Rising global temperatures increase the clinical and public-health need for effective monitoring of physiological responses to heat. Population-based indices provide environmental context but do not capture individual variability in heat tolerance and heat strain. The authors frame digital biomarkers—continuously measured, device-derived physiological and environmental features—as a promising approach to enable individualized, lab-independent, real-time monitoring during heat exposure. This study aimed to explore candidate digital biomarkers that could support person-centered heat health monitoring.
The investigation was a single-center, prospective, cross-sectional, cross-over, observational and explorative study. Twenty participants underwent exposures to a set of personalized heat stressors designed to represent three origins of heat exposure: increased ambient temperature, elevated relative humidity, and physical exertion. The cross-over element allowed assessment of differing stressor types within the same individuals to capture intraindividual variability in responses.
Physiological measurements were gathered with wearable sensors. Vital parameters reported in the study included heart rate and body temperature measures (core and close body temperature). Local sweat rate was quantified using technical absorbents. Environmental parameters relevant to heat exposure such as ambient temperature and relative humidity were incorporated into the dataset. The multimodal sensing strategy combined internal physiological signals with external environmental measures to form candidate digital biomarkers.
To explore whether multimodal digital biomarker candidates could distinguish the origin of heat exposure, the authors built a Random Forest classification model. The model used the collected physiological and environmental features to classify exposure conditions into the three stressor categories: ambient temperature change, relative humidity change, and physical exertion. The study reports the classifier’s overall performance and examines the relative importance of different predictive variables across stressor types.
The Random Forest model discriminated the three heat stressor types with an overall accuracy of 94.7%. Analysis of contributing variables showed distinct patterns by stressor category:
These differing variable distributions underscore the value of a multimodal approach: combining physiological signals with environmental context enhanced the model’s ability to attribute heat strain to its source.
Findings suggest that integrated sensing of physiological and environmental parameters can support comprehensive, automated, and lab-independent strategies for monitoring heat exposure and heat strain at the individual level. By differentiating whether heat burden arises from ambient temperature, humidity, or exertion, a multimodal digital biomarker framework could inform tailored mitigation strategies and more precise risk assessment during heat events.
The authors describe this work as explorative and note the need for further testing. Specifically, they recommend evaluation in more complex and real-life settings beyond the controlled exposures used here to assess generalizability and robustness of candidate digital biomarkers. The paper does not report detailed participant demographics, specific sensor models, or the full feature set and model parameters in the text provided; those details were not reported in the source excerpt and would need to be consulted in the full manuscript or data upon request.
The study received ethical approval from the Ethical Committee of Northwestern Switzerland (EKNZ ID 2022-01325). The authors declared no competing interests. Funding for the project was provided by Innosuisse — Swiss Innovation Agency (52023.1 IP-LS). All data produced in the study are available upon reasonable request to the authors.
This exploratory study with 20 participants indicates that a Random Forest model using combined physiological and environmental measures can distinguish ambient temperature, relative humidity, and exertion stressors with high reported accuracy (94.7%). The differing predictor patterns by stressor type highlight the potential of multimodal digital biomarkers to enable person-centered heat health monitoring, while acknowledging the need for validation in real-world and more complex conditions.