兽研摘要
目的
与宠物或儿童同睡很常见，但其与睡眠、心肺生理学及行为结局的关联研究不足。我们研究了与狗、猫或儿童同睡和睡眠特征、心肺生理学指标及次日体力活动之间的个体内关联。
方法
我们分析了11，733名佩戴WHOOP可穿戴设备的成年人共1，649，083人-天的数据。参与者通过设备配套智能手机应用中的每日日志自我报告夜间同睡情况。采用线性混合效应模型比较同一个体同睡与非同睡夜晚的差异。
结果
同睡与较低的静息心率（0.8–1.1次/分）、较低的呼吸频率（0.04–0.07次/分）和较高的心率变异性（1.41–1.95 ms）相关。睡眠结局结果不一，表现为睡眠时长更长（5.5–9.6分钟），但睡眠中断更多（0.36–0.40次），恢复性睡眠略少（0.23–1.17%）。各亚组间关联总体一致，但与儿童同睡与更明显的睡眠中断相关。与狗和猫同睡后次日体力活动更高（分别为7.6和7.4强度加权分钟），但与儿童同睡后更低（−2.6强度加权分钟）。尽管具有统计学显著性，所有标准化效应量均较小（β范围：0.008–0.045）。
结论
同睡与心肺生理学的微小差异（较低的静息心率、较高的心率变异性）以及轻度增加的睡眠中断相关。这些关联的幅度较小，提示同睡决策可能更多受个人和情境因素影响，而非受对实质性生理或睡眠影响的预期所驱动。
关键词：同床睡眠，宠物，可穿戴设备，心率变异性，自主神经功能
引言
睡眠对整体健康以及身体和认知功能至关重要，1然而西方化社会中相当比例的成年人睡眠不足或睡眠质量低下，约三分之一的人每天睡眠不足7小时，2近~30%报告夜间睡眠困难。3鉴于睡眠不足的高患病率，确定影响睡眠数量和质量的因素十分重要。由于睡眠发生在更广泛的社会和环境背景中，识别潜在影响因素（包括睡眠安排）是一个重要的研究领域。同睡，包括共用一张床和/或近距离睡眠（例如在同一房间内），很常见，30–50%的成年人与宠物、儿童或两者共用一张床。4，5尽管同睡做法普遍，但这些安排的生理学后果仍不完全清楚。
Andre等人最近的综述对同睡研究进行了全面概述。6与动物同睡时，物种特异性睡眠模式可能影响人类睡眠。狗是多相睡眠动物，通常夜间经历多次睡眠-觉醒转换，而猫是晨昏性动物，往往在傍晚和清晨更活跃，而人类在这些时段通常处于睡眠状态。7–9运动、过敏原暴露等额外因素可能进一步导致睡眠中断，10，11尽管宠物可能提供安全感或其他心理益处，从而改善睡眠质量。6，12实证研究反映了这种相互竞争影响的平衡。Hoffman等人报告夜间狗的活动与人类同床伴侣的活动增加相关。13同样，Patel等人发现狗共用床时睡眠质量比狗在房间内更差。14尽管研究猫同睡的研究较少，现有证据提示猫和狗主人的自报睡眠中断可能相似。15与儿童同睡可能特别复杂，因为近距离接触可能提供舒适感并便于夜间照护，同时增加照护者的觉醒和睡眠碎片化。这些影响也可能因性别而异，因为母亲通常承担更大比例的夜间照护责任，因此其睡眠指标可能更受干扰。6
同睡还可能通过睡眠时长或碎片化未能完全捕捉的机制影响心肺生理学。睡眠不佳与静息心率（RHR）升高和心率变异性（HRV）降低相关。16，17相反，感知安全感、依恋、社会联系和压力减轻可能增加副交感神经调节，可能有助于降低RHR和升高HRV。RHR和HRV反映心脏自主神经调节的各个方面，但也受众多生理、行为和情境因素影响。18，19因此，同睡与心肺功能之间的关联不能仅从同睡对睡眠的潜在影响来预测。呼吸频率提供了过夜心肺生理学的额外测量指标，并可能随自主神经状态、睡眠阶段和觉醒而变化。20
尽管人们对同床共眠的兴趣日益增长，但同床共眠与心肺生理学之间的关联受到的关注相对较少。现有文献受限于样本量小、依赖自我报告的睡眠结局，以及采用个体间设计比较习惯性同床共眠者与非同床共眠者，而非考察个体内差异。如果两组在人口统计学、睡眠习惯或其他生活方式因素上存在系统性差异，此类设计存在混杂风险。大规模、被动采集的可穿戴设备数据为在同一个体内重复考察这些关系提供了机会。因此，我们在大量可穿戴设备用户样本中，考察了与狗、猫或儿童同床共眠与夜间睡眠结局、心肺生理学指标以及次日体力活动之间的个体内关联。
材料与方法
研究设计
本研究为回顾性队列分析，使用可穿戴健身追踪器（WHOOP Inc.，马萨诸塞州波士顿）的成年用户（18–79岁）数据，按照流行病学观察性研究报告强化指南（STROBE）队列研究指南进行设计、实施和报告。21
参与者
参与者从完整数据库中随机选取，需符合预先设定的纳入标准，以确保数据完整性、组间人口统计学特征相似以及个体内变异性。符合条件的用户在2024年至少有14天（最多365天）的连续数据，且无睡眠或生物测量指标缺失。此外，参与者在WHOOP配套智能手机应用程序中完成每日日志记录，报告睡眠期间卧室中是否有狗、猫或儿童。为实现个体内比较，参与者需报告陪伴者（狗、猫或儿童）至少3天在场且至少3天不在场。
在确定原始样本后，出现在多个组中的个体（例如报告与狗和儿童同睡）被剔除，以分离每种独立效应。在匹配（后文描述）之前，样本包括61,347名个体（n=44,678名与狗同睡者，n=12,758名与猫同睡者，n=3,911名与儿童同睡者），贡献了9,966,572人日的数据。人口统计学特征（年龄和性别）和人体测量学指标（身高和体重）为自我报告，BMI使用最接近研究开始日期（2024年1月1日）记录的值计算。所有用户均同意将其匿名数据用于研究目的。本研究经Salus IRB审查和批准（方案编号#251121；德克萨斯州奥斯汀），并基于最小风险标准被认定免于持续监管。
可穿戴设备与测量指标
WHOOP手环结合加速度计和光电容积脉搏波描记法，检测持续时间至少一小时的主要睡眠时段。在每个检测到的睡眠时段内，WHOOP将每个30秒间隔分类为四种睡眠阶段之一：清醒、浅睡、慢波睡眠（SWS）和快速眼动（REM）睡眠。在整个睡眠时段内，使用逐搏间期以及频率变换衍生的特征来估算静息心率（RHR）。此外，WHOOP从光电容积脉搏波信号中提取峰值到峰值的心搏间期来估算心率变异性（HRV），计算为相邻间期差值的均方根（RMSSD）。为生成夜间RHR和HRV值，WHOOP首先滤除被识别为清醒或信号质量低的时段。其余时段则使用加权平均值进行聚合，其中慢波睡眠可能性更高的时段和更接近睡眠结束的时段被赋予更高的权重。睡眠时长定义为总睡眠时间（浅睡、SWS、REM）。恢复性睡眠定义为SWS和REM睡眠时间之和。WHOOP已在心率和两阶段睡眠分类方面分别针对心电图（99%一致性）和多导睡眠图（86–89%一致性）进行了验证。22,23
每日体力活动通过WHOOP手环的心率和运动数据自动检测，或由参与者在WHOOP智能手机应用程序上手动记录。最大心率使用非线性公式192–0.007*年龄^2估算，24或手动设定，或定义为高信号质量运动期间记录的最高心率。每次活动期间的连续心率数据用于计算五个区间的时间，区间以最大心率的百分比定义：区间1（50–59%）、区间2（60–69%）、区间3（70–79%）、区间4（80–89%）和区间5（90–100%）。每个区间的分钟数乘以区间特异性权重（即区间1=1，区间2=2，依此类推），然后求和得出复合的心率区间总分（SHRZS），我们将其称为体力活动（即强度加权分钟数）。25 2024年11月，平台过渡为基于心率储备的区间；如后文所述，我们使用过渡前（1月至10月）的数据进行了敏感性分析，发现纳入11月和12月数据未改变研究结局。
自我报告的与狗、猫或儿童同床共眠
共睡行为通过WHOOP智能手机应用程序的每日日志功能自我报告。每天早晨，用户被提示回答以下关于前一夜的是/否问题：“睡觉时房间里有狗吗？”、“睡觉时房间里有猫吗？”以及“卧室里有孩子和你一起睡吗？”。用户也可以跳过任何问题。仅明确回答“是”或“否”的天数被纳入主要分析。日志数据用于识别每组中有共睡暴露和无共睡暴露的夜晚。我们注意到，从日志条目功能中无法得知共睡同伴所在位置（例如，在床上还是婴儿床中）、共睡者数量（例如，无法区分是一只还是多只狗、猫或孩子）、孩子年龄、狗/猫的品种或体型、床的尺寸等细节。
统计分析
分析使用了报告与狗、猫或孩子共睡的参与者的日水平生理和行为数据。缺失所需睡眠、生物特征或日志数据的夜晚被排除在主要分析之外。
为提高共睡组间的可比性，参与者在年龄、性别、BMI以及个体水平平均RHR、HRV、睡眠时长和体力活动上进行匹配。使用这些结局的个体水平均值，使队列在习惯性水平上具有可比性，同时保留作为主要关注参数的每夜个体内偏差。由于儿童组最小，狗和猫参与者分别以1:1无放回方式使用最近邻马氏距离（无卡钳）与儿童参与者匹配。26 匹配后每组保留3911名参与者；未匹配的40,767名狗参与者和8,847名猫参与者被排除。使用标准化均数差评估平衡性（S1表）。
主要结局为夜间RHR和HRV、呼吸频率（RR）、总睡眠时长、觉醒次数、以每小时睡眠觉醒次数表示的觉醒指数、恢复性睡眠百分比（通过SWS + REM求和得出的指标）以及次日体力活动。共睡暴露被分解为个体均值项和每夜偏差项，以区分个体间关联与个体内关联。由于主要目的是比较同一个体内共睡夜与独睡夜，每夜偏差效应是主要关注参数。个体间结果报告于S2-S9表和补充文本中。
线性混合效应模型分别对狗、猫和儿童队列进行拟合。模型包括个体内和个体间共睡暴露、年龄、性别、BMI、季节、工作日与周末以及参与者特异性随机截距。评估了个体内共睡的随机斜率，但因模型不收敛而未保留。由于在考虑参与者特异性随机截距后，连续夜晚的残差可能仍存在相关性，所有24个主要模型还额外拟合了一阶自回归AR(1)残差相关结构（除随机截距外），自回归项以自每位参与者首次观察夜晚起经过的天数索引，以便将设备佩戴间隙视为时间间隙而非相邻夜晚。两种额外设定评估了稳健性：模型2额外调整了就寝时间和睡眠时间的每夜偏差，模型3进一步调整了前一日体力活动和前一夜睡眠时长。通过在每个模型设定中加入个体内共睡暴露与性别的交互项来评估性别差异。
使用方差膨胀因子以及残差和随机效应诊断对所有24个主要模型的模型假设进行了评估。HRV和体力活动模型显示出明显的非正态性偏离，因此使用基于秩的逆正态变换重复分析。结果的方向和统计学显著性未改变（S10表）。
为考虑多重检验，使用Benjamini-Hochberg错误发现率（FDR）程序调整p值，家族按共睡组和效应类型（个体内、个体间或性别交互）在每个模型设定内的八个结局中定义。正文中报告的p值为未调整值；FDR调整值以及标识调整后不再显著的估计值的标记提供于S3-S9表中。使用标准化β系数表征效应量，0.10、0.30和0.50分别解释为小、中和大效应。27
使用带有参与者特异性随机截距的混合效应逻辑回归检查了共睡日志应答缺失可能导致的偏倚。缺失情况作为年龄、性别、BMI、个体水平平均RHR和睡眠时长以及工作日与周末的函数进行建模（S11表）。使用两种敏感性分析评估了对缺失应答的稳健性：（1）将缺失应答编码为无共睡，同时将缺失应答指示变量作为协变量纳入，28 以及（2）链式方程多重插补（MICE；m = 5），使用Rubin规则合并估计值。29
由于确定心率区间（用于身体活动结局）的方法在11月发生了变化，我们同时使用2024年全部12个月的数据以及仅使用2024年心率区间变化前的10个月数据运行了模型。结果未发生变化；因此，主要分析使用了完整的12个月数据集。
所有分析均在R（版本4.4.2）中进行，统计显著性设定为α = 0.05。模型估计值以点估计值及相应的95%置信区间报告。结果呈现以个体内关联为主要重点，这直接针对研究目的，个体间关联则在补充材料中呈现。
结果
参与者特征
最终匹配数据集每组包含3911名唯一用户，总样本为11,733人，贡献1,649,083人日（犬：458,051共睡日[纳入分析的夜晚中有67.6%与犬同睡卧室]；猫：389,075共睡日[纳入夜晚的63.1%]；儿童：162,209共睡日[纳入夜晚的45.7%]）。
人口学特征见表1。各组间参与者平均年龄差异在两年以内。匹配还使BMI（差异在0.5 kg/m2以内）、RHR（差异在0.3次/分以内）和HRV（差异在2.4 ms以内）相似。有儿童组的睡眠时间约低8分钟，其活动评分也较低（约16-20强度加权分钟）。除SHRZS外，所有变量的组间标准化均数差（SMD）均<0.20，SHRZS在匹配后仍存在有意义的残余不平衡（犬 vs 儿童 SMD = 0.367；猫 vs 儿童 SMD = 0.300；S1表）。各组平均睡眠时长约为7.0小时（犬：420分钟，猫：420分钟，儿童：412分钟），接近成人推荐睡眠时长的低端。30 所有三组的平均静息心率约为61次/分，处于正常窦性范围内。31
Table 1. Demographics of Sample
| Dog Group (n=3,911) | Cat Group (n=3,911) | Child Group (n=3,911) | |
|---|---|---|---|
| Age (years) | 37.5 (10.3) | 35.8 (10.3) | 36.6 (6.5) |
| Body mass index (kg/m 2 ) | 26.6 (4.4) | 26.4 (4.7) | 26.8 (4.8) |
| RHR (beats/min) | 61.0 (8.0) | 61.2 (8.1) | 61.3 (8.2) |
| HRV (ms) | 52.0 (24.3) | 53.2 (24.9) | 50.8 (22.8) |
| Sleep time (min) | 420.0 (38.8) | 420.1 (39.2) | 412.0 (42.5) |
| Daily physical activity (SHRZS) | 77.1 (58.7) | 73.2 (57.5) | 56.8 (51.6) |
| % female | 35.8% | 35.8% | 35.8% |
共睡与较低的RHR和呼吸频率、较高的HRV以及更紊乱的睡眠相关
在个体内基础模型（图1-3和S2表）中，共睡与所有八项结局的差异相关（均p<0.001），尽管如下文所述，标准化效应量普遍非常小。在共睡夜晚，所有三组的RHR（次/分）均较低（犬：−1.14 [95% CI：−1.17, −1.11]，猫：−0.94 [95% CI：−0.97, −0.91]，儿童：−0.80 [95% CI：−0.84, −0.76]），而HRV（ms）较高（犬：1.95 [95% CI：1.86, 2.04]，猫：1.41 [95% CI：1.32, 1.50]，儿童：1.83 [95% CI：1.73, 1.93]）。RR（次/分）也较低（犬：−0.072 [95% CI：−0.076, −0.068]，猫：−0.066 [95% CI：−0.069, −0.062]，儿童：−0.044 [95% CI：−0.049, −0.040]），且在共睡夜晚所有三组的总睡眠时长（分钟）均较高（犬：9.58 [95% CI：9.11, 10.04]，猫：7.33 [95% CI：6.85, 7.82]，儿童：5.49 [95% CI：4.91, 6.08]）。在共睡夜晚，所有三组的觉醒次数（#）均较高（犬：0.38 [95% CI：0.36, 0.41]，猫：0.36 [95% CI：0.34, 0.39]，儿童：0.40 [95% CI：0.37, 0.44]），归一化觉醒次数（每小时睡眠）也较高（犬：0.019 [95% CI：0.015, 0.022]，猫：0.023 [95% CI：0.019, 0.026]，儿童：0.038 [95% CI：0.033, 0.042]）。在共睡夜晚，所有三组的恢复性睡眠百分比（%）均较低（犬：−0.35 [95% CI：−0.40, −0.29]，猫：−0.23 [95% CI：−0.28, −0.17]，儿童：−1.17 [95% CI：−1.24, −1.11]）。最后，在犬组和猫组中，共睡后的次日强度加权身体活动较高（犬：7.56 [95% CI：6.84, 8.28]，猫：7.35 [95% CI：6.62, 8.09]），但在儿童组中，共睡夜晚后次日身体活动较低（儿童：−2.57 [95% CI：−3.38, −1.77]）。
[图：与犬共睡 vs 不与犬共睡的个体内比较。否：未与犬共睡的夜晚。是：夜晚]
[图：与猫共睡 vs 不与猫共睡的个体内比较。否：未与猫共睡的夜晚。是：夜晚]
[图：与儿童共睡 vs 不与儿童共睡的个体内比较。否：未与儿童共睡的夜晚。是：夜]
对就寝时间（模型2；S3表）和行为协变量（模型3；S4表）的调整减弱了若干关联的幅度，但未改变其方向性或显著性，但有一个例外：儿童共睡与睡眠时长之间的个体内关联在完全调整后变得不显著（+0.4 ± 0.3分钟，p=0.106）。
共睡关联因性别而异
同睡与若干结局之间的关联因性别而异（S5表）。在犬组中，八个结局中有七个的交互作用显著（除干扰外，交互作用p=0.473）：与雌性相比，雄性在同睡夜中RHR和RR的下降幅度更大、HRV和总睡眠时间的增加幅度更大、标准化干扰的增加幅度更小、恢复性睡眠的下降幅度更不明显、体力活动的增加幅度更小（所有交互作用p≤0.022）。在猫组中，四个结局的交互作用显著（RR、干扰、标准化干扰和恢复性睡眠；所有交互作用p≤0.004），雄性RR下降幅度更大、干扰和标准化干扰的增加幅度更小、恢复性睡眠的下降幅度更不明显。在儿童组中，八个结局中有五个的交互作用显著（RHR、HRV、总睡眠时间、干扰和恢复性睡眠；所有交互作用p<0.001）：雄性RHR下降幅度更小、HRV增加幅度更小、总睡眠时间和干扰的增加幅度更大、恢复性睡眠的下降幅度更小，而RR、标准化干扰或体力活动未见性别差异（所有交互作用p≥0.200）。这些基础模型中描述为显著的所有交互作用在FDR校正后仍保持显著（S5表）。
对就寝时间（S6表）以及就寝时间加行为协变量（S7表）的校正减弱了若干交互作用，但未改变大多数结局的方向。最显著的例外是儿童同睡与总睡眠时间，在同睡夜中，雌性较低而雄性较高，这在就寝时间校正模型（雌性−2.64 vs 雄性+2.12分钟；交互作用p<0.001）和完全校正模型（雌性−3.00 vs 雄性+2.09分钟；交互作用p<0.001）中均如此。
研究结果对敏感性检验稳健
进行了两项敏感性分析，以评估主要个体内研究结果对缺失同睡应答之夜处理方式的稳健性（犬：约31%，猫：约34%，儿童：约55%）。敏感性分析A（S8表）在所有组和结局中比较了主要完整病例方法与将无效应答视为未同睡（null-as-no）的方法，且模型在所有组和结局中还将无效应答指示变量作为直接协变量与主要模型项一并纳入。所有24个个体内结局的关联方向和统计学显著性均完全保留，支持主要研究结果的稳健性。敏感性分析B（S9表）在所有组和结局中应用链式方程多重插补（MICE；m=5次插补），发现合并后所有24个个体内关联仍具有统计学显著性。日记缺失与参与者特征存在适度但系统性的关联（S11表）。在犬和猫队列中，较短的个体平均睡眠时长（每SD的比值比分别为0.76和0.79）、较高的个体平均RHR（1.17和1.21）和较年轻的年龄（0.82和0.77）与更高的缺失应答几率相关。在儿童队列中，这些关联较弱（睡眠时长0.89；RHR 0.94；年龄0.99），且男性参与者出现缺失应答的可能性较低（0.83）。三个队列中BMI均与缺失无关。以一阶自回归残差结构重新拟合所有24个主要模型，得到的估计残差自相关在各结局中为0.01–0.42，但每个个体内和个体间关联的方向、统计学显著性和幅度均未改变；所有24个个体内关联在FDR校正p<0.001时仍具有统计学显著性，标准误增加不超过13%，标准化β范围未改变（0.008–0.045）。
效应量非常小
为评估观察到的个体内关联的实际幅度，计算了个体内同睡关联的效应量（标准化β绝对值的范围：0.008–0.045；S1图）。为在统计学显著性之外评估实际效应幅度，所有个体内和个体间同睡关联的标准化β值及95%置信区间以森林图形式呈现于S1图。最大效应见于犬同睡对静息心率（β = −0.045）和睡眠时长（β = 0.043）的影响，以及儿童同睡对恢复性睡眠百分比（β = −0.045）的影响，其他结局和同睡条件下的效应均非常小。值得注意的是，所有效应量均低于小效应的常规β = 0.1阈值。
讨论
我们使用可穿戴设备评估了与狗、猫或儿童同睡对睡眠、心肺生理学以及次日体力活动的个体内关联。与独自睡眠相比，同睡相关的差异非常小，包括较低的静息心率和呼吸频率、较高的心率变异性和总睡眠时间、更多的睡眠 disturbances 以及较低的恢复性睡眠百分比。这些发现在各同睡组之间基本相似。对于一般健康成年人中可穿戴设备衍生的静息心率和心率变异性夜间变化的既定最小临床重要差异尚不可用。然而，本研究中观察到的绝对差异很小（静息心率：约0.8-1.1次/分；心率变异性：约1.4-2.0毫秒），标准化效应量一致地非常小（β = 0.008–0.045）。因此，尽管具有统计学显著性，这些关联不应被解释为同睡夜与独自睡眠夜之间存在临床意义生理差异的证据。此外，由于期刊条目仅记录狗、猫或儿童是否在卧室中，这些估计值平均了同床睡眠和同室睡眠的夜晚，可能低估了特别是同床睡眠相关的对比。
与宠物同睡相比，与儿童同睡显示出更具破坏性的模式。与宠物同睡相比，与儿童同睡与睡眠时间增加较小、睡眠 disturbances 增加较大以及恢复性睡眠百分比更大幅降低相关。儿童组的个体间关联与这些趋势一致，并在协变量调整后持续存在。先前的研究表明，与儿童同睡通常受照护需求、发育过渡或家庭情况的影响，而非仅由偏好决定，32 并与夜间觉醒增加、照护者参与更多以及较差的睡眠指标相关，其中许多与本研究的发现一致。33,34
与宠物同睡后次日体力活动较高（狗和猫的幅度相似），但与儿童同睡后的日子体力活动较低。过去的研究表明，养狗与较高的活动水平相关，35,36 而很少有研究直接考察养猫与体力活动的关系。与儿童同睡后体力活动较低可能反映疲劳或可用于活动的自由支配时间受限。37,38 例如，如果照护者与儿童同时醒来，与儿童同睡可能限制晨间锻炼的机会。我们推测，生理状态和日常时间限制的差异可能导致了宠物与儿童同睡后观察到的不同活动模式。
同睡关联因性别而异。男性通常在心肺测量中表现出较大的差异，而在睡眠 disturbances 中表现出较小的差异，尤其是在狗组中。最显著的是，与儿童同睡与男性睡眠时间较高但女性睡眠时间较低相关，这可能反映照护角色的差异。先前的研究发现，母亲是幼儿的主要夜间照护者。39 然而，与总体分析一样，性别分层效应非常小。
我们的研究具有若干重要优势，包括大样本、减少个体间差异混杂的个体内设计、仔细的匹配程序使三个同睡组在人口学和行为因素上相似，以及使用可穿戴设备对研究结果进行客观测量。
应考虑若干局限性。第一，尽管匹配在大多数特征上产生了可比组，但习惯性体力活动仍存在有意义的残余不平衡（狗vs儿童SMD=0.367；猫vs儿童SMD=0.300），儿童组比狗组和猫组活动更少，睡眠时长也存在较小的残余差异；然而，由于这种不平衡反映的是组间习惯性活动差异，它影响的是人与人之间的比较，而非本研究主要关注的人内关联。第二，我们缺乏若干潜在混杂因素的信息，包括社会经济状况、慢性疾病、用药、吸烟、酒精和咖啡因使用、睡眠障碍以及伴侣同睡。尽管人内设计控制了稳定的个体特征，但与同睡和研究结局均相关的未测量时变因素可能导致残余混杂。此外，尽管参与者特异性随机截距解释了个体内的重复观察，连续夜晚之间的残余相关仍然存在，其中心肺结局尤为明显；以一阶自回归残差结构重新拟合所有主要模型后，所有人与人内和人与人之间的关联在方向、统计显著性和效应量上均保持不变。其次，同睡为自我报告，相当一部分夜晚的日志回复缺失（狗约31%，猫约34%，儿童约55%），尽管敏感性分析证实人内发现在所有结局中均稳健；但在分析中仍不能完全排除数据非随机缺失的可能性。要求参与者同时提供同睡和独睡夜晚也可能选择出睡眠安排多变的个体，而排除与多种同伴类型同睡的个体可能限制对常见现实睡眠安排的代表性。另一个值得注意的局限性，且对解释有重大影响，是WHOOP日志不区分同床与同室（例如，宠物在地板上，或儿童在摇篮中），也未捕捉可能重要的因素，如床的大小、同睡者数量、宠物体型/品种或儿童年龄。因此，相对于同床而言，暴露错分可能很大，报告的关联应解释为两种安排的平均值。此外，尽管WHOOP提供客观结局评估，并已证明在心率测量方面具有良好的效度，23,40但可穿戴设备衍生的睡眠分期不如多导睡眠图准确，应据此解释。22,41,42最后，WHOOP用户往往比一般成年人群更年轻、更活跃，这可能限制对其他人群和可穿戴设备的推广性。
结论
这项针对保留日志记录表明同睡的WHOOP用户的研究发现，与独睡夜晚相比，与宠物或儿童同睡与心肺生理和睡眠方面非常小但一致的差异相关。虽然无法确定因果关系，但这些发现可能反映同睡夜晚HRV、RHR和RR的小幅同步差异以及睡眠碎片化增加。不同同伴类型的模式大致相似，但儿童同睡显示出更明显的睡眠中断，尤其是女性。尽管大样本量导致同睡夜晚与独睡夜晚之间有许多统计学显著差异，但效应量非常小，不太可能具有临床意义，提示同睡决定可能更多受个人因素（如舒适感、照护需求以及文化或家庭习惯）指导，而非对睡眠或心肺功能产生重大影响的担忧。
致谢
作者感谢WHOOP数据科学和数据工程团队在特征开发和数据基础设施方面的技术协助。我们还感谢更广泛的WHOOP员工群体对公司研究计划的持续支持。
资助声明
本工作由WHOOP, Inc.支持，通过向作者GJG和DJC提供薪资支持。本工作无特定资助编号。WHOOP, Inc.提供了总体机构支持，但在研究设计、数据收集和分析、发表决定或手稿准备中无任何角色。
AI使用声明
在本工作准备期间，作者使用Claude（Anthropic）进行语言编辑和代码审查。使用该工具后，作者根据需要审查和编辑了内容，并对发表文章的内容承担全部责任。
数据共享声明
支持本研究结果的数据和代码由于WHOOP, Inc.的知识产权考虑而不公开可用。去标识化的参与者数据和数据字典可在向WHOOP（通过science@whoop.com）提出合理请求时提供。获取需要提交方法学上合理的提案，经WHOOP研究团队批准，并签署数据使用协议。请求回复的时间范围为4周。
伦理批准和知情同意
所有用户均同意将其匿名化数据用于研究目的。本研究经Salus IRB（方案编号#251121；德克萨斯州奥斯汀）审查并批准，基于最小风险标准被认定免于持续监管。本研究按照《赫尔辛基宣言》的原则开展。
作者贡献
• AHKM：研究构思、方法学、数据分析与解释、可视化、撰写——初稿，以及撰写——审阅与编辑。
• DJC：软件、数据分析与解释、数据管理、可视化、撰写——初稿，以及撰写——审阅与编辑。
• GJG：研究构思、方法学、软件、数据管理、调查、资源、形式分析、项目管理、撰写——初稿，以及撰写——审阅与编辑。所有作者均对拟发表版本给予最终批准，同意文章所投稿的期刊，并对本工作的各个方面承担责任与问责。
利益披露
本研究由WHOOP, Inc.资助。GJG和DJC为WHOOP, Inc.的员工，GJG作为员工薪酬的一部分持有股票期权，这代表潜在的经济利益。所有分析均由研究团队独立完成。WHOOP, Inc.的营销、企业及其他非研究部门在研究批准、数据解释、稿件准备或投稿/发表决策中均未发挥作用。作者报告本工作中无其他利益冲突。本稿件的一个版本已作为预印本发布：https://www.biorxiv.org/content/10.64898/2026.06.13.732027v1。
参考文献
[内容请查看原文]
关联数据
数据可用性声明
支持本研究结果的数据和代码由于WHOOP, Inc.的知识产权考量未公开提供。去标识化的参与者数据及数据字典可在向WHOOP（通过science@whoop.com）提出合理请求后予以提供。获取权限需提交方法学上合理的提案，经WHOOP研究团队批准，并签署数据使用协议。请求回复的时间范围为4周。
查看英文原文▾
Abstract
Purpose
Co-sleeping with pets or children is common, yet its associations with sleep, cardiorespiratory physiology, and behavioral outcomes are understudied. We examined within-person associations between co-sleeping with a dog, cat, or child and sleep characteristics, measures of cardiorespiratory physiology, and next-day physical activity.
Methods
We analyzed 1,649,083 person-days from 11,733 adults wearing the WHOOP wearable device. Participants self-reported nightly co-sleeping via a daily journal in the device’s companion smartphone application. Linear mixed-effects models compared nights with and without co-sleeping within the same individual.
Results
Co-sleeping was associated with lower resting heart rate (0.8–1.1 beats/min), lower respiratory rate (0.04–0.07 breaths/min), and higher heart rate variability (1.41–1.95 ms). Sleep outcomes were mixed, with longer sleep duration (5.5–9.6 min) but more disturbances (0.36–0.40 instances) and slightly less restorative sleep (0.23–1.17%). Associations were generally consistent across groups, although child co-sleeping was associated with greater sleep disruption. Next-day physical activity was higher following dog and cat co-sleeping (7.6 and 7.4 intensity-weighted min, respectively) but lower following child co-sleeping (−2.6 intensity-weighted min). Despite statistical significance, all standardized effect sizes were small (β range: 0.008–0.045).
Conclusion
Co-sleeping was associated with small differences in cardiorespiratory physiology (lower resting heart rate, higher heart rate variability) alongside mildly greater sleep disruption. The small magnitude of these associations suggests that co-sleeping decisions may be influenced more by personal and contextual factors than by expectations of substantive physiologic or sleep-related impacts.
Keywords: bedsharing, pet, wearable device, heart rate variability, autonomic function
Introduction
Sleep is essential for overall health and both physical and cognitive function,1 yet a significant proportion of adults in Westernized societies have insufficient or low-quality sleep, with roughly one-third sleeping less than 7 hours per day2 and nearly ~30% reporting nighttime sleeping difficulties.3 With the high prevalence of inadequate sleep, it is important to determine factors that influence sleep quantity and quality. Because sleep occurs within a broader social and environmental context, identifying potentially influential factors, including sleeping arrangements, is an important area of inquiry. Co-sleeping, including the sharing of a bed and/or sleeping in close proximity (eg, within the same room) is common, with 30–50% of adults sharing a bed with a pet, child, or both.4,5 Despite the common practice of co-sleeping, the physiologic consequences of these arrangements remain incompletely understood.
A recent review by Andre et al provides a comprehensive overview of co-sleeping research.6 When co-sleeping with animals, species-specific sleep patterns may influence human sleep. Dogs are polyphasic and typically experience several sleep-wake transitions during the night, whereas cats are crepuscular and tend to be more active in the evening and early morning, when humans are more typically asleep.7–9 Additional factors such as movement and allergen exposure may further contribute to sleep disruption,10,11 though pets may provide security or other psychological benefits that could improve sleep quality.6,12 Empirical studies reflect this balance of competing influences. Hoffman et al reported that nighttime dog movements were associated with increased movement in human bed partners.13 Similarly, Patel et al found poorer sleep quality when dogs shared the bed compared to being present in the room.14 Although fewer studies have examined cat co-sleeping, available evidence suggests that self-reported sleep disruption may be similar for cat and dog owners.15 Co-sleeping with a child may be particularly complex because proximity may provide comfort and facilitate nighttime caregiving while simultaneously increasing caregiver awakenings and sleep fragmentation. These effects may also differ by sex, as mothers often assume a greater proportion of nighttime caregiving responsibilities and may therefore have more disrupted sleep measures.6
Co-sleeping may also influence cardiorespiratory physiology through mechanisms not fully captured by sleep duration or fragmentation. Poor sleep is associated with elevated resting heart rate (RHR) and lower heart rate variability (HRV).16,17 Conversely, perceived safety, attachment, social connection, and reduced stress may increase parasympathetic modulation, potentially contributing to lower RHR and higher HRV. RHR and HRV reflect aspects of cardiac autonomic regulation but are also influenced by numerous physiologic, behavioral, and contextual factors.18,19 Thus, associations between co-sleeping and cardiorespiratory function are not necessarily predictable from co-sleeping’s potential effects on sleep alone. Respiratory rate provides an additional measure of overnight cardiorespiratory physiology and may vary with autonomic state, sleep stage, and arousal.20
Despite growing interest in co-sleeping, associations between co-sleeping and cardiorespiratory physiology have received comparatively little attention. Existing literature is limited by small sample sizes, reliance on self-reported sleep outcomes, and between-person designs that compare habitual co-sleepers and non-co-sleepers rather than examining within-person differences. Such designs risk confounding if the two groups differ systematically in demographics, sleep habits, or other lifestyle factors. Large-scale, passively collected wearable data provide an opportunity to examine these relationships repeatedly within the same individual. Accordingly, we examined within-person associations between co-sleeping with a dog, cat, or child and nightly sleep outcomes, measures of cardiorespiratory physiology, and next-day physical activity in a large sample of wearable device users.
Materials and Methods
Study Design
This study was a retrospective cohort analysis using data from adult users (18–79 yrs) of a wearable fitness tracker (WHOOP Inc., Boston, MA), designed, conducted, and reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for cohort studies.21
Participants
Participants were randomly selected from the full database if they met predefined inclusion criteria designed to ensure both data completeness, similar between-group demographics, and within-person variability. Eligible users had at least 14 days (and up to 365) of continuous data in 2024, with no missing sleep or biometric measures. Additionally, participants completed daily journal entries in the WHOOP companion smartphone application about the presence of a dog, cat, or child in the bedroom during sleep. To enable within-person comparisons, participants were required to report the companion (dog, cat, or child) as present on at least three days and absent on at least three days.
After the original sample was identified, individuals who appeared in multiple groups (eg, reported sleeping with a dog and a child) were removed to isolate each individual effect. Prior to matching (described later) the sample included 61,347 individuals (n=44,678 dog co-sleepers, n=12,758 cat co-sleepers, n=3,911 child co-sleepers) contributing 9,966,572 person-days of data. Demographics (age and sex) and anthropometrics (height and weight) were self-reported, and BMI was calculated using the values recorded nearest to the study start date (January 1, 2024). All users consented for their anonymized data to be used for research purposes. The study was reviewed and approved by Salus IRB (Protocol #251121; Austin, TX) and deemed exempt from ongoing oversight based on minimal risk criteria.
Wearable Device and Measures
The WHOOP strap combines accelerometry and photoplethysmography to detect major sleep periods of at least one hour in duration. Within each detected sleep period, WHOOP classified each 30-second interval into one of four sleep stages: wake, light, slow wave sleep (SWS), and rapid eye movement (REM) sleep. Throughout each sleep period, beat-to-beat intervals, along with features derived from frequency transformations, were used to estimate RHR. Additionally, the WHOOP derived peak-to-peak heartbeat intervals from the photoplethysmography signal to estimate HRV, calculated as the root mean square of successive differences (RMSSD). To generate nightly RHR and HRV values, WHOOP first filtered out epochs identified as wake or as having low signal quality. The remaining epochs were then aggregated using a weighted average, in which epochs with higher likelihood of SWS and epochs closer to the end of the sleep were assigned higher weights. Sleep duration was defined as total sleep time (light, SWS, REM). Restorative sleep was defined as the sum of time spent in SWS and REM sleep. WHOOP has been validated against electrocardiography and polysomnography for heart rate (99% agreement) and 2-stage sleep categorization (86–89% agreement).22,23
Daily physical activity was detected automatically using heart rate and motion data from the WHOOP strap or logged manually on the WHOOP smartphone application by the participant. Maximum heart rate was estimated using the non-linear formula 192–0.007*age^2,24 manually set, or defined as the highest recorded heart rate during an exercise period with high signal quality. Continuous heart rate data during each activity were used to calculate time spent in five zones defined as percentages of heart rate maximum: zone 1 (50–59%), zone 2 (60–69%), zone 3 (70–79%), zone 4 (80–89%), and zone 5 (90–100%). Minutes in each zone were multiplied by zone-specific weights (ie, zone 1=1, zone 2=2, etc), then summed to yield a composite, summated heart rate zone score (SHRZS), which we refer to as physical activity (ie, intensity-weighted min).25 In November 2024, the platform transitioned to heart rate reserve-based zones; as discussed later, we performed a sensitivity analysis using data from prior to this transition (January-October) and found that inclusion of November and December data did not change study outcomes.
Self-Reported Co-Sleeping with Dog, Cat, or Child
Co-sleep behavior was self-reported via the WHOOP smartphone application’s daily journal feature. Each morning, users were prompted to answer the following yes/no questions regarding the prior night: “Had a dog in the room while sleeping?”, “Had a cat in the room while sleeping?”, and “Slept with a child in your bedroom?”. Users could also skip any question. Only days with explicit “yes” or “no” responses were included in the primary analysis. Journal data were used to identify nights with and without co-sleeping exposure for each group. We note that, from the journal entry feature, it is not possible to know details such as where the co-sleeping companion was located (eg, in the bed vs in a crib), number of co-sleepers (eg, cannot specify if it is one or multiple dogs, cats, or children), age of children, breed or size of dog/cat, bed size, etc.
Statistical Analysis
Analyses used day-level physiological and behavioral data from participants reporting co-sleeping with a dog, cat, or child. Nights missing required sleep, biometric, or journal data were excluded from the primary analysis.
To improve comparability across co-sleeping groups, participants were matched on age, sex, BMI, and person-level mean RHR, HRV, sleep duration, and physical activity. Person-level means of these outcomes were used so that cohorts were comparable in habitual levels while the nightly within-person deviations that were the primary parameters of interest were preserved. Because the child group was smallest, dog and cat participants were separately matched 1:1 without replacement to child participants using nearest-neighbor Mahalanobis distance without a caliper.26 Matching retained 3911 participants in each group; the 40,767 dog and 8847 cat participants without a match were excluded. Balance was assessed using standardized mean differences (S1 Table).
Primary outcomes were nocturnal RHR and HRV, respiratory rate (RR), total sleep duration, number of disturbances, a disturbance index expressed as disturbances per hour of sleep, restorative sleep percentage (a measure derived by summing SWS + REM), and next-day physical activity. Co-sleeping exposure was decomposed into person-mean and nightly deviation terms to distinguish between-person from within-person associations. Because the primary aim was to compare co-sleeping and solitary nights within the same individual, nightly deviation effects were the primary parameters of interest. Between-person results are reported in S2-S9 Tables and Supplemental Text.
Linear mixed-effects models were fit separately for the dog, cat, and child cohorts. Models included within- and between-person co-sleeping exposure, age, sex, BMI, season, weekday vs weekend, and participant-specific random intercepts. Random slopes for within-person co-sleeping were evaluated but not retained because of model nonconvergence. Because residuals from consecutive nights may remain correlated after accounting for participant-specific random intercepts, all 24 primary models were also refit with a first-order autoregressive, AR(1), residual correlation structure in addition to the random intercept, with the autoregressive term indexed by days elapsed since each participant’s first observed night so that gaps in device wear were treated as time gaps rather than as adjacent nights. Two additional specifications assessed robustness: Model 2 additionally adjusted for bedtime and nightly deviations in sleep timing, and Model 3 further adjusted for prior-day physical activity and prior-night sleep duration. Sex differences were evaluated by adding an interaction between within-person co-sleeping exposure and sex to each model specification.
Model assumptions were evaluated for all 24 primary models using variance inflation factors and residual and random-effect diagnostics. HRV and physical activity models showed notable departures from normality and were therefore repeated using rank-based inverse normal transformations. Results were unchanged in direction and statistical significance (S10 Table).
To account for multiple testing, p-values were adjusted using the Benjamini-Hochberg false discovery rate (FDR) procedure, with families defined by co-sleeping group and effect type (within-person, between-person, or sex interaction) across the eight outcomes within each model specification. P-values reported in the main text are unadjusted; FDR-adjusted values, and flags identifying estimates that did not remain significant after adjustment, are provided in S3–S9 Tables. Standardized β coefficients were used to characterize effect sizes, with 0.10, 0.30, and 0.50 interpreted as small, medium, and large, respectively.27
Potential bias from missing co-sleeping journal responses was examined using mixed-effects logistic regression with participant-specific random intercepts. Missingness was modeled as a function of age, sex, BMI, person-level mean RHR and sleep duration, and weekday vs weekend (S11 Table). Robustness to missing responses was evaluated using two sensitivity analyses: (1) coding missing responses as no co-sleeping while including a missing-response indicator as a covariate,28 and (2) multiple imputation by chained equations (MICE; m = 5), with estimates pooled using Rubin’s rules.29
Because the method of determining heart rate zones (for the physical activity outcome) changed in November, we ran our models both with all 12 months of 2024 data and with only the 10 months from 2024 leading up to the heart rate zone change. Findings were unchanged; therefore, the full 12-month dataset was used for the primary analyses.
All analyses were conducted in R (version 4.4.2), with statistical significance set at α = 0.05. Model estimates are reported as point estimates with corresponding 95% confidence intervals. Results are presented with primary emphasis on within-person associations, which directly address the study aim, and between-person associations are presented in the Supplementary Material.
Results
Participant Characteristics
The final matched dataset consisted of 3911 unique users per group, for a total sample of 11,733 individuals contributing 1,649,083 person-days (dog: 458,051 co-sleeper days [slept with dog in bedroom on 67.6% of nights included in analysis]; cat: 389,075 co-sleeper days [63.1% of included nights]; child: 162,209 co-sleeper days [45.7% of included nights]).
Demographic characteristics are presented in Table 1. Average participant ages were within two years across groups. Matching also resulted in similar BMI (within 0.5 kg/m2), RHR (within 0.3 beats/min), and HRV (within 2.4 ms). Sleep time was ~8 minutes lower in the group with a child, who also had a lower activity score (~16-20 intensity-weighted minutes). Standardized mean differences (SMDs) between groups were <0.20 for all variables except SHRZS, for which meaningful residual imbalance remained after matching (dog vs child SMD = 0.367; cat vs child SMD = 0.300; S1 Table). Mean sleep duration was approximately 7.0 hours across groups (dog: 420 min, cat: 420 min, child: 412 min), near the low end of recommended adult sleep duration.30 Mean resting heart rate was approximately 61 beats per minute across all three groups, within the normal sinus range.31
Table 1. Demographics of Sample
| Dog Group (n=3,911) | Cat Group (n=3,911) | Child Group (n=3,911) | |
|---|---|---|---|
| Age (years) | 37.5 (10.3) | 35.8 (10.3) | 36.6 (6.5) |
| Body mass index (kg/m 2 ) | 26.6 (4.4) | 26.4 (4.7) | 26.8 (4.8) |
| RHR (beats/min) | 61.0 (8.0) | 61.2 (8.1) | 61.3 (8.2) |
| HRV (ms) | 52.0 (24.3) | 53.2 (24.9) | 50.8 (22.8) |
| Sleep time (min) | 420.0 (38.8) | 420.1 (39.2) | 412.0 (42.5) |
| Daily physical activity (SHRZS) | 77.1 (58.7) | 73.2 (57.5) | 56.8 (51.6) |
| % female | 35.8% | 35.8% | 35.8% |
Co-Sleeping was Associated with Lower RHR and Respiratory Rate, Higher HRV, and More Disrupted Sleep
In the within-person base models (Figures 1–3 and S2 Table), co-sleeping was associated with differences across all eight outcomes (all p<0.001), although standardized effect sizes were uniformly very small, as described below. On co-sleeping nights, RHR (bpm) was lower in all three groups (dog: −1.14 [95% CI: −1.17, −1.11], cat: −0.94 [95% CI: −0.97, −0.91], child: −0.80 [95% CI: −0.84, −0.76]), whereas HRV (ms) was higher (dog: 1.95 [95% CI: 1.86, 2.04], cat: 1.41 [95% CI: 1.32, 1.50], child: 1.83 [95% CI: 1.73, 1.93]). RR (breaths/min) was also lower (dog: −0.072 [95% CI: −0.076, −0.068], cat: −0.066 [95% CI: −0.069, −0.062], child: −0.044 [95% CI: −0.049, −0.040]) and total sleep duration (min) was higher in all three groups on co-sleep nights (dog: 9.58 [95% CI: 9.11, 10.04], cat: 7.33 [95% CI: 6.85, 7.82], child: 5.49 [95% CI: 4.91, 6.08]). Disturbances (#) were higher in all three groups (dog: 0.38 [95% CI: 0.36, 0.41], cat: 0.36 [95% CI: 0.34, 0.39], child: 0.40 [95% CI: 0.37, 0.44]), and normalized disturbances (per hour of sleep) were also higher (dog: 0.019 [95% CI: 0.015, 0.022], cat: 0.023 [95% CI: 0.019, 0.026], child: 0.038 [95% CI: 0.033, 0.042]) on co-sleeping nights. Restorative sleep percentage (%) was lower in all three groups on co-sleeping nights (dog: −0.35 [95% CI: −0.40, −0.29], cat: −0.23 [95% CI: −0.28, −0.17], child: −1.17 [95% CI: −1.24, −1.11]). Finally, next-day intensity-weighted physical activity was higher following co-sleeping in the dog and cat groups (dog: 7.56 [95% CI: 6.84, 8.28], cat: 7.35 [95% CI: 6.62, 8.09]) but was lower in the child group (child: −2.57 [95% CI: −3.38, −1.77]) following co-sleeping nights.
[图：Within-person comparison of co-sleeping vs not co-sleeping with a dog. No: nights not co-sleeping with dog. Yes: nights ]
[图：Within-person comparison of co-sleeping vs not co-sleeping with a cat. No: nights not co-sleeping with cat. Yes: nights ]
[图：Within-person comparison of co-sleeping vs not co-sleeping with a child. No: nights not co-sleeping with child. Yes: nig]
Adjustment for bedtime (Model 2; S3 Table) and behavioral covariates (Model 3; S4 Table) attenuated the magnitude of several associations but did not change their directionality or significance, with one exception: the within-person association between child co-sleeping and sleep duration became non-significant after full adjustment (+0.4 ± 0.3 min, p=0.106).
Co-Sleeping Associations Differed by Sex
Associations between co-sleeping and several outcomes differed by sex (S5 Table). In the dog group, interactions were significant for seven of eight outcomes (all except disturbances, interaction p=0.473): relative to females, males showed larger reductions in RHR and RR, larger increases in HRV and total sleep time, a smaller increase in normalized disturbances, a less pronounced decrease in restorative sleep, and a smaller increase in physical activity (all interaction p≤0.022) on co-sleeping nights. In the cat group, interactions were significant for four outcomes (RR, disturbances, normalized disturbances, and restorative sleep; all interaction p≤0.004), with males showing a larger reduction in RR, smaller increases in disturbances and normalized disturbances, and a less pronounced decrease in restorative sleep. In the child group, interactions were significant for five of eight outcomes (RHR, HRV, total sleep time, disturbances, and restorative sleep; all interaction p<0.001): males showed smaller decreases in RHR, smaller increases in HRV, larger increases in total sleep time and disturbances, and a smaller decrease in restorative sleep, whereas no sex differences were observed for RR, normalized disturbances, or physical activity (all interaction p≥0.200). All interactions described as significant in these base models remained significant after FDR adjustment (S5 Table).
Adjustment for bedtime (S6 Table) and for bedtime plus behavioral covariates (S7 Table) attenuated several interactions but did not alter their direction for most outcomes. The most notable exception was child co-sleeping and total sleep time, which was lower in females but was higher in males in both the bedtime-adjusted model (females −2.64 vs males +2.12 min; interaction p<0.001) and the fully adjusted model (females −3.00 vs males +2.09 min; interaction p<0.001) on co-sleeping nights.
Findings Were Robust to Sensitivity Checks
Two sensitivity analyses were conducted to assess robustness of the primary within-person findings to the handling of nights with missing co-sleeping responses (dog: ~31%, cat: ~34%, child: ~55%). Sensitivity analysis A (S8 Table) compared the primary complete-case approach to treating null responses as no co-sleeping (null-as-no) across all groups and outcomes, and the model also included a null-response indicator as a direct covariate alongside the primary model terms across all groups and outcomes. The direction and statistical significance of associations were fully preserved across all 24 within-person outcomes, supporting the robustness of primary findings. Sensitivity analysis B (S9 Table) applied multiple imputation by chained equations (MICE; m=5 imputations) across all groups and outcomes, finding that all 24 within-person associations remained statistically significant after pooling. Journal missingness was modestly but systematically related to participant characteristics (S11 Table). In the dog and cat cohorts, shorter person-mean sleep duration (odds ratio per SD 0.76 and 0.79), higher person-mean RHR (1.17 and 1.21), and younger age (0.82 and 0.77) were associated with greater odds of a missing response. In the child cohort, these associations were weaker (sleep duration 0.89; RHR 0.94; age 0.99), and male participants were less likely to have missing responses (0.83). BMI was unrelated to missingness in all three cohorts. Refitting all 24 primary models with a first-order autoregressive residual structure yielded estimated residual autocorrelations of 0.01–0.42 across outcomes, but the direction, statistical significance, and magnitude of every within- and between-person association were unchanged; all 24 within-person associations remained statistically significant at FDR-adjusted p < 0.001, standard errors increased by no more than 13%, and the standardized β range was unchanged (0.008–0.045).
Effect Sizes Were Very Small
To assess the practical magnitudes of the observed within-person associations, effect sizes for within-person co-sleeping associations were calculated (absolute value of the standardized β range: 0.008–0.045; S1 Figure). To evaluate practical effect magnitude alongside statistical significance, standardized β values and 95% confidence intervals for all within-person and between-person co-sleeping associations are presented as forest plots in S1 Figure. The largest effects were observed for dog co-sleeping on resting heart rate (β = −0.045) and sleep duration (β = 0.043) and for child co-sleeping on restorative sleep percentage (β = −0.045), with very small effects observed across other outcomes and co-sleeping conditions. Notably, all effect sizes fell below the conventional β = 0.1 threshold for a small effect.
Discussion
We evaluated the within-person associations of co-sleeping with a dog, cat, or child on sleep, cardiorespiratory physiology, and next-day physical activity using a wearable device. Compared to sleeping alone, co-sleeping was associated with very small differences, including lower RHR and RR, higher HRV and total sleep time, more sleep disturbances, and lower restorative sleep percentage. These findings were largely similar across co-sleeping groups. Established minimal clinically important differences for night-to-night changes in wearable-derived RHR and HRV in generally healthy adults are not available. However, the absolute differences observed in the present study were small (RHR: ~0.8-1.1 beats/min; HRV: ~1.4-2.0 ms), and standardized effect sizes were uniformly very small (β = 0.008–0.045). Thus, although statistically significant, these associations should not be interpreted as evidence of clinically meaningful physiologic differences between co-sleeping nights and nights sleeping alone. In addition, because the journal item records only whether a dog, cat, or child was in the bedroom, these estimates average over bed-sharing and room-sharing nights and are likely to understate the contrast associated with bed-sharing specifically.
Child co-sleeping showed a more disruptive pattern than pet co-sleeping. Child co-sleeping was associated with smaller increases in sleep duration and greater increases in sleep disturbances and more substantially reduced restorative sleep percentage relative to pet co-sleeping. Between-person associations in the child group were consistent with these trends and persisted after covariate adjustment. Prior work has shown that co-sleeping with children is often influenced by caregiving needs, developmental transitions, or family circumstances rather than preference alone,32 and is associated with increased nighttime awakenings, greater caregiver involvement, and poorer sleep indices, many of which agree with the findings in the present study.33,34
Next-day physical activity was higher following co-sleeping with a pet (with similar magnitudes in dogs and cats) but was lower in days following co-sleeping with a child. Past research has shown that dog ownership is associated with higher activity levels,35,36 whereas little research has directly examined cat ownership and physical activity. Lower physical activity following co-sleeping with children may reflect fatigue or constrained discretionary time to be active.37,38 For example, co-sleeping with a child may limit opportunities for early-morning exercise if caregivers wake simultaneously with the child. We speculate that differences in physiologic state and daily time constraints may contribute to the different activity patterns observed following pet vs child co-sleeping.
Co-sleeping associations varied by sex. Males generally tended to show larger differences in cardiorespiratory measures and smaller differences in sleep disturbances than females, particularly in the dog group. Most notably, co-sleeping with a child was associated with higher sleep time in males but lower sleep time in females, which may reflect differences in caregiving roles. Previous research has found that mothers are the primary night-time caregivers to young children.39 However, as with the overall analysis, sex-stratified effects were very small.
Our study has several important strengths including a large sample, a within-person design that reduces confounding from between-person differences, a careful matching procedure that yielded similar demographic and behavioral factors across the three co-sleeping groups, and objective measurement of study outcomes with the wearable device.
Several limitations should be considered. First, although matching produced comparable groups on most characteristics, meaningful residual imbalance remained for habitual physical activity (dog vs child SMD = 0.367; cat vs child SMD = 0.300), with the child group being less active than the dog and cat groups, and smaller residual differences remained for sleep duration; however, because this imbalance reflects a between-group difference in habitual activity, it bears on the between-person comparisons rather than on the within-person associations that are the primary focus of this study. Second, we lacked information on several potential confounders, including socioeconomic status, chronic disease, medication use, smoking, alcohol and caffeine use, sleep disorders, and partner co-sleeping. Although the within-person design controls for stable individual characteristics, unmeasured time-varying factors associated with both co-sleeping and study outcomes may have resulted in residual confounding. In addition, although participant-specific random intercepts accounted for repeated observations within individuals, residual correlation across consecutive nights remained, particularly for the cardiorespiratory outcomes; refitting all primary models with a first-order autoregressive residual structure left every within- and between-person association unchanged in direction, statistical significance, and effect size. Next, co-sleeping was self-reported and a substantial portion of nights had missing journal responses (~31% for dog, ~34% for cat, ~55% for child), though sensitivity analyses confirmed that within-person findings were robust across all outcomes; still, the possibility of data being missing-not-at-random cannot be fully ruled out in the analysis. Requiring participants to contribute both co-sleeping and solitary nights may also select for individuals with variable sleeping arrangements, while excluding individuals who co-slept with multiple companion types may limit representation of common real-world sleeping arrangements. Another notable limitation, and one that bears substantially on interpretation, is that the WHOOP journal does not distinguish between bed-sharing and room-sharing (eg, pet was on the floor, or child was in a bassinet), nor does it capture potentially important factors such as bed size, number of co-sleepers, pet size/breed, or child age. Substantial exposure misclassification relative to bed-sharing specifically is therefore likely, and the reported associations should be interpreted as an average across both arrangements. Furthermore, although WHOOP provides objective outcome assessment and has demonstrated good validity for heart rate measures,23,40 wearable-derived sleep staging is less accurate than polysomnography and should be interpreted accordingly.22,41,42 Finally, WHOOP users tend to be younger and more active than the general adult population, which may limit generalizability to other populations and wearable devices.
Conclusions
This study of WHOOP users who kept journal entries indicating co-sleeping found that co-sleeping with a pet or child was associated with very small but consistent differences in cardiorespiratory physiology and sleep as compared to nights when sleeping alone. While causality cannot be determined, these findings may reflect small concurrent differences in HRV, RHR, and RR alongside increased sleep fragmentation on co-sleeping nights. Patterns were broadly similar across companion types, though child co-sleeping showed more pronounced sleep disruption, particularly in females. While the large sample size resulted in many statistically significant differences between co-sleeping nights and nights sleeping alone, effect sizes were very small and are unlikely to be of clinical significance, suggesting that co-sleeping decisions may be guided more by personal factors such as comfort, caregiving needs, and cultural or family routines than by concerns about large impacts on sleep or cardiorespiratory function.
Acknowledgments
The authors wish to acknowledge the technical assistance of the WHOOP Data Science and Data Engineering teams for their contributions to feature development and data infrastructure. We also thank the broader WHOOP employee community for their continued support of the company’s research initiatives.
Funding Statement
This work was supported by WHOOP, Inc. through salary support provided to authors GJG and DJC. No specific grant number is associated with this work. WHOOP, Inc. provided general institutional support but had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
AI Diligence Statement
During the preparation of this work the authors used Claude (Anthropic) for language editing and code review. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.
Data Sharing Statement
The data and code that support the findings of this study are not publicly available due to intellectual property concerns of WHOOP, Inc. Deidentified participant data and a data dictionary may be made available upon reasonable request to WHOOP (via science@whoop.com). Access will require submission of a methodologically sound proposal, approved by WHOOP’s research team, and a signed data use agreement. The timeframe for response to requests will be 4 weeks.
Ethics Approval and Informed Consent
All users consented for their anonymized data to be used for research purposes. The study was reviewed and approved by Salus IRB (Protocol #251121; Austin, TX) and deemed exempt from ongoing oversight based on minimal-risk criteria. The study was conducted in accordance with the principles of the Declaration of Helsinki.
Author Contributions
• AHKM: study conception, methodology, data analysis and interpretation, visualization, writing – original draft, and writing – review & editing.
• DJC: software, data analysis and interpretation, data curation, visualization, writing – original draft, and writing – review & editing.
• GJG: study conception, methodology, software, data curation, investigation, resources, formal analysis, project administration, writing – original draft, and writing – review & editing. All authors gave final approval of the version to be published, have agreed on the journal to which the article has been submitted, and take responsibility and accountability for all aspects of the work.
Disclosure
The study was funded by WHOOP, Inc. GJG and DJC are employees of WHOOP, Inc., and GJG holds stock options as part of his employee compensation, which represents a potential financial interest. All analysis was performed solely by the research team. Marketing, enterprise, and other non-research groups at WHOOP, Inc. had no role in study approval, data interpretation, manuscript preparation, or decisions regarding submission or publication. The authors report no other conflicts of interest in this work. A version of this manuscript was made available as a pre-print: https://www.biorxiv.org/content/10.64898/2026.06.13.732027v1.
References
[内容请查看原文]
Associated Data
Data Availability Statement
The data and code that support the findings of this study are not publicly available due to intellectual property concerns of WHOOP, Inc. Deidentified participant data and a data dictionary may be made available upon reasonable request to WHOOP (via science@whoop.com). Access will require submission of a methodologically sound proposal, approved by WHOOP’s research team, and a signed data use agreement. The timeframe for response to requests will be 4 weeks.