CoChat
Sign Up
Cited by 0Diabetes Management and ResearchArtificial Intelligence in HealthcareNon-Invasive Vital Sign Monitoring

Sleep and temperature data from wearable devices support noninvasive detection of diabetes mellitus in a large-scale, retrospective analysis

Open access
Viswanath, Varun K., Navaneethan, Shreenithi, Burks, Jamison H., Soltani, Severine, Kasl, Patrick, Hartogensis, Wendy, Dilchert, Stephan, Hecht, Frederick M. +3 more·2026·Communications Medicine
DOI 10.1038/s43856-026-01501-0PMC PMC13079903
Open this paper in CoChat — free accountOpen PDF

Abstract

Background: Diabetes Mellitus is a common, chronic metabolic disorder affecting the cardiovascular system, autonomic nervous system, and sleep quality. Diabetes affects diverse physiological data including heart rate variability, distal body temperature, and sleep duration. We hypothesized that biologically informed features from wearable device data, combined with appropriate application of longitudinal data, can capture physiological covariates of diabetes and support the noninvasive detection of diabetes. Methods: We obtained 4 months and 7 days of wearables data (Oura Ring) from 389 individuals self-reporting diabetes and 10,820 people self-reporting no diabetes diagnosis from the TemPredict database. We selected 36 features of sleep, circadian disruption, and distal body temperature from literature and evaluated whether time windows of these features could be classified to be from individuals self-reporting diabetes (N = 236) or self-reporting no diabetes diagnosis (N = 282). Results: Here we show longer time windows of input perform better, with the best algorithm (21-nights) achieving 0.88 Area under ROC (AUROC) and 0.80 Area under Precision Recall (AUPRC) (0.30 improvement over random). Feature analyses reveal the importance of further derived distal body temperature features (increase AUROC by 0.0724), especially to differentiate other chronic conditions from diabetes. The model achieves 0.80 AUROC and 0.28 improvement over random in AUPRC in an imbalanced cohort drawn from 6,658 individuals, emulating a general population. Conclusions: These results indicate the value of biologically informed features and longitudinal data for identifying people with diabetes and further, suggest that these methods could make such separations possible for other chronic conditions that affect sleep and inflammation.

AI snapshot

Written from the paper’s full text.

TL;DR

In a large retrospective TemPredict/Oura Ring analysis, biologically informed sleep, circadian, and distal body temperature features classified multi-night windows as from people self-reporting diabetes versus no diabetes, with the best 21-night XGBoost model achieving AUROC 0.88 and AUPRC 0.80 in a balanced cohort. Performance remained above chance in an imbalanced general-population-like cohort (AUROC 0.80, AUPRC 0.46).

Key findings

  • The best model, a 21-night XGBClassifier, achieved AUROC = 0.88 and AUPRC = 0.80 (0.30 improvement over random) in the balanced cohort; the 14-night XGBClassifier was similar (AUROC = 0.88, AUPRC = 0.79), while 1-night models were worse (AUROC = 0.73–0.74, AUPRC = 0.67).
  • At the optimal threshold, the 21-night XGBClassifier had TPR = 81.6% and FPR = 19.1% (1,746/2,140 diabetes windows; 394/3,823 no-diabetes windows). In the imbalanced cohort (87 self-reported diabetes, 6,571 no diabetes; 17.9% class balance), it achieved AUROC = 0.80, AUPRC = 0.46 (0.28 over random baseline AUPRC = 0.179), sensitivity = 80.6%, and specificity = 63.1%.
  • HR/HRV features and secondary distal body temperature features were most impactful: HR/HRV delta AUROC = 0.0729 and distal body temperature delta AUROC = 0.0724. Adding temperature features improved separation from other chronic conditions; without feature sets 4 and 5, people self-reporting no diabetes with other chronic conditions had 0.082 higher predicted probabilities.
  • Population-level common features differed significantly (p < 0.0001): HR mean 62.713 vs 67.819 BPM (Cohen’s d = 0.607), RMSSD 42.944 vs 30.141 ms (d = −0.555), total sleep time 7.946 vs 7.649 hrs (d = −0.192), and distal body temperature circadian amplitude 7.261 vs 6.008 °C (d = −0.455).

Methods

  • Retrospective TemPredict study using Oura Ring Gen2 data from January–December 2020, with the analyzed window April 22–August 29, 2020 (4 months and 7 days). After filtering, the dataset contained 24,486 nights from 315 people self-reporting diabetes and 764,315 nights from 9,063 people self-reporting no diabetes diagnosis, with 36 features; the machine-learning balanced cohort included 236 self-reported diabetes and 282 no-diabetes individuals (31,725 windows).
  • Extracted 36 features across HR/HRV, primary sleep, secondary sleep, primary temperature, and secondary temperature. Population differences were tested with two-sided Mann–Whitney U tests. LogisticRegression and XGBClassifier were trained on rolling 1-, 3-, 7-, 14-, and 21-night windows with an 80/20 individual-level split, 3-fold cross-validation, and evaluation by AUROC/AUPRC, feature importance/ablation, other chronic conditions, and an imbalanced cohort (87 diabetes, 6,571 no diabetes; 100 repeated samplings at 17.9% class balance).

Limitations

  • The paper states: the approach is limited to the modalities measured by the wearable used; sensor accuracy can vary with skin tone and ethnicity, and the population was majority white (83.55%); ground-truth labels are self-reported, likely imprecise, and do not distinguish diabetes types; deep learning was not explored; physiological effects vary within and across individuals, and false-negative windows resembled true negatives, so repeated re-testing would likely be needed; climate differences were not accounted for; data are private commercial data and not reproducible without qualified request; and other behavioral/health factors can produce similar physiological effects.
  • The paper also states: it lacked CGM/glucose data, could not assess the repeated re-testing assertion, and the model was not tuned for optimal general-population performance.
  • Evident from the text: the study is retrospective with self-reported diagnosis rather than clinical confirmation; balanced-cohort AUPRC (0.80) dropped to 0.46 in the imbalanced cohort; there was no external validation or prospective clinical study; and daytime HR missingness prevented exploration of circadian HR patterns.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s43856-026-01501-0

Cites

55 of 55 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

65 of 109 cited sources checked · 22 supported · 1 flagged · accuracy not yet measured

  • Premisereasoning

    Neither prior literature was able to test their efficacy in large, real-world cohorts representing a practical assessment.

    “Though interesting, neither literature was able to test their efficacy in large, real-world cohorts representing a practical assessment.”
    Quote in the paperSubstantial claim4 of 7 sources checked
    • cites [29] Bogue-Jimenez, B., Huang, X., Powell, D. & Doblas, A. Selection of noninvasive features in wrist-based wearable sensors to predict blood glucose concentrations using machine learning algorithms. Sensors 22, 3534 (2022). source
    • cites [30] Schechter, A. et al. A prototype of a new noninvasive device to detect nocturnal hypoglycemia in adolescents with type 1 diabetes—a pilot study. Diab. Technol. Ther. 14, 683–689 (2012). source
    • cites [31] Niramitmahapanya, S. & Yotha, D. Physiologic change in hypoglycemia in diabetes patients in Rajavithi Hospital using non-invasive hypoglycemic warning wearable devices: a pilot study. Int. J. Innov. Res. Med. Sci. 6, 414–417 (2021). source
    • cites [32] Piersanti, A. et al. Variability of the skin temperature from wrist-worn device for definition of novel digital biomarkers of glycemia. Sensors 25, 4038 (2025). source

      The Discussion (section 4.3, page unavailable) says this study evaluated hypoglycemia metrics in only six of fifteen subjects and calls for larger populations and further evaluation for real applications. That supports the lack of a large-cohort practical assessment for this study, but the passage does not establish the same limitation for both pieces of prior literature described in the claim. (LLM, conf=0.78)

      Experimental — accuracy not yet measured
    • cites [35] Amiya, E. et al. Diurnal body temperature rise is reduced in diabetes with autonomic neuropathy. Clin. Auton. Res. 24, 95–97 (2014). source
    • cites [38] Brooks, T. G. et al. Diurnal rhythms of wrist temperature are associated with future disease risk in the UK Biobank. Nat. Commun. 14, 5172 (2023). source

      The Methods (Actigraphy in the UK Biobank; Proportional hazards models) describe an analysis of 91,462 UK Biobank participants, directly contrary to the claim that this work did not assess its approach in a large, real-world cohort. The paper tests associations between wrist-temperature rhythms and future disease, rather than the efficacy of an intervention. (LLM, conf=0.84)

      Checked by a reasoning model — may be wrong.

      Experimental — accuracy not yet measured
    • cites [39] Mason, A. E. et al. Elevated body temperature is associated with depressive symptoms: results from the TemPredict Study. Sci. Rep. 14, 1884 (2024). source

      The Introduction says prior studies of body temperature and depressive symptoms had fewer than 300 participants and were conducted in controlled settings; the Discussion says this study examined the association in a larger sample outside the laboratory. That supports the claim about a gap in large, real-world assessments, but the paper does not say those earlier studies tested efficacy, and it describes its own work as observational rather than an efficacy test. No page numbers are available in our copy. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
  • Premisecitation

    Diabetes mellitus can worsen the quality of sleep.

    “A bidirectional relationship exists between DM risk and sleep quality: chronic sleep disturbances can elevate the risk of developing insulin resistance, while DM can worsen the quality of sleep10.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [10] Chattu, V. et al. The interlinked rising epidemic of insufficient sleep and diabetes mellitus. Healthcare 7, 37 (2019). source
  • Premisecitation

    Chronic sleep disturbances can elevate the risk of developing insulin resistance.

    “A bidirectional relationship exists between DM risk and sleep quality: chronic sleep disturbances can elevate the risk of developing insulin resistance, while DM can worsen the quality of sleep10.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [10] Chattu, V. et al. The interlinked rising epidemic of insufficient sleep and diabetes mellitus. Healthcare 7, 37 (2019). source
  • Premisecitation

    Within a night, the complexity of the distal body temperature time series is related to the autonomic nervous system's health during sleep.

    “Within a night, the complexity of the distal body temperature time series has been related to the autonomic nervous system’s (ANS) health during sleep6,34.”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [6] Kenny, G. P., Sigal, R. J. & McGinn, R. Body temperature regulation in diabetes. Temperature 3, 119–145 (2016). source

      The Basic temperature regulation > Heat stress and Type 1 diabetes > Heat stress passages discuss skin blood flow, sweating, and neuropathy, but not the complexity of distal temperature time series during sleep. Searches for sleep, nocturnal temperature, time-series complexity, and entropy found no passage addressing the claimed relationship; page numbers were unavailable in our copy. (LLM, conf=0.76)

      Experimental — accuracy not yet measured
    • cites [34] Harding, E. C., Franks, N. P. & Wisden, W. The temperature dependence of sleep. Front. Neurosci. 13, 336 (2019). source

      The “Warm Bath Effect” section links distal-skin temperature and vasodilation to sleep initiation, but does not discuss the complexity of a within-night temperature time series or its relationship to ANS health. Searches for those specific concepts found no passage addressing the claimed relationship. (LLM, conf=0.74)

      Experimental — accuracy not yet measured
  • Premisecitation

    Features of sleep-wake time series can directly capture insomnia-like patterns in sleep related to diabetes mellitus.

    “Similarly, features of sleep-wake time series can directly capture insomnia-like patterns in sleep related to DM6,34,43,44.”
    Quote in the paperSubstantial claim4 of 4 sources checked
    • cites [6] Kenny, G. P., Sigal, R. J. & McGinn, R. Body temperature regulation in diabetes. Temperature 3, 119–145 (2016). source

      The Type 1 diabetes > Heat stress passages discuss sweating, skin blood flow, and body heat storage, not sleep. Searches for “sleep,” “insomnia,” and “actigraphy sleep-wake time series” found no passage in our copy addressing whether time-series features capture diabetes-related insomnia-like patterns. (LLM, conf=0.75)

      Experimental — accuracy not yet measured
    • cites [34] Harding, E. C., Franks, N. P. & Wisden, W. The temperature dependence of sleep. Front. Neurosci. 13, 336 (2019). source

      The Introduction links insomnia to disruption of the body-cooling cycle, but does not discuss diabetes mellitus or features extracted from sleep-wake time series. Searches for diabetes, insomnia, and time series found no passage addressing whether such features capture diabetes-related insomnia-like patterns. (LLM, conf=0.75)

      Experimental — accuracy not yet measured
    • cites [43] Katori, M. et al. ‘The 103,200-arm acceleration dataset in the UK Biobank revealed a landscape of human sleep phenotypes’. Proc. Natl. Acad. Sci. USA 119, https://doi.org/10.1073/pnas.2116729119 (2022). source

      The Discussion (page not provided) says the sleep/wake classification algorithm identified insomnia-like phenotypes characterized by midawake. The cited paper does not establish that those patterns are related to diabetes mellitus; the diabetes-specific part of the claim is unsupported. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites [44] Viswanath, V. K. et al. Five million nights: temporal dynamics in human sleep phenotypes. npj Digit. Med. 7, 150 (2024). source
  • May combine two claims. 33
  • Classified as probably not a substantial claim. 54
  • The quote only partly states this claim. 2

  • May combine two claims.

    Premisecitation

    People with diabetes mellitus have lower skin blood flow and dysregulated thermoregulatory responses during heat exposure.

    “Furthermore, people with DM have been reported to have lower skin blood flow and dysregulated thermoregulatory responses during heat exposure, which is associated with dysregulated thermal, cardiovascular, and glycemic control6–8.”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites [6] Kenny, G. P., Sigal, R. J. & McGinn, R. Body temperature regulation in diabetes. Temperature 3, 119–145 (2016). source
    • cites [7] Semenza, J. Excess hospital admissions during the July 1995 heat wave in Chicago. Am. J. Prev. Med. 16, 269–277 (1999). source
    • cites [8] Schwartz, J. Who is sensitive to extremes of temperature? Epidemiology 16, 67–72 (2005). source

  • May combine two claims.

    Premisecitation

    Diabetes mellitus is related to chronic inflammation, which can lead to increases in body temperature over time.

    “DM is also intimately related to chronic inflammation9, which can lead to noticeable increases in body temperature, especially as a pattern of change across time.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [9] Tsalamandris, S. et al. The role of inflammation in diabetes: current concepts and future perspectives. Eur. Cardiol. Rev. 14, 50–59 (2019). source

      The Historical Perspectives section links inflammation with diabetes, and the Adipose Tissue and Obesity section describes chronic low-grade inflammation in type 2 diabetes. The paper does not report that this inflammation causes noticeable increases in body temperature or a rising temperature pattern over time; searches for body temperature and related terms found no such passage. Page numbers were not available in our copy. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    Both diabetes mellitus and prediabetes are independently associated with lower heart rate variability (HRV).

    “Both DM and prediabetes are independently associated with lower heart rate variability (HRV), which suggests that cardiac autonomic dysfunction may be characteristic of prediabetes2–5.”
    Quote in the paperSubstantial claim4 of 4 sources checked
    • cites [2] Hsueh, W. et al. ‘Therapeutic interventions to reduce the risk of progression from prediabetes to type 2 diabetes mellitus’, Therapeutics and Clinical Risk Management, 173, https://doi.org/10.2147/tcrm.s39564 (2014). source

      The “Prediabetes is a disease” section reports increased polyneuropathy in people with impaired fasting glucose or impaired glucose tolerance, but does not discuss heart rate variability. Searches for HRV and cardiac autonomic dysfunction found no passage establishing an independent association between either diabetes or prediabetes and lower HRV. (LLM, conf=0.74)

      Experimental — accuracy not yet measured
    • cites [3] Benichou, T. et al. ‘Heart rate variability in type 2 diabetes mellitus: a systematic review and Meta–analysis’, PLoS ONE 13, https://doi.org/10.1371/journal.pone.0195166 (2018). source

      The Discussion and Conclusion report substantially lower HRV in patients with type 2 diabetes than in those without it. Our copy does not report a corresponding finding for prediabetes or establish that both associations are independent (searched: “prediabetes” and “independent association”). (LLM, conf=0.87)

      Experimental — accuracy not yet measured
    • cites [4] Coopmans, C. et al. Both prediabetes and type 2 diabetes are associated with lower heart rate variability: the Maastricht study. Diab. Care 43, 1126–1133 (2020). source
    • cites [5] Verma, S. et al. Effect of glycemic control and disease duration on cardiac autonomic function and oxidative stress in type 2 diabetes mellitus. J. Diab. Metab. Disord. 17, 149–158 (2018). source

      The “Effect of DM” section (page unavailable) reports reduced HRV measures in people with type 2 diabetes compared with non-diabetic controls. The cited paper does not report a separate prediabetes group or establish an independent association between prediabetes and lower HRV. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    Prior works developed approaches from at most 100 individuals and in most cases recorded less than a day of data.

    “These works developed approaches from at most 100 individuals and in most cases, recorded less than a day of data.”
    Quote in the paperSubstantial claim1 of 10 sources checked
    • cites [12] Nitzan, M., Babchenko, A., Khanokh, B. & Landau, D. The variability of the photoplethysmographic signal-a potential method for the evaluation of the autonomic nervous system. Physiol. Meas. 19, 93 (1998). source
    • cites [13] Usman, S., Mohamad Rozi, R., Reaz, M. B. I. & Mohd Ali, M. A. “Analysis of area under curve of PPG and its relation with HbA1c.” In 2012 IEEE-EMBS Conference on Biomedical Engineering and Sciences, pp. 260–263 (IEEE, 2012). source
    • cites [14] Wu, H.-T. et al. Multiscale cross-approximate entropy analysis as a measure of complexity among the aged and diabetic. Comput. Math. Methods Med. 2013, 324325 (2013). source

      Methods §2.3 reports 30-minute measurements, supporting the claim that this study recorded less than a day of data. But Methods §2.2 lists four groups totaling 123 participants, so this study does not support the asserted maximum of 100 individuals. This paper alone cannot establish what happened in most of the other works. (LLM, conf=0.90)

      Experimental — accuracy not yet measured
    • cites [15] Wu, H.-T., Lee, C.-Y., Liu, C.-C. & Liu, A.-B. Multiscale cross-approximate entropy analysis as a measurement of complexity between ECG R-R interval and PPG pulse amplitude series among the normal and diabetic subjects. Comput. Math. Methods Med. 2013, 231762 (2013).

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [16] Pilt, K., Meigas, K., Ferenets, R., Temitski, K. & Viigimaa, M. Photoplethysmographic signal waveform index for detection of increased arterial stiffness. Physiol. Meas. 35, 2027 (2014). source
    • cites [17] Bharati, S. & Gidveer, G. Waveform analysis of pulse wave detected in the fingertip with PPG. Int. J. Adv. Eng. Technol. 3, 92 (2012).

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [18] Sarkar, S., Kumar Bhoi, A. & Savita, G. Fingertip pulse wave (PPG signal) analysis and heart rate detection. Int. J. Emerg. Technol. Adv. Eng. 2, 404–408 (2012).

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [19] Monte-Moreno, E. Non-invasive estimate of blood glucose and blood pressure from a photoplethysmograph by means of machine learning techniques. Artif. Intell. Med. 53, 127–138 (2011). source
    • cites [20] Okkesim, Ş et al. Comparison of pulse rate variability and heart rate variability for hypoglycemia syndrome. Methods Inf. Med. 55, 250–257 (2016). source
    • cites [21] Moreno, E. M. et al. Type 2 diabetes screening test by means of a pulse oximeter. IEEE Trans. Biomed. Eng. 64, 341–351 (2016).

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Features of heart rate, heart rate variability, movement, and electrodermal activity extracted from wearable data have been applied to detect hypoglycemic events in cohorts ranging from 1 to 36 individuals.

    “Features of HR, HRV, movement, and electrodermal activity (EDA; a measure of perspiration) extracted from wearable data have been applied to detect hypoglycemic events in cohorts ranging from 1 to 36 individuals5,26–28.”
    Quote in the paperNot a substantial claim3 of 4 sources checked
    • cites [5] Verma, S. et al. Effect of glycemic control and disease duration on cardiac autonomic function and oxidative stress in type 2 diabetes mellitus. J. Diab. Metab. Disord. 17, 149–158 (2018). source

      The Cardiac autonomic function section (page unavailable) describes HR and HRV measurements from ECG during an exercise test, not features extracted from wearable data to detect hypoglycemic events. Searches for wearable hypoglycemia detection and electrodermal activity found no passage addressing that application or cohorts of 1–36 individuals. (LLM, conf=0.75)

      Experimental — accuracy not yet measured
    • cites [26] Maritsch, M. et al. “Towards wearable-based hypoglycemia detection and warning in diabetes.” In Extended Abstracts of the 2020 CHI Conference on Human Factors in Computing Systems, pp. 1–8. (ACM, 2020). source
    • cites [27] Lehmann, V. et al. Noninvasive hypoglycemia detection in people with diabetes using smartwatch data. Diab. Care 46, 993–997 (2023). source
    • cites [28] Mendez, C. et al. Toward detection of nocturnal hypoglycemia in people with diabetes using consumer-grade smartwatches and a machine learning approach. J. Diabetes Sci. Technol. 25, 19322968251319800 (2025). source

      The Results (Baseline characteristics; page unavailable) report 36 participants, and Methods (Data preprocessing; page unavailable) describe heart-rate, heart-rate-variability, and movement data used for hypoglycemia detection. The Introduction discusses other wearable studies using electrodermal activity, but this study did not use it, and our copy does not establish the lower bound of one individual. (LLM, conf=0.84)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Skin temperature collected from wrist-worn wearables has been assessed for changes with hyperglycemic events in cohorts ranging from 10 to 20 individuals.

    “Skin temperature was additionally collected from wrist-worn wearables was assessed for changes with hyperglycemic events in cohorts ranging from 10 to 20 individuals29–32.”
    Quote in the paperNot a substantial claim2 of 4 sources checked
    • cites [29] Bogue-Jimenez, B., Huang, X., Powell, D. & Doblas, A. Selection of noninvasive features in wrist-based wearable sensors to predict blood glucose concentrations using machine learning algorithms. Sensors 22, 3534 (2022). source

      Methods §2.1 identifies skin temperature as a feature measured by wristbands, and Results §3.1 assesses its relationship with glucose levels. But Methods §2.2 says the OhioT1DM dataset had 12 participants while this study evaluated only six; Discussion §4 says testing for hyperglycemic events remains future work. The source therefore does not support describing this as an assessment of skin-temperature changes during hyperglycemic events in a cohort of 10–20. (LLM, conf=0.88)

      Experimental — accuracy not yet measured
    • cites [30] Schechter, A. et al. A prototype of a new noninvasive device to detect nocturnal hypoglycemia in adolescents with type 1 diabetes—a pilot study. Diab. Technol. Ther. 14, 683–689 (2012). source
    • cites [31] Niramitmahapanya, S. & Yotha, D. Physiologic change in hypoglycemia in diabetes patients in Rajavithi Hospital using non-invasive hypoglycemic warning wearable devices: a pilot study. Int. J. Innov. Res. Med. Sci. 6, 414–417 (2021). source
    • cites [32] Piersanti, A. et al. Variability of the skin temperature from wrist-worn device for definition of novel digital biomarkers of glycemia. Sensors 25, 4038 (2025). source

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Changes in statistical measures of distal body temperature can be used for identifying hypoglycemic events and predicting blood glucose concentration.

    “These works analyzed how changes in statistical measures of distal body temperature can be used for identifying hypoglycemic events and prediction blood glucose concentration29–32.”
    Quote in the paperSubstantial claim2 of 4 sources checked
    • cites [29] Bogue-Jimenez, B., Huang, X., Powell, D. & Doblas, A. Selection of noninvasive features in wrist-based wearable sensors to predict blood glucose concentrations using machine learning algorithms. Sensors 22, 3534 (2022). source

      The UofM Results (§3.2; page unavailable) report that skin temperature correlated with blood glucose in one subject, supporting its potential relevance to glucose prediction. The cited passages do not show that changes in statistical measures of distal body temperature identify hypoglycemic events; the paper instead evaluates several wearable-sensor features for predicting glucose concentration. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
    • cites [30] Schechter, A. et al. A prototype of a new noninvasive device to detect nocturnal hypoglycemia in adolescents with type 1 diabetes—a pilot study. Diab. Technol. Ther. 14, 683–689 (2012). source
    • cites [31] Niramitmahapanya, S. & Yotha, D. Physiologic change in hypoglycemia in diabetes patients in Rajavithi Hospital using non-invasive hypoglycemic warning wearable devices: a pilot study. Int. J. Innov. Res. Med. Sci. 6, 414–417 (2021). source
    • cites [32] Piersanti, A. et al. Variability of the skin temperature from wrist-worn device for definition of novel digital biomarkers of glycemia. Sensors 25, 4038 (2025). source

      The Results report that wrist skin-temperature metrics differed significantly during hypoglycemic conditions, supporting their potential to identify such events. The Conclusions describe possible use in non-invasive glycemia assessment, but this study did not demonstrate prediction of blood glucose concentration; the Discussion (4.3) calls for further evaluation of hypoglycemia detection. (LLM, conf=0.87)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim.

    Premisecitation

    Diurnal features of distal and core body temperature are associated with future disease risk for a breadth of conditions, including Type 2 Diabetes.

    “Diurnal features of distal and core body temperature have been central to analyses of circadian rhythm33,34; studied as measures for fever, depression, muscular work33,35–37; and revealed to be associated with future disease risk for a breadth of conditions (including Type 2 Diabetes) in a study of the UKBioBank38.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [38] Brooks, T. G. et al. Diurnal rhythms of wrist temperature are associated with future disease risk in the UK Biobank. Nat. Commun. 14, 5172 (2023). source

      The Introduction reports that diurnal wrist-temperature amplitude in the UK Biobank was associated with future diagnoses across many conditions, including type 2 diabetes. The study measured peripheral wrist temperature, however; its discussion of core temperature does not establish an association between core-temperature features and future disease risk. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Diurnal amplitude in distal body temperature measured by an Oura Ring is associated with depression.

    “Diurnal amplitude in distal body temperature measured by an Oura Ring was associated with depression39, possibly due to its sensitivity to thermoregulatory dysregulation.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [39] Mason, A. E. et al. Elevated body temperature is associated with depressive symptoms: results from the TemPredict Study. Sci. Rep. 14, 1884 (2024). source

      The Methods identify the Oura Ring as the device measuring distal body temperature, and the Results report that diurnal amplitude generally decreased as depressive symptom severity increased. However, the Abstract says the overall association only tended in that direction and did not reach statistical significance. The unqualified claim that amplitude “was associated” omits that important caveat. (LLM, conf=0.87)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim.

    Premisecitation

    Multiscale analyses and features of entropy elucidate more information related to the complexity of a data stream than statistics alone.

    “Multiscale analyses and features of entropy elucidate more information related to the complexity of a data stream than statistics alone40–42.”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites [40] Batista, G. E. A. P. A., Wang, X. & Keogh, E. J. A complexity-invariant distance measure for time series. In Proc. 2011 SIAM International Conference on Data Mining. 699–710, https://doi.org/10.1137/1.9781611972818.60 (2011). source
    • cites [41] Costa, M. & Healey, J. A. Multiscale entropy analysis of complex heart rate dynamics: discrimination of age and heart failure effects. Comput. Cardiol. Thessalon. Greece 2003, 705–708 (2003). source
    • cites [42] Burks, J. H. et al. Chronobiologically-informed features from CGM data provide unique information for XGBoost prediction of longer-term glycemic dysregulation in 8,000 individuals with type-2 diabetes. PLoS Digit. Health 4, e0000815 (2025). source

      The Results section on model performance reports that multiscale complexity and across-day variance features added information beyond statistical features in CGM data. The Discussion describes sample entropy as a measure of signal complexity, but the paper used a different complexity index; it does not establish the claim for entropy features or data streams generally. No page number is available in our copy. (LLM, conf=0.84)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductcitation

    For each 24-h period, distal body temperature values below the 5th percentile and above the 95th percentile were discarded to exclude periods of non-wear and clinical fever and to minimize influence from non-representative extreme values.

    “For each 24-h period, we discarded wearable sensor-assessed distal body temperature values below the 5th percentile value and above the 95th percentile. This served to exclude periods of non-wear and periods of clinical fever and to minimize influence from non-representative extreme values (e.g., outliers), as done in prior research with wearable sensor-assessed metrics51,52.”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [51] Kasl, P. et al. Utilizing wearable device data for syndromic surveillance: a fever detection approach. Sensors 24, 1818 (2024). source

      The Materials and Methods section (page unavailable) describes classifying fever days using self-reported symptoms and a temperature threshold of 38 °C. Neither it nor the retrieved passages describes discarding wearable temperature values below the 5th or above the 95th percentile for each 24-hour period. Searches for those percentiles, non-wear, and outlier removal did not establish the attributed method. (LLM, conf=0.74)

      Experimental — accuracy not yet measured
    • cites [52] Smarr, B. L. et al. Feasibility of continuous fever monitoring using wearable devices. Sci. Rep. 10, 21640 (2020). source

      The Methods > Data preparation passage says the paper used medians of the lowest and highest 360 minutes in each 24-hour window to reduce the influence of outliers and acute extremes. It does not describe discarding values below the 5th or above the 95th percentile, or doing so to exclude non-wear and clinical fever. The cited work supports the general outlier-minimization rationale, but not the specific procedure attributed to prior research. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation

    The Oura Ring heart rate and heart rate variability data have been validated by prior work.

    “The HR and HRV data have been validated by prior work48.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [48] Cao, R. et al. Accuracy assessment of oura ring nocturnal heart rate and heart rate variability in comparison with electrocardiography in time and frequency domains: comprehensive analysis. J. Med. Internet Res. 24, e27487 (2022). source

  • Classified as probably not a substantial claim. May combine two claims.

    Findingcitation

    The TemPredict study provided physiological, sleep, and self-report data from 33,374 individuals, of whom 389 self-reported a diabetes mellitus diagnosis (1.16%), 10,820 self-reported never receiving a diabetes mellitus diagnosis (32.4%), and 21,402 did not self-report this information (64.1%).

    “We obtained physiological, sleep, and self-report data from 33,374 individuals through the TemPredict study54, of whom 389 self-reported having received a DM DX from a healthcare professional (1.16%), 10,820 self-reported having never received a DM DX from a healthcare professional (32.4%), and 21,402 individuals did not self-report this information (64.1%; see Fig. 1a).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [54] Soltani, S. et al. Testing the impact of intensive, longitudinal sampling on assessments of statistical power and effect size within a heterogeneous human population: natural experiment using change in heart rate on weekends as a surrogate intervention. J. Med. Internet Res. 27, e60284 (2025). source

      The Introduction and Methods > Data and Device identify TemPredict as the source of sleep-time physiological measurements and report participant self-reported demographics. The paper analyzes a different cohort (46,217 participants in Results > Participant Characteristics) and does not report the claimed 33,374-person cohort or its diabetes-diagnosis counts and percentages. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • Rests on an assumption: Self-reported diabetes mellitus diagnosis accurately indicates the presence of diabetes mellitus.
    • Rests on an assumption: Self-reporting no diabetes mellitus diagnosis accurately indicates absence of diabetes mellitus.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Commonly used features are supported by prior work to be relevant to the effects of DM.

    “commonly used features supported by prior work to be relevant to the effects of DM31–33,35–38,45,49”
    Quote in the paperNot a substantial claim6 of 9 sources checked
    • cites [31] Niramitmahapanya, S. & Yotha, D. Physiologic change in hypoglycemia in diabetes patients in Rajavithi Hospital using non-invasive hypoglycemic warning wearable devices: a pilot study. Int. J. Innov. Res. Med. Sci. 6, 414–417 (2021). source
    • cites [32] Piersanti, A. et al. Variability of the skin temperature from wrist-worn device for definition of novel digital biomarkers of glycemia. Sensors 25, 4038 (2025). source

      The Introduction (p. unavailable) reports prior evidence linking skin temperature to glycemic variation, and Discussion §4.2 says earlier studies used standard statistical temperature features. But the paper describes its engineered features as novel and studied people at risk of diabetes, not people with overt diabetes; it does not establish those features as commonly used measures of the effects of DM. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
    • cites [33] Vellei, M., Chinazzo, G., Zitting, K. M. & Hubbard, J. Human thermal perception and time of day: a review. Temperature 8, 320–341 (2021). source

      The Discussion (Individual differences in circadian timing; Controlling for light) identifies chronotype, sleep schedules, and light exposure as relevant to detecting time-of-day effects on thermal perception. It does not establish that the citing document’s unspecified ‘commonly used features’ are relevant to the effects of DM, so that broader attribution cannot be verified from this source. (LLM, conf=0.68)

      Experimental — accuracy not yet measured
    • cites [35] Amiya, E. et al. Diurnal body temperature rise is reduced in diabetes with autonomic neuropathy. Clin. Auton. Res. 24, 95–97 (2014). source
    • cites [36] Miller A. C., Koeneman S. H., Suneja M., Cavanaugh J. E. & Polgreen P.M. Diurnal temperature variation and the implications for diagnosis and infectious disease screening: a population-based study. Diagnosis (Berl). 11, 54–62 (2023). source

      The Background describes a study of diurnal body-temperature variation and fever detection by age and sex. Searches for “DM,” “diabetes mellitus,” and features relevant to the effects of DM found no passage in our copy addressing the attributed claim. (LLM, conf=0.76)

      Experimental — accuracy not yet measured
    • cites [37] Hagan, R. D. & Horvath, S. M. Effect of diurnal rhythm of body temperature on muscular work. J. Therm. Biol. 3, 235–239 (1978). source
    • cites [38] Brooks, T. G. et al. Diurnal rhythms of wrist temperature are associated with future disease risk in the UK Biobank. Nat. Commun. 14, 5172 (2023). source
    • cites [45] de Zambotti, M. et al. The sleep of the ring: comparison of the ŌURA sleep tracker against polysomnography. Behav. Sleep. Med. 17, 124–136 (2019). source

      The Discussion (page not provided) describes sleep parameters and heart-rate variability as useful features for distinguishing sleep stages. It does not link those features to the effects of DM. The Methods > Participants section (page not provided) says participants with diabetes were excluded. (LLM, conf=0.82)

      Experimental — accuracy not yet measured
    • cites [49] Inaishi, J. et al. Associations between glycemic variability, sleep quality, and daily steps in subjects without diabetes using wearable devices. Metab. Open 20, 100263 (2023). source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    With a larger, heterogeneous, negative population, more false positives should arise because people differ biologically.

    “We expected with a larger, heterogeneous, negative population, more false positives should arise because people differ biologically55.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [55] Parameswaran, G. & Ray, D. W. Sleep, circadian rhythms, and type 2 diabetes mellitus. Clin. Endocrinol. 96, 12–20 (2021). source

      Table 1 (Introduction; page unavailable) discusses large, diverse populations in sleep research, but not negative populations or false-positive results. Searches for “false positive” and diagnostic specificity found no passage linking biological differences to an increase in false positives. (LLM, conf=0.75)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Findingcitation

    The balanced cohort is 5 times and the imbalanced cohort is 91 times the largest prior cohort of wearable data.

    “balanced cohort is 5× and imbalanced cohort is 91× the largest prior cohort of wearable data”
    Quote in the paperSubstantial claim3 of 5 sources checked
    • cites [5] Verma, S. et al. Effect of glycemic control and disease duration on cardiac autonomic function and oxidative stress in type 2 diabetes mellitus. J. Diab. Metab. Disord. 17, 149–158 (2018). source

      The Sample size section reports 123 participants—60 with type 2 diabetes and 63 controls—and the Limitations and future perspective section calls this a small sample. Searches for the stated 5× and 91× comparisons and for balanced, imbalanced wearable-data cohorts found no passage addressing them. Page numbers were not available in our copy. (LLM, conf=0.78)

      Experimental — accuracy not yet measured
    • cites [29] Bogue-Jimenez, B., Huang, X., Powell, D. & Doblas, A. Selection of noninvasive features in wrist-based wearable sensors to predict blood glucose concentrations using machine learning algorithms. Sensors 22, 3534 (2022). source

      Section 2.2 describes the OhioT1DM and UofM datasets, including their participant counts, but does not identify a balanced or imbalanced cohort or compare either with the largest prior wearable-data cohort. Searches for the stated 5× and 91× comparisons found no passage addressing them. (LLM, conf=0.76)

      Experimental — accuracy not yet measured
    • cites [30] Schechter, A. et al. A prototype of a new noninvasive device to detect nocturnal hypoglycemia in adolescents with type 1 diabetes—a pilot study. Diab. Technol. Ther. 14, 683–689 (2012). source
    • cites [31] Niramitmahapanya, S. & Yotha, D. Physiologic change in hypoglycemia in diabetes patients in Rajavithi Hospital using non-invasive hypoglycemic warning wearable devices: a pilot study. Int. J. Innov. Res. Med. Sci. 6, 414–417 (2021). source
    • cites [32] Piersanti, A. et al. Variability of the skin temperature from wrist-worn device for definition of novel digital biomarkers of glycemia. Sensors 25, 4038 (2025). source

      The Results report 15 analyzed subjects, and Discussion §4.2 mentions other wearable datasets, but neither section identifies a largest prior wearable cohort or compares it with balanced and imbalanced cohorts. Searches for the 5× and 91× ratios and for cohort-size comparisons found no supporting passage in our copy; page numbers were unavailable. (LLM, conf=0.73)

      Experimental — accuracy not yet measured

  • The quote only partly states this claim.

    Findingown data

    In the balanced cohort, longer time windows of input perform better.

    “Here we show longer time windows of input perform better, with the best algorithm (21-nights) achieving 0.88 Area under ROC (AUROC) and 0.80 Area under Precision Recall (AUPRC) (0.30 improvement over random).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The model achieves 0.80 AUROC and 0.28 improvement over random in AUPRC in an imbalanced cohort drawn from 6,658 individuals, emulating a general population.

    “The model achieves 0.80 AUROC and 0.28 improvement over random in AUPRC in an imbalanced cohort drawn from 6,658 individuals, emulating a general population.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The study extracted 32 other features from prior work and found significant differences in these features as well.

    “We extracted 32 other features (see Feature Extraction in “Methods”) from prior work and found significant differences in these features as well (shown in Supplementary Table 4).”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The observed differences in physiological features between people self-reporting diabetes mellitus and people self-reporting no diabetes mellitus diagnosis are attributable to diabetes status rather than to age, sex, or other confounders.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingnone given

    Rebalancing in the larger dataset led to an increased number of individuals self-reporting DM in the testing data.

    “rebalancing in the larger dataset led to an increased number of individuals self-reporting DM in the testing data.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Findingown data

    After data filtering, the dataset contained 24,486 nights from 315 people self-reporting diabetes mellitus and 764,315 nights from 9,063 people self-reporting no diabetes mellitus diagnosis, with 36 features and 84.3 nights per individual (65.3% data completeness).

    “After data filtering, we obtained 24,486 nights from 315 people self-reporting DM and 764,315 nights from 9063 people self-reporting no DM DX with 36 features, 84.3 nights per individual (65.3% data completeness).”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The data filtering and completeness do not introduce systematic bias that affects the comparisons between people self-reporting diabetes mellitus and people self-reporting no diabetes mellitus diagnosis.

  • Classified as probably not a substantial claim.

    Findingown data

    The imbalanced cohort testing dataset consisted of 87 individuals who self-reported DM and 6571 individuals who self-reported no DM DX, not in the training dataset.

    “The imbalanced cohort testing dataset consisted of 87 individuals who self-reported DM and 6571 individuals who self-reported no DM DX, not in the training dataset.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingreasoning

    Each curve was generated from 870 windows from people self-reporting DM and 3990 windows from people self-reporting no DM DX.

    “So, each curve was generated from 870 windows from people self-reporting DM and 3990 windows from people self-reporting no DM DX.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data

    Data completeness was 84.6 nights per individual (65.6%) for individuals self-reporting no diabetes mellitus diagnosis and 77.7 nights per individual (60.2%) for individuals self-reporting diabetes mellitus.

    “This corresponds to 84.6 nights per individual (65.6% data completeness) from individuals who self-reported no DM DX and 77.7 nights from individuals (60.2% data completeness) who self-reported DM.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The data filtering and completeness do not introduce systematic bias that affects the comparisons between people self-reporting diabetes mellitus and people self-reporting no diabetes mellitus diagnosis.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    The distributions of the older population self-reporting diabetes mellitus tend to be less Gaussian, with bimodal distributions in the 60-69 and 70-79 age groups.

    “The darker red lines that represent the older population self-reporting DM tend to be less Gaussian across the figures (e.g., Fig. 1f: both the 60–69 and 70–79 distributions are bimodal; similar pattern in Fig. 1c)”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    For the analysis of true positive, false negative, and true negative time windows, the study used 3,093 true negatives, 394 false negatives, and 1,746 true positive time windows.

    “For the analysis of true positive, false negative, and true negative time windows, we used 3093 true negatives, 394 false negatives, and 1746 true positive time windows to conduct a Kruskal–Wallis test, followed by a 2-sided Posthoc Dunn’s Test.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Our testing data contained 18 individuals self-reporting no DM DX while reporting a chronic condition other than DM (18 out of the 56 individuals self-reporting no DM DX in the test dataset).

    “Our testing data contained 18 individuals self-reporting no DM DX while reporting a chronic condition other than DM (18 out of the 56 individuals self-reporting no DM DX in the test dataset).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    For the analysis of other chronic conditions, the tests used 47 people self-reporting diabetes mellitus and 38 people self-reporting no diabetes mellitus diagnosis with no chronic conditions, and 18 people self-reporting no diabetes mellitus diagnosis with another chronic condition.

    “Both tests used 47 people self-reporting DM and 38 people self-reporting no DM DX with no chronic conditions, and 18 people self-reporting no DM DX with another chronic condition (more detailed sample sizes in Supplementary Table 1).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    The held-out testing data for the machine learning models consisted of 5,963 time windows drawn from 103 individuals.

    “Figure 2 shows these models’ results on the held-out testing data, 5963 time windows drawn from 103 individuals (see Table 3 for detailed sample size data).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    All age groups self-reporting no diabetes mellitus diagnosis had similar kernel density estimates, and distributions of groups self-reporting diabetes mellitus were consistently shifted away from the blue lines.

    “All the age groups self-reporting no DM DX had similar kernel density estimates, and distributions of groups self-reporting DM were consistently shifted away from the blue lines.”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingown data

    The distributions of predicted probabilities of people self-reporting no DM DX without other chronic conditions were significantly different from the distribution of predicted probabilities for the cohort self-reporting DM (Posthoc Dunn's; 2-sided; p-value = 6.9e-12).

    “People Self-Reporting DM vs. People Self-Reporting No Chronic Conditions, p-value = 6.9e-12”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Features derived from HR and HRV (hr_lowest, rmssd) had the highest average feature importance and were the most informative in identifying people self-reporting DM among all features used in this study.

    “the highest average feature importance for individual features was found in two items in feature set 1 (hr_lowest, rmssd), demonstrating that features derived from HR and HRV were the most informative in identifying people self-reporting DM out of all the features used in this study”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The 21-day XGBClassifier achieved a median AUPRC of 0.46 on the imbalanced cohort, which represents a 0.28 improvement over a random classifier (0.179 AUPRC).

    “As shown in Fig. 5, the 21-day XGBClassifier achieved a median AUPRC of 0.46, which represents a 0.28 improvement over a random classifier (0.179 AUPRC).”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Self-reported diabetes diagnosis accurately identifies individuals with diabetes.
    • Rests on an assumption: The imbalanced dataset constructed from 6571 individuals self-reporting no DM DX is representative of the general US population.

  • May combine two claims.

    Findingown data

    In both models (using only the first three features and using all five features), the predicted probabilities of both cohorts self-reporting no DM DX were significantly different from the predicted probabilities of the cohort self-reporting DM, but were not significantly different from each other.

    “In both models, when using only the first three features as well as using all five features, the predicted probabilities of both cohorts self-reporting no DM DX were significantly different from the predicted probabilities of the cohort self-reporting DM, but were not significantly different from each other.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Supplementary Table 1

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • Rests on an assumption: Self-reported diabetes diagnosis accurately identifies individuals with diabetes.

  • May combine two claims.

    Findingown data

    In all three groups, the addition of temperature features resulted in a higher probability for the cohort self-reporting DM to be predicted as DM-positive and a lower probability for the two cohorts self-reporting no DM DX to be predicted as DM-positive.

    “In all three groups, the addition of temperature features resulted in a higher probability for the cohort self-reporting DM to be predicted as DM-positive and a lower probability for the two cohorts self-reporting no DM DX to be predicted as DM-positive.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Self-reported diabetes diagnosis accurately identifies individuals with diabetes.

  • May combine two claims.

    Findingown data

    The improvement over a random classifier in the imbalanced cohort (0.28) is comparable to that in the balanced cohort, where the model's AUPRC (0.80) represented a 0.30 improvement over a random classifier.

    “This improvement over a random classifier is comparable to that of the analysis in the balanced cohort, where the model’s AUPRC on the balanced cohort (0.80) represented a 0.30 improvement over a random classifier.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Self-reported diabetes diagnosis accurately identifies individuals with diabetes.
    • Rests on an assumption: The imbalanced dataset constructed from 6571 individuals self-reporting no DM DX is representative of the general US population.

  • May combine two claims.

    Findingown data

    The work demonstrates in naturalized contexts that features derived to capture biological dynamics (secondary features) lead to gains in separability of physiology consistent with DM (delta AUROC = 0.09), complementary to primary features.

    “Our work demonstrates in naturalized contexts that features derived to capture biological dynamics (secondary features) lead to gains in separability of physiology consistent with DM (delta AUROC = 0.09), complementary to primary features.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Self-reported diabetes diagnosis accurately identifies individuals with diabetes.
    • Rests on an assumption: Wearable device data streams (HR, HRV, sleep, distal body temperature) are valid measures of the physiological effects of DM.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data

    Our data is gathered from all over the US and represents a variety of climates with differing environmental temperatures.

    “Our data is gathered from all over the US and so represents a variety of climates with differing environmental temperatures.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisenone given

    All prior works either used short recordings from a single day as input or evaluated their methods with small cohorts of 23-100 subjects.

    “Finally, all prior works either used short recordings (usually only a couple minutes) from a single day as input or evaluated their methods with small cohorts (23–100 subjects), raising the question of whether these methods maintain their performance over time (over weekly and monthly timescales) and across individuals (as the effects of DM can differ with ages and comorbidities).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisereasoning

    The combination of features from multiple data streams is essential because it holds potential for improved separation between physiology consistent with diabetes mellitus and other states.

    “First, the combination of features from multiple data streams is essential because it holds the potential for improved separation between physiology consistent with DM and other states.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    The UCSF Institutional Review Board and the U.S. Department of Defense Human Research Protections Office approved all study activities, and all research was performed in accordance with relevant guidelines and regulations and the Declaration of Helsinki.

    “The University of California, San Francisco (UCSF) Institutional Review Board (IRB, IRB# 20-30408) and the U.S. Department of Defense (DOD) Human Research Protections Office (HRPO, HRPO# E01877.1a) approved of all study activities, and all research was performed in accordance with relevant guidelines and regulations and the Declaration of Helsinki.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductown data

    The study performed stratified sampling based on age and sex to obtain a balanced number of individuals self-reporting diabetes mellitus and individuals self-reporting no diabetes mellitus diagnosis in each stratum.

    “We then performed stratified sampling based on age and sex to obtain a balanced number of individuals self-reporting DM and individuals self-reporting no DM DX in each stratum, for a total of 300 individuals self-reporting DM and 300 individuals self-reporting no DM DX.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    These methods could make such separations possible for other chronic conditions that affect sleep and inflammation.

    “These results indicate the value of biologically informed features and longitudinal data for identifying people with diabetes and further, suggest that these methods could make such separations possible for other chronic conditions that affect sleep and inflammation.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductown data

    The controlled dataset was split into 80% training/validation and 20% held-out testing data by individual, with balanced numbers of individuals self-reporting DM and no DM DX, and balanced age bins in each dataset.

    “We split our controlled dataset into 80% training/validation data and 20% held-out testing data by individual, with a balanced number of individuals self-reporting DM and individuals self-reporting no DM DX, as well as a balanced number of individuals in each age bin in each dataset.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    All models, each a combination of different window lengths and algorithms, were tested on the same set of individuals and start dates of time windows.

    “Thus, all models, each a combination of different window lengths and algorithms, were tested on the same set of individuals and start dates of time windows.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductreasoning

    Samples were excluded if more than 2 nights in each week were missing, ensuring similar data quality as the window size changed.

    “Samples were excluded if more than 2 nights in each week were missing (so valid samples included ≤2 nights missing for 3-night windows, ≤2 nights missing for 7-night windows, ≤4 nights missing for 14-night windows, and ≤6 nights missing for 21-night windows). This ensured similar data quality as the window size changed.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    All participants provided informed consent (electronic).

    “All participants provided informed consent (electronic).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    The study controlled for time of year by limiting data extraction to 2020-04-22 to 2020-08-29.

    “We controlled for time of year by limiting data extraction to data falling within a specified date range: 2020-04-22 to 2020-08-29.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Premisenone given

    Metrics that depend on the proportion of each class (e.g., AUPRC) should be expected to vary in populations with different proportions of people with DM.

    “Metrics that depend on the proportion of each class (e.g., AUPRC) should be expected to vary in populations with different proportions of people with DM.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    Kruskal-Wallis tests and Mann-Whitney U tests were used instead of ANOVAs and t-tests to avoid assumptions of normality in the tested distributions.

    “We used Kruskal–Wallis tests and Mann–Whitney U tests instead of ANOVAs and t-tests to avoid assumptions of normality in the tested distributions.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    All analyses of other chronic conditions were performed with the testing dataset.

    “All these analyses were performed with our testing dataset.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    Hyperparameters were optimized with 3-fold cross-validation prior to final evaluation on the held-out testing data.

    “We optimized hyperparameters with 3-fold cross-validation prior to the final evaluation of our algorithms on the held-out testing data.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisereasoning

    AUROC and AUPRC consider the probabilities outputted by the model, whereas metrics like F1 score convert the probabilities to a binary variable based on a specific threshold, reducing the information considered in the metric.

    “We select the optimal model based on AUROC and AUPRC because we analyze the probabilities outputted by the model in later analyses. AUROC and AUPRC consider these probabilities, whereas metrics like F1 score convert the probabilities to a binary variable based on a specific threshold, reducing the information considered in the metric.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    These outcomes represent the degree to which this model improves the separability of physiology consistent with DM in a general population.

    “These outcomes represent the degree to which this model improves the separability of physiology consistent with DM in a general population.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    The model was trained on the training and validation data from the balanced cohort and evaluated using a combination of the testing data from the balanced cohort and additional data from individuals excluded when constructing the balanced cohorts.

    “This model was trained on the training and validation data from the balanced cohort and evaluated using a combination of the testing data from the balanced cohort and additional data from individuals excluded when constructing the balanced cohorts.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The US distribution of people with DM is 17.9% as estimated by the AHA.

    “we chose to match the US distribution of people with DM (17.9% as estimated by the AHA1).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites AHA1

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    All feature importance and ablation analyses were performed with the testing dataset.

    “All these analyses were performed with our testing dataset.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    The model is not tuned for optimal performance, so the outcomes should not be considered the maximal performance achievable in the general population.

    “Further, this model is not tuned for optimal performance, so the outcomes of this experiment should not be considered the maximal performance that could be achieved in the general population.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    The study used Kruskal-Wallis tests and Mann-Whitney U tests instead of ANOVAs and t-tests to avoid assumptions of normality in the tested distributions.

    “We used Kruskal–Wallis tests and Mann–Whitney U tests instead of ANOVAs and t-tests to avoid assumptions of normality in the tested distributions.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The US distribution of people with DM is 17.9%, as estimated by the AHA.

    “set to match the US distribution of people with DM (17.9% as estimated by the AHA1).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites AHA1

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    All statistical test results shown in Fig. 3 were generated from Dunn's Test, using Benjamini-Hochberg p-value threshold correction.

    “All statistical test results shown in this figure were generated from Dunn’s Test, using Benjamini–Hochberg p-value threshold correction.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The model was able to differentiate people self-reporting DM from people self-reporting no DM DX, regardless of whether they had other chronic conditions, and this differentiation was strengthened by the addition of temperature features.

    “These results demonstrate that we were able to differentiate people self-reporting DM from people self-reporting no DM DX, regardless of whether they had other chronic conditions, and that this differentiation was strengthened by the addition of temperature features.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Self-reported diabetes diagnosis accurately identifies individuals with diabetes.

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Premisenone given

    Unlike in the detection of acute infections, for chronic conditions, the physiological changes might be relatively small but persistent, and so easier to separate by resampling over time.

    “Unlike in the detection of acute infections, for chronic conditions, the physiological changes might be relatively small but persistent, and so easier to separate by resampling over time.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The model recovers comparable nonrandom information in an imbalanced cohort to the balanced cohort, indicating the robustness of sleep and distal body temperature features and highlighting the importance of measuring across time and not relying on single time points or days.

    “Nevertheless, it recovers comparable nonrandom information in an imbalanced cohort to the balanced cohort, indicating the robustness of sleep and distal body temperature features and highlighting the importance of measuring across time and not relying on single time points or days.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Self-reported diabetes diagnosis accurately identifies individuals with diabetes.
    • Rests on an assumption: The imbalanced dataset constructed from 6571 individuals self-reporting no DM DX is representative of the general US population.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationown data

    Secondary features are likely also essential to developing highly specific algorithms for other tasks requiring identification of physiology consistent with DM in natural settings.

    “Secondary features are likely also essential to developing highly specific algorithms for other tasks requiring identification of physiology consistent with DM in natural settings.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Self-reported diabetes diagnosis accurately identifies individuals with diabetes.
    • Rests on an assumption: Wearable device data streams (HR, HRV, sleep, distal body temperature) are valid measures of the physiological effects of DM.

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    Our work moves the needle towards passive screening for DM risk from a broad array of non-specialized devices, as well as passive, noninvasive monitoring to complement CGM for those in need early in their condition.

    “Our work moves the needle towards passive screening for DM risk from a broad array of non-specialized devices, as well as passive, noninvasive monitoring to complement CGM for those in need early in their condition.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductreasoning

    The ground truth labels are self-reported and do not distinguish between types of DM, which likely carries some imprecision.

    “Third, our ground truth labels are self-reported, which likely carries some imprecision, and they do not distinguish between types of DM, suggesting some of our time windows may not have been representative of their class.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    We attempted to account for seasonal changes by limiting data to a single season (spring), but could not account for climate differences between geographical areas.

    “While we attempted to account for seasonal changes in environmental temperature by limiting our data to span a single season (spring), we couldn’t account for climate differences between geographical areas.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    The findings cannot be reproduced due to the data being private commercial data only available upon request.

    “Another limitation to our findings is that they cannot be reproduced due to the data being private commercial data that is only available upon a reasonable or qualified request.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    This is consistent with some individuals managing their DM more effectively at some times than others.

    “This is consistent with some individuals managing their DM more effectively at some times than others, as might be expected.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationown data

    Despite limitations, our work suggests that a useful screening or detection tool is possible and may be further adapted and developed in a clinical study.

    “Despite these limitations, we believe our work suggests that a useful screening or detection tool is possible and may be further adapted and developed in a clinical study.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: A wide breadth of behavioral and health factors orthogonal to DM can lead to similar physiological and behavioral effects.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

    Interpretationreasoning

    The variance in physiological effect unavoidably limits the effectiveness of any physiologically based detection tool working in isolation.

    “The variance in physiological effect unavoidably limits the effectiveness of any physiologically based detection tool working in isolation.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The deleterious physiological effects of hyperglycemia vary within an individual over time and across individuals.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    We did not explore deep learning algorithms.

    “Fourth, we did not explore deep learning algorithms as there are numerous combinations of architectures and training methods, and differences between these combinations don’t meaningfully test our hypotheses regarding sleep and skin temperature features.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The study's approach is limited to modalities measured by the wearable device used.

    “First, our approach is limited to modalities measured by the wearable device used.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Different sensors have different margins of error in measurement.
74 of 74

  • The paper relies on this without support.

    Decreased blood flow to the skin and decreased peripheral uptake of glucose lead to reductions in skin temperature in response to environmental heat or food intake.

    “Decreased blood flow to the skin and decreased peripheral uptake of glucose lead to reductions in skin temperature in response to environmental heat or food intake.”

  • The paper relies on this without support.

    Distal body temperature is a physiological function of core body temperature, vascularization in the extremities, coupled with sleep cycles.

    “Distal body temperature is a physiological function of core body temperature, vascularization in the extremities, coupled with sleep cycles.”

  • The paper relies on this without support.

    Diabetes mellitus affects physiological control mechanisms related to nocturnal thermoregulation.

    “DM affects physiological control mechanisms related to nocturnal thermoregulation”

  • The paper relies on this without support.

    Circadian rhythmicity is affected by diabetes mellitus.

    “circadian rhythmicity is affected by DM”

  • The paper relies on this without support.

    Measures of complexity in heart rate variability are related to differences across ages as well as between healthy and disease states.

    “It has been shown that measures of complexity in HRV are related to differences across ages as well as between healthy and disease states.”

  • The paper relies on this without support.

    Other chronic conditions perturb the physiological data streams measured.

    “Other chronic conditions also perturb the physiological data streams we measured”

  • The paper relies on this without support.

    LogisticRegression and XGBClassifier have a standard method to quantify feature importance.

    “Both the algorithms we used, LogisticRegression and XGBClassifier, have a standard method to quantify feature importance, or how much each feature contributes to predictions.”

  • The paper relies on this without support.

    Feature ablation is a method for evaluating the value of features in performance.

    “Another method for evaluating the value of features in performance is feature ablation.”

  • The paper relies on this without support.

    Many wearable health devices do not have access to a temperature sensor.

    “because many wearable health devices do not have access to this sensor”
    “many wearable devices still do not have temperature sensors”

  • The paper relies on this without support.

    Distal body temperature is a more stable physiological indicator than heart rate.

    “Distal body temperature is a more stable physiological indicator than HR”

  • The paper relies on this without support.

    The exact proportion of individuals with DM varies from country to country.

    “The exact proportion of individuals with DM varies from country to country”

  • The paper relies on this without support.

    Self-reported diabetes mellitus diagnosis accurately indicates the presence of diabetes mellitus.

    “389 self-reported having received a DM DX from a healthcare professional (1.16%)”

  • The paper relies on this without support.

    Self-reporting no diabetes mellitus diagnosis accurately indicates absence of diabetes mellitus.

    “10,820 self-reported having never received a DM DX from a healthcare professional (32.4%)”

  • The paper relies on this without support.

    The Oura Ring's proprietary algorithms provide valid measurements of heart rate, heart rate variability, respiratory rate, respiratory rate variability, and distal body temperature.

    “We obtained nightly-level physiological data of HR, HRV, respiratory rate, and respiratory rate variability that were generated by the proprietary Oura algorithm.”

  • The paper relies on this without support.

    The observed differences in physiological features between people self-reporting diabetes mellitus and people self-reporting no diabetes mellitus diagnosis are attributable to diabetes status rather than to age, sex, or other confounders.

    “The commonly used feature values from the population self-reporting DM differed significantly from those of the population self-reporting no DM DX in HR, RMSSD, TST, and DDBTA”

  • The paper relies on this without support.

    The data filtering and completeness do not introduce systematic bias that affects the comparisons between people self-reporting diabetes mellitus and people self-reporting no diabetes mellitus diagnosis.

    “After data filtering, we obtained 24,486 nights from 315 people self-reporting DM and 764,315 nights from 9063 people self-reporting no DM DX with 36 features, 84.3 nights per individual (65.3% data completeness).”

  • The paper relies on this without support.

    Some physiological effects of DM are likely shared by other chronic conditions (e.g., hypertension, sleep apnea).

    “Some of the physiological effects of DM are likely shared by other chronic conditions (e.g., hypertension, sleep apnea)”

  • The paper relies on this without support.

    Self-reported diabetes diagnosis accurately identifies individuals with diabetes.

    “Individuals self-reporting DM differed significantly from individuals self-reporting no DM DX in average nightly HR, average nightly HRV, TST, and DDBTA with moderate to large effect sizes.”

  • The paper relies on this without support.

    The imbalanced dataset constructed from 6571 individuals self-reporting no DM DX is representative of the general US population.

    “This imbalanced dataset was constructed with all possible individuals self-reporting DM not in the training data (N = 87) and a random sample (N = 399) drawn from 6571 individuals self-reporting no DM DX, set to match the US distribution of people with DM (17.9% as estimated by the AHA1).”

  • The paper relies on this without support.

    Wearable device data streams (HR, HRV, sleep, distal body temperature) are valid measures of the physiological effects of DM.

    “Commercially available wearable devices continuously monitor data streams related to physiology (e.g., HR, HRV, sleep, and distal body temperature), and researchers may be able to use these data to measure the physiological effects of DM and/or related conditions on the cardiovascular and autonomic nervous systems, as well as the effects of DM on sleep.”

  • The paper relies on this without support.

    Sensor accuracy can vary with skin tone and ethnicity.

    “Second, sensor accuracy can vary with skin tone and ethnicity.”

  • The paper relies on this without support.

    The deleterious physiological effects of hyperglycemia vary within an individual over time and across individuals.

    “Fifth, the deleterious physiological effects of hyperglycemia vary within an individual over time, based on environmental factors, as well as across individuals, possibly based on genetic factors.”

  • The paper relies on this without support.

    A wide breadth of behavioral and health factors orthogonal to DM can lead to similar physiological and behavioral effects.

    “Finally, we note that a wide breadth of behavioral and health factors orthogonal to DM can lead to similar physiological and behavioral effects.”

  • The paper relies on this without support.

    Longer time horizons than 21 days contain more information but more complex longitudinal patterns.

    “In addition, longer time horizons than 21-days contain more information but more complex longitudinal patterns.”

  • The paper relies on this without support.

    The effects of behavioral variability and intrinsic physiological cycles on longitudinal wearable data have not yet been well studied.

    “two factors whose effects on longitudinal wearable data have not yet been well studied”

  • The paper relies on this without support.

    Different sensors have different margins of error in measurement.

    “especially since different sensors have different margins of error in measurement”
26 of 26

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Reported false-positive rate uses the wrong count

    Major

    If 394 of 3,823 no-DM windows were false positives, the false-positive rate would be 10.3%, not 19.1%. Elsewhere the paper identifies 394 as the number of false negatives.

    Reported 19.1 · recomputed 10.306042

  • Table 1's no-DM night total does not sum

    Minor

    The five age-bin subtotal rows contain 662,147 nights from people reporting no DM diagnosis, rather than the reported total of 764,315.

    Reported 764315 · recomputed 662147

  • Third complexity scale has the wrong stated resolution

    Minor

    The first scale groups two minute-level samples and each subsequent scale doubles the grouping. The third scale therefore spans 8 minutes, not 16.

    Reported 16 · recomputed 8

  • A p-value of 0.100 is described as significant

    Minor

    The text says the model distributions differed significantly for the listed groups, including the other-chronic-condition group with p = 0.100. The paper's stated significance threshold is p < 0.05.

  • Table 1's individual totals exceed the age-bin subtotals

    Minor

    The five age-bin subtotal rows sum to 7,709 people reporting no DM diagnosis, not 9,063, and to 308 people reporting DM, not 315. The table describes its final rows as totals across the age groups.

    Reported 9063 · recomputed 7709

5 of 5

Cites

Flagged first, then by citations

Research works printed but not linked: 2

  1. Bharati, S. & Gidveer, G. Waveform analysis of pulse wave detected in the fingertip with PPG. Int. J. Adv. Eng. Technol. 3, 92 (2012).
  2. Sarkar, S., Kumar Bhoi, A. & Savita, G. Fingertip pulse wave (PPG signal) analysis and heart rate detection. Int. J. Emerg. Technol. Adv. Eng. 2, 404–408 (2012).

Cited by

Still being checked.

Authors
Viswanath, V. K., Navaneethan, S., Burks, J. H., Soltani, S., Kasl, P., Hartogensis, W., Dilchert, S., Hecht, F. M., Mason, A. E., Wang, E. J., & Smarr, B. L.
Affiliations
  • University of California San Diego (US)
  • University of California, San Francisco (US)
  • Baruch College (US)
Corresponding author
Viswanath, Varun K.
ORCID
Venue
Communications Medicine
Publisher
Springer Science and Business Media LLC
Published
2026-03-16
Volume
6
Issue
1
ISSN
2730-664X (ISSN-L)
DOI
10.1038/s43856-026-01501-0
PMC
PMC13079903
OpenAlex
W7137255072
Semantic Scholar
575f87eebd450521d75aff18b5d6d97d819377fd
Cited by
0
References
55
FWCI
0.00
Type
journal article
Language
en
Open access
Licence
Locations
  • Communications Medicine (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • PubMed Central (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • U.S. Department of Defense (7029991, MTEC-20-12-COVID19-D.-023, 134650)
  • Osher Center for Integrative Medicine
  • National Metal and Materials Technology Center
  • University of California, San Diego
  • University of California, San Francisco
  • Jacobs School of Engineering, University of California, San Diego
  • U.S. Army Medical Research and Development Command (MTEC-20-12-COVID19-D.-023)
Topics
  • Health Sciences › Medicine › Endocrinology, Diabetes and Metabolism › Diabetes Management and Research
  • Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
  • Physical Sciences › Engineering › Biomedical Engineering › Non-Invasive Vital Sign Monitoring
Keywords
diabetes mellitus, non-invasive disease detection, wearable devices, sleep, body temperature, circadian rhythm disruption, Oura Ring, chronic disease classification, imbalanced classification

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review