Key Points
Question
Among children evaluated in a clinical sleep laboratory, is outpatient melatonin use associated with differences in polysomnographic sleep architecture?
Findings
In this cross-sectional study of 684 children (342 propensity score–matched pairs), melatonin users had a small reduction in REM sleep percentage compared with matched nonusers (16.7% vs 19.0%). Total sleep time, non-REM sleep stages, respiratory parameters, arousal indices, and periodic limb movements did not differ.
Meaning
In this study, outpatient melatonin use was associated with a small REM reduction whose clinical significance is uncertain.
This cross-sectional study examines whether outpatient melatonin use in children is associated with differences in objectively measured sleep architecture using polysomnography.
Abstract
Importance
Melatonin is the most widely used sleep aid in children, yet no large study has examined its association with objective sleep architecture measured by polysomnography (PSG).
Objective
To assess whether outpatient melatonin use is associated with differences in PSG-derived sleep architecture in a pediatric sleep clinic population.
Design, Setting, and Participants
This cross-sectional, propensity score–matched study used the Nationwide Children’s Hospital Sleep DataBank, a deidentified dataset of PSG recordings and linked electronic health record data from children evaluated at a single academic pediatric sleep laboratory between 2017 and 2019. Children with at least 1 technically adequate PSG and complete electronic health record data were included; children older than 18 years were excluded. Data were analyzed from January to March 2026.
Exposures
Outpatient melatonin prescription documented in the medication list.
Main Outcomes and Measures
The prespecified primary outcome was percentage of rapid eye movement (REM) sleep. Fourteen additional PSG parameters (sleep stages, respiratory indices, arousal index, periodic limb movements, oxygen desaturation index) were exploratory. Propensity score matching (1:1 nearest-neighbor, caliper 0.2 SD of the logit propensity score) on age, sex, body mass index percentile, comorbidity burden, obstructive sleep apnea, and epilepsy yielded 342 matched pairs (684 children). Outcomes were compared with 2-sided Wilcoxon signed-rank tests at α = .05; Benjamini–Hochberg false discovery rate (FDR) correction was applied across the 15 outcomes.
Results
The analytic cohort constituted 3392 children (mean [SD] age, 8.0 [5.0] years; 1924 male [56.7%]), including 346 melatonin users and 3046 nonusers. The matched cohort constituted 684 children (342 users and 342 nonusers). After matching, all 6 covariates achieved standardized mean differences below 0.10. Melatonin users had a lower median percentage of REM sleep than matched nonusers (16.7% [IQR, 11.6%-22.0%] vs 19.0% [IQR, 14.6%-23.2%]; Hedges g = −0.22 [95% CI, −0.32 to −0.11]; FDR-corrected P = .003). Total sleep time, sleep efficiency, non-REM sleep stages, respiratory indices, arousal index, periodic limb movements, and oxygen desaturation index did not differ (FDR-corrected P > .05 for all 14 exploratory outcomes). The REM association was robust to 1:k matching (each melatonin user matched to 2 or 3 nonusers) and to averaging across repeated PSG nights. After adjusting for 3 psychiatric diagnoses (attention-deficit/hyperactivity disorder, anxiety, depression), REM reduction was preserved with attenuated magnitude (16.6% [IQR, 11.6%-22.0%] vs 18.4% [IQR, 13.4%-22.8%]; Hedges g = −0.17 [95% CI, −0.27 to −0.06]; P = .01).
Conclusions and Relevance
In this cross-sectional study of 684 propensity score–matched children, outpatient melatonin use was associated with a small reduction in REM sleep percentage and was not associated with differences in total sleep time, non-REM sleep stages, respiratory parameters, arousal architecture, or periodic limb movements. The REM association was attenuated when psychiatric diagnoses were added to the propensity model. Residual confounding cannot be excluded, and prospective studies with documented dose, timing, and adherence are needed to clarify the directionality of this association.
Introduction
Melatonin has become the most commonly used sleep supplement in children, with recent surveys indicating that up to 19% of school-age children have used melatonin in the past 30 days.1,2 In the US, melatonin is classified as a dietary supplement, is available without prescription, and is widely recommended as a pharmacologic option for pediatric insomnia, even though behavioral intervention remains the recommended first-line treatment.3,4 However, the evidence base for its effects on objective sleep architecture remains limited.
Despite widespread use, little is known about how melatonin is associated with objective sleep architecture in children. The existing literature is dominated by small randomized clinical trials that relied on actigraphy or parent-reported outcomes rather than polysomnography (PSG), the criterion standard for assessing sleep stages.5,6 Malow et al7 conducted a dose-escalation trial of melatonin in 24 children with autism spectrum disorder (ASD) using actigraphy. A systematic review of melatonin in ASD identified 18 studies, none of which reported PSG sleep architecture outcomes.8
This evidence gap is clinically relevant. Rapid eye movement (REM) sleep is implicated in neurodevelopment, memory consolidation, and emotional regulation in children.9,10 Pharmacologic agents that suppress REM sleep, including serotonergic antidepressants and benzodiazepines, are used cautiously in pediatric populations.11 Melatonin’s interaction with serotonergic pathways raises questions about potential REM modulation, yet no adequately powered observational or randomized study has examined this possibility.12 Other domains of sleep architecture warrant the same scrutiny: slow-wave sleep is closely linked to growth hormone secretion and restorative function,13,14 and sleep-disordered breathing, arousal indices, and periodic limb movements are clinically relevant to any sleep-modifying agent yet remain unstudied at scale for melatonin. We used the Nationwide Children’s Hospital Sleep DataBank (NCHSDB), one of the largest publicly available pediatric PSG databases in North America,15 to compare 15 PSG parameters between propensity score–matched melatonin users and nonusers, with percentage of REM sleep prespecified as the primary outcome.
Methods
Data Source and Ethics
In this cross-sectional study, we used NCHSDB version 0.3.0, a deidentified, publicly available dataset of PSG recordings and linked electronic health record (EHR) data from Nationwide Children’s Hospital (Columbus, Ohio), accessed through the National Sleep Research Resource (NSRR).15,16 The database contains 3984 PSG studies from patients evaluated in the hospital’s accredited sleep laboratory between 2017 and 2019. All PSGs were recorded with standard pediatric montages and scored according to American Academy of Sleep Medicine (AASM) criteria by sleep technologists certified at the originating institution; equipment, montages, and scoring procedures are described in the NCHSDB data descriptor.15,17 As described by the dataset creators, the deidentified NCHSDB does not meet the definition of human participant research under US Department of Health and Human Services and Food and Drug Administration regulations. The original data collection was approved by the Nationwide Children’s Hospital institutional review board with informed consent obtained from parents or guardians; the secondary analysis of deidentified data was conducted under a Health Insurance Portability and Accountability Act waiver without additional consent requirements.15 This study follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cross-sectional studies.18
Study Population and PSG Indications
Eligibility required (1) at least 1 technically adequate PSG passing institutional quality control for staging completeness and recording artifacts; (2) complete linked EHR data; and (3) age 18 years or younger at the time of PSG. Sequential exclusion gates were applied from the source database to the analytic sample (eFigure 1 in Supplement 1). From 3984 studies, 130 were excluded for recording artifacts or incomplete staging; 276 duplicate PSGs were excluded by retaining the first technically adequate study per patient; and 186 patients older than 18 years were excluded, leaving a total of 3392 PSGs. The NCHSDB does not document the clinical indication for individual referrals; typical pediatric PSG indications include suspected obstructive sleep apnea, nocturnal events in children with epilepsy, sleep complaints in children with neurodevelopmental disorders, and pre- or post-adenotonsillectomy assessment.15 The cohort reflects a referred clinical population rather than a community sample.
Race, Ethnicity, and Exposure Definition
Race and ethnicity were extracted as recorded in the originating EHR using PCORnet Common Data Model categories19; the source of classification (clinician documentation vs patient or caregiver self-report) is not distinguishable in the deidentified release. Race categories were reported as Asian, Black or African American, White, or other race (ie, American Indian or Alaska Native, Native Hawaiian or Other Pacific Islander, multiple races, and race not reported). Melatonin exposure was defined as any outpatient melatonin prescription documented in the patient’s medication list (the ever outpatient rule). This rule classified 346 patients as melatonin users and 3046 as nonusers.
Outcomes
The prespecified primary outcome was percentage of REM sleep, selected for its role in neurodevelopment and plausible modulation by melatonin’s serotonergic interaction.9,12 Fourteen exploratory outcomes were also assessed: total sleep time (minutes); sleep efficiency (percentages); sleep onset latency (minutes); REM latency (minutes); wake after sleep onset (WASO; minutes); stages N1, N2, and N3 (percentages); apnea-hypopnea index (AHI; events per hour); obstructive AHI (events per hour); central apnea index (events per hour); arousal index (events per hour); periodic limb movement index (events per hour); and oxygen desaturation index (events per hour). All outcomes were derived from AASM-scored PSG.17
Propensity Score Matching
We performed 1:1 nearest-neighbor propensity score matching on the logit of the propensity score with a caliper of 0.2 SDs.20 Propensity scores were estimated using logistic regression with 6 covariates selected a priori based on documented associations with both melatonin prescribing and sleep architecture: age (years; continuous, untransformed), sex (male = 1; binary), body mass index (BMI) percentile (continuous, untransformed), comorbidity burden (continuous, untransformed; defined as the count of unique International Classification of Diseases and Other Related Health Problems, Tenth Revision [ICD-10] chapters represented in the patient’s diagnosis history), obstructive sleep apnea (binary), and epilepsy (binary).21 Covariate balance was assessed with absolute standardized mean differences (SMDs); SMD<0.10 was considered adequate.20 Matching yielded 342 pairs; 4 melatonin users had no nonuser within the caliper distance and were therefore unmatched. Propensity score matching addresses imbalance in the covariates in the model within the matched sample, including confounding by indication as a particularly relevant form of imbalance.
Statistical Analysis
Data were analyzed from January to March 2026. The primary outcome (REM sleep percentage) was tested with a 2-sided Wilcoxon signed-rank test at α = .05. The 14 exploratory outcomes were tested identically; P values across all 15 outcomes were corrected with the Benjamini–Hochberg false discovery rate (FDR) procedure.22 Bonferroni-corrected P values are also reported for reference. Effect sizes are presented as Hedges g with 95% CIs derived from the SD of within-pair differences.23 Descriptive statistics are reported as medians and IQRs for skewed variables and means and SDs for approximately normal variables. E-values were computed using the method of VanderWeele and Ding24; Hedges g was converted to an approximate odds ratio (OR) using Chinn’s formulation25 (OR ≈ exp[gπ/√3]). Matching, comparison, and effect-size routines were deterministic, with no random sampling step.
Sensitivity Analyses and Missing Data
Seven prespecified sensitivity analyses repeated the full matching and outcome comparison within these subgroups: ASD (eMethods 1 in Supplement 1), age (≤6, 6-12, and >12 years), exclusion of patients on any psychotropic medication, epilepsy, and attention-deficit/hyperactivity disorder (ADHD). Four post hoc analyses were not prespecified: (1) 2 expanded propensity models, the first adding ADHD, anxiety, and depression and the second adding those 3 plus insomnia, benzodiazepines, and alpha-2 agonists (clonidine and guanfacine); (2) 1:k matching at k = 2 and 3 within the same caliper; (3) using the mean of PSG metrics across repeated studies before matching; and (4) replacing the comorbidity-chapter count with 11 binary diagnosis-type indicators (14 covariates total, retaining age, sex, and BMI percentile). Missingness per outcome was below 2% (eTable 6 in Supplement 1); analyses were complete-case within each outcome. REM latency data were missing differentially, with higher rates of epilepsy, melatonin use, and comorbidity burden among affected children (several SMDs >0.10; eTable 7 in Supplement 1), whereas the primary outcome had no missing data (eTable 6 in Supplement 1). All analyses were conducted in Python, version 3.14 using SciPy, version 1.16; statsmodels, version 0.14; and scikit-learn, version 1.8 (Python Software Foundation).
Results
Cohort and Baseline Characteristics
The 3392 children in the analytic cohort had a mean (SD) age of 8.0 (5.0) years; 1924 (56.7%) were male and 1468 (43.3%) were female. Among the 346 melatonin users, the mean (SD) age was 9.5 (4.7) years, 200 (57.8%) were male, and 146 (42.2%) were female; among the 3046 nonusers, the mean (SD) age was 7.9 (5.0) years, 1724 (56.6%) were male, and 1322 (43.4%) were female. For race and ethnicity, 6 (1.7%) and 82 (2.7%) were Asian, 81 (23.4%) and 604 (19.8%) were Black or African American, 13 (3.8%) and 161 (5.3%) were Hispanic, 214 (61.8%) and 2025 (66.5%) were White, and 45 (13.0%) and 335 (11.0%) were other race in the melatonin users and nonuser groups, respectively. Before matching, melatonin users had a higher comorbidity burden (mean [SD], 10.0 [2.8] vs 7.1 [3.2] unique ICD-10 chapters; P < .001) and higher prevalences of epilepsy (124 [35.8%] vs 384 [12.6%]; P < .001), ADHD (155 [44.8%] vs 452 [14.8%]; P < .001), anxiety (139 [40.2%] vs 452 [14.8%]; P < .001), and depression (65 [18.8%] vs 224 [7.4%]; P < .001). The rate of obstructive sleep apnea was lower among melatonin users (237 [68.5%] vs 2324 [76.3%]; P = .002). BMI percentile and sex distribution did not differ (Table 1). These baseline differences underscore the importance of propensity score matching to minimize confounding by indication.
Table 1. Baseline Characteristics of the Unmatched Cohort and the Matched Subcohorta.
| Characteristic | Unmatched | Matched | |||
|---|---|---|---|---|---|
| Melatonin (n = 346) | No melatonin (n = 3046) | P value | Melatonin (n = 342) | No melatonin (n = 342) | |
| Age, mean (SD), y | 9.5 (4.7) | 7.9 (5.0) | <.001 | 9.5 (4.7) | 9.2 (4.9) |
| Sex, No. (%) | |||||
| Male | 200 (57.8) | 1724 (56.6) | .71 | 198 (57.9) | 196 (57.3) |
| Female | 146 (42.2) | 1322 (43.4) | 144 (42.1) | 146 (42.7) | |
| Race, No. (%) | |||||
| Asian | 6 (1.7) | 82 (2.7) | .16 | 6 (1.8) | 7 (2.0) |
| Black or African American | 81 (23.4) | 604 (19.8) | 80 (23.4) | 78 (22.8) | |
| White | 214 (61.8) | 2025 (66.5) | 212 (62.0) | 213 (62.3) | |
| Otherb | 45 (13.0) | 335 (11.0) | 44 (12.9) | 44 (12.9) | |
| Hispanic ethnicity, No. (%) | 13 (3.8) | 161 (5.3) | .28 | 13 (3.8) | 16 (4.7) |
| BMI percentile, mean (SD) | 74.0 (31.5) | 72.0 (31.0) | .27 | 74.1 (31.5) | 73.0 (31.1) |
| Comorbidity burden, mean (SD)c | 10.0 (2.8) | 7.1 (3.2) | <.001 | 10.0 (2.8) | 9.9 (2.9) |
| Selected comorbidities, No. (%) | |||||
| Obstructive sleep apnea | 237 (68.5) | 2324 (76.3) | .002 | 234 (68.4) | 233 (68.1) |
| Asthma | 158 (45.7) | 785 (25.8) | <.001 | 156 (45.6) | 152 (44.4) |
| Epilepsy or seizures | 124 (35.8) | 384 (12.6) | <.001 | 122 (35.7) | 124 (36.3) |
| ADHD | 155 (44.8) | 452 (14.8) | <.001 | 154 (45.0) | 67 (19.6) |
| Depression | 65 (18.8) | 224 (7.4) | <.001 | 64 (18.7) | 41 (12.0) |
| Anxiety | 139 (40.2) | 452 (14.8) | <.001 | 138 (40.4) | 83 (24.3) |
| GERD | 160 (46.2) | 803 (26.4) | <.001 | 158 (46.2) | 154 (45.0) |
| Down syndrome | 15 (4.3) | 173 (5.7) | .36 | 15 (4.4) | 32 (9.4) |
Abbreviations: ADHD, attention-deficit/hyperactivity disorder; BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); GERD, gastroesophageal reflux disease.
P values are for the unmatched comparison only (independent samples t test or Mann-Whitney U test for continuous variables; χ2 test for categorical variables). The matched columns are descriptive; balance after matching is shown in Table 2 and was achieved on the 6 propensity model covariates only.
Other category includes American Indian or Alaska Native, Native Hawaiian or Other Pacific Islander, multiple races, and race not reported as recorded in the source electronic health record using PCORnet Common Data Model categories.
Defined as the count of unique chapters from the International Classification of Diseases and Other Related Health Problems, Tenth Revision represented in the patient’s diagnosis history.
Propensity score matching successfully matched 342 of 346 melatonin users (98.8%) to 342 nonusers; 4 melatonin users had no nonuser within the 0.2-SD logit caliper and were dropped. After matching, all 6 covariates achieved an SMD under 0.10 (Table 2): age (from 0.332 to 0.070), comorbidity burden (from 0.959 to 0.044), epilepsy (from 0.562 to 0.006), and obstructive sleep apnea (from 0.178 to <.001). Sex (from 0.025 to 0.018) and BMI percentile (from 0.065 to 0.025) remained well balanced. Baseline characteristics of the matched cohort are shown in Table 1. In the matched cohort of 342 pairs, melatonin users had a median REM sleep percentage of 16.7% (IQR, 11.6%-22.0%) compared with 19.0% (IQR, 14.6%-23.2%) among nonusers.
Table 2. Covariate Balance Before and After Propensity Score Matching.
| Covariate | Matching | Balanceda | |
|---|---|---|---|
| SMD before | SMD after | ||
| Age, y | 0.33 | 0.07 | Yes |
| Male sex | 0.03 | 0.02 | Yes |
| BMI percentile | 0.07 | 0.03 | Yes |
| Comorbidity burdenb | 0.96 | 0.04 | Yes |
| Obstructive sleep apnea | 0.18 | <.001 | Yes |
| Epilepsy or seizures | 0.56 | 0.01 | Yes |
Abbreviations: BMI, body mass index; SMD, absolute standardized mean difference.
Balance defined as SMD<0.10. All 6 covariates achieved adequate balance after matching.
Defined as the count of unique chapters from the International Classification of Diseases and Other Related Health Problems, Tenth Revision represented in the patient’s diagnosis history.
Melatonin Use, REM Sleep Percentage, and PSG Outcomes
Melatonin users had a lower percentage of REM sleep than matched nonusers (median [IQR], 16.7% [11.6%-22.0%] vs 19.0% [14.6%-23.2%]; Hedges g = −0.22; 95% CI, −0.32 to −0.11), with a raw P < .001 and a FDR-corrected P = .003 (Table 3; Figure 1). REM latency was scoreable in 316 of 342 pairs; the 26 pairs with missing data reflect studies with no recorded REM epochs.
Table 3. Polysomnographic Outcomes Among Propensity Score–Matched Pairsa.
| Outcome | No. of pairs | Median (IQR) | Hedges g (95% CI) | P value | |||
|---|---|---|---|---|---|---|---|
| Melatonin | No melatonin | Raw | FDR | Bonferroni | |||
| % Stage REM (primary) | 342 | 16.7 (11.6 to 22.0) | 19.0 (14.6 to 23.2) | −0.22 (−0.32 to −0.11) | <.001 | .003 | .003 |
| Total sleep time, min | 342 | 391.8 (346.8 to 432.5) | 395.8 (345.5 to 434.8) | −0.04 (−0.14 to 0.07) | .46 | .67 | >.99 |
| Sleep efficiency, % | 342 | 86.6 (76.0 to 93.1) | 86.6 (75.4 to 92.4) | 0.01 (−0.10 to 0.12) | .62 | .72 | >.99 |
| Sleep onset latency, min | 342 | 12.3 (2.8 to 31.0) | 12.4 (4.7 to 27.8) | 0.07 (−0.04 to 0.17) | .60 | .72 | >.99 |
| REM latency, min | 316 | 132.2 (79.4 to 191.8) | 115.8 (74.4 to 172.9) | 0.10 (−0.01 to 0.21) | .07 | .14 | .99 |
| WASO, min | 342 | 30.8 (13.5 to 68.9) | 37.2 (16.6 to 74.2) | −0.08 (−0.18 to 0.03) | .04 | .13 | .63 |
| % Stage N1 | 342 | 3.3 (1.6 to 5.5) | 3.3 (1.5 to 6.0) | −0.05 (−0.15 to 0.06) | .26 | .49 | >.99 |
| % Stage N2 | 342 | 48.9 (42.2 to 57.4) | 48.1 (41.2 to 55.0) | 0.14 (0.03 to 0.24) | .02 | .06 | .23 |
| % Stage N3 | 342 | 27.8 (21.5 to 35.8) | 27.6 (20.7 to 34.3) | 0.00 (−0.11 to 0.10) | .90 | .93 | >.99 |
| AHI, events/h | 342 | 0.6 (0.1 to 1.8) | 0.8 (0.2 to 2.5) | −0.03 (−0.13 to 0.08) | .01 | .06 | .20 |
| Obstructive AHI, events/h | 342 | 0.4 (0.0 to 1.4) | 0.7 (0.1 to 2.2) | −0.03 (−0.14 to 0.08) | .01 | .06 | .15 |
| Central apnea index, events/h | 342 | 0.0 (0.0 to 0.2) | 0.0 (0.0 to 0.2) | 0.00 (−0.11 to 0.10) | .93 | .93 | >.99 |
| Arousal index, events/h | 342 | 4.5 (3.1 to 6.7) | 5.0 (3.2 to 7.3) | −0.09 (−0.20 to 0.01) | .06 | .14 | .90 |
| PLMS index, events/h | 342 | 0.0 (0.0 to 0.1) | 0.0 (0.0 to 0.8) | −0.05 (−0.15 to 0.06) | .49 | .67 | >.99 |
| O2 desaturation index, events/h | 342 | 3.3 (0.9 to 9.6) | 4.2 (1.4 to 9.9) | −0.03 (−0.13 to 0.08) | .38 | .63 | >.99 |
Abbreviations: AHI, apnea-hypopnea index; FDR, false discovery rate; O2, oxygen; PLMS, periodic limb movements of sleep; REM, rapid eye movement; WASO, wake after sleep onset.
The prespecified primary outcome is shown in the first row. Effect sizes are Hedges g with 95% CIs derived from the SD of within-pair differences (positive Hedges g = higher value in the melatonin group). Paired comparisons used a 2-sided Wilcoxon signed-rank test at α = .05. FDR correction was applied across all 15 outcomes by the Benjamini–Hochberg procedure. REM latency was scoreable in 316 of 342 pairs because 26 studies recorded no REM epochs.
Figure 1. Forest Plot Showing Polysomnographic Outcomes in Propensity Score–Matched Melatonin Users vs Nonusers.
The figure illustrates Hedges g effect sizes for 15 polysomnography parameters comparing 342 melatonin users with 342 matched nonusers. The prespecified primary outcome (% REM sleep) and remaining exploratory outcomes are presented. The dashed line indicates g = 0; values to the left of 0 indicate lower values in the melatonin group. Error bars represent 95% CIs. AHI indicates apnea-hypopnea index; O2, oxygen; PLMS, periodic limb movements of sleep; REM, rapid eye movement; WASO, wake after sleep onset.
After FDR correction, none of the 14 exploratory PSG outcomes differed between groups (all FDR-corrected P > .05). No significant difference was observed with total sleep time, sleep efficiency, or sleep onset latency. WASO was lower among melatonin users compared with nonusers (median [IQR] time, 30.8 [13.5 to 68.9] vs 37.2 [16.6 to 74.2] minutes; Hedges g = −0.08; 95% CI, −0.18 to 0.03; raw P = .04) but not significant after FDR correction (P = .13). Non-REM (NREM) sleep stages, respiratory indices (AHI, obstructive AHI, central apnea index, oxygen desaturation index), arousal index, and periodic limb movements all crossed the null, except percentage of stage N2 sleep. Full descriptive statistics and effect sizes are presented in Table 3.
Sensitivity Analyses of the REM Sleep Association
The REM reduction was directionally consistent across all 7 prespecified subgroups (Figure 2; subgroup effect-size detail in eTable 1 in Supplement 1). Effect sizes ranged from a Hedges g of −0.21 in the no-psychotropics (n = 253 pairs; FDR-corrected P = .02) and epilepsy (n = 114 pairs) subgroups to a Hedges g of −0.03 in the older than 12 years subgroup (n = 118 pairs). In the ASD subgroup (n = 64 pairs), melatonin users had a higher percentage of stage N2 sleep (Hedges g = 0.30; 95% CI, 0.06-0.55), but it did not survive correction for multiple comparisons (FDR-corrected P = .17) and needs confirmation. In the ADHD subgroup (n = 152 pairs), the REM difference was small and not significant (Hedges g = −0.05; 95% CI, −0.21 to 0.10; FDR-corrected P = .78).
Figure 2. Forest Plot Showing Sensitivity and Post Hoc Analyses of the Rapid Eye Movement (REM) Sleep Association.
The figure illustrates Hedges g for REM sleep percentage across the prespecified primary analysis, 7 prespecified subgroup analyses, and 4 post hoc analyses requested at peer review (1:2 matching, 1:3 matching, mean across repeated polysomnographies [PSGs], and an expanded propensity model adding psychiatric diagnoses and concurrent sleep medications). The dashed line indicates g = 0; values to the left of 0 indicate a lower REM percentage in the melatonin group. Error bars represent 95% CIs. ADHD indicates attention-deficit/hyperactivity disorder; ASD, autism spectrum disorder.
The REM reduction persisted across all 4 post hoc sensitivity analyses, with effect sizes ranging from a Hedges g of −0.16 to −0.21 (all raw P < .05), close to the primary estimate of Hedges g of −0.22. In 1:2 matching, 325 treated participants were matched to 640 controls; Hedges g was −0.19 (95% CI, −0.30 to −0.08; raw P = .002), and in 1:3 matching with 314 treated participants matched to 908 controls, Hedges g was −0.20 (95% CI, −0.31 to −0.08; raw P < .001) (eTable 3 in Supplement 1). Averaging across repeated PSG nights with 342 pairs yielded a Hedges g of −0.21 (95% CI, −0.32 to −0.11; raw P < .001) (eTable 4 in Supplement 1), and replacing the comorbidity-chapter count with 11 binary diagnosis-type flags, using 14 total covariates with 318 pairs, yielded a Hedges g of −0.16 (95% CI, −0.27 to −0.05; raw P = .009) (eTable 5 in Supplement 1).
The REM association persisted but was attenuated under 2 parallel expanded propensity models (eTable 2 in Supplement 1). Adjusting for ADHD, anxiety, and depression alone, Hedges g was −0.17 (95% CI, −0.27 to −0.06; raw P = .01; n = 344 pairs). Retaining those 3 psychiatric diagnoses and adding insomnia plus 2 sleep medications (benzodiazepines and alpha-2 agonists), Hedges g attenuated further to −0.11 (95% CI, −0.22 to 0.01; raw P = .12; n = 295 pairs), with the upper confidence interval bound just crossing the null.
Developmental Gradient
Within the matched cohort (n = 342 pairs), a linear regression model tested whether the melatonin effect on REM sleep percentage varied with age. The melatonin-by-age interaction estimate was 0.20 percentage points per year (SE, 0.13; P = .11) (eTable 8 in Supplement 1), indicating that the REM reduction was on average about 0.20 percentage points smaller for each additional year of age, without reaching statistical significance. A per-pair-difference regression confirmed the same direction (slope = 0.20; P = .25). Age-stratified subgroup estimates of the melatonin-associated REM reduction, restricted to same-bin matched pairs (114 of 342 pairs) showed a nonmonotonic pattern: the effect size was largest in middle childhood (6-12 years: Hedges g = −0.52; 95% CI, −0.84 to −0.19; n = 41 pairs) and smaller at the extremes (≤6 years: Hedges g = −0.18; 95% CI, −0.56 to 0.20; n = 27 pairs; 12–18 years: Hedges g = −0.02; 95% CI, −0.31 to 0.27; n = 46 pairs) (eFigure 2 in Supplement 1). On average, the interaction pointed toward a melatonin effect that weakens as children grow older. However, the age groups did not show a steady decline; the reduction was largest in middle childhood, and neither test of this age dependence reached statistical significance.
Discussion
In this cross-sectional, propensity score–matched analysis of 342 pediatric pairs from the NCHSDB, outpatient melatonin use was associated with a small reduction in REM sleep percentage. None of the 14 exploratory PSG outcomes differed between groups after correction for multiple comparisons.
Magnitude of the REM Association
Outpatient melatonin use was associated with a modest reduction in REM sleep. Melatonin users spent a median of 16.7% of sleep time in REM, compared with 19.0% among matched nonusers, a difference of 2.3 percentage points (Hedges g = −0.22). By Cohen’s conventions, this finding represents a small effect, and its clinical meaning is genuinely uncertain. A difference of this size sits within the night-to-night variability that characterizes REM sleep even in healthy children, and no established threshold defines how large a REM change must be before it carries functional consequences for neurodevelopment, memory, or emotional regulation. The absolute REM percentages in both groups fell below normative pediatric values, most likely reflecting the high burden of sleep disorders in this clinical referral population rather than anything specific to melatonin. Earlier randomized trials of pediatric melatonin could not address this question because they relied on actigraphy or parent report rather than polysomnography, leaving REM architecture effectively unmeasured.
Robustness and Residual Confounding
The REM association was stable across 4 alternative analytic specifications: 1:2 matching, 1:3 matching, averaging across repeated PSG nights, and matching on expanded diagnosis-type covariates, with the point estimate remaining within 0.06 of the primary value in each case. It attenuated, however, when psychiatric diagnoses and concurrent sleep medications were added to the propensity model. The E-value of 1.74 for the confidence interval bound nearest the null indicates that unmeasured confounding of modest strength, beyond the matched covariates, could account for the association. Reverse causation is also possible: children with intrinsically lower REM sleep may be more likely to be prescribed melatonin, generating an association that does not reflect a pharmacologic effect.
Sleep Onset Latency, WASO, and Sleep Efficiency
Reduced sleep onset latency is the most frequently cited effect of melatonin.26 WASO was significantly lower in the primary analysis but did not survive FDR correction; sleep onset latency and sleep efficiency showed no nominal difference. The absence of FDR-significant group differences in sleep onset latency, WASO, or sleep efficiency may reflect 3 factors: an outpatient prescription does not establish that melatonin was taken on the night of the PSG; dose, formulation, and timing are not captured; and in a clinical PSG, the laboratory environment, electrodes, and parental presence may dominate any small pharmacologic contribution to sleep onset.
Developmental Pattern and Subgroups
Age-stratified analyses showed a nonmonotonic pattern across childhood: the largest REM reduction appeared in middle childhood (6-12 years: Hedges g = −0.52 within 41 same-bin pairs) rather than in the youngest or oldest children. Neither test of the melatonin-by-age interaction was statistically significant (P = .11, linear regression; P = .25, per-pair-difference model). The linear-interaction signal is directionally consistent with developmental attenuation on average, but the 3 age groups do not fall into a steady decline, and the sample is too small to establish the true age pattern; larger studies designed to evaluate age-related differences are needed.
Clinical Interpretation
This analysis offers 2 messages for clinicians who prescribe melatonin to children. First, it did not detect differences in NREM sleep stages, respiratory parameters, arousal architecture, or periodic limb movements between melatonin users and matched nonusers; because the design is observational, this finding is informative but does not establish that melatonin has no effect on these parameters. Second, melatonin users showed a small reduction in REM sleep of uncertain origin and clinical significance, which may warrant awareness among clinicians treating children with neurodevelopmental disorders who may already have reduced baseline REM sleep.
Limitations
This study has several limitations. First, the cross-sectional design does not support causal inference; residual confounding by unmeasured variables, including severity of underlying sleep complaints and psychiatric symptom load not captured by diagnosis codes, cannot be excluded, as the observed attenuation under expanded adjustment supports this concern. Second, the binary ever-prescribed exposure indicator did not capture dose, timing, formulation, or duration; did not confirm melatonin was taken on the PSG night; and could not detect over-the-counter use in the control group—each of which could bias associations toward the null.27 Third, the cohort is a referred clinical population overrepresenting children with sleep-disordered breathing, obesity, neurodevelopmental disorders, and epilepsy15; generalization to community samples requires caution. Fourth, NCHSDB is a single-center dataset; replication is needed.15 Fifth, PSG captures a single laboratory night, which may not reflect habitual sleep patterns. Sixth, REM latency was scoreable in 316 of 342 pairs (26 studies recorded no REM epochs).
Conclusions
In this cross-sectional study of 684 propensity score–matched children, to our knowledge the largest PSG analysis of pediatric melatonin use, outpatient melatonin use was associated with a small reduction in REM sleep percentage and was not associated with measurable differences in NREM sleep stages, respiratory parameters, arousal architecture, or periodic limb movements. Whether the residual REM association reflects a pharmacologic effect of melatonin, residual confounding, reverse causation, or some combination cannot be determined from these cross-sectional data. Prospective studies with documented dose, timing, and adherence are needed to clarify the directionality and clinical significance of the observed REM association.
eMethods. Definitions of Subgroup and Sensitivity-Analysis Covariates
eFigure 1. Study Flow Diagram
eFigure 2. Age-Stratified Hedges g for Percentage of REM Sleep With Formal Interaction Test
eTable 1. Summary of Prespecified Sensitivity Analyses (Subgroup-Level Top 3 Effect Sizes)
eTable 2. Expanded Propensity Score Model Adding Psychiatric Diagnoses and Concurrent Sleep Medications (Post Hoc)
eTable 3. 1:k Matching Sensitivity (Post Hoc)
eTable 4. Mean-of-Repeated-PSGs Sensitivity (Post Hoc)
eTable 5. Diagnosis-Type Matching: Replacing the Count of ICD-10 Chapters With 11 Binary Diagnosis Flags (14 Covariates Total, Post Hoc)
eTable 6. Missingness Per Polysomnographic Outcome
eTable 7. Baseline Comparison of Children With and Without Complete REM-Latency Data
eTable 8. Developmental Gradient: OLS Interaction Model and Per-Pair-Difference Regression
Data Sharing Statement
References
- 1.Hartstein LE, Garrison MM, Lewin D, Boergers J, LeBourgeois MK. Characteristics of melatonin use among US children and adolescents. JAMA Pediatr. 2024;178(1):91-93. doi: 10.1001/jamapediatrics.2023.4749 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Esposito S, Laino D, D’Alonzo R, et al. Pediatric sleep disturbances and treatment with melatonin. J Transl Med. 2019;17(1):77. doi: 10.1186/s12967-019-1835-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Owens JA, Mindell JA. Pediatric insomnia. Pediatr Clin North Am. 2011;58(3):555-569. doi: 10.1016/j.pcl.2011.03.011 [DOI] [PubMed] [Google Scholar]
- 4.Morgenthaler TI, Owens J, Alessi C, et al. ; American Academy of Sleep Medicine . Practice parameters for behavioral treatment of bedtime problems and night wakings in infants and young children. Sleep. 2006;29(10):1277-1281. [PubMed] [Google Scholar]
- 5.Bruni O, Alonso-Alconada D, Besag F, et al. Current role of melatonin in pediatric neurology: clinical recommendations. Eur J Paediatr Neurol. 2015;19(2):122-133. doi: 10.1016/j.ejpn.2014.12.007 [DOI] [PubMed] [Google Scholar]
- 6.Gringras P, Gamble C, Jones AP, et al. ; MENDS Study Group . Melatonin for sleep problems in children with neurodevelopmental disorders: randomised double masked placebo controlled trial. BMJ. 2012;345:e6664. doi: 10.1136/bmj.e6664 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Malow B, Adkins KW, McGrew SG, et al. Melatonin for sleep in children with autism: a controlled trial examining dose, tolerability, and outcomes. J Autism Dev Disord. 2012;42(8):1729-1737. doi: 10.1007/s10803-011-1418-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Rossignol DA, Frye RE. Melatonin in autism spectrum disorders: a systematic review and meta-analysis. Dev Med Child Neurol. 2011;53(9):783-792. doi: 10.1111/j.1469-8749.2011.03980.x [DOI] [PubMed] [Google Scholar]
- 9.Diekelmann S, Born J. The memory function of sleep. Nat Rev Neurosci. 2010;11(2):114-126. doi: 10.1038/nrn2762 [DOI] [PubMed] [Google Scholar]
- 10.Marks GA, Shaffery JP, Oksenberg A, Speciale SG, Roffwarg HP. A functional role for REM sleep in brain maturation. Behav Brain Res. 1995;69(1-2):1-11. doi: 10.1016/0166-4328(95)00018-O [DOI] [PubMed] [Google Scholar]
- 11.Owens JA, Moturi S. Pharmacologic treatment of pediatric insomnia. Child Adolesc Psychiatr Clin N Am. 2009;18(4):1001-1016. doi: 10.1016/j.chc.2009.04.009 [DOI] [PubMed] [Google Scholar]
- 12.Monti JM. Serotonin control of sleep-wake behavior. Sleep Med Rev. 2011;15(4):269-281. doi: 10.1016/j.smrv.2010.11.003 [DOI] [PubMed] [Google Scholar]
- 13.Van Cauter E, Leproult R, Plat L. Age-related changes in slow wave sleep and REM sleep and relationship with growth hormone and cortisol levels in healthy men. JAMA. 2000;284(7):861-868. doi: 10.1001/jama.284.7.861 [DOI] [PubMed] [Google Scholar]
- 14.Dijk DJ. Regulation and functional correlates of slow wave sleep. J Clin Sleep Med. 2009;5(2)(suppl):S6-S15. doi: 10.5664/jcsm.5.2S.S6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Lee H, Li B, DeForte S, et al. A large collection of real-world pediatric sleep studies. Sci Data. 2022;9(1):421. doi: 10.1038/s41597-022-01545-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Zhang GQ, Cui L, Mueller R, et al. The National Sleep Research Resource: towards a sleep data commons. J Am Med Inform Assoc. 2018;25(10):1351-1358. doi: 10.1093/jamia/ocy064 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Berry RB, Albertario CL, Harding SM, et al. ; American Academy of Sleep Medicine . The AASM Manual for the Scoring of Sleep and Associated Events: Rules, Terminology and Technical Specifications. Version 2.5. American Academy of Sleep Medicine; 2018. [Google Scholar]
- 18.von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP; STROBE Initiative . The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet. 2007;370(9596):1453-1457. doi: 10.1016/S0140-6736(07)61602-X [DOI] [PubMed] [Google Scholar]
- 19.Forrest CB, McTigue KM, Hernandez AF, et al. PCORnet® 2020: current state, accomplishments, and future directions. J Clin Epidemiol. 2021;129:60-67. doi: 10.1016/j.jclinepi.2020.09.036 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Austin PC. An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivariate Behav Res. 2011;46(3):399-424. doi: 10.1080/00273171.2011.568786 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Gregory AM, Sadeh A. Annual Research Review: sleep problems in childhood psychiatric disorders–a review of the latest science. J Child Psychol Psychiatry. 2016;57(3):296-317. doi: 10.1111/jcpp.12469 [DOI] [PubMed] [Google Scholar]
- 22.Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc B. 1995;57(1):289-300. doi: 10.1111/j.2517-6161.1995.tb02031.x [DOI] [Google Scholar]
- 23.Hedges LV. Distribution theory for Glass’s estimator of effect size and related estimators. J Educ Stat. 1981;6(2):107-128. doi: 10.3102/10769986006002107 [DOI] [Google Scholar]
- 24.VanderWeele TJ, Ding P. Sensitivity analysis in observational research: introducing the E-value. Ann Intern Med. 2017;167(4):268-274. doi: 10.7326/M16-2607 [DOI] [PubMed] [Google Scholar]
- 25.Chinn S. A simple method for converting an odds ratio to effect size for use in meta-analysis. Stat Med. 2000;19(22):3127-3131. doi: 10.1002/1097-0258(20001130)19:22<3127::AID-SIM784>3.0.CO;2-M [DOI] [PubMed] [Google Scholar]
- 26.Zhdanova IV. Melatonin as a hypnotic: pro. Sleep Med Rev. 2005;9(1):51-65. doi: 10.1016/j.smrv.2004.04.003 [DOI] [PubMed] [Google Scholar]
- 27.Rothman KJ, Greenland S, Lash TL. Modern Epidemiology. 3rd ed. Lippincott Williams & Wilkins; 2008. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
eMethods. Definitions of Subgroup and Sensitivity-Analysis Covariates
eFigure 1. Study Flow Diagram
eFigure 2. Age-Stratified Hedges g for Percentage of REM Sleep With Formal Interaction Test
eTable 1. Summary of Prespecified Sensitivity Analyses (Subgroup-Level Top 3 Effect Sizes)
eTable 2. Expanded Propensity Score Model Adding Psychiatric Diagnoses and Concurrent Sleep Medications (Post Hoc)
eTable 3. 1:k Matching Sensitivity (Post Hoc)
eTable 4. Mean-of-Repeated-PSGs Sensitivity (Post Hoc)
eTable 5. Diagnosis-Type Matching: Replacing the Count of ICD-10 Chapters With 11 Binary Diagnosis Flags (14 Covariates Total, Post Hoc)
eTable 6. Missingness Per Polysomnographic Outcome
eTable 7. Baseline Comparison of Children With and Without Complete REM-Latency Data
eTable 8. Developmental Gradient: OLS Interaction Model and Per-Pair-Difference Regression
Data Sharing Statement
