Efficacy of Melatonin in Children With Postconcussive Symptoms: A Randomized Clinical Trial

We conducted a single-center, randomized, double-blind, placebo-controlled trial at Alberta Children’s Hospital. The rationale for the trial, design, and analysis plan have been published previously.²³  This study was approved by the university research ethics board. Enrollment occurred between February 2014 and April 2017. Changes to inclusion criteria (ie, lower age limit decreased from 13 to 8 years and time since last concussion decreased from 12 to 3 months) were made in December 2014 to improve recruitment. The trial was conducted in accordance with the standards of Good Clinical Practice. Potential participants were children seen in the emergency department with a medically diagnosed concussion and/or mTBI who consented to telephone follow-up. Concussion and/or mTBI was defined by using the American Academy of Neurology criteria.²⁴  Outcome assessments were performed just before treatment commenced, weekly during treatment, at the end of treatment (posttreatment), and again at 4 and 6 months postinjury. Follow-up was completed in October 2017.

Participants were enrolled by using a 2-step process: first by telephone at 2 to 4 weeks and then in person 2 weeks later. We enrolled children ages 8 to 18 years if they had PPCS and a ≥10-point increase in their total symptom score on the Post-Concussion Symptom Inventory (PCSI) postinjury when compared with their preinjury score (assessed at enrollment).²⁵  Children were ineligible if they had a significant medical or psychiatric history, a previous concussion within the last 3 months, persistent symptoms after a previous concussion, or a more severe TBI previously. Other exclusions included lactose intolerance, use of neuroactive drugs, and inability to complete questionnaires. All participants and their guardians provided written consent. At the time of enrollment, a standardized interview and medical examination were performed. Figure 1 outlines the trial details. An independent trial monitoring board periodically reviewed safety data. No interim analyses of efficacy were performed.

After enrollment, participants were randomly assigned by using a random-block-size design (block sizes 3, 6, and 9) to 3 parallel treatment groups with a 1:1:1 allocation: placebo, melatonin 3 mg, and melatonin 10 mg. The computer-generated randomization list was created and held by an external statistician. An independent pharmacist created sequentially numbered, identical treatment packages. Melatonin and placebo tablets were identical in appearance and flavor. All investigators, outcome assessors, parents, and children were blinded to treatment groups.

The study drug was taken sublingually one hour before sleep time at night for 28 days and was continued even if symptom resolution occurred. No restrictions were placed on the use of other medications. Participants were advised to avoid analgesia overuse, abstain from contact sports, perform light exercise, and gradually return to school. Compliance and adverse events were monitored weekly.

The primary outcome was the change in Post-Concussion Symptom Inventory–Youth (PSCI-Y) total score for youth self-report at the end of 28 days of treatment, which was calculated as pretreatment score minus end-of-treatment score. This standardized, 26-item questionnaire provides an overall rating of PPCS (total score range: 0–156). The PCSI-Y has 4 specific domains derived from factor analysis (physical, cognitive, emotional, and fatigue) and a high level of internal consistency and reliability (α = .92).^26,27 

Secondary outcomes were behavioral, cognitive, and sleep problems and functional impairment, which are common complaints among PPCS. The Post-Concussion Symptom Inventory–Parent (PCSI-P), a parent-proxy questionnaire, was completed (total score range: 0–104).²⁶  Health-related quality of life was assessed by using the Child Health Questionnaire (Parent Form 50 [parent]; child form 87 [child]).²⁸  Behavioral adjustment and everyday executive functioning were assessed by using the parent-report versions of the Behavior Assessment System for Children, Second Edition and the Behavior Rating Inventory of Executive Function, respectively. Scores for the Behavior Assessment System for Children, Second Edition and the Behavior Rating Inventory of Executive Function are age- and sex-adjusted T-scores (mean = 50; SD = 10). Neurocognitive ability was assessed by using a computerized assessment battery (CNS Vital Signs). The Neurocognition Index was used as a global score of neurocognition. The validity of test performance was assessed by using the Test of Memory Malingering (TOMM).²⁹  These measures have demonstrated validity in TBI and concussion.^30–33 

Sleep was assessed by using a wrist-worn accelerometer (Actiwatch² ) on the nondominant wrist for 5 to 7 days before and after treatment.³⁴  Actigraphy was used to measure sleep-activity patterns, including total sleep time, onset latency, efficiency, and wake after sleep onset (WASO) (amount of time). Epochs were set at 15 seconds.

Overnight urinary 6-sulphatoxymelatonin (aMT6s), the major metabolite of melatonin, was analyzed before treatment, midtreatment, and at the end of treatment by using a solid-phase enzyme-linked immunosorbent assay (Cat#RE54031; IBL International, Hamburg, Germany). Individual levels were converted to a ratio normalized by using urinary creatinine concentration.³⁵ 

Data from a previous epidemiological study¹  were used to calculate a Reliable Change Index.³⁶  A 10-point change (SD = 14.7) in PCSI score indicated a reliable change for subjects who have PPCS at one month and is similar to previous reports.^37,38  On the basis of medical record review, the relevance of 10-point improvement in PCSI score is a difference of ∼30% in overall clinical symptoms. Clinically meaningful examples of this would be a 50% reduction in posttraumatic headache burden (ie, daily headaches becoming intermittent and less severe or a 50% decrease in analgesic use), returning to full-time education with modifications, or a significant decrease in fatigue with associated increased participation in activities. The sample size (33 per group) was calculated a priori and powered (80%) to detect a 10-point (SD = 14.7) PCSI-Y score change at a significance level of α = .05.¹ 

Masked analysis in the differences in the change in PCSI-Y and PCSI-P scores between groups was performed by using the nonparametric Kruskal-Wallis test (KW) because parametric assumptions were not met. Bias-corrected and accelerated bootstrapped 95% confidence intervals (CIs) (1000 samples) for means and medians were computed.³⁹  All analyses were performed with an intent-to-treat principle. Time to symptom resolution (as defined by a PCSI-Y total score equal to or less than preinjury) was examined using Kaplan-Meier survival curves. Hierarchical regression was used to predict symptom improvement by entering sleep efficiency (step 1) and treatment group (step 2) to determine how predictive treatment was above and beyond any effects of sleep efficiency. A secondary efficacy per-protocol analysis was performed, excluding participants with protocol violations. The per-protocol analysis included only participants who missed no more than 2 consecutive doses or <5 doses in total (n = 92).

Secondary outcomes were analyzed by using repeated-measures analysis of variance or the KW when assumptions were violated (change in sleep parameters) with significance α set to .01 to allow for multiple comparisons. Frequency and severity of side effects were evaluated by using Fisher’s exact test. Subsequent sensitivity analyses using quantile regression were performed to examine any effect of outliers, protocol deviations, and compliance on the primary outcome. All assumptions were checked and sufficed. Statistical analyses were performed by using IBM SPSS version 24 (IBM SPSS Statistics, IBM Corporation) and Stata release 14 (Stata Corp, College Station, TX).

The funding sources had no input in the study design, data analysis, interpretation, report generation, or submission for publication.

Ninety-nine participants (mean age = 13.8 years; SD = 2.6; 58% girls) were enrolled and randomly assigned to the placebo,³³  melatonin 3-mg,³³  or melatonin 10-mg³³  groups. Enrollment details are displayed in Fig 1. Two participants withdrew after random assignment without starting treatment, and 3 were lost to follow-up. Ninety-four participants completed assessments immediately after intervention, and 92 completed the trial according to protocol.

Participant demographics, injury details, and clinical characteristics are summarized in Table 1. The majority of injuries (65%) were sport related, and 95% were witnessed. Treatment started at a mean of 38 days (SD = 6) postinjury. The median PCSI-Y total score before treatment was 36 (range: 6–127); Supplemental Table 5 presents details of symptom complaints.

There was, on average, a decrease in scores in all groups over time, but there was no significant difference between the groups (Table 2). Participants reported an overall mean decrease in PCSI-Y score of 21 points (95% CI: 16 to 27; P < .001) after 28 days of treatment. Parents also reported an overall mean decrease in PCSI-P score of 13 points (95% CI: 10 to 16; P < .001). Placebo, melatonin 3-mg, and melatonin 10-mg groups showed a similar median change in PCSI-Y scores (16, 16, and 27, respectively; KW = 2.024; P = .36) and PCSI-P scores (12.5, 15, and 9.5, respectively; KW = 1.692; P = .43; Fig 2).

Kaplan-Meier survival curves in Fig 3 illustrate no significant group differences in the probability of symptom survival during treatment or 6-month follow-up. The Cox proportional hazards model showed no significant difference in hazard ratios of symptoms returning to baseline between the placebo and melatonin 3-mg (1.00 [95% CI: 0.49 to 2.06; P = .99]) groups and the melatonin 10-mg group (1.11 [95% CI: 0.54 to 2.29; P = .77]). The proportionality assumption was not violated. Regression revealed no effect of melatonin after any effect of sleep efficiency (pretreatment) was considered (F[1,84] = 0.158; P = .69). Per-protocol analysis did not show any evidence of a favorable effect of melatonin on PCSI-Y change (KW = 2.811; P = .24).

Paired sleep actigraphy parameters before and after treatment were available for 66 participants: placebo,²³  melatonin 3 mg,¹⁶  and melatonin 10 mg.²⁵  Missing data were due to watch unavailability,¹⁶  watch malfunction,¹⁰  and loss of watch.²  No group differences on actigraphy sleep parameters were apparent after treatment (Table 3).

Cognitive and behavioral outcomes are reported in Table 4. Five participants displayed noncredible effort on the TOMM and were excluded from the analysis of cognitive function. Groups did not differ on changes in cognitive testing, executive function, or internalizing problems after treatment. Externalizing problems (hyperactivity, conduct, and aggression) were significantly different across groups after treatment (F[2, 84] = 4.928; P = .009). Post hoc paired comparisons demonstrated decreased externalizing problems in the 3-mg melatonin group (P = .008). This had a large effect size (−0.797 [95% CI: −1.327 to −0.266]) compared with the placebo.

Youth and parents in all groups reported improved quality of life at the end of treatment (Supplemental Table 6). Mean parental rating of physical functioning significantly improved from 34.55 to 45.46 (change 95% CI: 8.12 to 13.70; t[86] = 7.77; P < .001), and psychosocial functioning increased from 42.07 to 47.50 (change 95% CI: 3.78 to 7.99; t[86] = 5.55; P < .001). Changes in quality of life significantly correlated with changes in PCSI-P total scores (Supplemental Fig 4). However, the treatment groups did not differ significantly.

Pretreatment overnight urinary aMT6s levels were not different between groups (mean = 54.59; SD = 33.40; F[2,67] = 0.87; P = .42). Forty-two participants missed at least 1 dose, and 5 participants missed 5 doses or more. Midtreatment urinary aMT6s levels increased in the melatonin groups only: placebo (mean = 36.85; SD = 20.94), melatonin 3 mg (mean = 2595.82; SD = 2891.34), and melatonin 10 mg (mean = 9889.22; SD = 11 372.35; F[2,56] = 9.256; P < .001). Sensitivity analyses revealed no effect of compliance or urinary melatonin levels on outcome.

Thirty-two adverse events were reported in 28 participants: placebo (n = 8), melatonin 3 mg (n = 13), and melatonin 10 mg (n = 11). One participant had a serious adverse event (appendicitis) unrelated to the study drug. Eight events involved a known melatonin side effect, and 5 were potentially related (Supplemental Table 7). Ten events were associated with a mild functional impact. The frequency (χ² = 1.755; P = .42) and severity (χ⁶ = 6.619; P = .36) of adverse events did not differ between groups.

In children with PPCS, the administration of melatonin at 4 weeks postinjury for 28 days did not significantly improve postconcussion symptoms compared with a placebo nor was there any effect of melatonin apparent on cognitive or health-related quality of life outcomes. Similarly, most measures of behavior were unchanged. We observed wide CIs in the primary outcome measure, however, suggesting that the sample size was insufficient to definitively conclude an absence of effect. Interestingly, parents reported fewer externalizing problems (mainly decreased hyperactivity) in those participants treated with melatonin, as reported previously in children with neurodevelopmental disability.⁴⁰  The mean change of 2.7 T-scores seen in our study, however, is not likely to be clinically significant. Melatonin is often recommended in the management of pediatric PPCS, and both melatonin and a derivative, ramelteon, have been reported to improve sleep complaints in moderate or severe adult TBI.^{12,19,21,41–43}  Pretreatment sleep actigraphy parameters in our study, however, were not predictive of PPCS improvement with melatonin treatment.

This study has several strengths. Although many studies have examined recovery from concussion, few have systematically examined treatments for PPCS. Previous trials have been small, examined a variety of therapies,^2,13–15  and compared interventions to routine care or no treatment and so are susceptible to attribution bias.⁴⁴  This is the first randomized placebo-controlled trial of a pharmacologic intervention. It is also one of the largest studies evaluating treatment in children with PPCS. The results are generalizable to typically developing children with mTBI from a variety of etiologies, including sport-related concussion. Children were enrolled and carefully phenotyped at a similar and tight time frame postinjury. As well as recording postconcussive symptoms, we also examined other traditional neurocognitive and behavioral outcomes that may be more robust than symptom ratings and allow for comparisons between treatment groups and other populations.

This trial provides useful information about PPCS in children. Participants at enrollment had a high symptom burden and significant impairments in their quality of life. Despite this, neurocognitive, executive, and behavioral function remained well within normal ranges. Participants performed well on symptom validity testing, and only 5% failed to demonstrate credible effort on the TOMM, suggesting that aberrant psychological responses to injury, if present, were not likely to have affected neuropsychological performance.^45,46 

A high proportion of participants displayed significant clinical improvement during the intervention period. This could be due to the natural history of PPCS, comprehensive standardized care, or a placebo effect. Given our previous experience in studies involving similar populations, this preponderance for improvement is unlikely to be due to natural history alone^1,47  but could be artificially inflated by the exclusion of children with a significant psychiatric history.⁴⁸  More likely, however, the potential treatment effect is due to the supportive multidisciplinary care model employed in the brain injury clinic and is congruent with outcomes in a collaborative care model for the management of PPCS.⁴⁹  Although neither of these factors should affect the results of this trial, the placebo effect could lead to an underestimation of the effect of melatonin.^50,51 

Melatonin doses were chosen to stay within common clinical practice parameters to minimize risks to participants and achieve supraphysiological levels sufficient to target both receptor-mediated and subcellular processes.⁵²  We achieved this goal, observing excellent tolerability and treatment compliance while demonstrating supraphysiological overnight urinary melatonin levels. The observed melatonin levels and our sensitivity analyses, adjusting for compliance, suggested that treatment failure was unlikely to be due to insufficient melatonin dosing at night.

The lack of an observable effect of melatonin in our study was somewhat surprising, but several possibilities could explain this outcome. Posttraumatic oxidative stress peaks between 3 and 5 days postinjury, although its temporal course and role in chronic mTBI have not been fully elucidated.^53,54  Melatonin may need to be used within the first few days of injury, when oxidative stress is at its highest, to capitalize on its antioxidant properties and reduce the biological consequences of mTBI.^55,56  The excellent tolerability and side-effect profile of melatonin seen in this study provides reassurance that it could be administered early at low risk to the patient. Conversely, the antiinflammatory effects of melatonin may be more evident with longer treatment durations if persistent neuroinflammation is playing a role in PPCS.^1,5  Future studies are needed to evaluate these hypotheses.

This study has some limitations. The sample size for this study was determined on the basis of 80% power, resulting in the 20% chance that our study was underpowered. The wide CIs found for our primary outcomes indicate that our sample size was too small. Therefore, the results of this study must be interpreted with caution and need to be replicated in a larger sample. Obtaining retrospective reports of preinjury symptoms at enrollment could have introduced a “good-old-days” bias^57,58  that may result in recovery rates being underestimated; however, this should not have affected any response to the intervention. To minimize bias due to attrition, we used an intention-to-treat analysis, imputing missing data using the last observation carried forward approach. Although this technique could introduce bias into the analysis, because the primary outcome data were missing in only 3 participants (1 per treatment group), it is unlikely to have affected our results. A specific pediatric sleep questionnaire would have improved the characterization of sleep disturbances.

The administration of melatonin at 4 weeks postinjury did not significantly improve postconcussion symptoms in children with PPCS compared with a placebo. Seventy-eight percent of children with PPCS demonstrated significant recovery between 1 and 3 months postinjury. The evidence from this randomized controlled trial does not support the use of melatonin for the treatment of PPCS during this time period postinjury.

aMT6s

urinary 6-sulphatoxymelatonin

CI

confidence interval

KW

Kruskal-Wallis test

mTBI

mild traumatic brain injury

PCSI

Post-Concussion Symptom Inventory

PCSI-P

Post-Concussion Symptom Inventory–Parent

PCSI-Y

Post-Concussion Symptom Inventory–Youth

PPCS

persistent postconcussive symptoms

TBI

traumatic brain injury

TOMM

Test of Memory Malingering

WASO

wake after sleep onset