Wearable Sensors

Burks, J. H., Bruce, L. K., Kasl, P., Soltani, S., Viswanath, V., Hartogensis, W., Dilchert, S., Hecht, F. M., Dasgupta, S., Altintas, I., Gupta, A., Mason, A. E., & Smarr, B. L.
npj Women’s Health, 2(1)
(2024)

In tasks involving human health condition data, feature selection is heavily affected by data types, the complexity of the condition manifestation, and the variability in physiological presentation. One type of variability often overlooked or oversimplified is the effect of biological sex. As females have been chronically underrepresented in clinical research, we know less about how conditions manifest in females. Innovations in wearable technology have enabled individuals to generate high temporal resolution data for extended periods of time. With millions of days of data now available, additional feature selection pipelines should be developed to systematically identify sex-dependent variability in data, along with the effects of how many per-person data are included in analysis. Here we present a set of statistical approaches as a technique for identifying sex-dependent physiological and behavioral manifestations of complex diseases starting from longitudinal data, which are evaluated on diabetes, hypertension, and their comorbidity.

Kasl, P., Soltani, S., Bruce, L. K., Viswanath, V. K., Hartogensis, W., Gupta, A., Altintas, I., Dilchert, S., Hecht, F. M., Mason, A., & Smarr, B. L.
Proceedings of Machine Learning Research, 248
(2024)

Large-scale wearable datasets are increasingly being used for biomedical research and to develop machine learning (ML) models for longitudinal health monitoring applications. However, it is largely unknown whether biases in these datasets lead to findings that do not generalize. Here, we present the first comparison of the data underlying multiple longitudinal, wearable-device-based datasets. We examine participant-level resting heart rate (HR) from four studies, each with thousands of wearable device users. We demonstrate that multiple regression, a community standard statistical approach, leads to conflicting conclusions about important demographic variables (age vs resting HR) and significant intra- and interdataset differences in HR. We then directly test the cross-dataset generalizability of a commonly used ML model trained for three existing day-level monitoring tasks: prediction of testing positive for a respiratory virus, flu symptoms, and fever symptoms. Regardless of task, most models showed relative performance loss on external datasets; most of this performance change can be attributed to concept shift between datasets. These findings suggest that research using large-scale, pre-existing wearable datasets might face bias and generalizability challenges similar to research in more established biomedical and ML disciplines. We hope that the findings from this study will encourage discussion in the wearable-ML community around standards that anticipate and account for challenges in dataset bias and model generalizability.

Viswanath, V. K., Hartogenesis, W., Dilchert, S., Pandya, L., Hecht, F. M., Mason, A. E., Wang, E. J., & Smarr, B. L.
npj Digital Medicine, 7(150)
(2024)

Sleep monitoring has become widespread with the rise of affordable wearable devices. However, converting sleep data into actionable change remains challenging as diverse factors can cause combinations of sleep parameters to differ both between people and within people over time. Researchers have attempted to combine sleep parameters to improve detecting similarities between nights of sleep. The cluster of similar combinations of sleep parameters from a night of sleep defines that night’s sleep phenotype. To date, quantitative models of sleep phenotype made from data collected from large populations have used cross-sectional data, which preclude longitudinal analyses that could better quantify differences within individuals over time. In analyses reported here, we used five million nights of wearable sleep data to test (a) whether an individual’s sleep phenotype changes over time and (b) whether these changes elucidate new information about acute periods of illness (e.g., flu, fever, COVID-19). We found evidence for 13 sleep phenotypes associated with sleep quality and that individuals transition between these phenotypes over time. Patterns of transitions significantly differ (i) between individuals (with vs. without a chronic health condition; chi-square test; p-value < 1e−100) and (ii) within individuals over time (before vs. during an acute condition; Chi-Square test; p-value < 1e−100). Finally, we found that the patterns of transitions carried more information about chronic and acute health conditions than did phenotype membership alone (longitudinal analyses yielded 2–10× as much information as cross-sectional analyses). These results support the use of temporal dynamics in the future development of longitudinal sleep analyses.

Jacob M. Ryan, Shreenithi Navaneethan, Natalie Damaso,
Stephan Dilchert, Wendy Hartogensis, Joseph L. Natale,
Frederick M. Hecht, Ashley E. Mason, & Benjamin L. Smarr
Frontiers in Network Physiology
2024

Algorithms for the detection of COVID-19 illness from wearable sensor devices tend to implicitly treat the disease as causing a stereotyped (and therefore recognizable) deviation from healthy physiology. In contrast, a substantial diversity of bodily responses to SARS-CoV-2 infection have been reported in the clinical milieu. This raises the question of how to characterize the diversity of illness manifestations, and whether such characterization could reveal meaningful relationships across different illness manifestations. Here, we present a framework motivated by information theory to generate quantified maps of illness presentation, which we term “manifestations,” as resolved by continuous physiological data from a wearable device (Oura Ring). We test this framework on five physiological data streams (heart rate, heart rate variability, respiratory rate, metabolic activity, and sleep temperature) assessed at the time of reported illness onset in a previously reported COVID-19-positive cohort (N = 73). We find that the number of distinct manifestations are few in this cohort, compared to the space of all possible manifestations. In addition, manifestation frequency correlates with the rough number of symptoms reported by a given individual, over a several-day period prior to their imputed onset of illness. These findings suggest that information-theoretic approaches can be used to sort COVID-19 illness manifestations into types with real-world value. This proof of concept supports the use of information-theoretic approaches to map illness manifestations from continuous physiological data. Such approaches could likely inform algorithm design and real-time treatment decisions if developed on large, diverse samples.

Kasl, P., Bruce, L. K., Hartogensis, W., Dasgupta, S., Pandya, L. S., Dilchert, S., Hecht, F. M., Gupta, A., Altintas, I., Mason, A. E., & Smarr, B. L.
Sensors, 24(6)
(2024)

Commercially available wearable devices (wearables) show promise for continuous physiological monitoring. Previous works have demonstrated that wearables can be used to detect the onset of acute infectious diseases, particularly those characterized by fever. We aimed to evaluate whether these devices could be used for the more general task of syndromic surveillance. We obtained wearable device data (Oura Ring) from 63,153 participants. We constructed a dataset using participants’ wearable device data and participants’ responses to daily online questionnaires. We included days from the participants if they (1) completed the questionnaire, (2) reported not experiencing fever and reported a self-collected body temperature below 38 ◦C (negative class), or reported experiencing fever and reported a self-collected body temperature at or above 38 ◦C (positive class), and (3) wore the wearable device the nights before and after that day. We used wearable device data (i.e., skin temperature, heart rate, and sleep) from the nights before and after participants’ fever day to train a tree-based classifier to detect self-reported fevers. We evaluated the performance of our model using a five-fold cross-validation scheme. Sixteen thousand, seven hundred, and ninety-four participants provided at least one valid ground truth day; there were a total of 724 fever days (positive class examples) from 463 participants and 342,430 non-fever days (negative class examples) from 16,687 participants. Our model exhibited an area under the receiver operating characteristic curve (AUROC) of 0.85 and an average precision (AP) of 0.25. At a sensitivity of 0.50, our calibrated model had a false positive rate of 0.8%. Our results suggest that it might be possible to leverage data from these devices at a public health level for live fever surveillance. Implementing these models could increase our ability to detect disease prevalence and spread in real-time during infectious disease outbreaks.

Mason, A. E., Kasl, P., Soltani, S., Green, A., Hartogensis, W., Dilchert, S., Chowdhary, A., Pandya, L. S., Siwik, C. J., Foster, S. L., Nyer, M., Lowry, C. A., Raison, C. L., Hecht, F. M., & Smarr, B. L.
Scientific Reports, 14(1)
(2024)

Correlations between altered body temperature and depression have been reported in small samples; greater confidence in these associations would provide a rationale for further examining potential mechanisms of depression related to body temperature regulation. We sought to test the hypotheses that greater depression symptom severity is associated with (1) higher body temperature, (2) smaller differences between body temperature when awake versus asleep, and (3) lower diurnal body temperature amplitude. Data collected included both self-reported body temperature (using standard thermometers), wearable sensor-assessed distal body temperature (using an off-the-shelf wearable sensor that collected minute-level physiological data), and self-reported depressive symptoms from > 20,000 participants over the course of ~ 7 months as part of the TemPredict Study. Higher self-reported and wearable sensor-assessed body temperatures when awake were associated with greater depression symptom severity. Lower diurnal body temperature amplitude, computed using wearable sensor-assessed distal body temperature data, tended to be associated with greater depression symptom severity, though this association did not achieve statistical significance. These findings, drawn from a large sample, replicate and expand upon prior data pointing to body temperature alterations as potentially relevant factors in depression etiology and may hold implications for development of novel approaches to the treatment of major depressive disorder.

Bruce, L. K., Kasl, P., Soltani, S., Viswanath, V. K., Hartogensis, W., Dilchert, S., Hecht, F. M., Chowdhary, A., Anglo, C., Pandya, L., Dasgupta, S., Altintas, I., Gupta, A., Mason, A. E., & Smarr, B.
Biology of Sex Differences, 14(76)
(2023)

Females have been historically excluded from biomedical research due in part to the documented presumption that results with male subjects will generalize effectively to females. This has been justified in part by the assumption that ovarian rhythms will increase the overall variance of pooled random samples. But not all variance in samples is random. Human biometrics are continuously hanging in response to stimuli and biological rhythms; single measurements taken sporadically do not easily support exploration of variance across time scales. Recently we reported that in mice, core body temperature measured longitudinally shows higher variance in males than cycling females, both within and across individuals at multiple time scales. Here, we explore longitudinal human distal body temperature, measured by a wearable sensor device (Oura Ring), for 6 months in females and males ranging in age from 20 to 79 years. In this study, we did not limit the comparisons to female versus male, but instead we developed a method for categorizing individuals as cyclic or acyclic depending on the presence of a roughly monthly pattern to their nightly temperature. We then compared structure and variance across time scales using multiple standard instruments. Sex differences exist as expected, but across multiple statistical comparisons and timescales, there was no one group that consistently exceeded the others in variance. When variability was assessed across time, females, whether or not their temperature contained monthly cycles, id not significantly differ from males both on daily and monthly time scales. These findings contradict the viewpoint that human females are too variable across menstrual cycles to include in biomedical research. Longitudinal temperature of females does not accumulate greater measurement error over time than do males and the majority of unexplained variance is within sex category, not between them.

Shiba, S. K., Temple, C. A., Krasnoff, J., Dilchert, S., Smarr, B. L., Robishaw, J., & Mason, A. E.
Cureus, 15(9)
(2023)

Identifying early signs of a SARS-CoV-2 infection in healthcare workers could be a critical tool in reducing disease transmission. To provide this information, both daily symptom surveys and wearable device monitoring could have utility, assuming there is a sufficiently high evel of participant adherence. The aim of this study is to evaluate adherence to a daily symptom survey and a wearable device (Oura Ring) among healthcare professionals (attending physicians and other clinical staff) and trainees (residents and medical students) in a hospital etting during the early stages of the COVID-19 pandemic. Wearable device adherence was significantly higher than the daily symptom survey adherence for most participants. Overall, participants were highly adherent to the wearable device, wearing the device an average of 87.8 1.6% of study nights compared to survey submission, showing an average of 63.8 ± 27.4% of study days. In subgroup analysis, we found that healthcare professionals (HCPs) and medical students had the highest adherence to wearing the wearable device, while medical residents had ower adherence in both wearable adherence and daily symptom survey adherence.These results indicated high participant adherence to wearable devices to monitor for impending infection in the course of a research study conducted as part of clinical practice. Subgroup analysis ndicated HCPs and medical students maintained high adherence, but residents’ adherence was lower, which is likely multifactorial, with differences in work demands and stress contributing to the findings. These results can guide the development of adherence strategies for a wearable device to increase the quality of data collection and assist in disease detection in this and future pandemics.

Klein, A., Puldon, K., Dilchert, S., Hartogensis, W., Chowdhary, A., Anglo, C., Pandya, L. S., Hecht, F. M., Mason, A. E., & Smarr, B. L.
Frontiers in Big Data, 5.
(2022)

Daily symptom reporting collected via web-based symptom survey tools holds the potential to improve disease monitoring. Such screening tools might be able to not only discriminate between states of acute illness and non-illness, but also make use of additional demographic information so as to identify how illnesses may differ across groups, such as biological sex. These capabilities may play an important role in the context of future disease outbreaks. We used daily symptom profiles to plot symptom progressions for COVID-19, influenza (flu), and the common cold. We then built a Bayesian network to discriminate between these three illnesses based on daily symptom reports. We identified key symptoms that contributed to a COVID-19 prediction in both males and females using a logistic regression model. Although the Bayesian model performed only moderately well in identifying a COVID-19 diagnosis (71.6% true positive rate), the model showed promise in being able to differentiate between COVID-19, flu, and the common cold, as well as periods of acute illness vs. non-illness. Additionally, COVID-19 symptoms differed between the biological sexes; specifically, fever was amore important symptom in identifying subsequent COVID-19 infection among males than among females.

Mason, A. E., Hecht, F. M., Davis, S. K., Natale, J. L., Hartogensis, W., Damaso, N., Claypool, K. T., Dilchert, S., […] & Smarr, B. L.
Scientific Reports, 12(1), 3463.
(2022)

Early detection of diseases such as COVID-19 could be a critical tool in reducing disease transmission by helping individuals recognize when they should self-isolate, seek testing, and obtain early medical intervention. Consumer wearable devices that continuously measure physiological metrics hold promise as tools for early illness detection. We gathered daily questionnaire data and physiological data using a consumer wearable (Oura Ring) from 63,153 participants, of whom 704 self-reported possible COVID-19 disease. We selected 73 of these 704 participants with reliable confirmation of COVID-19 by PCR testing and high-quality physiological data for algorithm training to identify onset of COVID-19 using machine learning classification. The algorithm identified COVID-19 an average of 2.75 days before participants sought diagnostic testing with a sensitivity of 82% and specificity of 63%. The receiving operating characteristic (ROC) area under the curve (AUC) was 0.819 (95% CI [0.809, 0.830]). Including continuo s temperature yielded an AUC 4.9% higher than without this feature. For further validation, we obtained SARS CoV-2 antibody in a subset of participants and identified 10 additional participants who self-reported COVID-19 disease with antibody confirmation. The algorithm had an overall ROC AUC of 0.819 (95% CI [0.809, 0.830]), with a sensitivity of 90% and specificity of 80% in these additional participants. Finally, we observed substantial variation in accuracy based on age and biological sex. Findings highlight the importance of including temperature assessment, using continuous physiological features for alignment, and including diverse populations in algorithm development to optimize accuracy in COVID-19 detection from wearables.

Mason, A. E., Kasl, P., Hartogensis, W., Natale, J. L., Dilchert, S., Dasgupta, S., Purawat, S., Chowdhary, A., Anglo, C., Veasna, D., Pandya, L. S., Fox, L. M., Puldon, K. Y., Prather, J. G., Gupta, A., Altintas, I., Smarr, B. L., & Hecht, F. M.
Vaccines, 10(2), 264.
(2022)

There is significant variability in neutralizing antibody responses (which correlate with immune protection) after COVID-19 vaccination, but only limited information is available about predictors of these responses. We investigated whether device-generated summaries of physiological metrics collected by a wearable device correlated with post-vaccination levels of antibodies to the SARS-CoV-2 receptor-binding domain (RBD), the target of neutralizing antibodies generated by existing COVID-19 vaccines. One thousand, one hundred and seventy-nine participants wore an off-the-shelf wearable device (Oura Ring), reported dates of COVID-19 vaccinations, and completed testing for antibodies to the SARS-CoV-2 RBD during the U.S. COVID-19 vaccination rollout. We found that on the night immediately following the second mRNA injection (Moderna-NIAID and Pfizer-BioNTech) increases in dermal temperature deviation and resting heart rate, and decreases in heart rate variability (a measure of sympathetic nervous system activati n) and deep sleep were each statistically significantly correlated with greater RBD antibody responses. These associations were stronger in models using metrics adjusted for the pre-vaccination baseline period. Greater temperature deviation emerged as the strongest independent predictor of greater RBD antibody responses in multivariable models. In contrast to data on certain other vaccines, we did not find clear associations between increased sleep surrounding vaccination and antibody responses.

Smarr, B. L., Aschbacher, K., Fisher, S. M., Chowdhary, A., Dilchert, S., Puldon, K., Rao, A., Hecht, F. M., & Mason, A. E.
Scientific Reports, 10(1), 21640.
(2020)

Elevated core temperature constitutes an important biomarker for COVID-19 infection; however, no standards currently exist to monitor fever using wearable peripheral temperature sensors. Evidence that sensors could be used to develop fever monitoring capabilities would enable largescale health-monitoring research and provide high-temporal resolution data on fever responses across heterogeneous populations. We launched the TemPredict study in March of 2020 to capture continuous physiological data, including peripheral temperature, from a commercially available wearable device during the novel coronavirus pandemic. We coupled these data with symptom reports and COVID-19 diagnosis data. Here we report findings from the first 50 subjects who reported COVID- 19 infections. These cases provide the first evidence that illness-associated elevations in peripheral temperature are observable using wearable devices and correlate with self-reported fever. Our analyses support the hypothesis that wearable sensors can detec illnesses in the absence of symptom recognition. Finally, these data support the hypothesis that prediction of illness onset is possible using continuously generated physiological data collected by wearable sensors. Our findings should encourage further research into the role of wearable sensors in public health efforts aimed at illness detection, and underscore the importance of integrating temperature sensors into commercially available wearables.

Wiernik, B. M., Ones, D. S., Marlin, B. M., Giordano, C., Dilchert, S., Mercado, B. K., Stanek, K. C., Birkland, A., Wang, Y., Ellis, B., Yazar, Y., Kostal, J. W., Kumar, S., Hnat, T., Ertin, E., Sano, A., Ganesan, D. K., Choudhoury, T., & al’Absi, M.
European Journal of Psychological Assessment, 36(6), 935–947.
(2020)

Research interest in personality dynamics over time is rapidly growing. Passive personality assessment via mobile sensors offers an intriguing new approach for measuring a wide variety of personality dynamics. In this paper, we address the possibility of integrating sensorbased assessments to enhance personality dynamics research. We consider a variety of research designs that can incorporate sensor-based measures and address pitfalls and limitations in terms of psychometrics and practical implementation. We also consider analytic challenges related to data quality and model evaluation that researchers must address when applying machine learning methods to translate sensor data into composite personality assessments.