach may be operationally useful for campus health surveillance. The study supports the feasibility of wearable-derived illness monitoring while noting that broader release of processing scripts and peer-reviewed publication will provide further methodological transparency.