A Systematic Review and Network Meta-Analysis of Randomized Controlled Trials Evaluating the Evidence Base of Melatonin, Light Exposure, Exercise, and Complementary and Alternative Medicine for Patients with Insomnia Disorder

Insomnia is a prevalent disorder and it leads to relevant impairment in health-related quality of life. Recent clinical guidelines pointed out that Cognitive-Behavior Therapy for Insomnia (CBT-I) should be considered as first-line intervention. Nevertheless, many other interventions are commonly used by patients or have been proposed as effective for insomnia. These include melatonin, light exposure, exercise, and complementary and alternative medicine. Evaluation of comparable effectiveness of these interventions with first-line intervention for insomnia is however still lacking. We conducted a systematic review and network meta-analysis on the effects of these interventions. PubMed, PsycInfo, PsycArticles, MEDLINE, and CINAHL were systematically searched and 40 studies were included in the systematic review, while 36 were entered into the meta-analysis. Eight network meta-analyses were conducted. Findings support effectiveness of melatonin in improving sleep-onset difficulties and of meditative movement therapies for self-report sleep efficiency and severity of the insomnia disorder. Some support was observed for exercise, hypnotherapy, and transcranial magnetic resonance, but the number of studies for these interventions is still too small. None of the considered interventions received superior evidence to CBT-I, which should be more widely disseminated in primary care.


Introduction
Insomnia is one of the most frequent psychophysiological disorders [1] and it leads to clinically relevant impairment in diverse areas of health-related quality of life [2] as well as to high societal costs [3]. The Diagnostic and Statistical Manual of Mental Disorders (DSM 5) [4] defines insomnia as 1.
Melatonin is generally not recommended because evidence shows low efficacy.

2.
Light therapy and physical activity may be useful as adjunct therapy, but recommendation is weak due to low-quality evidence. 3.
Complementary and alternative interventions are not recommended because of poor evidence.
In the area of complementary and alternative interventions, several treatments for insomnia have been suggested, including acupuncture, acupressure, aromatherapy, reflexology, homeopathy, meditative movement therapies, moxibustion, music therapy, and yoga [40].
Updating systematic evaluation of comparable efficacy of these treatments is needed for two reasons. First, as patients often recur to these therapies, evidence-based data and comparisons with current recommended interventions is a priority. Second, although the efficacy of CBT-I is well proved (for a review see [23]), still many patients retain sleep disturbance after CBT-I (see, e.g., in [41]). Thus, it is necessary to identify treatments which may be alternative to CBT-I for those patients who do not respond to the main clinical procedures.
The aim of the present work was to provide a systematic review on the comparable efficacy of melatonin, light exposure therapy, exercise, and complementary and alternative medicine for insomnia disorder with respect to CBT-I, hypnotic medication or placebo/waiting list control conditions. We considered all age ranges including adult and pediatric populations. Acupuncture/acupressure was not included in our list, because of the extensive work previously published through the Cochrane Library [38,39].
In the context of evidence-based medicine, network meta-analysis aims to compare a number of available treatments for a given diagnosis by combining direct and indirect evidence on treatment effects based on a common comparator [42,43]. Network meta-analysis is a valid statistical method which allows for simultaneous analysis of both direct and indirect comparisons among multiple treatments across multiple studies. This method has advantages over pairwise meta-analysis which allows comparison of only two interventions, including (a) borrowing strength from indirect evidence to compare all treatments, (b) estimating comparative effects that have not been investigated head-to-head in randomized clinical trials, and (c) comparing between more than two interventions for one condition which informs clinical practice. To explain the concept underpinning network meta-analysis, and specifically the meaning of direct and indirect evidence, we suppose we compare two active treatments A and B and a control condition C. Given direct evidence from studies regarding the difference of treatment effects for A and C and evidence regarding the difference of treatment effects for B and C from other studies, these studies also provide indirect evidence for treatments A and B. If there is direct evidence for A vs. B, direct and indirect evidences are combined to get the most precise estimate of treatment difference. Therefore, the aim of network meta-analysis is to estimate the treatment differences and associated standard errors combining direct and indirect evidence [44].

Methods
The present study was conducted according to the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines for systematic review incorporating a network meta-analysis [45] (see Document S1: PRISMA check-list). The study protocol was reviewed and approved for funding from the German Ministry for Education and Research (Bundesministerium für Bildung und Forschung, BMBF, 01KG1111).

Search Procedure
We used several strategies to identify our final study sample. First, we conducted computer-based searches using PubMed, PsycInfo, PsycArticles, MEDLINE, and CINAHL using the following keywords, capturing the title and the abstract: 'insomnia' AND 'ayurveda' OR 'chelation' OR 'diet based therapy' OR 'energy healing therapy' OR 'exercise' OR 'folk medicine' OR 'homeopathic' OR 'hypnosis' OR 'light exposure' OR 'massage' OR 'meditation' OR 'melatonin' OR 'music therapy' OR 'natural herbs' OR 'naturopathy' OR 'qi gong' OR 'reiki' OR 'tai chi' OR 'transcranial magnetic stimulation' OR 'valerian' OR 'vitamin' OR 'yoga'. The search was conducted from January 1994, the date in which the DSM-IV [46] was published and September 2019. All abstracts found through keywords search were saved in a project in CITAVI software. CITAVI is a software for references management and organization (https://www.citavi.com). All abstracts were screened by four independent raters coordinated by the second author (Z.B.) collaborating together with the first author (C.B.) whenever the inclusion or exclusion of one study was doubtful. Similar procedures were followed for full-text screening.
Second, we expanded our search through identifying further studies from the references of the screened full-texts.
Third, we contacted authors in the field through the mailing list of the European Insomnia Network (a group of experts in insomnia promoted by the European Sleep Research Society (ESRS)), to obtain further published studies and, if needed, to obtain additional information.
Fourth, unpublished data were searched through contact with experts in the field (as described in the previous paragraph) and through screening of all abstracts of poster and oral presentations to the

Eligibility Criteria
We included all clinical studies evaluating melatonin, light exposure therapy, exercise, and complementary and alternative medicine interventions for insomnia based on the following eligibility criteria and the Population, Intervention, Comparison, Outcomes and Study design (PICOS) approach [45]. PICOS is a structured approach for framing questions which assess five components and may help facilitate the process: the patient population or the disease being addressed (P); the interventions of exposure (I); the comparator group (C); the outcome or endpoint (O); and study design (S): (1) Population: Individuals with insomnia disorder of all ages (including adult and pediatric populations) and of both gender with or without any mental, somatic or sleep comorbidity. (2) Intervention: Experimental interventions including one of the following administered alone (i.e., not in combination with recommended therapies for insomnia): ayurveda, chelation, diet-based therapy, energy healing therapy, exercise, folk medicine, homeopathy, hypnosis, light exposure, massage, meditation, melatonin, music therapy, natural herbs, naturopathy, qi gong, reiki, tai chi, transcranial magnetic stimulation, valerian, vitamin, and yoga.

Data Extraction
Descriptive information related to sample demographic and clinical characteristics, clinical aspects of insomnia disorder, intervention characteristics and methodological aspects of the study design was collected. Descriptive information included (a) age and sex of participants in the experimental and control interventions; (b) diagnostic tool/s according to which participants met the definition of insomnia disorder; (c) past medical conditions; (d) information about mental and somatic comorbidities; (e) number of treatment sessions and duration; (f) treatment setting and conducting personnel; (g) type of control intervention; (h) insomnia duration.

Assessment of Risk of Bias
Methodological variables were assessed through the Cochrane Collaboration's tool for assessing risk of bias in randomized controlled trials [50]. In this tool five main domains are assessed with a judgement of high, low or unclear risk.
(1) Selection bias. This domain refers to systematic differences between baseline characteristics and covers two parts: (1) Did the investigators use a random sequence generation process? (2) Could intervention allocations have been foreseen in advance of enrolment? (2) Performance bias. This domain judges whether participants and personnel were blinded. Because blinding therapists and patients is not desirable in some form of interventional studies (such as psychotherapy or mindfulness or yoga), performance biases for these types of studies was systematically scored as "low risk". (3) Detection bias.
This domain refers to whether outcome assessors are aware of intervention assignments. (4) Attrition bias. This domain refers to systematic differences between groups in withdrawals from a study. Amount, nature, and handling of incomplete outcome data are evaluated. (5) Reporting bias. This domain refers to selective outcome reporting. For each included clinical study a search was conducted to find registered protocols in order to check the consistency between the planned and the reported analyses.
Each study was rated by two independent raters (Z.B. and V.B.) as at low risk (1), moderate risk (2), and high risk (3) with respect to five areas of methodological risk: randomization and allocation methods (selection bias), blinding of participants and personnel (performance bias) and of outcome assessors (detection bias), incomplete outcome data (attrition bias), and selective reporting (reporting bias). For selection bias, two different scores were applied for randomization and allocation methods. Divergences were discussed with the first (C.B.) and the fourth (F.B.) authors.

Statistical Analyses
Analyses were planned for an application for funding by the German Ministry for Education and Research (Bundesministerium für Bildung und Forschung, BMBF, 01KG1111). The plan was structured together with a biostatistician with excellent expertise in network meta-analyses (G.R.).
Data were extracted both for sleep and insomnia related variables as well as for daytime symptoms. Data are available in the online Supplementary Materials online (Document S2: Data extraction). Nevertheless, neither statistical analysis nor qualitative summary was conducted for data related to daytime symptoms, as these were too few and too different between studies.
Eight network meta-analyses were conducted considering experimental and control interventions categories as explained above (Data extraction) for the following categories of outcomes.
(1) Self-reported sleep efficiency: defined as a sleep efficiency index from sleep diaries or, if this was not reported, as sleep quality perception from sleep diaries; (2) Sleep efficiency measured through physiological indices: defined as sleep efficiency index measured by polysomnography, or, if this was not reported, by actigraphy; (3) Self-reported daytime sleepiness: defined as total score of the Epworth Sleepiness Scale [51], or, if this was not available, of the Children's Sleep Habits Scale (daytime symptoms which includes a scale of daytime sleepiness [52]); (4) Self-reported severity of sleep problems: defined as total score from Insomnia Severity Index [53], or, if this was not available, from the Pittsburgh Sleep Quality Index [54]. (5) Self-reported sleep onset latency: defined as sleep onset latency measured through sleep diaries; (6) Sleep onset latency measured through physiological indices: defined as sleep onset latency measured by polysomnography or, if this was not reported, by actigraphy. (7) Self-reported wake time during the night: defined as wake after sleep onset latency measured through sleep diaries; (8) Wake time during the night measured through physiological indices: defined as wake after sleep onset latency measured by polysomnography or, if this was not reported, by actigraphy.
Meta-analytic calculations were performed with the statistical software package R [55] (http: //www.Rproject.org/). A frequentist network meta-analysis was performed using the R-package "netmeta" [56,57]. Through network meta-analysis it is possible to graphically represent the comparisons included in the network. The function "netgraph" was used to visualize a net-graph evidencing the geometry of the network. The network graph consists of nodes that represent the treatments and edges that represent direct evidence, i.e., treatments that are directly compared in at least one study.
The width of those lines is based on sample size for those comparisons. Finally, shadows indicate multi-arm studies.
Second, we calculated effect sizes as standardized mean differences (Cohens'd) between post-treatment assessment and baseline. We referred to data of participants who completed the post-treatment assessment. All classes of interventions were compared against placebo/waiting list interventions because this category of control treatment was used most frequently. A random-effects model was used because of expected considerable heterogeneity between studies (e.g., comorbidities, different treatment variables, and different populations).
To test the heterogeneity of the network, Cochran's Q and Higgin's I 2 were calculated. Cochran's Q is computed as a weighted sum of squared differences between single study effects and the pooled effect across studies. Significant values indicate a high level of heterogeneity between studies that needs to be further investigated. Higgin's I 2 calculates the variability in effect estimates that is due to between-study heterogeneity rather than due to chance. Low percentages of I 2 are indicative of low heterogeneity while percentages over 75% represent considerable levels of heterogeneity [58].
In network meta-analyses, net heat plots can be plotted to investigate potential sources of heterogeneity [59]. These are graphical tools that represent changes in heterogeneity due to relaxing the consistency assumption for single designs in a matrix visualization. In the matrix, areas of the grey squares indicate the contribution of pooled direct evidence of each single design in the column to each network estimate in the row. This tool also graphically represents changes in Cochran's Q statistic due to relaxing the consistency assumption for single designs in matrix visualization. "Hot spots" of inconsistency are indicated by red colors while blue colors indicate consistency.
For psychological/behavioral interventions, number of sessions ranged from a minimum of 1 [84] to a maximum of 117 [86], with a range duration of the all experiments from a minimum of 1 day [84], to a maximum of 26 weeks [86]. For other interventions duration ranged from a minimum of 1 week with a dose of once a day [94], to a maximum of 6 months with a dose of once a day [64].
J. Clin. Med. 2020, 9, x FOR PEER REVIEW 8 of 27 Figure 1 illustrates the search flow of the studies included in the present systematic review and network meta-analysis. Literature search between January 1994 -September 2019 yielded 4589 abstracts (EBSCO n= 1686; PubMed n= 1265; Web of Science n= 1628). After removing duplicates and adding articles found through search in other sources a total of 2474 studies were found. Of these, 40 studies were included in the systematic review while 36 were entered in the network meta-analysis (four studies were excluded because data were not available). Details are shown in Figure 1.   Average insomnia duration ranged from 1 month to 11.3 years, fifteen studies did not report this information. Descriptive information on sample characteristics and outcome measure is reported in Table 1. Further information on study characteristics is reported in the online supplemental material Document S3: Study characteristics. Figure 2 illustrates the studies' evaluation for risk of bias. In supplemental material online Document S4 the risk of bias evaluation is reported in detail. Only two studies received two "High risk" judgments [67,79] in detection bias and attrition bias and selective reporting, and not a single study received all "Low risk" judgment. Furthermore, two studies received all "Unclear" risk judgment [91,92].

Risk of Bias
J. Clin. Med. 2020, 9, x FOR PEER REVIEW 13 of 27 Figure 2 illustrates the studies' evaluation for risk of bias. In supplemental material online Document S4 the risk of bias evaluation is reported in detail. Only two studies received two "High risk" judgments [67,79] in detection bias and attrition bias and selective reporting, and not a single study received all "Low risk" judgment. Furthermore, two studies received all "Unclear" risk judgment [91,92].

Network Meta-Analysis
Eight network meta-analyses were conducted considering four categories of outcomes (explained in the description of data-analyses): (1) Self-reported sleep efficiency; (2) Sleep efficiency measured through physiological indices; (3) Self-reported sleepiness; (4) Self-reported severity of sleep problems; (5) Self-reported sleep onset latency; (6) Sleep onset latency measured through

Network Meta-Analysis
Eight network meta-analyses were conducted considering four categories of outcomes (explained in the description of data-analyses): (1) Self-reported sleep efficiency; (2) Sleep efficiency measured through physiological indices; (3) Self-reported sleepiness; (4) Self-reported severity of sleep problems; (5) Self-reported sleep onset latency; (6) Sleep onset latency measured through physiological indices; (7) Self-reported wake time during the night; (8) Wake time during the night measured through physiological indices. Extracted data are given in Document S2: Data extraction.

Self-Reported Sleep Efficiency
Sixteen studies measured subjective sleep quality. The network consisted of 11 interventions including experimental and control conditions, based on 18 pairwise comparisons and 10 different designs. Figure 3 presents the netgraph and the forest plot presenting the effects of all available classes compared with placebo/waiting list.

Self-Reported Sleep Efficiency
Sixteen studies measured subjective sleep quality. The network consisted of 11 interventions including experimental and control conditions, based on 18 pairwise comparisons and 10 different designs. Figure 3 presents the netgraph and the forest plot presenting the effects of all available classes compared with placebo/waiting list.
A large significant effect could be detected for CBT-I ( Cochran's Q and Higgins I 2 tests revealed very high heterogeneity between studies (tau^2 = 0.3396; I^2 = 95.7%). The net heat plot (see Supplementary Materials online Document S5) suggests that the designs that are together responsible for the majority of between-design heterogeneity are the following ones; (1) CBT-I vs. meditative movement therapies, (2) CBT-I vs. meditative movement therapies which is involved in a three-arm study: CBT-I, meditative movement therapies, and Placebo. Furthermore, the net heat plot suggests that the evidence for the treatment comparison CBT-I vs. meditative movement therapies from the designs CBT-I vs. meditative movement therapies (design from the three-arm study: CBT-I, meditative movement therapies, and Placebo) and CBT-I vs. Placebo (design from the three-arm study: CBT-I, meditative movement therapies, and Placebo) is inconsistent with the other evidence.  Cochran's Q and Higgins I 2 tests revealed very high heterogeneity between studies (tauˆ2 = 0.3396; Iˆ2 = 95.7%). The net heat plot (see Supplementary Materials online Document S5) suggests that the designs that are together responsible for the majority of between-design heterogeneity are the following ones; (1) CBT-I vs. meditative movement therapies, (2) CBT-I vs. meditative movement therapies which is involved in a three-arm study: CBT-I, meditative movement therapies, and Placebo. Furthermore, the net heat plot suggests that the evidence for the treatment comparison CBT-I vs. meditative movement therapies from the designs CBT-I vs. meditative movement therapies (design from the three-arm study: CBT-I, meditative movement therapies, and Placebo) and CBT-I vs. Placebo (design from the three-arm study: CBT-I, meditative movement therapies, and Placebo) is inconsistent with the other evidence.

Sleep Efficiency Measured Through Physiological Indices
Eighteen studies measured sleep quality through physiological indices The network consisted of 10 interventions including experimental and control conditions, based on 24 pairwise comparisons and 12 different designs. Figure 4 presents the netgraph and the forest plot presenting the effects of all available classes compared with placebo. melatonin which is involved in a three-arm study, (2) CBT-I vs. placebo which is involved in a three-arm study and (3) Melatonin vs. placebo. Moreover, the net heat plot indicates "hot spots" of inconsistency (see the designs in red color that are not on the top-left to bottom-right diagonal, e.g., the evidence for the treatment comparison "melatonin vs. placebo" from the designs "Melatonin vs. placebo" and "CBT-I vs. melatonin" from the three-arm study "CBT-I, melatonin, placebo" is inconsistent with the other evidence).  No significant result could be detected. Cochran's Q and Higgins I 2 tests revealed very high heterogeneity between studies (tauˆ2 = 0.2759; Iˆ2 = 87.9%). The net heat plot (see Document S5) suggests that the designs that are together responsible for the majority of between-design heterogeneity are the following ones; (1) CBT-I vs. melatonin which is involved in a three-arm study, (2) CBT-I vs. placebo which is involved in a three-arm study and (3) Melatonin vs. placebo. Moreover, the net heat plot indicates "hot spots" of inconsistency (see the designs in red color that are not on the top-left to bottom-right diagonal, e.g., the evidence for the treatment comparison "melatonin vs. placebo" from the designs "Melatonin vs. placebo" and "CBT-I vs. melatonin" from the three-arm study "CBT-I, melatonin, placebo" is inconsistent with the other evidence). Cochran's Q and Higgins I 2 tests revealed very high heterogeneity between studies (tau^2 = 0.5109; I 2 = 94.4%). The net heat plot (see Document S5) suggests that the designs that are together responsible for the majority of between-design heterogeneity are the following ones; (1) meditative movement therapies vs. placebo, (2) exercise vs. waitlist, and (3) Placebo vs. rTMS. Again, the net heat plot indicates "hot spots" of inconsistency (see the designs in red color that are not on the top-left to bottom-right diagonal).  Cochran's Q and Higgins I 2 tests revealed very high heterogeneity between studies (tauˆ2 = 0.5109; I 2 = 94.4%). The net heat plot (see Document S5) suggests that the designs that are together responsible for the majority of between-design heterogeneity are the following ones; (1) meditative movement therapies vs. placebo, (2) exercise vs. waitlist, and (3) Placebo vs. rTMS. Again, the net heat plot indicates "hot spots" of inconsistency (see the designs in red color that are not on the top-left to bottom-right diagonal).

Daytime Sleepiness
Six studies measured daytime sleepiness. The network consisted of 5 interventions including experimental and control condition, based on 8 pairwise comparisons and 5 different designs. Figure 6 presents the netgraph and the forest plot presenting the effects of all available classes compared with placebo. J. Clin. Med. 2020, 9, x FOR PEER REVIEW 17 of 27

Daytime Sleepiness
Six studies measured daytime sleepiness. The network consisted of 5 interventions including experimental and control condition, based on 8 pairwise comparisons and 5 different designs. Figure 6 presents the netgraph and the forest plot presenting the effects of all available classes compared with placebo.
Significant results could be detected for melatonin only (d = −0.27, 95% CI: [ −0.48; −0.06]). Cochran's Q and Higgins I 2 tests revealed low heterogeneity between studies (tau^2 =0; I^2 = 0%). The net heat plot (see Document S5) shows no relevant between-design heterogeneity and does not indicate "hot spots" of inconsistency.    Cochran's Q and Higgins I 2 tests revealed low heterogeneity between studies (tauˆ2 =0; Iˆ2 = 0%). The net heat plot (see Document S5) shows no relevant between-design heterogeneity and does not indicate "hot spots" of inconsistency.

Self-Reported Sleep Onset Latency
Fourteen studies measured self-reported sleep onset latency. The network consisted of 9 interventions including experimental and control condition, based on 16 pairwise comparisons and 8 different designs. Figure 7 presents the netgraph and the forest plot presenting the effects of all available classes compared with placebo.

Daytime Sleepiness
Six studies measured daytime sleepiness. The network consisted of 5 interventions including experimental and control condition, based on 8 pairwise comparisons and 5 different designs. Figure 6 presents the netgraph and the forest plot presenting the effects of all available classes compared with placebo.

Sleep Onset Latency Measured Through Physiological Indices
Nineteen studies measured sleep onset latency through physiological indices. The network consisted of 9 interventions including experimental and control conditions, based on 25 pairwise comparisons and 10 different designs. Figure 8 presents the netgraph and the forest plot presenting the effects of all available classes compared with placebo.

Sleep Onset Latency Measured Through Physiological Indices
Nineteen studies measured sleep onset latency through physiological indices. The network consisted of 9 interventions including experimental and control conditions, based on 25 pairwise comparisons and 10 different designs. Figure 8 presents the netgraph and the forest plot presenting the effects of all available classes compared with placebo.

Self-Reported Wake During the Night
Eight studies measured self-reported wake during the night. The network consisted of 8 interventions including experimental and control condition, based on 10 pairwise comparisons and 7 different designs. Figure 9 presents the netgraph and the forest plot presenting the effects of all available classes compared with placebo.
No significant results could be observed. Cochran's Q and Higgins I2 tests revealed very high heterogeneity between studies (tau^2 =4.1301; I^2 = 99.4%). The net heat plot (see Document S5) suggests that the designs that are together responsible for the majority of between-design heterogeneity are the following ones; (1) CBT-I vs. meditative movement therapies and (2) CBT vs. Meditative (from the three-arm study: CBT:Meditative:Placebo/WL).  Cochran's Q and Higgins I 2 tests revealed high heterogeneity between studies (tauˆ2 =0.1446; I 2 = 78.8%). The net heat plot (see Document S5) suggests that the designs that are together responsible for the majority of between-design heterogeneity are the following ones; (1) Meditative movement therapies vs. placebo/waitlist, (2) CBT:Placebo/WL (from the three-arm study: CBT:Meditative:Placebo/WL) and (3) CBT:Placebo/WL (from the three-arm study CBT:Melatonin:Placebo/WL). There are no "hot spots" of inconsistency.

Self-Reported Wake During the Night
Eight studies measured self-reported wake during the night. The network consisted of 8 interventions including experimental and control condition, based on 10 pairwise comparisons and 7 different designs. Figure 9 presents the netgraph and the forest plot presenting the effects of all available classes compared with placebo. J. Clin. Med. 2020, 9, x FOR PEER REVIEW 18 of 27

Sleep Onset Latency Measured Through Physiological Indices
Nineteen studies measured sleep onset latency through physiological indices. The network consisted of 9 interventions including experimental and control conditions, based on 25 pairwise comparisons and 10 different designs. Figure 8 presents the netgraph and the forest plot presenting the effects of all available classes compared with placebo.

Self-Reported Wake During the Night
Eight studies measured self-reported wake during the night. The network consisted of 8 interventions including experimental and control condition, based on 10 pairwise comparisons and 7 different designs. Figure 9 presents the netgraph and the forest plot presenting the effects of all available classes compared with placebo.
No significant results could be observed. Cochran's Q and Higgins I2 tests revealed very high heterogeneity between studies (tau^2 =4.1301; I^2 = 99.4%). The net heat plot (see Document S5) suggests that the designs that are together responsible for the majority of between-design heterogeneity are the following ones; (1) CBT-I vs. meditative movement therapies and (2) CBT vs. Meditative (from the three-arm study: CBT:Meditative:Placebo/WL).  No significant results could be observed. Cochran's Q and Higgins I2 tests revealed very high heterogeneity between studies (tauˆ2 =4.1301; Iˆ2 = 99.4%). The net heat plot (see Document S5) suggests that the designs that are together responsible for the majority of between-design heterogeneity are the following ones; (1) CBT-I vs. meditative movement therapies and (2) CBT vs. Meditative (from the three-arm study: CBT:Meditative:Placebo/WL).

Wake during the Night Measured Through Physiological Indices
Fourteen studies measured wake during the night measured through physiological indices. The network consisted of 8 interventions including experimental and control condition, based on 18 pairwise comparisons and 9 different designs. Figure 10 presents the netgraph and the forest plot presenting the effects of all available classes compared with placebo.

Wake During the Night Measured Through Physiological Indices
Fourteen studies measured wake during the night measured through physiological indices. The network consisted of 8 interventions including experimental and control condition, based on 18 pairwise comparisons and 9 different designs. Figure 10 presents the netgraph and the forest plot presenting the effects of all available classes compared with placebo.
No significant results could be observed. Cochran's Q and Higgins I 2 tests revealed high heterogeneity between studies (tau^2 = 0.0848; I 2 = 71.0%). The net heat plot (see Document S5) suggests that the designs that are together responsible for the majority of between-design heterogeneity are the following ones; (1) CBT vs. Melatonin (from the three-arm study CBT:Melatonin:Placebo/WL) and (2) CBT vs. Placebo/WL (from the three-arm study CBT:Melatonin:Placebo/WL).

Discussion
In the present study, network meta-analysis was used to compare the efficacy of different interventions for insomnia disorder, which have been suggested as effective, but for which clinical guidelines do not give recommendations. Specifically, the focus of this study was to compare the effectiveness of melatonin, light exposure, exercise and complementary and alternative medicine for insomnia disorder. Within this last category, many different interventions were searched, but for most of them no study was identified and selected for the systematic review and the network meta-analysis. Thus, it was possible to compare the following groups of interventions; melatonin, meditative movements therapies, exercise, natural herbal pharmacotherapies, light exposure, transcranial magnetic resonance, homeopathy, hypnotherapy, and dietary supplement.
Melatonin was evaluated in the largest group of studies in our network meta-analysis (12 of 36). Results support melatonin as an effective intervention to ameliorate sleep onset latency, both measured with physiological (i.e., polysomnography or actigraphy) and self-report (i.e., sleep diaries or questionnaires) indices, with a large effect for the physiological indices, but a rather small effect for the self-report ones. Moreover, melatonin was found to be effective in improving daytime sleepiness, although the observed effect was rather small. Nevertheless, consumption of melatonin was not associated with improved self-report or physiological measured sleep quality, perceived severity of sleep problem, and self-report or physiological measured wake during the night. Interestingly, the only five studies based on pediatric populations all tested the efficacy of melatonin. Although no group analysis on age groups was possible to conduct due to scarcity of data, this work suggests that melatonin could be effective for sleep-onset problems in both pediatric and adult populations with specific problems in initiating sleep. For children, use of melatonin has received better clinical favor [100]. As melatonin was not effective considering sleep-efficiency index and severity of the disorder, it is likely that the use of melatonin should be combined with CBT-I to obtain significant clinical benefits. The study conducted by Cortesi and colleagues [60] provides No significant results could be observed. Cochran's Q and Higgins I 2 tests revealed high heterogeneity between studies (tauˆ2 = 0.0848; I 2 = 71.0%). The net heat plot (see Document S5) suggests that the designs that are together responsible for the majority of between-design heterogeneity are the following ones; (1) CBT vs. Melatonin (from the three-arm study CBT:Melatonin:Placebo/WL) and (2) CBT vs. Placebo/WL (from the three-arm study CBT:Melatonin:Placebo/WL).

Discussion
In the present study, network meta-analysis was used to compare the efficacy of different interventions for insomnia disorder, which have been suggested as effective, but for which clinical guidelines do not give recommendations. Specifically, the focus of this study was to compare the effectiveness of melatonin, light exposure, exercise and complementary and alternative medicine for insomnia disorder. Within this last category, many different interventions were searched, but for most of them no study was identified and selected for the systematic review and the network meta-analysis. Thus, it was possible to compare the following groups of interventions; melatonin, meditative movements therapies, exercise, natural herbal pharmacotherapies, light exposure, transcranial magnetic resonance, homeopathy, hypnotherapy, and dietary supplement.
Melatonin was evaluated in the largest group of studies in our network meta-analysis (12 of 36). Results support melatonin as an effective intervention to ameliorate sleep onset latency, both measured with physiological (i.e., polysomnography or actigraphy) and self-report (i.e., sleep diaries or questionnaires) indices, with a large effect for the physiological indices, but a rather small effect for the self-report ones. Moreover, melatonin was found to be effective in improving daytime sleepiness, although the observed effect was rather small. Nevertheless, consumption of melatonin was not associated with improved self-report or physiological measured sleep quality, perceived severity of sleep problem, and self-report or physiological measured wake during the night. Interestingly, the only five studies based on pediatric populations all tested the efficacy of melatonin. Although no group analysis on age groups was possible to conduct due to scarcity of data, this work suggests that melatonin could be effective for sleep-onset problems in both pediatric and adult populations with specific problems in initiating sleep. For children, use of melatonin has received better clinical favor [100]. As melatonin was not effective considering sleep-efficiency index and severity of the disorder, it is likely that the use of melatonin should be combined with CBT-I to obtain significant clinical benefits. The study conducted by Cortesi and colleagues [60] provides some preliminary evidence that combining behavioral treatment for insomnia with melatonin in children (age range: 4-10 years) may be more effective than melatonin or behavioral intervention only. Nevertheless, more research on this is needed.
Meditative movement therapies were evaluated in nine studies. Results showed large effects supporting the offer of these therapies for ameliorating perceived sleep efficiency and severity of the sleep problem. Instead, no support was evidenced for the other measured variables. Comparability data did not evidence superior effects for meditative movement therapies with respect to CBT for insomnia. These results suggest that meditative movement therapies may be important as additional interventions or strategies in the treatment of insomnia, but that they may be not adequate to be offered alone. Meditative movement therapies mainly target emotion regulation processes. The role of emotional processes in the consolidation of the disorder of insomnia has been stressed [101]. The introduction of mindfulness modules in the standard protocol for CBT-I has been already used in many recent large clinical trials [102]. It would be interesting to test whether these interventions targeting emotion regulation and self-regulation are specifically effective for daytime symptoms. As sleep is strictly associated with socioemotional development [103], it seems to be important to evaluate combined therapies including CBT-I and meditative movement therapies in children and adolescents.
Exercise and hypnotherapy resulted efficacious in ameliorating self-reported sleep onset latency with small to medium effects. Nevertheless, only two studies were available for exercise, including in total 61 adults with insomnia disorder randomized to experimental intervention (N = 30) or control condition (N = 31). Hypnotherapy was tested in one single study of our sample, including 60 adults with insomnia randomized to experimental (N = 30) or control (N = 30) conditions. Similarly, transcranial magnetic resonance was observed to have a large positive effect for perceived severity of the sleep problem, but was also tested in only two studies including in total 116 adults with insomnia disorder randomized to experimental intervention (N = 58) or control condition (N = 58). Thus, no strong conclusions or hypotheses may be driven by these results.
No supportive evidence for natural herbal pharmacotherapies, light exposure, homeopathy, and dietary supplement was observed.
High to very high heterogeneity was found for all outcomes with the exception of self-reported sleep onset latency and daytime sleepiness. Therefore, results must be interpreted with caution.
Some limitations of the present work should be considered. Some interventions have been considered in a larger number of studies compared to others. Especially, most studies focused on melatonin and meditative movement therapies, whereas other therapies were considered in a relative small number of trials. That resulted in networks with unbalanced designs. Quality of the study is questionable, as no study was included for which a total judgment of "low risk" could be applied. It is strongly encouraged that clinical trials evaluating possible alternative interventions for insomnia disorder to those recommended carefully consider all possible biases which may limit interpretation of the results. Moreover, subgroup analyses were not conducted because data were too few for many of the considered interventions and study characteristics were too different between trials. Furthermore, diversity in studies clinical procedures and populations was extremely high. Another limitation is that publication bias could not be analyzed. It is still debated in literature how to conduct publication bias analyses in the specific context of network meta-analyses. Funnel plots which are commonly used in traditional meta-analyses to assess publication bias are not recommended with regard to network meta-analyses where the direction of effects of small studies cannot be assumed. That is, the results presented in this work may be susceptible to the publication bias, thus their interpretation should be done carefully. Finally, no follow-up data could be summarized in the network meta-analysis, due to excessive diversity of study designs.

Clinical Implications
The results of our network meta-analysis do not evidence any of the considered interventions to be effective for both nighttime sleep symptoms and perception of severity of insomnia. Thus, our findings do not support any of the selected therapies to be recommended for insomnia disorder. Instead, our work supports the notion that much more should be done to offer first-line treatment of insomnia, i.e., cognitive-behavioral therapy (CBT), in primary care. Interesting findings associated with the use of melatonin, meditative movement therapies, and exercise should be considered. Future research should be conducted to assess the efficacy of protocols combining melatonin and CBT for insomnia for those patients who mostly complain of sleep onset difficulties. In addition, more trials are encouraged to evaluate the efficacy of protocols combining standard CBT for insomnia with modules including meditative movement therapies or exercise. This should be done by considering efficacy for both nighttime and daytime symptoms.

Conclusions
The results of this study do not support other interventions to replace or to be comparable to the first-line treatment of insomnia (i.e., CBT for insomnia disorder). As a consequence, it is considered necessary that much more effort is done to better implement and disseminate CBT for insomnia in primary care. A recent paper of the European Academy for Cognitive-Behavioral Therapy for Insomnia (CBT for insomnia) has provided guidelines for improving clinician's training and therapy dissemination [21]. Some important hypotheses should be further tested, considering both nighttime and daytime symptoms. These would be specifically directed to consider the role of melatonin as an integrative treatment for patients with severe sleep onset difficulties, and to consider the role of meditative movement therapies and exercise as integrative therapies for daytime symptoms and perception of sleep quality and severity of the sleep problem. A lifespan approach considering pediatric, adult, and elderly populations should be encouraged.