Music for insomnia in adults

Kira V Jespersen; Julian Koenig; Poul Jennum; Peter Vuust

doi:10.1002/14651858.CD010459.pub2

Música para el insomnio en adultos

Declaraciones de intereses de los autores

Versión publicada: 13 agosto 2015 Historial de versiones

https://doi.org/10.1002/14651858.CD010459.pub2

Contraer todo Desplegar todo

Resumen

disponible en

Antecedentes

El insomnio es un trastorno del sueño frecuente en la sociedad moderna. Provoca una reducción en la calidad de vida y se asocia con deterioro de la salud física y mental. Escuchar música es un método muy utilizado como ayuda para el sueño, pero todavía no está claro si en realidad puede mejorar el insomnio en adultos.

Objetivos

Evaluar los efectos de escuchar música para el insomnio en adultos y evaluar la influencia de variables específicas que puedan moderar el efecto.

Métodos de búsqueda

Se hicieron búsquedas en CENTRAL, PubMed, Embase, en otras nueve bases de datos y en dos registros de ensayos en mayo 2015. Además, se realizaron búsquedas manuales en revistas de musicoterapia específicas, listas de referencias de los estudios incluidos y se estableció contacto con autores de estudios publicados para identificar estudios adicionales elegibles para inclusión, incluido cualquier ensayo no publicado o en curso.

Criterios de selección

Ensayos controlados aleatorizados y ensayos controlados cuasialeatorizados que compararon los efectos de escuchar música con ningún tratamiento o tratamiento habitual sobre la mejoría del sueño en adultos con insomnio.

Obtención y análisis de los datos

Dos autores de la revisión examinaron de forma independiente los resúmenes, seleccionaron los estudios, evaluaron el riesgo de sesgo y extrajeron los datos de todos los estudios elegibles para la inclusión. Se realizaron metanálisis con los datos de las medidas de resultado predefinidas cuando se informaron de forma consistente por al menos dos estudios. Los metanálisis se realizaron mediante modelos de efectos fijos y de efectos aleatorios. La heterogeneidad entre los estudios incluidos se evaluó con la estadística I².

Resultados principales

Se incluyeron seis estudios con un total de 314 participantes. Los estudios examinaron el efecto de escuchar diariamente música pregrabada, de 25 a 60 minutos, durante un período de tres días a cinco semanas.

Según el enfoque Grades of Recommendations, Assessment, Development and Evaluation (GRADE), la evidencia de cinco estudios que midieron el efecto de escuchar música sobre la calidad del sueño se calificaron de calidad moderada. La evidencia de un estudio que examinó otros aspectos del sueño (ver a continuación) se calificó de calidad baja. La calidad de la evidencia se disminuyó principalmente debido a limitaciones en el diseño o por ser el único estudio publicado. En cuanto al riesgo de sesgo, la mayoría de los estudios presentaban un alto riesgo de sesgo en al menos un ámbito: un estudio presentaba un alto riesgo de sesgo de selección y uno se consideró de riesgo poco claro; seis estudios presentaban un alto riesgo de sesgo de rendimiento; tres estudios presentaban un alto riesgo de sesgo de detección; un estudio presentaba un alto riesgo de sesgo de desgaste y uno se consideró de riesgo poco claro; dos estudios se consideraron de riesgo poco claro de sesgo de notificación; y cuatro estudios presentaban un alto riesgo de otro tipo de sesgo.

Cinco estudios (n = 264) que informaron sobre la calidad de sueño evaluada según el Pittsburgh Sleep Quality Index (PSQI) se incluyeron en el metanálisis. Los resultados de un meta‐análisis de efectos aleatorios revelaron un efecto a favor de la escucha de música (diferencia de medias (DM) ‐2,80; intervalo de confianza (IC) del 95%: ‐3,42 a ‐2,17; Z = 8,77, P < 0,00001; evidencia de calidad moderada). El tamaño del efecto indica un aumento en la calidad de sueño de un tamaño cercano a una desviación estándar a favor de la intervención en comparación con ningún tratamiento o tratamiento habitual.

Sólo un estudio (N = 50; evidencia de baja calidad) informó datos sobre la latencia de aparición del sueño, el tiempo total de sueño, la interrupción del sueño y la eficiencia del sueño. Sin embargo, no encontró evidencia que indicara que la intervención tuvo efectos beneficiosos sobre estos resultados. Ninguno de los estudios informó ningún evento adverso.

Conclusiones de los autores

Los resultados de esta revisión aportan evidencia de que la música puede ser efectiva para mejorar la calidad subjetiva del sueño en adultos con síntomas de insomnio. La intervención es segura y fácil de administrar. Se necesitan más estudios de investigación para establecer el efecto de escuchar música sobre otros aspectos del sueño, así como las consecuencias diurnas del insomnio.

PICO

Population

Intervention

Comparison

Outcome

El uso y la enseñanza del modelo PICO están muy extendidos en el ámbito de la atención sanitaria basada en la evidencia para formular preguntas y estrategias de búsqueda y para caracterizar estudios o metanálisis clínicos. PICO son las siglas en inglés de cuatro posibles componentes de una pregunta de investigación: paciente, población o problema; intervención; comparación; desenlace (outcome).

Para saber más sobre el uso del modelo PICO, puede consultar el Manual Cochrane.

Resumen en términos sencillos

disponible en

Música para el insomnio en adultos

Pregunta de la revisión

Esta revisión evaluó los efectos de escuchar música sobre el insomnio en adultos y la repercusión de los factores que pueden influir en el efecto.

Antecedentes

En todo el mundo, millones de personas presentan insomnio. Las personas pueden tener dificultades para dormir, permanecer dormido o pueden experimentar una calidad del sueño deficiente.

El sueño deficiente afecta la salud física y mental de las personas. Las consecuencias del sueño deficiente son costosas para los individuos y para la sociedad. Muchas personas escuchan música para mejorar el sueño, pero el efecto de escuchar música no está claro.

Características de los estudios

Se efectuaron búsquedas en las bases de datos electrónicas y las revistas de musicoterapia para identificar los estudios relevantes. Se incluyeron seis estudios, con un total de 314 participantes. Los estudios compararon el efecto de escuchar música solo o con atención estándar con atención estándar sola o ningún tratamiento. Los estudios examinaron el efecto de escuchar música pregrabada diariamente, durante 25 a 60 minutos, por un período de tres días a cinco semanas. La evidencia está actualizada hasta el 22 de mayo de 2015.

Resultados clave

Cinco estudios midieron la calidad del sueño. Los hallazgos sugieren que escuchar música puede mejorar la calidad del sueño. Solamente un estudio informó datos sobre otros aspectos del sueño, incluido el período de tiempo que tarda en quedar dormido, la cantidad de sueño real que logra y el número de veces que la persona despierta. Este estudio no encontró ninguna evidencia que sugiera que escuchar música beneficia a estos resultados. Ninguno de los estudios informó ningún efecto secundario negativo causado por escuchar música.

Calidad de la evidencia

La calidad de la evidencia de los cinco estudios que examinaron la calidad del sueño era moderada. La calidad de la evidencia para los otros aspectos del sueño fue baja. Se necesitan más estudios de investigación de alta calidad para investigar y establecer el efecto de escuchar música sobre otros aspectos del sueño, además de la calidad del sueño, y sobre medidas diurnas relevantes.

Authors' conclusions

Implications for practice

The findings of the meta‐analysis suggest that listening to music may improve sleep quality in different populations experiencing insomnia symptoms. Furthermore, the results indicate that the intervention is safe and easy to administer. No conclusions can be drawn on the effect of music listening on other aspects of sleep or on related physiological and psychological aspects of daytime function, since no trials or only single trials reported these outcomes. More research is needed to clarify the effect of the intervention on outcomes beyond self‐reports of sleep quality. Since the studies report limited information on the nature of participants' sleep problems, it is not possible to draw any conclusions with regard to the effect on insomnia subtypes such as difficulties with sleep initiation, sleep maintenance or non‐restorative sleep. None of the participants were diagnosed with insomnia, and we do not know if listening to music can improve sleep in adults diagnosed with insomnia disorder.

All included trials used music that was characterized as sedative or relaxing. However, these included a number of different musical styles (e.g. classical, new age, jazz, etc.) and at this point, it is not clear if some types of music may be more effective than others. In the literature, it is often recommended that participants are allowed to choose their own preferred music. In this review, there was no difference in the effect on sleep quality between trials using researcher‐selected music and trials giving the participants a choice among a number pre‐selected types of music. Very few participants were offered the possibility to bring their own preferred music, and the effect of purely participant‐selected music could not be investigated.

Implications for research

More high‐quality randomised controlled trials are needed to assess the effectiveness of music listening for treating insomnia. Future trials need to pay close attention to reducing risk of bias. Randomisation needs to be properly concealed, and although blinding of participants is not possible with a music intervention, blinding of researchers and outcome assessors should be prioritised to minimise performance and detection biases.

Future research should consider a wider range of outcomes. In particular, more research should include objective measures of sleep, such as polysomnography and actigraphy, that are less sensitive to detection bias. The use of objective measures of sleep would reduce the impact of any placebo effect. Furthermore, there is a lack of knowledge of the effect of listening to music on daytime consequences and waking correlates of insomnia. It is important to know if the reported changes in sleep patterns or sleep quality are sufficient to affect daytime function. Furthermore, longer follow‐up periods are important to genuinely establish the effectiveness of music and its long‐term effect.

More research is needed to establish the effect of the intervention on different insomnia groups. Insomnia is a highly heterogenous disease with different aetiology and severity. Future trials should take care to define and appropriately measure sleep disturbances and provide detailed information on the cause, duration and severity of symptoms, as well as any comorbid conditions. Participants should be screened for other sleep disorders to clarify the nature of the sleep complaint. It is also recommended that researchers employ well‐defined criteria for insomnia such as the diagnostic criteria of the Diagnostic and Statistical Manual of Mental Disorders (DSM) or International Classification of Sleep Disorders (ICSD). This would improve the precision of the clinical diagnosis and hence improve comparability across trials.

The music therapy literature recommends that music used for sedative purposes should be characterised by a slow tempo and an absence of abrupt changes and rhythmic complexity (Wigram 2002). These recommendations are supported by experimental research in the field of music psychology, but more controlled clinical trials are needed to examine which aspects of music are important to achieve an improvement in sleep. In addition, the relationship between the objective characteristics of the music and the subjective preferences of the individual remain unclear, and more trials are needed to investigate potential differences in effect between music selected by the researcher and that selected by the participant. Another aspect of the intervention that remains unclear is the optimal frequency, timing, and duration of the intervention. Further research into these domains is important for assessing the effectiveness of the intervention and for providing the best treatment options for people with insomnia.

Summary of findings

Open in table viewer

Summary of findings for the main comparison. Listening to music compared to no treatment or treatment‐as‐usual (TAU) for adults with insomnia

Listening to music compared to no treatment or treatment‐as‐usual (TAU) for adults with insomnia
Patient or population: adults with insomnia Settings: home, sleep lab or rehabilitation centre Intervention: listening to music Comparison: no treatment or TAU
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	Number of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	No treatment or TAU	Listening to music
Sleep quality ‐ immediately post‐treatment PSQI¹. Scale from: 0 to 21 Follow‐up: 21 to 35 days		The mean score in the intervention groups was 2.8 lower (3.42 lower to 2.17 lower)		264 (5 studies)	⊕⊕⊕⊝ moderate^2,3	A lower score indicates better sleep quality (i.e. fewer sleep problems). The change is about the size of one standard deviation which is considered a clinically relevant change. The studies included participants with a complaint of poor sleep (PSQI > 5)¹.
Sleep onset latency ‐ immediately post‐treatment PSG and morning questionnaire Follow‐up: 3 days	See comment	See comment		50 (1 study)	⊕⊕⊝⊝ low^4,5	The one trial reporting this outcome found no evidence of an effect of the intervention. The data were not available for analysis. The study included participants that had reported poor sleep for at least one month (PSQI > 5)¹.
Total sleep time ‐ immediately post‐treatment PSG and morning questionnaire Follow‐up: 3 days	See comment	See comment		50 (1 study)	⊕⊕⊝⊝ low^4,5	The one study reporting this outcome found no evidence of an effect of the intervention. The data were not available for analysis. The study included participants that had reported poor sleep for at least one month (PSQI > 5)¹.
Sleep interruption ‐ immediately post‐treatment PSG and morning questionnaire Follow‐up: 3 days	See comment	See comment		50 (1 study)	⊕⊕⊝⊝ low^4,5	The one study reporting this outcome found no evidence of an effect of the intervention. The data are not available for analysis. The study included participants that had reported poor sleep for at least one month (PSQI > 5)¹.
Sleep efficiency ‐ immediately post‐treatment PSG Follow‐up: 3 days	See comment	See comment		50 (1 study)	⊕⊕⊝⊝ low^4,5	The one study reporting this outcome found no evidence of an effect of the intervention. The data were not available for analysis. The study included participants that had reported poor sleep for at least one month (PSQI > 5)¹.
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI) CI: confidence interval; PSG: polysomnography; PSQI: Pittsburgh Sleep Quality Index
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹ Pittsburgh Sleep Quality Index. 0 indicates good sleep quality and 21 indicates severe sleep problems. Clinical cut off > 5 (Buysse 1989). ² Limitations in the design such as inadequate allocation concealment procedures and lack of blinding. ³ The estimated effect is about one standard deviation and can be considered large. ⁴ Limitations in the design such as lack of blinding. ⁵ Only one trial measured this outcome.

Background

Description of the condition

Insomnia is one of the most common sleep disorders and affects millions of people worldwide. Insomnia can be defined as a subjective complaint of disturbed sleep in the presence of adequate opportunity and circumstance for sleep (NIH 2005). It is characterised by dissatisfaction with the quality, duration or continuity of sleep, such as problems falling asleep, maintaining sleep, early morning awakenings or complaints of non‐restorative sleep (Morin 2013). The diagnostic criteria for insomnia disorder as a clinical diagnosis vary depending on the diagnostic manual used. The fourth edition of the American Psychiatric Association's (APA) Diagnostic and Statistical Manual for Mental Disorders (DSM‐IV) defines primary insomnia as a subjective complaint of difficulty in initiating or maintaining sleep, or the presence of non‐restorative sleep, lasting for at least one month (APA 1994). According to these criteria, disturbed sleep causes clinically significant distress or impairment in social, occupational or other important areas of functioning. According to the APA criteria, primary insomnia is not the consequence of any substance use and is not linked to a major general medical condition, any mental disorder or an otherwise specified sleep disorder (APA 1994). However, the distinction between primary and secondary insomnia has been removed in the most recent update. Instead, 'insomnia disorder' is recognized as a condition requiring clinical attention independently of other medical conditions that may be present (APA 2013). Other diagnostic manuals, such as the the World Health Organization's (WHO) International Classification of Diseases, 10ͭʰ Revision (ICD‐10) and the International Classification of Sleep Disorders (ICSD‐2) distinguish between different insomnia subtypes and introduce different duration and frequency criteria (AASM 2005; WHO 1992).

Insomnia is common in people with medical or psychiatric illness and trials have found consistent relationships between insomnia and depression, anxiety disorders, and other psychological disorders, as well as substance abuse and dependence. Furthermore, insomnia is associated with a number of medical problems such as decreased immune functioning (Taylor 2003), cardiovascular disorders, hypertension, chronic pain, breathing difficulties, and gastrointestinal and urinary problems (Taylor 2007). Insomnia itself can have a number of negative daytime consequences and it is well recognised that people with insomnia experience impairments in everyday life such as fatigue and greater irritability (Riedel 2000; Shekleton 2010). People with insomnia report significantly lower quality of life than those without insomnia, and the reduction in quality of life is correlated with symptom severity (Léger 2001). Insomina affects occupational function and social relations and is associated with higher work absenteeism and increased risk of accidents, and therefore represents a condition with great costs for both the individual and society (Walsh 2004).

Estimates of the prevalence of insomnia vary according to the definitions used. A review of epidemiological trials revealed that about one third of the general population experiences symptoms of insomnia, such as difficulties initiating or maintaining sleep. When adding daytime consequences to the definition of insomnia, the prevalence rate drops to about 9% to 15% (Ohayon 2002). Sleep dissatisfaction is reported by 8% to 18% of the general population; using the DSM‐IV criteria for insomnia, a point‐prevalence rate of 6% is reported within the literature, with primary insomnia as the most frequent diagnosis (2% to 4%), followed by insomnia related to another mental disorder (1% to 3%) (Ohayon 2002).

Description of the intervention

Available treatments for insomnia include pharmacotherapy (medication), psychological and behavioural interventions, and a variety of complementary and alternative therapies. Basic education in sleep hygiene is commonly used to inform patients about lifestyle and environmental factors that may interfere with or promote sleep (Morin 2005). Although good sleep hygiene can be helpful, it is not considered sufficient to improve severe and chronic insomnia. Therefore, the most common treatment of insomnia is the use of pharmacotherapeutic interventions, including over‐the‐counter and prescribed medications (NIH 2005). Benzodiazepine receptor agonists have been approved for the treatment of insomnia by the U.S. Food and Drug Administration (FDA) and trials report beneficial short‐term effects of these agents. With the exception of eszopiclone, approved use is limited to 35 days or less (NIH 2005). However, there are some concerns about the long‐term use of these medications, due to risk of abuse, dependence, and adverse effects such as residual daytime sedation, cognitive impairments, and reduced motor co‐ordination. Furthermore, the use of hypnotics has been associated with increased mortality (Frandsen 2014; Kripke 2012).

Psychological and behavioural treatments show good efficacy in the treatment of persistent insomnia (Morin 2006b). These treatments include cognitive behavioural therapy, stimulus control therapy, sleep restriction therapy, and relaxation training, often combined in multi‐component approaches. In spite of good results, psychological and behavioural treatments appear to be under‐utilised, perhaps because they require considerable time and effort for the patient (Krystal 2004). Furthermore, there is a problem of availability, with relatively few well‐trained professionals in the field (Wilson 2010).

Given the current limitations of psychological and pharmacological treatments, different alternative and complementary interventions, one of which is music, are available to combat insomnia in adults. Trials show that listening to music is often used by adults as a self‐help intervention to improve sleep (Aritake‐Okada 2009; Morin 2006a; Urponen 1988), and a simple Google search on 'music' and 'sleep' reveals a huge market of music that is promoted for its sleep‐inducing properties. However, the efficiency of music as an intervention for insomnia remains unclear. Music is used as a therapeutic intervention in a number of areas, including pain relief (Cepeda 2006), psychiatry (Maratos 2008; Mössler 2011), neuro‐rehabilitation (Bradt 2010), and for improving psychological outcomes in medical conditions such as cancer and heart disease (Bradt 2011; Bradt 2013). Experimental and clinical studies show that music can affect autonomous nervous system functioning (Hodges 2009) as well as psychological factors such as mood and attention (Garza‐Villarreal 2014; Juslin 2001); as such, it may potentially impact sleep (see How the intervention might work). Research on the impact of music on sleep has evolved during the past 20 years, and methods for applying music listening to improve sleep quality vary across trials. Generally, the intervention involves the use of pre‐recorded music in relation to sleep initiation. Music listening can be used passively, or it can be used actively with specific instructions (e.g. relaxation instructions). The duration of the intervention period and the time spent listening to music may vary. The choice of music may be determined by the researcher or by the participants themselves. Relatively few trials on music and insomnia are randomised controlled trials (RCTs). Since the impact of the music intervention may differ depending on the design of the intervention, there is a need to evaluate the results of both RCTs and quasi‐randomised controlled trials (qRCTS) in order to retain valuable and reliable evidence.

How the intervention might work

Music has been found to influence human beings on many levels (Juslin 2001) and the impact of music listening on sleep has been attributed to different mechanisms. Several authors argue that improvement of sleep is obtained because slow soothing music enhances relaxation (Deshmukh 2009; Hernández‐Ruiz 2005; Jespersen 2012; Lai 2005). This suggestion is substantiated by trials showing reduced levels of cortisol as an effect of music listening (Koelsch 2011; Nilsson 2009), and changes in autonomous measures such as heart‐rate and blood pressure (Korhan 2011; Su 2013; Trappe 2010). These trials show that music can affect various physiological measures that reflect autonomic nervous system responses, and as such, slow soothing music may lead to a decrease in sympathetic arousal and thus improve sleep (Su 2013). From a psychological perspective, trials have shown that listening to music can reduce anxiety and stress responses (Dileo 2007; Zhang 2012), which can lead to greater relaxation and improvement of sleep. Another possible mechanism for the effect of music on sleep is the distracting power of music. Hernández‐Ruiz 2005 suggests that music can function as a focal point of attention that distracts from stressful thoughts and thereby improves sleep. A number of individual factors are also likely to influence the music experience, such as age and gender (Juslin 2011; Nieminen 2012), music preference (Vuust 2010), musical training (Brattico 2009; Vuust 2006), and culture (Hargreaves 1997). Different effects may therefore be found depending on the type of music used, the etiology of insomnia symptoms, and the length and duration of the intervention.

Why it is important to do this review

Music is commonly used to relieve sleep problems and the use of music as a non‐pharmacological intervention offers potential advantages of easy administration, low cost, and safety. Clinical trials have been performed to investigate the effect of music on sleep, but it remains unclear if the existing evidence is rigorous enough to reach conclusions about the general efficacy of the intervention. A systematic review is needed to establish the efficacy of music listening for improvement of sleep quality and thereby refute or validate the popular belief that music is helpful to promote sleep.

Objectives

To assess the effects of listening to music on sleep in adults with insomnia and to assess the influence of specific variables that may moderate the effect.

Methods

Criteria for considering studies for this review

Types of studies

The methods of this review were pre‐specified in the protocol (Jespersen 2013a).

We considered randomised controlled trials (RCTs) and quasi‐randomised controlled trials (qRCTs) that allocated participants to groups on the basis of a quasi‐random process, such as date of birth or alternate numbers for inclusion, for inclusion in the review. Since it is not possible to blind participants to the treatment (music), we included unblinded or single‐blinded trials.

Types of participants

We included adults with insomnia as documented by standardised measures (e.g. Pittsburgh Sleep Quality Index (PSQI; Buysse 1989)), objective measures (e.g. polysomnography or actigraphy), reports or diaries kept by participants, relatives or other informants; or individuals diagnosed with an insomnia disorder by standard diagnostic criteria, such as the International Classification of Diseases (ICD; WHO 1992), the Diagnostic and Statistical Manual of Mental Disorders (DSM; APA 1994) or International Classification of Sleep Disorders (ICSD; AASM 2005); or with an individual's complaint of sleep difficulties.

Types of interventions

We included any intervention that comprised listening to pre‐recorded music with or without relaxation instructions. The intervention could be self‐administered or administered by research or clinical personnel. Interventions included music listening compared with a no music control group or treatment‐as‐usual (TAU), and music therapy added to TAU compared to TAU alone.

Types of outcome measures

Primary outcomes

Our outcomes of interest were sleep‐ and insomnia‐related symptoms as measured by sleep diaries, polysomnography, actigraphy, or by standardised scales for the assessment of sleep and sleep‐related symptoms (e.g. PSQI or Insomnia Severity Index (ISI)). Furthermore, to establish the safety of the intervention, we considered the reporting of adverse events as a primary outcome. The primary outcomes were:

Sleep quality.
Sleep onset latency.
Total sleep time.
Sleep interruption (number of awakenings and waking after sleep onset).
Sleep efficiency (percent of time in bed spent asleep).
Adverse events (as reported by trialists; e.g. discomfort or hearing loss).

Secondary outcomes

Secondary outcomes of interest were waking‐related correlates and daytime consequences of insomnia. The relevant measures included:

Psychological outcomes.
1. Depression.
2. Anxiety.
3. Quality of life.
Physical outcomes.
1. Fatigue.
2. Daytime sleepiness.
3. Pain.
Physiological outcomes.
1. Heart rate.
2. Heart rate variability.
3. Blood pressure.

We included trials that measured psychological outcomes by standardized questionnaires with established reliability and validity (e.g. Beck Depression Inventory (BDI; Beck 1996), State‐Trait Anxiety Inventory (STAI; Spielberger 1983), Short‐Form 36 (SF‐36) health survey (Ware 1992)). We included trials that measured physical outcomes with standardised procedures such as the Multiple Sleep Latency Test (MSLT) or validated rating scales. We included trials that measured physiological outcomes with standardized procedures such as an electrocardiogram (ECG).

We considered the trial period and follow‐up as described in the included trials. When assessing outcomes in relation to time points, we grouped the data as: immediate post‐intervention, short‐term (post‐intervention to one month), medium‐term (between one and three months follow‐up), and long‐term (more than three months follow‐up) effects.

We reported all primary outcomes in summary of findings Table for the main comparison.

Search methods for identification of studies

We did not apply any restrictions on date, language, or publication status when searching for trials or when deciding on trial inclusion.

Electronic searches

To identify the relevant trials, we searched the following electronic databases on 22 May 2015.

Cochrane Central Register of Controlled Trials (CENTRAL), 2015, Issue 4, part of theCochrane Library, and includes the Cochrane Developmental Psychosocial and Learning Problems Group Specialised Register.
PubMed (ncbi.nlm.nih.gov/pubmed), 1950 to 22 May 2015.
Embase (Elsevier), 1980 to 22 May 2015.
CINAHL (EBSCOhost), 1982 to 22 May 2015.
PsycINFO (ProQuest), 1967 to 22 May 2015.
Web of Science (Science Citation Index Expanded, Social Sciences Citation Index, Arts and Humanities Citation Index, Conference Proceedings Citation Index ‐ Science, and Conference Proceedings Citation Index ‐ Social Science and Humanities), 1980 to 22 May 2015.
SCOPUS (Elsevier), 1960 to 22 May 2015.
ClinicalTrials.gov (ClinicalTrials.gov), all available years searched 22 May 2015.
Current Controlled Trials (controlled‐trials.com/), all available years searched 22 May 2015.
Répertoire International de Littérature Musicale (RILM; EBSCOhost), 1969 to 22 May 2015.

The search strategies for the databases can be found in Appendix 1.

Searching other resources

We handsearched 15 journals, which are listed in Appendix 2, and searched the bibliographies of retrieved articles and relevant reviews to identify potential trials missed by the electronic searches. We also performed a citation search in ISI Web of Science to identify reports that had cited any relevant trials retrieved by the search to discover more recent trials that could have been missed. In addition, we contacted experts in the field to identify any unpublished trials.

Data collection and analysis

Selection of studies

Two authors (KVJ and JK) independently screened all titles and abstracts. All papers for which the title or abstract referred to a trial on music and sleep were retrieved in full. In cases where there was insufficient information in the title or abstract to determine the relevance of a paper, we retrieved the full text. Both authors independently reviewed the full‐text papers against a previously prepared inclusion criteria form to assess the trial's eligibility for inclusion. Disagreements were discussed and a third author (PJ) was involved in making the final decision. We kept a record of excluded articles and the reason for their exclusion (Characteristics of excluded studies).

Data extraction and management

Using a standardized coding form, the first two authors (KVJ and JK), who were blinded to each other’s assessment, extracted the data. Disagreements were resolved by consensus. If outcome data were not available, we contacted the authors of the trial.

From each trial, we extracted the following information.

1. General information

Author
Year of publication
Title
Journal (title, volume, pages) or if unpublished source
Country
Language of publication

2. Trial design

Design (e.g. parallel or cross‐over design)
Method of randomisation (and concealment)
Nature of the control group (e.g. no treatment or TAU)
Losses to follow‐up
Blinding of trial evaluators
Washout period in cross‐over design
Inclusion criteria
Exclusion criteria

3. Participants

Total sample size
Number in experimental group
Number in control group
Age
Gender
Ethnicity
Diagnosis
Comorbidities
Sleep quality (and reason for poor sleep)
Duration of disorder
Previous or additional treatments

4. Intervention

Type of music employed (characteristics)
Music selection (selected by participant or researcher)
Who provided the music (participant or research personal)
Length and frequency of intervention sessions
Intervention period (duration of intervention)
How participants were exposed to music (e.g. headphones or loudspeakers)
Listening instructions

5. Outcomes

Methods of sleep assessment
Secondary outcome measures
Pre‐test means and post‐test means or change scores and standard deviations, for all groups for all outcomes specified above
Baseline differences
Follow‐up period

Assessment of risk of bias in included studies

Two authors (KVJ and JK) independently assessed the risk of bias by using the tool described (and the criteria outlined) in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). They solved disagreements by discussion with the fourth author (PV). When information for evaluating methodological criteria was absent, we contacted the authors to obtain further information.

We graded each trial for risk of bias in each of the following domains.

1. Random sequence generation (checking for possible selection bias)

We assessed the method used to generate the allocation sequence for each included trial in sufficient detail to allow an assessment of whether it produced comparable groups.

We rated the risk of bias as follows.

Low risk of bias (adequate method of random sequence generation: e.g. any truly random process such as random number table; computer random number generator).
High risk of bias (inadequate method of random sequence generation: e.g. any non‐random process such as odd or even date of birth; hospital or clinic record number).
Unclear risk of bias (insufficient information about the method of random sequence generation to permit a judgement of low risk or high risk of bias).

2. Allocation concealment (checking for possible selection bias)

We assessed the method used to conceal the allocation sequence for each included trial and determined whether intervention allocation could have been foreseen in advance of, or during recruitment, or changed after assignment.

We rated the risk of bias as follows.

Low risk of bias (adequate method of allocation concealment: e.g. telephone or central randomisation; consecutively numbered sealed opaque envelopes).
High risk of bias (inadequate method of allocation concealment: e.g. open random allocation; unsealed or non‐opaque envelopes; alternation; date of birth).
Unclear risk of bias (insufficient information to permit a judgement of low risk or high risk of bias).

3. Blinding of participants and personnel (checking for possible performance bias)

We assessed the different methods used to blind personnel from knowledge of which intervention a participant received for each included trial. Since it is not possible to blind a participant to the treatment (music), we assessed trials at low risk of bias if we judged that the lack of blinding was not affecting the results. We assessed blinding of participants and personnel separately for different outcomes or classes of outcomes, since we expected certain outcomes (e.g. laboratory measurements and physiological data such as heart rate or blood pressure) to be unaffected by blinding of participants and personnel.

We rated the risk of bias as follows.

Low risk of bias (adequate method of blinding; outcome not likely to be influenced by lack of blinding).
High risk of bias (inadequate method of blinding; outcome likely to be influenced by lack of blinding).
Unclear risk of bias (insufficient information to permit a judgement of low risk or high risk of bias).

4. Blinding of outcome assessment (checking for possible detection bias)

We assessed the methods used to blind outcome assessment for each included trial. We assessed blinding separately for different outcomes or classes of outcomes, as stated above.

We rated the risk of bias as follows.

Low risk of bias (adequate method of blinding; outcome not likely to be influenced by lack of blinding).
High risk of bias (inadequate method of blinding; outcome likely to be influenced by lack of blinding).
Unclear risk of bias (insufficient information to permit a judgement of low risk or high risk of bias).

5. Incomplete outcome data (checking for possible attrition bias through withdrawals, dropouts, protocol deviations)

We assessed data on attrition, exclusions, and reasons to determine if they introduced bias. We described the completeness of data for each included trial and for each outcome or class of outcomes. We assessed whether attrition and exclusions were reported, the numbers of participants included at each stage of the analysis (compared with the total participants randomised), whether reasons for attrition or exclusion were reported, and whether missing data were balanced across groups or were likely to be related to outcomes.

We judged whether incomplete data was dealt with adequately and rated the risk of bias as follows.

Low risk of bias (no missing outcome data; balanced missing outcome data; appropriate methods of imputing missing data).
High risk of bias (missing outcome data; unbalanced missing outcome data; inappropriate methods of imputing missing data).
Unclear risk of bias (insufficient information to permit a judgement of low risk or high risk of bias).

6. Selective reporting bias

We investigated the possibility of selective outcome reporting bias for each included trial. We conducted electronic searches to identify protocols of respective trials as a source to judge selective reporting.

We rated the risk of bias as follows.

Low risk of bias (all pre‐specified and expected outcomes were reported).
High risk of bias (not all pre‐specified and expected outcomes were reported; outcome that was not pre‐specified was reported; outcome was reported incompletely).
Unclear risk of bias (insufficient information to permit a judgement of low risk or high risk of bias).

7. Other bias

We assessed other risks of bias, specifically a risk of bias from baseline differences and a risk of bias from carry‐over or period effects for cross‐over trials.

We rated the risk of bias as follows.

Low risk of bias (the trial appeared to be free of other sources of bias).
High risk of bias (there was at least one important risk of bias).
Unclear risk of bias (insufficient information to permit a judgement of low risk or high risk of bias).

With reference to each of the above domains, we assessed the likely magnitude and direction of the bias and whether we considered it likely to impact the findings. We explored the impact of the level of bias by undertaking sensitivity analyses – see subsection on Sensitivity analysis.

Measures of treatment effect

Two authors (KVJ and JK) independently extracted data from trials identified for inclusion to ensure accuracy. We used Review Manager 5 (RevMan; RevMan 2014) for data entry and analyses.

Continuous data

We analysed continuous outcomes measured on the same scale between trials (e.g. PSQI) using the mean difference (MD).

Ordinal data

We analysed ordinal data measured on scales (i.e. sleep quality on visual analogue scales) as continuous data and the intervention effect was expressed as a difference in means.

Please see Jespersen 2013 and Appendix 3 for additional methods archived for future updates of this review.

Unit of analysis issues

Cluster‐randomised trials

We did not identify any cluster‐randomised trials. For further information on how these kinds of studies will be dealt with in future updates of this review, see Jespersen 2013a and Appendix 3.

Cross‐over trials

We did not identify any cross‐over trials. For further information on how these kinds of studies will be dealt with in future updates of this review, see Jespersen 2013a and Appendix 3.

Trials with more than two treatment arms

If a trial reported multiple treatment arms, we only used comparisons between the music intervention and the control or TAU group. For further information on how we will deal with other trials with more than two treatment arms, see Jespersen 2013a and Appendix 3 .

Dealing with missing data

For included trials, we noted levels of attrition in the incomplete outcome data section of the 'Risk of bias' tables beneath the Characteristics of included studies tables. Where information about the presented data set was missing in the trial reports, or if there was a lack of detail or a discrepancy between different reports, or clarification was needed, we tried to retrieve relevant information by contacting the authors of the trial. Where data were missing due to loss to follow‐up or dropout, we attempted to obtain complete outcome data from trial authors to include all participants randomised to each group in the analyses. If any outcome data remained missing, or if trial authors did not respond within a reasonable time, we analysed data on an available case basis, based on the numbers of participants for whom outcome data (continuous and dichotomous) were known. We did not impute missing data. For more information on how we will deal with missing data in future updates of this review, see Jespersen 2013a and Appendix 3.

Assessment of heterogeneity

We assessed clinical and methodological heterogeneity by examining the characteristics of the trials. The similarities between interventions (e.g. dose, frequency), participants (e.g. age), trial design (e.g. allocation concealment, blinding, losses to follow‐up), and the outcomes are reported in the Included studies subsection. Heterogeneity of treatment response was assessed visually from the forest plot of the mean difference (MD) and the Chi² test. In addition, heterogeneity was assessed statistically according to the standard method using the I² statistic, calculated for each comparison on each outcome. Substantial heterogeneity was assumed if I² was greater than 50%, indicating that 50% of the variability in the outcome cannot be explained by sampling variation. For further information on how we will deal with heterogeneity in future updates of this review, see Jespersen 2013a and Appendix 3.

Assessment of reporting biases

We attempted to minimize the potential for publication bias by our comprehensive search strategy that included evaluating published and unpublished literature.

Where we suspected reporting bias, we contacted trial authors asking them to provide missing outcome data.

For further information on how we will deal with reporting bias in future updates of this review, see Jespersen 2013a and Appendix 3.

Data synthesis

We entered all trials included in the systematic review into Review Manager 5 (RevMan 2014) and checked for data entry errors. We conducted a meta‐analysis when there were data from at least two included trials and substantial heterogeneity was absent. We undertook meta‐analyses using both fixed‐effect and random‐effects models. Where there was agreement between the results of both analyses, we reported the results from random‐effects models as it conveys the variability better. If fixed‐effect and random‐effect models revealed different results, we investigated possible sources of heterogeneity or inconsistency among trials in the magnitude or direction of effects.

Assessing the quality of the evidence

Using the approach developed by the Grading of Recommendation, Assessment, Development and Evaluation (GRADE) working group, we rated the quality of the evidence as 'high', 'moderate', 'low' or 'very low' (GRADE 2013). Evidence from RCTs were given an initial 'high quality' rating, but the assessment was downgraded if the trial methodology had a risk of bias, if there was substantial inconsistency among the results, if the evidence was indirect or imprecise, and if there was evidence of publication bias. The assessment was upgraded if a large effect was found. The results of the GRADE assessment for the primary outcomes is found in summary of findings Table for the main comparison. The table was constructed using the GRADEpro software (GRADEpro 2014).

Subgroup analysis and investigation of heterogeneity

We carried out the following subgroup analyses (ranked in order of importance).

Researcher‐selected music versus participant‐selected music.
Music listening alone versus music listening with relaxation instructions.

The subgroup analyses were exploratory and conducted as recommended in section 9.6 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011).

For further information on other intended subgroup analyses, see Jespersen 2013a and Appendix 3.

Sensitivity analysis

We conducted the following sensitivity analyses to determine the impact of trial quality and risks of bias on the results of the meta‐analyses.

Excluded trials with inadequate methods of random sequence generation.
Excluded trials using inadequate methods of allocation concealment.
Excluded trials using inadequate methods of blinding outcome assessment.

For additional information on other intended sensitivity analyses, see Jespersen 2013a and Appendix 3.

Results

Description of studies

Results of the search

After removing duplicates, we identified 465 records during the search conducted in May 2015 (see Figure 1). After screening titles and abstracts, we identified 71 references that were considered potentially relevant and retrieved them for further examination. One potentially relevant trial was published by two of the authors of this review. To avoid the influence of dual authorship, two authors with no involvement in the trial (JK and PJ) assessed the eligibility and risks of bias of this trial. We excluded 55 trials (from 61 reports) (see Excluded studies). We included six trials (eight reports) that matched the inclusion criteria. We categorised one trial (two reports) as 'awaiting classification', since there was insufficient information to assess inclusion and exclusion. We identified two relevant ongoing trials.

Figure 1

Study flow diagram

Included studies

Six trials met the inclusion criteria for this review (see Characteristics of included studies). Five were RCTs (Chang 2012; Harmat 2008; Kullich 2003; Lai 2005; Shum 2014) and one was a qRCT (Jespersen 2012). All trials used a parallel group design. One trial examined the effects of music listening on sleep quality in adults with insomnia related to pain (Kullich 2003), and one on its effects on insomnia related to psychological trauma (Jespersen 2012). Two trials included older adults with sleep problems (Lai 2005; Shum 2014), and two trials provided no information on the reason for poor sleep quality (Chang 2012; Harmat 2008).

The trials were conducted in five different countries. Two were conducted in Taiwan (Chang 2012; Lai 2005), one in Singapore (Shum 2014), one in Hungary (Harmat 2008), one in Denmark (Jespersen 2012), and one in Austria (Kullich 2003).

Trial size

The six included trials comprised a total of 314 participants. Trial sample sizes ranged from 15 to 65 participants with an average sample size of 52 (median 60). One trial had a small sample size of 15 participants (Jespersen 2012), and the remaining five trials had sample sizes between 50 and 65 participants (Chang 2012; Harmat 2008; Kullich 2003; Lai 2005; Shum 2014).

Setting

Four of the included trials measured the effect of the music listening intervention in the participants' own home (Harmat 2008; Jespersen 2012; Lai 2005; Shum 2014). In three of these trials, data were collected during a weekly visit to the participants' homes (Jespersen 2012; Lai 2005; Shum 2014), and in two of these trials, participants were also telephoned once or twice a week to ensure compliance with the protocol (Lai 2005; Shum 2014). In the fourth trial, the intervention group, but not the control group, was telephoned once a week to assess compliance (Harmat 2008).

Of the remaining two trials, one was conducted in a sleep laboratory (Chang 2012), and one implemented the intervention at an inpatient rehabilitation facility for persons with low‐back pain (Kullich 2003).

Participants

The participants in the included trials were between 19 and 83 years of age. One trial did not report gender (Lai 2005), but in the remaining five trials, the majority of participants were women (70%).

Five studies stated that they recruited participants with complaints of insomnia, and all trials used scores greater than five on the Pittsburgh Sleep Quality Index (PSQI) as evidence of insomnia symptoms. One trial did not include sleep difficulties as inclusion criteria, but all participants had PSQI scores greater than five, indicating insomnia problems (Kullich 2003). The severity of the sleep difficulties varied, with mean PSQI scores at baseline ranging from 6.8 to 16.0. Four of the trials reported mean baseline scores around 10 (Chang 2012; Kullich 2003; Lai 2005; Shum 2014).

The target populations in the six trials were diverse. One trial recruited traumatized refugees (Jespersen 2012), and another recruited patients undergoing rehabilitation for low‐back pain (Kullich 2003). Four trials recruited persons with insomnia that were otherwise healthy (Chang 2012; Harmat 2008; Lai 2005; Shum 2014), and two of these trials focused on older adults (Lai 2005; Shum 2014).

Interventions

All included trials used listening to pre‐recorded music as the main intervention. Four trials examined the effects of listening to music only (Chang 2012; Harmat 2008; Jespersen 2012; Shum 2014), and two trials examined music listening with relaxation instructions (Kullich 2003; Lai 2005). In all trials, music was used once a day. Four trials instructed participants to listen to the music at bedtime (Chang 2012; Harmat 2008; Jespersen 2012; Lai 2005). The remaining two trials did not specify what time of the day to listen to the music (Kullich 2003; Shum 2014). The length of the music listening sessions ranged from 25 to 60 minutes with an average of 43 minutes (median 45). The duration of the intervention period ranged from 3 to 35 days, with four trials having an intervention period of 21 days (Harmat 2008; Jespersen 2012; Kullich 2003; Lai 2005), one trial having an intervention period of three days (Chang 2012), and one trial assessing five weeks of intervention (Shum 2014).

The music used in the intervention was selected by the researchers in five trials (Harmat 2008; Jespersen 2012; Kullich 2003; Lai 2005; Shum 2014). Two of these trials gave participants a choice among four or six types of music in different genres (Lai 2005; Shum 2014 respectively). The remaining three trials used the same music for all participants. The sixth trial encouraged participants to bring their own preferred music for bedtime listening (Chang 2012). Those who did not bring their own music (more than half) listened to music prepared by the researchers. In total, 10 participants listened to their own preferred music and 149 participants listened to music chosen by the researcher.

All trials provided information on the music used in the study. The genres reported were Western and Chinese classical music, new age, eclectic, popular oldies, and jazz. Four trials gave information on the specific recordings used (composition title and composer) (Harmat 2008; Jespersen 2012; Kullich 2003; Lai 2005). Two trials stated the specific pieces of music used, but did not give performance‐specific information (Chang 2012; Shum 2014). Four trials described characteristics of the music (Chang 2012; Jespersen 2012; Lai 2005; Shum 2014). These shared common features such as low tempo (52 to 85 beats per minute), stable dynamic structure, and no strong rhythmic accentuation.

Four trials compared the music listening intervention to a no‐treatment control group (Chang 2012; Harmat 2008; Lai 2005; Shum 2014), and two trials compared music listening adjunctive to treatment‐as‐usual (TAU) versus TAU alone (Jespersen 2012; Kullich 2003). One trial had two active intervention groups (Harmat 2008), but only data from the music listening group compared to the no‐treatment control group were included in this review.

Outcomes

Five trials reported on sleep quality using the PSQI (Harmat 2008; Jespersen 2012; Kullich 2003; Lai 2005; Shum 2014). One trial used the polysomnography to objectively measure sleep onset latency, total sleep time, sleep interruption, and sleep efficiency (Chang 2012). The same trial also used a morning questionnaire to report subjective measures of sleep onset latency, total sleep time, and sleep interruption. No trials reported adverse events or deterioration of outcomes during the intervention period. None of the trials reported any secondary outcomes of interest to this review (see the Methods section). Harmat 2008 did measure levels of depression, but only in the intervention group and not in the control group.

Funding sources

Four trials were funded or partly funded by a grant from a national research council, university, government or foundation (Chang 2012; Harmat 2008; Jespersen 2012; Kullich 2003). Two trials reported no information on funding sources (Lai 2005; Shum 2014).

Excluded studies

We excluded 55 trials (from 61 reports) from this review. We excluded 19 trials because they did not have an RCT or qRCT design (e.g. no control group or no randomisation procedure); 19 trials because the participants were not adults with insomnia as defined in the Types of participants subsection (some trials used participants with no sleep problems, some included both good and poor sleepers, and some had no clear documentation of the insomnia problems). We excluded six trials because the intervention was not music listening (e.g. choir singing); six trials because they compared two active interventions (e.g. music versus muscle relaxation techniques); two trials because they had no sleep outcome measure, and three trials because they were not clinical trials (e.g. reflections on practice). We listed the reasons for exclusion for each individual trial in the Characteristics of excluded studies table.

Ongoing studies

Two relevant studies were still ongoing when this review was written (NCT02321826; NCT02376686); see Characteristics of ongoing studies.

Trials awaiting classification

One potentially relevant but unpublished trial was identified (Miller 2002). The trial is referred to in published material (Bernatzky 2011), but there is insufficient information to assess the trial for inclusion or exclusion. The author has been contacted, but has yet to respond (see Characteristics of studies awaiting classification).

Risk of bias in included studies

We assessed the six included trials for risk of bias across the following domains: 1) random sequence generation (selection bias); 2) allocation concealment (selection bias); 3) blinding of participants and personnel (performance bias); 4) blinding of outcome assessment (detection bias); 5) incomplete outcome data (attrition bias); 6) selective reporting (reporting bias); and 7) other bias. The results are depicted in Figure 2. Figure 3 provides a summary of the risk of bias results for each of the included trials. Reasons for the judgement are described in the risk of bias tables beneath the Characteristics of included studies tables.

Figure 2

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included trials

Figure 3

Risk of bias summary: review authors' judgements about each risk of bias item for each included trial

Allocation

All trials described the randomisation procedures applied. Five trials used a true randomisation procedure (e.g. using a computerized randomisation table), and were judged to be at low risk of bias (Chang 2012; Harmat 2008; Kullich 2003; Lai 2005; Shum 2014). One trial used a quasi‐randomised alternation procedure and was judged to be at high risk of bias (Jespersen 2012).

Four trials described measures taken to conceal allocation and were rated as being at low risk of bias (Chang 2012; Kullich 2003; Lai 2005; Shum 2014). One trial had limited information on allocation concealment and was judged to be unclear for a risk of bias (Harmat 2008), and one trial had no concealment procedure and was judged at high risk of bias (Jespersen 2012).

Blinding

Due to the nature of the intervention, it is not possible to blind participants and it can be difficult to blind personnel or researchers. Bias was considered likely to have occurred in those trials using subjective reports of sleep quality when participants were not blinded (Harmat 2008; Jespersen 2012; Kullich 2003; Lai 2005; Shum 2014). Bias was considered less likely to have occurred in the one trial involving objective measures of sleep, although this trial also included subjective measures of sleep (Chang 2012). Therefore, all trials were judged to be at high risk of performance bias.

In one trial, the technician scoring the polysomnography and the researchers doing the statistical analyses were blinded to group allocation (Chang 2012). One trial reported that group allocation was coded to blind the outcome assessors (Harmat 2008, and another trial reported that the data assessment was done by administrative personnel who were unaware of group allocation (Kullich 2003). These three trials were judged to be at low risk of detection bias. However, the three remaining trials were unblinded and rated as being at high risk of detection bias (Jespersen 2012; Lai 2005; Shum 2014).

Incomplete outcome data

Four trials reported no attrition and no missing data and were rated as being at low risk of attrition bias (Chang 2012; Harmat 2008; Kullich 2003; Shum 2014). In two of these studies (Harmat 2008; Kullich 2003) the information on attrition was not included in the published report but was provided by the authors at our request. One trial had unclear information on attrition and was rated as being unclear for this risk of bias (Lai 2005). Another trial reported the dropout of four participants resulting in an uneven distribution between groups, and was consequently rated as being at high risk of attrition bias (Jespersen 2012). We conducted a sensitivity analysis and found that including or excluding this study did not change the results. Therefore, we included it in successive analyses.

Selective reporting

One trial did not include outcome measures for the no‐treatment control group in the published report, but this information was provided by the authors at our request and did not change the results or conclusions of the published paper (Harmat 2008). Hence, we did not suspect reporting bias and judged the trial to be at low risk of reporting bias. In two trials, there was some uncertainty about the full reporting of outcomes, so these trials were rated as being unclear for the risk of reporting bias (Chang 2012; Shum 2014). There was no evidence of selective reporting in the remaining three trials, which were judged to be at low risk of reporting bias (Jespersen 2012; Kullich 2003; Lai 2005).

Other potential sources of bias

Other potential sources of bias were detected in four trials (Chang 2012; Harmat 2008; Jespersen 2012; Lai 2005). Three trials reported baseline differences between the intervention and control group (Chang 2012; Jespersen 2012; Lai 2005), and in one trial, methods of data collection differed between the intervention and control group (Harmat 2008). Thus, these trials were judged to be at high risk for other biases. No other risks of bias was detected in the final two trials and so these trials were rated as being at low risk for other bias (Kullich 2003; Shum 2014).