Insomnia severity modifies diary-based sleep-continuity and sleep-initiation associations with next-day daytime HRV in a real-world wearable dataset

Abstract

Objectives

This study examined whether baseline insomnia severity modifies within-person associations of diary-based sleep-continuity and sleep-initiation metrics with next-day daytime heart rate variability (HRV) in a wearable-based dataset.

Methods

We conducted a secondary analysis of a publicly available dataset combining daily sleep diaries with wearable-derived HRV over 4 weeks. The analytic dataset was constructed from the publicly released raw files. Daytime heart rate variability (HRV) was aggregated from eligible 5-minute windows recorded between 09:00 and 23:00, and the primary outcome was daytime RMSSD on the natural log scale. Sleep efficiency was operationalized using a Consensus Sleep Diary (CSD)–style definition that explicitly incorporated wake after sleep onset (WASO). Within-person associations were estimated using participant fixed-effects linear models with standard errors clustered at the participant level. Effect modification was tested using sleep metric × baseline Insomnia Severity Index (ISI), mean-centered, interaction terms. All analyses were restricted to observations meeting prespecified coverage thresholds and day-level complete-case criteria.

Results

The analytic dataset comprised 1,372 diary-days from 49 participants. The primary sleep-efficiency model included 451 person-days from 39 participants. Baseline insomnia symptoms were largely non-severe, and no participant met criteria for severe baseline insomnia. The CSD-style sleep efficiency × baseline ISI interaction was positive (β=0.0060; 95% CI, 0.0023 to 0.0097; P=0.002), and the sleep latency × baseline ISI interaction was negative (β=−0.0034; 95% CI, −0.0051 to −0.0018; P<0.001). Interactions involving WASO and CSD-style sleep duration were not statistically significant.

Conclusions

Baseline insomnia symptom burden modified the associations of diary-based sleep efficiency and sleep latency with daytime RMSSD. The findings support symptom-dependent sleep–HRV coupling in a predominantly healthy cohort, but higher-ISI estimates were based on few participants in the moderate-ISI range and on person-days meeting daytime HRV coverage criteria. These findings should not be extrapolated to populations with severe or clinically diagnosed insomnia.

Keywords

insomnia sleep continuity sleep initiation heart rate variability wearable devices digital phenotyping

Introduction

Sleep health is increasingly defined as a multidimensional construct of population relevance that includes sleep duration, continuity, timing/regularity, and satisfaction and is associated with cardiometabolic and mental health, and public health outcomes.^1–4 Clinical guidelines for chronic insomnia and trials of digital cognitive behavioral therapy for insomnia provide context for management and digital interventions, but the present study is observational rather than interventional.^5,6 In digital health research, the interpretability of wearable monitoring depends not only on quantifying sleep and physiology as separate streams, but also on establishing when changes in self-reported sleep correspond to interpretable shifts in autonomic physiology in daily life. Wearable sleep technology is increasingly deployed in clinical and research settings; accordingly, device outputs require validation and interpretation with respect to measurement conditions and behavioral context.⁷ Standardized prospective sleep diary frameworks, including the Consensus Sleep Diary (CSD), support consistent recording of key sleep parameters such as sleep onset, wake time, sleep efficiency, and wake after sleep onset (WASO).⁸

Heart rate variability (HRV), particularly time-domain indices such as the root mean square of successive differences (RMSSD), is often used in ambulatory settings as a proxy for vagally mediated cardiac autonomic regulation.^9,10 Lower HRV has been associated with adverse outcomes across population and clinical studies, and wearable-derived HRV is increasingly incorporated into digital health research.¹¹ In free-living monitoring, HRV data are susceptible to missingness, variable signal quality, behavioral confounding, and device-specific preprocessing; this makes it necessary to prespecify eligible windows, aggregation rules, and coverage thresholds to support transparent inference.¹² The public dataset analyzed in this study was designed to pair wearable-based HRV summaries with daily sleep diaries, providing a setting in which diary–HRV coupling can be examined under real-world monitoring constraints.¹³ The present study was not intended to validate the device-level accuracy of wearable HRV measurement. Instead, it used the released feature-level HRV summaries to examine whether diary-reported sleep–HRV coupling varied according to baseline insomnia symptom burden.

Insomnia symptoms provide a clinically relevant stratification variable for sleep-HRV analyses.^14,15 In the present study, sleep continuity is operationalized using diary-derived indices of sleep maintenance and consolidation after sleep is attempted (including sleep efficiency and WASO), whereas sleep latency is treated as a sleep-initiation metric rather than a continuity metric. The hyperarousal model characterizes insomnia as persistent cognitive and physiological activation that may influence autonomic regulation both at night and during the day.^14,16,17 Because RMSSD is commonly interpreted as an index of vagally mediated cardiac regulation, the present analysis is framed in terms of cardiac vagal physiology rather than sympathetic activation per se. Within this framework, no unidirectional causal assumption is required; altered cardiac vagal modulation may precede, follow, or co-occur with sleep disruption.

Against this backdrop, the primary question was whether the day-to-day coupling between diary-based sleep metrics and daytime RMSSD was stronger among individuals with greater baseline insomnia symptom burden. Clarifying symptom-dependent coupling may improve the interpretation of wearable HRV by identifying conditions under which diary-reported sleep metrics more closely track cardiac vagal physiology. To address this question, we conducted a secondary analysis of a publicly available dataset pairing daily sleep diaries with wearable-derived HRV.¹³ We hypothesized that sleep-continuity and sleep-initiation measures would show stronger within-person coupling with daytime RMSSD among participants with higher baseline ISI scores.

Methods

Study design and data source

We performed a secondary analysis of a publicly available, de-identified dataset that paired daily sleep diaries with wearable-derived HRV feature summaries collected over four weeks in South Korea.¹³ The study’s reporting adhered to the STROBE statement for observational studies.¹⁸ The protocol for the source study was approved by the Institutional Review Board of KAIST (KH2020-027), and all participants provided written informed consent. In the present secondary analysis, we used only de-identified records made available through Figshare (https://doi.org/10.6084/m9.figshare.28509740) no new data were collected, and participants were not recontacted. This secondary analysis of publicly available, de-identified data was reviewed by the Ethics Committee of Wuxi Huishan District People’s Hospital, which granted an exemption from full ethics review (Approval No. HYEC-Exempt-202601). In the original study, data collection was conducted in 2021 and included an in-person orientation, followed by four weeks of free-living monitoring and device return. Participants were recruited in South Korea from a university and a research institution via online postings and flyers.

Participants and original data collection

In the original study, 49 healthy adults aged 21–43 years (mean age 28.35 ± 5.87 years; 51% female), who were not undergoing hospital treatment for acute medical, surgical, or psychiatric illness, were monitored for four weeks. For the present secondary analysis, we included participants who provided at least one day of both sleep diary data and eligible daytime HRV windows, yielding an analytic sample of 49 participants. Participants attended an in-person orientation and completed four weeks of free-living monitoring, with daily reminders to facilitate diary completion. The smartwatch was worn on the non-dominant wrist during waking hours and generally removed for charging at night; hence, our analyses focused on daytime HRV.

Study size and sample-size justification

Because this study was a secondary analysis of a fixed, publicly available dataset, no recruitment-based sample-size calculation was performed. The available sample size was therefore determined by the released data, which included 49 participants and 1,372 diary-days. The primary analytic sample was subsequently defined by prespecified HRV coverage thresholds and day-level complete-case criteria. The analyses should therefore be interpreted as exploratory, with the aim of estimating within-person associations and effect modification while explicitly accounting for uncertainty, rather than producing definitive clinical estimates.

Wearable device and HRV feature generation

Physiological sensing in the source study was performed using a commercially available smartwatch and a custom research application.¹³ PPG signals were segmented into consecutive 5-minute windows, and the source data descriptor reported HeartPy-based HRV feature generation after preprocessing and quality filtering.^13,19 The present analysis used the released feature-level HRV summaries rather than raw photoplethysmography waveforms. Wearable-derived RMSSD was therefore interpreted with caution, and the emphasis was placed on within-person association patterns rather than on absolute clinical thresholds for HRV.²⁰

Sleep diary measures

Participants completed a daily sleep diary recording bedtime, sleep-onset time, wake time, nocturnal awakenings, and WASO. In this analysis, sleep continuity was operationalized using diary-derived indices of sleep maintenance and consolidation after sleep onset, specifically CSD-style sleep efficiency and WASO. Sleep latency was analyzed separately as a sleep-initiation metric.

CSD-style total sleep time was derived as time in bed minus sleep latency and WASO, and CSD-style sleep efficiency was then calculated as CSD-style total sleep time divided by time in bed. The originally released sleep efficiency variable was retained for sensitivity analyses.

The primary sleep exposures were CSD-style sleep efficiency, sleep latency, WASO, and CSD-style sleep duration. For interpretability, sleep efficiency was scaled per 10-percentage-point increase, sleep latency per 30-minute increase, WASO per 10-minute increase, and sleep duration per 1-hour increase.

Insomnia severity and other questionnaires

Insomnia symptoms were assessed using the Insomnia Severity Index (ISI), administered three times at approximately 2-week intervals.¹⁵ Baseline ISI was specified as the primary effect modifier because exact questionnaire administration dates were not available in the released dataset; therefore, baseline status provided a stable stratification variable for within-person diary–HRV analyses. ISI categories were defined as 0–7, no clinically significant insomnia; 8–14, subthreshold insomnia; 15–21, moderate insomnia; and 22–28, severe insomnia.¹⁵

The dataset also included the Patient Health Questionnaire-9 (PHQ-9) and Generalized Anxiety Disorder-7 (GAD-7), which were incorporated into sensitivity analyses to assess whether the findings were robust to adjustment for depressive and anxiety symptoms.¹³ The present secondary analysis did not administer questionnaires. No questionnaire items or forms were reproduced; only de-identified scores released in the public dataset were analyzed.

Outcome, exposures, and diary–HRV alignment

Daytime HRV was aggregated from eligible 5-minute wearable HRV windows recorded between 09:00 and 23:00 local time. The primary outcome was log-transformed daytime RMSSD. The primary analysis required at least 24 eligible daytime HRV windows per diary day, a threshold selected to ensure a minimum level of daytime HRV coverage while retaining sufficient repeated observations for within-person modeling.

Sleep diary variables were aligned with daytime HRV summaries from the corresponding diary day, so that the HRV summaries represented daytime autonomic function after the reported sleep episode.

Covariates and missing data

Models were adjusted for mean daytime heart rate, and a pedometer-derived activity summary aggregated over the same daytime window. Because participant fixed effects absorb all time-invariant participant characteristics, age, sex, baseline body mass index, and other stable individual-level characteristics were not separately estimated in the primary within-person models. Baseline lifestyle variables, including habitual coffee consumption, alcohol consumption frequency, smoking, and exercise frequency, are summarized in Table 1. Daily caffeine and alcohol intake were not available in the released sleep diary file and, therefore, could not be included as time-varying covariates. Analyses used complete-case model samples after applying the prespecified HRV coverage threshold and excluding observations with missing values for the outcome, exposure, baseline ISI, or covariates.

Table 1.

Participant-level characteristics of all participants and the primary analytic sample.

Characteristic	All participants	Primary analytic participants	Not in primary analytic sample
Participant count
Participants, n	49	39	10
Demographic characteristics
Age, years, mean ± SD	28.3 ± 5.9	28.4 ± 5.9	28.2 ± 6.2
Sex, n (%)
Male	24 (49.0%)	22 (56.4%)	2 (20.0%)
Female	25 (51.0%)	17 (43.6%)	8 (80.0%)
Height, cm, mean ± SD	168.8 ± 6.4	169.5 ± 6.0	165.2 ± 8.0
Weight, kg, mean ± SD	62.9 ± 11.5	64.7 ± 11.3	53.2 ± 7.1
BMI, kg/m2, mean ± SD	21.9 ± 3.1	22.4 ± 3.2	19.4 ± 1.0
Occupation, n (%)
Undergraduate student	15 (30.6%)	11 (28.2%)	4 (40.0%)
Graduate student	17 (34.7%)	15 (38.5%)	2 (20.0%)
Office worker	17 (34.7%)	13 (33.3%)	4 (40.0%)
Lifestyle factors
Lifestyle regularity, n (%)
Very regular	2 (4.1%)	2 (5.1%)	0 (0.0%)
Regular	28 (57.1%)	22 (56.4%)	6 (60.0%)
Average	12 (24.5%)	9 (23.1%)	3 (30.0%)
Irregular	5 (10.2%)	4 (10.3%)	1 (10.0%)
Very irregular	2 (4.1%)	2 (5.1%)	0 (0.0%)
Exercise frequency, n (%)
>5 times/week	5 (10.2%)	5 (12.8%)	0 (0.0%)
3-4 times/week	12 (24.5%)	11 (28.2%)	1 (10.0%)
2 times/week	16 (32.7%)	11 (28.2%)	5 (50.0%)
≤1 time/week	8 (16.3%)	6 (15.4%)	2 (20.0%)
Rarely	8 (16.3%)	6 (15.4%)	2 (20.0%)
Coffee consumption, n (%)
No coffee	15 (30.6%)	12 (30.8%)	3 (30.0%)
<1 cup/day	8 (16.3%)	5 (12.8%)	3 (30.0%)
1 cup/day	13 (26.5%)	9 (23.1%)	4 (40.0%)
2 cups/day	9 (18.4%)	9 (23.1%)	0 (0.0%)
≥3 cups/day	4 (8.2%)	4 (10.3%)	0 (0.0%)
Alcohol consumption, n (%)
Barely/not drink	18 (36.7%)	16 (41.0%)	2 (20.0%)
1-2 times/month	15 (30.6%)	11 (28.2%)	4 (40.0%)
1-2 times/week	15 (30.6%)	11 (28.2%)	4 (40.0%)
3-5 times/week	1 (2.0%)	1 (2.6%)	0 (0.0%)
Almost daily	0 (0.0%)	0 (0.0%)	0 (0.0%)
Smoking frequency, n (%)
Non-smoker	47 (95.9%)	37 (94.9%)	10 (100.0%)
Occasional	1 (2.0%)	1 (2.6%)	0 (0.0%)
<0.5 pack/day	0 (0.0%)	0 (0.0%)	0 (0.0%)
0.5-1 pack/day	1 (2.0%)	1 (2.6%)	0 (0.0%)
>1 pack/day	0 (0.0%)	0 (0.0%)	0 (0.0%)
Clinical questionnaires
Baseline ISI, mean ± SD	7.4 ± 4.6	6.9 ± 4.4	9.3 ± 5.4
Baseline ISI, median (IQR)	8.0 (4.0, 10.0)	7.0 (4.0, 9.0)	10.0 (6.5, 10.8)
Midpoint ISI, mean ± SD	7.3 ± 4.6	6.7 ± 4.2	9.6 ± 5.4
Endpoint ISI, mean ± SD	7.0 ± 4.6	7.0 ± 4.6	7.3 ± 4.9
Mean of three ISI measurements, mean ± SD	7.2 ± 4.3	6.9 ± 4.1	8.7 ± 4.8
Baseline ISI category, n (%)
0-7 no clinically significant insomnia	24 (49.0%)	21 (53.8%)	3 (30.0%)
8-14 subthreshold insomnia	22 (44.9%)	17 (43.6%)	5 (50.0%)
15-21 moderate insomnia	3 (6.1%)	1 (2.6%)	2 (20.0%)
22-28 severe insomnia	0 (0.0%)	0 (0.0%)	0 (0.0%)
Baseline PHQ-9, mean ± SD	3.7 ± 3.6	3.5 ± 3.7	4.3 ± 3.4
Baseline GAD-7, mean ± SD	3.0 ± 3.5	2.8 ± 3.3	3.9 ± 4.1
MEQ score, mean ± SD	43.6 ± 8.0	44.0 ± 8.2	41.8 ± 7.4
Baseline PHQ-9 category, n (%)
0-4 none/minimal	32 (65.3%)	27 (69.2%)	5 (50.0%)
5-9 mild	13 (26.5%)	9 (23.1%)	4 (40.0%)
10-14 moderate	4 (8.2%)	3 (7.7%)	1 (10.0%)
15-19 moderately severe	0 (0.0%)	0 (0.0%)	0 (0.0%)
20-27 severe	0 (0.0%)	0 (0.0%)	0 (0.0%)
Baseline GAD-7 category, n (%)
0-4 minimal	37 (75.5%)	30 (76.9%)	7 (70.0%)
5-9 mild	9 (18.4%)	8 (20.5%)	1 (10.0%)
10-14 moderate	3 (6.1%)	1 (2.6%)	2 (20.0%)
15-21 severe	0 (0.0%)	0 (0.0%)	0 (0.0%)
Data contribution
Diary days per participant, median (IQR)	28 (28, 28)	28 (28, 28)	28 (28, 28)
Days with any daytime HRV, median (IQR)	25 (22, 26)	26 (24, 27)	20 (19, 21)
Days with coverage ≥ 24, median (IQR)	6 (1, 18)	11 (3, 20)	0 (0, 0)
Primary analytic days per participant, median (IQR)	6 (1, 18)	10 (2, 20)	0 (0, 0)

Notes. Values are presented as mean ± SD, median (IQR), or n (%), as appropriate. The primary analytic sample included participants who contributed at least one person-day to the primary analytic model using recalculated CSD-style sleep efficiency, for which daytime HRV coverage was ≥24 eligible 5-minute windows.

Statistical analysis

Within-person associations were estimated using participant-fixed effects linear models with participant-level, cluster-robust standard errors.^21,22 Effect modification by baseline ISI was assessed by including sleep metric × baseline ISI interaction terms. Because baseline ISI was time-invariant, its main effect was absorbed by participant fixed effects and was not estimated separately. The main effect of each sleep metric, therefore, reflects the within-person association at the sample mean of baseline ISI. For interpretability, coefficients from log-RMSSD models were expressed as percent changes, calculated as 100 × [exp(β) − 1]. Marginal effects for CSD-style sleep efficiency and sleep latency were estimated across the observed baseline ISI range and summarized in both graphical and tabular form. Statistical significance was assessed using a two-sided threshold of P < 0.05. Because this was a secondary analysis of a fixed dataset, no a priori sample-size calculation was performed; all eligible observations available in the publicly accessible dataset were analyzed.

Sensitivity analyses were conducted to evaluate whether the findings depended on modeling choices, data completeness, or variable definitions. These analyses examined alternative HRV coverage thresholds, alternative covariate sets, released versus recalculated sleep-efficiency definitions, exclusion of extreme sleep-diary values, exclusion of participants contributing only one analytic day, approximate repeated questionnaire-based measures, heart-rate outcomes, reverse-direction models, and between-person associations between baseline ISI and HRV. Participants with at least one complete analytic day were retained in the primary analysis to avoid discarding released observations at the outset; because single-day contributors provide limited within-person information, their exclusion was evaluated as a sensitivity analysis.

Analyses were conducted in Python using pandas and NumPy for data handling, statsmodels for regression modeling, and matplotlib for visualization.

Results

After the analytic criteria were applied, the final data set comprised 1372 sleep diary-days from 49 participants. The primary model for CSD-style sleep efficiency included 451 person-days from 39 participants, and the corresponding sleep-latency model included 454 person-days from 39 participants. Baseline insomnia symptoms were largely within the absent-to-subthreshold range: 24 participants (49.0%) had baseline ISI scores of 0–7, 22 (44.9%) had scores of 8–14, and 3 (6.1%) had scores of 15–21. No participant met the ISI criterion for severe insomnia (22–28) (Table 1). Compared with the full cohort, inclusion in the primary analytic sample was determined mainly by daytime HRV coverage and complete-case requirements. The implications of this reduction in analyzable observations are addressed in the Discussion and summarized in Supplementary Tables S1–S2.

In participant fixed-effects models, baseline ISI modified the within-person association between CSD-style sleep efficiency and daytime RMSSD. The interaction between CSD-style sleep efficiency and baseline ISI was positive (β, 0.0060; 95% CI, 0.0023 to 0.0097; P=0.002), indicating that each 1-point higher baseline ISI was associated with a 0.60% (95% CI, 0.24% to 0.97%) greater sleep-efficiency slope on the percent-change scale. The corresponding interaction for sleep latency was in the opposite direction. The sleep latency × baseline ISI interaction was negative (β, -0.0034; 95% CI, -0.0051 to -0.0018; P<0.001), corresponding to a -0.34% (95% CI, -0.51% to -0.18%) change in the latency slope for each 1-point higher baseline ISI (Table 2).

Table 2.

Participant fixed-effects models for baseline ISI modification of sleep diary–daytime RMSSD associations.

Exposure	Effect unit	Model term	Beta (95% CI)	P Value	Percent change in RMSSD (95% CI)	Participants, n	Person-days, n	Mean baseline ISI used for centering
CSD-style sleep efficiency	per +10 percentage points	Sleep metric main effect at sample-mean ISI	-0.0012 (-0.0246, 0.0222)	0.917	-0.12% (-2.43, 2.24)	39	451	6.76
CSD-style sleep efficiency	per +10 percentage points	Sleep metric × baseline ISI interaction	0.0060 (0.0023, 0.0097)	0.002	0.60% (0.24, 0.97)	39	451	6.76
Sleep latency	per +30 min	Sleep metric main effect at sample-mean ISI	0.0037 (-0.0149, 0.0223)	0.689	0.37% (-1.48, 2.25)	39	454	6.79
Sleep latency	per +30 min	Sleep metric × baseline ISI interaction	-0.0034 (-0.0051, -0.0018)	<0.001	-0.34% (-0.51, -0.18)	39	454	6.79
WASO	per +10 min	Sleep metric main effect at sample-mean ISI	-0.0034 (-0.0078, 0.0010)	0.130	-0.34% (-0.78, 0.10)	39	451	6.76
WASO	per +10 min	Sleep metric × baseline ISI interaction	0.0006 (-0.0008, 0.0020)	0.369	0.06% (-0.08, 0.20)	39	451	6.76
CSD-style sleep duration	per +1 h	Sleep metric main effect at sample-mean ISI	0.0030 (-0.0125, 0.0184)	0.699	0.30% (-1.24, 1.86)	39	451	6.76
CSD-style sleep duration	per +1 h	Sleep metric × baseline ISI interaction	0.0034 (-0.0004, 0.0072)	0.079	0.34% (-0.04, 0.72)	39	451	6.76

Notes. The outcome was log-transformed daytime RMSSD. Models were estimated with participant fixed effects and participant-clustered standard errors, and were adjusted for mean daytime heart rate and pedometer-derived activity. Baseline ISI was mean-centered separately within each analytic model. Because baseline ISI was time-invariant, its main effect was absorbed by the participant fixed effects. For interaction rows, percent-change estimates represent the change in the sleep-metric slope per 1-point higher baseline ISI. Abbreviations: CSD, Consensus Sleep Diary; ISI, Insomnia Severity Index; RMSSD, root mean square of successive differences; WASO, wake after sleep onset.

No comparable interaction with baseline ISI was observed for WASO. The WASO × baseline ISI interaction was small and not statistically significant (β, 0.0006; 95% CI, -0.0008 to 0.0020; P=0.369). The CSD-style sleep duration × baseline ISI interaction was positive, although the confidence interval crossed zero and the association did not reach the conventional threshold for statistical significance (β, 0.0034; 95% CI, -0.0004 to 0.0072; P=0.079) (Table 2). The four primary sleep metric × baseline ISI interaction estimates are also summarized graphically in Supplementary Figure S1.

Marginal-effect estimates within the observed range of ISI values provided a more direct expression of these interaction terms. A 10-percentage-point higher CSD-style sleep efficiency was associated with a 4.33% (95% CI, 1.86% to 6.86%) higher daytime RMSSD at an ISI value of 14 and an 8.16% (95% CI, 3.84% to 12.67%) higher daytime RMSSD at the observed maximum ISI value. In contrast, each 30-minute longer sleep latency was associated with a -2.09% (95% CI, -3.32% to -0.85%) lower daytime RMSSD at an ISI value of 14 and a -4.10% (95% CI, -5.49% to -2.68%) lower daytime RMSSD at the observed maximum ISI value (Figure 1 and Supplementary Table S3).

Figure 1.

Baseline insomnia severity modifies day-to-day sleep–HRV coupling. Predicted percent changes in daytime RMSSD across baseline ISI values were estimated from participant fixed-effects models. Panel A presents the estimated RMSSD change per 10-percentage-point higher CSD-style sleep efficiency, whereas Panel B presents the estimated RMSSD change per 30-minute longer sleep latency. Models used daytime HRV values aggregated from 09:00 to 23:00 and required ≥24 eligible 5-minute daytime HRV windows per day. Estimates are restricted to the observed baseline ISI range of 0–20. Shaded bands indicate 95% confidence intervals, and dashed vertical lines mark ISI cut points at 7 and 14. CSD-style sleep efficiency reflects sleep continuity, whereas sleep latency reflects sleep initiation.

The main interaction patterns were generally retained in sensitivity analyses that used alternative HRV coverage thresholds, different covariate specifications, recalculated rather than released sleep-efficiency values, exclusion of extreme sleep-diary values, exclusion of single-day contributors, and additional adjustment for PHQ-9 and GAD-7 measures (Supplementary Tables S4–S7 and Supplementary Figures S2–S3). Models using mean heart rate as the outcome did not reproduce the same interaction pattern, suggesting that the primary findings were more closely tied to RMSSD-based HRV than to mean heart rate alone (Supplementary Table S8). In exploratory reverse-direction models, daytime log-transformed RMSSD also showed ISI-dependent associations with subsequent sleep-diary measures. The daytime log-transformed RMSSD × baseline ISI interaction was positive for next-night CSD-style sleep efficiency (β=0.1482; 95% CI, 0.0255 to 0.2710; P=0.019) and negative for next-night sleep latency (β=−0.2905; 95% CI, −0.5625 to −0.0185; P=0.037), whereas the corresponding interactions for next-night WASO and next-night CSD-style sleep duration were not statistically significant (Supplementary Table S9). These exploratory findings are consistent with stronger coupling between daytime cardiac vagal physiology and subsequent diary-reported sleep among participants with higher baseline ISI scores. Between-person models of baseline ISI and HRV were exploratory and are reported in Supplementary Tables S10.

Discussion

In this secondary analysis of a free-living wearable dataset, baseline insomnia symptom burden modified the within-person associations of diary-based sleep-continuity and sleep-initiation measures with daytime RMSSD. Higher recalculated sleep efficiency was associated with higher daytime RMSSD, and this association was stronger among participants with higher baseline ISI scores. A similar but opposite pattern was observed for sleep latency: longer sleep latency, a diary-based marker of difficulty initiating sleep, was more strongly associated with lower daytime RMSSD within the same observed ISI range.

The findings are consistent with a sleep health framework in which sleep maintenance and sleep initiation are treated as clinically relevant dimensions rather than interchangeable summary measures.¹ This interpretation is further supported by literature linking insomnia symptoms to hyperarousal, altered cardiac autonomic and vagal regulation, and ecological or wearable approaches to insomnia assessment.^23–26 The present results therefore do not indicate a uniform association between diary-based sleep metrics and HRV across all participants. Instead, they suggest that the physiological relevance of sleep continuity measures may vary according to baseline insomnia symptom burden.

Several non-mutually exclusive mechanisms may explain the observed effect modification, although the temporal sequence cannot be determined from this analysis. Because RMSSD is commonly interpreted as an index of vagally mediated cardiac regulation, these findings should be interpreted as relating primarily to cardiac vagal modulation rather than as direct evidence of sympathetic activation. Higher ISI scores may reflect a symptom profile characterized by greater cognitive or physiological arousal,^14,16,17 in whom sleep initiation, sleep maintenance, and cardiac vagal regulation may be more closely linked. Under this interpretation, poorer sleep continuity could be accompanied by lower next-day vagally mediated HRV. The exploratory reverse-direction models are consistent with this interpretation because daytime RMSSD showed stronger associations with subsequent sleep-efficiency and sleep-latency measures among participants with higher baseline ISI scores. Nevertheless, altered cardiac vagal modulation may precede poorer sleep, follow sleep disruption, or co-occur with sleep disturbance as part of a shared arousal-related process. For these reasons, the findings should be interpreted as evidence of symptom-dependent sleep–HRV coupling, not as evidence that sleep disruption causes next-day autonomic change.

A strength of this analysis is the use of a publicly available wearable dataset that combined repeated sleep diaries with standardized HRV feature extraction. The participant fixed-effects design reduced confounding by time-invariant participant characteristics, and segment-level quality filtering and prespecified HRV coverage thresholds made the availability of usable physiological data explicit.^13,21,22 This issue is particularly relevant in free-living wearable studies, where missingness and signal-quality filtering are integral to the measurement process rather than secondary technical issues.

The distinction between within-person coupling and between-person HRV differences should also be emphasized. In between-person sensitivity analyses, baseline ISI was not statistically significantly associated with participant-mean daytime HRV. The main finding therefore does not appear to reflect a simple pattern in which participants with higher ISI had uniformly lower HRV. Rather, it concerns how day-to-day deviations in sleep continuity were associated with daytime RMSSD within the same participant, conditional on baseline insomnia symptom burden.

Several limitations should be considered. The cohort consisted of healthy adults and did not include participants with severe baseline insomnia; therefore, these results should not be extrapolated to patients with severe insomnia or clinically diagnosed sleep disorders. Monitoring was also conducted during the COVID-19 era, when sleep, activity, stress, and cardiac vagal physiology may have been affected by broader contextual factors. Because the released dataset did not include individual-level variables that captured these contextual influences, they could not be modeled directly. The reduction from the full diary dataset to the primary analytic sample was substantial and reflected real-world HRV coverage and complete-case requirements. Selection bias may have occurred if participants or days with sufficient daytime HRV coverage differed systematically from those excluded. Possible contributors include non-wear, device removal during daytime activities, motion artifacts, physiological or proprietary signal-quality filtering, and other behavioral factors.

Additional limitations relate to daytime wearable HRV measurement and unmeasured time-varying behaviors. This analysis relied on publicly available feature-level HRV summaries and could not independently validate device-specific accuracy of daytime RMSSD. Daytime HRV is sensitive to physical activity, posture, psychological stress, caffeine intake, alcohol use, and related behaviors. Although the models adjusted for mean daytime heart rate and pedometer-derived activity, and for baseline alcohol, coffee, smoking, and exercise patterns, daily alcohol and caffeine intake were not available. Residual behavioral confounding, therefore, remains possible. Sensitivity analyses generally supported the main interaction patterns across alternative HRV coverage thresholds, covariate specifications, recalculated versus released sleep-efficiency values, exclusion of extreme sleep-diary values, and adjustment for PHQ-9 and GAD-7 measures. Because exact questionnaire administration dates were unavailable, analyses incorporating approximate time-varying questionnaire variables should be regarded as exploratory. Reverse-direction models were also exploratory and should not be interpreted as causal evidence.

Clinical reviews of insomnia, guidelines, and reviews of behavioral and digital cognitive behavioral therapy provide a clinical context.^27–29 The present secondary analysis did not evaluate insomnia interventions. Accordingly, the findings should not be used to infer treatment effects or to guide intervention selection.

Conclusion

In this free-living wearable dataset, baseline insomnia symptom burden modified the within-person associations of diary-reported sleep efficiency and sleep latency with next-day daytime RMSSD. Higher sleep efficiency showed stronger positive coupling with RMSSD, whereas longer sleep latency showed stronger negative coupling with RMSSD among participants with higher observed baseline ISI scores. These findings provide exploratory evidence of symptom-dependent sleep–HRV coupling in a predominantly healthy cohort spanning the range from absent to non-severe insomnia. Further validation is needed in larger cohorts enriched for clinically diagnosed insomnia and a broader range of sleep-disorder severity, particularly because the higher-ISI estimates in this dataset were based on a small number of participants with moderate insomnia symptoms and on a daytime HRV analytic sample defined by coverage criteria.

Supplemental material

Supplemental material - Insomnia severity modifies diary-based sleep-continuity and sleep-initiation associations with next-day daytime HRV in a real-world wearable dataset

Supplemental material for Insomnia severity modifies diary-based sleep-continuity and sleep-initiation associations with next-day daytime HRV in a real-world wearable dataset by Xiangming Meng, and Mingjing Cai in DIGITAL HEALTH.

Footnotes

Acknowledgments

We thank the original dataset authors for making the de-identified data publicly available.

ORCID iD

Xiangming Meng

Ethical considerations

This secondary analysis of a publicly available, de-identified dataset was approved by the Ethics Committee of Wuxi Huishan District People’s Hospital with exemption from full ethics review (Approval No. HYEC-Exempt-202601). No new data were collected, no participants were contacted, and no questionnaires or proprietary tools were administered by the present investigators. The original study protocol received Institutional Review Board approval from KAIST (KH2020-027).

Consent to participate

Participants provided informed consent during the original data collection.

Author contributions

Xiangming Meng: Conceptualization; Methodology; Formal analysis; Data curation; Visualization; Writing – original draft; Writing – review & editing; Supervision; Project administration. Mingjing Cai: Methodology; Software; Validation; Writing – review & editing.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

The dataset analyzed in this study is publicly available and de-identified. It can be accessed via Figshare ()³⁰. The dataset and associated data descriptor were released under a Creative Commons Attribution 4.0 International License, permitting reuse with appropriate attribution.

Supplemental material

Supplemental material for this article is available online.

Appendix

References

Buysse

. Sleep health: can we define it? Does it matter? Sleep 2014; 37: 9–17. https://doi.org/10.5665/sleep.3298

Grandner

. Sleep, Health, and Society. Sleep Med Clin 2017; 12: 1–22. https://doi.org/10.1016/j.jsmc.2016.10.012

St-Onge

Grandner

Brown

, et al. Sleep Duration and Quality: Impact on Lifestyle Behaviors and Cardiometabolic Health: A Scientific Statement From the American Heart Association. Circulation 2016; 134: e367–e386. https://doi.org/10.1161/cir.0000000000000444

Huang

Duncan

Cistulli

, et al. Sleep and physical activity in relation to all-cause, cardiovascular disease and cancer mortality risk. Br J Sports Med 2022; 56: 718–724. https://doi.org/10.1136/bjsports-2021-104046

Qaseem

Kansagara

Forciea

, et al. Management of Chronic Insomnia Disorder in Adults: A Clinical Practice Guideline From the American College of Physicians. Ann Intern Med 2016; 165: 125–133. https://doi.org/10.7326/m15-2175

Espie

Emsley

Kyle

, et al. Effect of Digital Cognitive Behavioral Therapy for Insomnia on Health, Psychological Well-being, and Sleep-Related Quality of Life: A Randomized Clinical Trial. JAMA Psychiatry 2019; 76: 21–30. https://doi.org/10.1001/jamapsychiatry.2018.2745

de Zambotti

Cellini

Goldstone

, et al. Wearable Sleep Technology in Clinical and Research Settings. Med Sci Sports Exerc 2019; 51: 1538–1557. https://doi.org/10.1249/mss.0000000000001947

Carney

Buysse

Ancoli-Israel

, et al. The consensus sleep diary: standardizing prospective sleep self-monitoring. Sleep 2012; 35: 287–302. https://doi.org/10.5665/sleep.1642

Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology . Heart rate variability: standards of measurement, physiological interpretation and clinical use. Circulation 1996; 93: 1043–1065. https://doi.org/10.1161/01.cir.93.5.1043

10.

Shaffer

Ginsberg

. An Overview of Heart Rate Variability Metrics and Norms. Front Public Health 2017; 5: 258. https://doi.org/10.3389/fpubh.2017.00258

11.

Jarczok

Weimer

Braun

, et al. Heart rate variability in the prediction of mortality: A systematic review and meta-analysis of healthy and patient populations. Neurosci Biobehav Rev 2022; 143: 104907. https://doi.org/10.1016/j.neubiorev.2022.104907

12.

Davis-Wilson

Hegarty-Craver

Gaur

, et al. Effects of Missing Data on Heart Rate Variability Measured From A Smartwatch: Exploratory Observational Study. JMIR Form Res 2025; 9: e53645. https://doi.org/10.2196/53645

13.

Baigutanova

Park

Constantinides

, et al. A continuous real-world dataset comprising wearable-based heart rate variability alongside sleep diaries. Sci Data 2025; 12: 1474. https://doi.org/10.1038/s41597-025-05801-3

14.

Riemann

Spiegelhalder

Feige

, et al. The hyperarousal model of insomnia: a review of the concept and its evidence. Sleep Med Rev 2010; 14: 19–31. https://doi.org/10.1016/j.smrv.2009.04.002

15.

Morin

Belleville

Bélanger

, et al. The Insomnia Severity Index: psychometric indicators to detect insomnia cases and evaluate treatment response. Sleep 2011; 34: 601–608. https://doi.org/10.1093/sleep/34.5.601

16.

Vgontzas

Fernandez-Mendoza

Liao

, et al. Insomnia with objective short sleep duration: the most biologically severe phenotype of the disorder. Sleep Med Rev 2013; 17: 241–254. https://doi.org/10.1016/j.smrv.2012.09.005

17.

Dodds

Miller

Kyle

, et al. Heart rate variability in insomnia patients: A critical review of the literature. Sleep Med Rev 2017; 33: 88–100. https://doi.org/10.1016/j.smrv.2016.06.004

18.

von

Altman

Egger

, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Ann Intern Med 2007; 147: 573–577. https://doi.org/10.7326/0003-4819-147-8-200710160-00010

19.

Van Gent

Farah

Van Nes

, et al. HeartPy: A novel heart rate algorithm for the analysis of noisy signals. Transportation research part F: traffic psychology and behaviour 2019; 66: 368–378. https://doi.org/10.1016/j.trf.2019.09.015

20.

Dudarev

Barral

Zhang

, et al. On the Reliability of Wearable Technology: A Tutorial on Measuring Heart Rate and Heart Rate Variability in the Wild. Sensors (Basel) 2023; 23: 5863. https://doi.org/10.3390/s23135863

21.

Abadie

Athey

Imbens

, et al.

When Should You Adjust Standard Errors for Clustering?

The Quarterly Journal of Economics 2023; 138: 1–35. https://doi.org/10.1093/qje/qjac038

22.

Cameron

Miller

. A Practitioner’s Guide to Cluster-Robust Inference. Journal of Human Resources 2015; 50: 317–372. https://doi.org/10.3368/jhr.50.2.317

23.

Zhao

Van Someren

EJW

, et al. Identifying the insomnia-related psychological issues associated with hyperarousal: A network perspective. Int J Psychophysiol 2024; 195: 112276. https://doi.org/10.1016/j.ijpsycho.2023.112276

24.

Zhao

Jiang

. Heart rate variability in patients with insomnia disorder: a systematic review and meta-analysis. Sleep Breath 2023; 27: 1309–1313. https://doi.org/10.1007/s11325-022-02720-0

25.

Menghini

Yuksel

Prouty

, et al. Wearable and mobile technology to characterize daily patterns of sleep, stress, presleep worry, and mood in adolescent insomnia. Sleep Health 2023; 9: 108–116. https://doi.org/10.1016/j.sleh.2022.11.006

26.

Baglioni

Johann

Benz

, et al. Interactions between insomnia, sleep duration and emotional processes: An ecological momentary assessment of longitudinal influences combining self-report and physiological measures. J Sleep Res 2024; 33: e14001. https://doi.org/10.1111/jsr.14001

27.

Perlis

Posner

Riemann

, et al. Insomnia. Lancet 2022; 400: 1047–1060. https://doi.org/10.1016/s0140-6736(22)00879-0

28.

Edinger

Arnedt

Bertisch

, et al. Behavioral and psychological treatments for chronic insomnia disorder in adults: an American Academy of Sleep Medicine clinical practice guideline. J Clin Sleep Med 2021; 17: 255–262. https://doi.org/10.5664/jcsm.8986

29.

Hwang

Lee

Woo

, Systematic review and meta-analysis on fully automated digital cognitive behavioral therapy for insomnia. NPJ Digit Med 2025; 8: 157. https://doi.org/10.1038/s41746-025-01514-4

30.

Baigutanova

Park

Constantinides

, et al. In-situ wearable-based dataset of continuous heart rate variability monitoring accompanied by sleep diaries. figshare. Dataset, 2025. Vol 1. https://doi.org/10.6084/m9.figshare.28509740

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.51 MB