Objective prediction of siesta based on machine learning and association with obesity

IF 4.8 2区医学 Q2 CLINICAL NEUROLOGY

Sleep Health Pub Date : 2026-06-01 Epub Date: 2026-04-09 DOI:10.1016/j.sleh.2026.02.007

María Rodríguez-Martín MSc , Fernando Moreno Caballero BSc , Hassan S. Dashti PhD, RD , Richa Saxena PhD , Frank A.J.L. Scheer PhD , Jesualdo T. Fernández Breis PhD , Marta Garaulet PhD

{"title":"Objective prediction of siesta based on machine learning and association with obesity","authors":"María Rodríguez-Martín MSc , Fernando Moreno Caballero BSc , Hassan S. Dashti PhD, RD , Richa Saxena PhD , Frank A.J.L. Scheer PhD , Jesualdo T. Fernández Breis PhD , Marta Garaulet PhD","doi":"10.1016/j.sleh.2026.02.007","DOIUrl":null,"url":null,"abstract":"<div><h3>Objectives</h3><div>To predict siesta behavior using machine learning models trained on self-reported and objective data—temperature (T), activity (A), position (P), and the integrated TAP variable—and to explore its associations with obesity-related traits.</div></div><div><h3>Methods</h3><div>From ONTIME-MT, 889 adults wore wrist sensors for 7 days to continuously record temperature, activity, and position, and self-reported daily siesta. Machine learning models were developed to classify 30-second epoch siesta data, to reconstruct weekly siesta behavior. Anthropometric and metabolic parameters were assessed. Associations were analyzed using linear and logistic regression. Model generalizability was evaluated in an independent Mediterranean cohort (n = 70).</div></div><div><h3>Results</h3><div>The machine learning model allowed to obtain 83% of success in siesta patterns prediction. Among the input variables, activity was the most discriminative by the decision tree (threshold: 27 Δ°/min), followed by TAP (0.51 AU) and position (4.7°). In an independent external validation cohort, success in prediction reached 77%, indicating strong alignment between algorithm-based and self-reported siesta patterns detection. Predicted siesta—but not self-reported alone—was significantly associated with obesity-related traits. Later siesta timing was linked to increased waist circumference in women (β = 0.769 cm per hour; P = 0.026). Longer siesta duration was associated with increased obesity risk (OR=2.081; P=0.002), BMI (β=0.013 kg/m²/h; P = 0.034), and systolic blood pressure (β = 3.540mmHg/h; P = 0.049). Greater siesta frequency was associated with lower corrected insulin response (β = −0.037 AU/day; P = 0.012).</div></div><div><h3>Conclusion</h3><div>Objective data from temperature, activity, position, and TAP, combined with ML models, accurately predict siesta behavior and its metabolic relevance. These findings support the use of machine learning approaches based on temperature, activity, position, and the integrated TAP, to assess siesta under free-living conditions.</div></div><div><h3>ClinicalTrials.gov identifier</h3><div>NCT03036592</div></div>","PeriodicalId":48545,"journal":{"name":"Sleep Health","volume":"12 3","pages":"Pages 504-512"},"PeriodicalIF":4.8000,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Sleep Health","FirstCategoryId":"3","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2352721826000185","RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"2026/4/9 0:00:00","PubModel":"Epub","JCR":"Q2","JCRName":"CLINICAL NEUROLOGY","Score":null,"Total":0}

引用次数: 0

Abstract

Objectives

To predict siesta behavior using machine learning models trained on self-reported and objective data—temperature (T), activity (A), position (P), and the integrated TAP variable—and to explore its associations with obesity-related traits.

Methods

From ONTIME-MT, 889 adults wore wrist sensors for 7 days to continuously record temperature, activity, and position, and self-reported daily siesta. Machine learning models were developed to classify 30-second epoch siesta data, to reconstruct weekly siesta behavior. Anthropometric and metabolic parameters were assessed. Associations were analyzed using linear and logistic regression. Model generalizability was evaluated in an independent Mediterranean cohort (n = 70).

Results

The machine learning model allowed to obtain 83% of success in siesta patterns prediction. Among the input variables, activity was the most discriminative by the decision tree (threshold: 27 Δ°/min), followed by TAP (0.51 AU) and position (4.7°). In an independent external validation cohort, success in prediction reached 77%, indicating strong alignment between algorithm-based and self-reported siesta patterns detection. Predicted siesta—but not self-reported alone—was significantly associated with obesity-related traits. Later siesta timing was linked to increased waist circumference in women (β = 0.769 cm per hour; P = 0.026). Longer siesta duration was associated with increased obesity risk (OR=2.081; P=0.002), BMI (β=0.013 kg/m²/h; P = 0.034), and systolic blood pressure (β = 3.540 mmHg/h; P = 0.049). Greater siesta frequency was associated with lower corrected insulin response (β = −0.037 AU/day; P = 0.012).

Conclusion

Objective data from temperature, activity, position, and TAP, combined with ML models, accurately predict siesta behavior and its metabolic relevance. These findings support the use of machine learning approaches based on temperature, activity, position, and the integrated TAP, to assess siesta under free-living conditions.

ClinicalTrials.gov identifier

NCT03036592

查看原文本刊更多论文

基于机器学习的午睡客观预测及其与肥胖的关联。

目的：利用基于自我报告和客观数据(温度(T)、活动（A)、位置(P)和综合TAP变量）训练的机器学习模型来预测午睡行为，并探索其与肥胖相关特征的关联。方法：通过ONTIME-MT， 889名成年人佩戴腕部传感器7天，连续记录体温、活动和体位，以及自我报告的每日午睡情况。开发了机器学习模型来分类30秒的午睡数据，以重建每周的午睡行为。评估人体测量和代谢参数。使用线性和逻辑回归分析相关性。在一个独立的地中海队列（n = 70）中评估模型的通用性。结果：机器学习模型在午睡模式预测中获得了83%的成功率。在输入变量中，决策树最具判别性的是活动（阈值：27 Δ°/min），其次是TAP （0.51 AU）和位置（4.7°）。在一个独立的外部验证队列中，预测成功率达到77%，表明基于算法和自我报告的午睡模式检测之间有很强的一致性。预测的午睡——但不是自我报告的——与肥胖相关的特征显著相关。较晚的午睡时间与女性腰围增加有关（β = 0.769厘米/小时；P = 0.026）。较长的睡眠时间与肥胖风险增加（OR=2.081； P=0.002）、BMI （β=0.013 kg/m²/h； P= 0.034）和收缩压（β= 3.540mmHg/h； P= 0.049）相关。较高的睡眠频率与较低的校正胰岛素反应相关（β = -0.037 AU/天；P = 0.012）。结论：温度、活动、体位、TAP等客观数据结合ML模型可准确预测午睡行为及其代谢相关性。这些发现支持使用基于温度、活动、位置和综合TAP的机器学习方法来评估自由生活条件下的午睡。临床试验：政府标识符：NCT03036592。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Sleep Health CLINICAL NEUROLOGY-

CiteScore

6.30

自引率

9.80%

发文量

114

审稿时长

54 days

期刊介绍： Sleep Health Journal of the National Sleep Foundation is a multidisciplinary journal that explores sleep''s role in population health and elucidates the social science perspective on sleep and health. Aligned with the National Sleep Foundation''s global authoritative, evidence-based voice for sleep health, the journal serves as the foremost publication for manuscripts that advance the sleep health of all members of society.The scope of the journal extends across diverse sleep-related fields, including anthropology, education, health services research, human development, international health, law, mental health, nursing, nutrition, psychology, public health, public policy, fatigue management, transportation, social work, and sociology. The journal welcomes original research articles, review articles, brief reports, special articles, letters to the editor, editorials, and commentaries.