Overview

The systematic review involved a number of steps:

• Literature Search
• Development of Inclusion Criteria
• Study Selection
• Data Extraction
• Assessment of Study Quality
• Data Analysis

Literature Search

The research librarian, in collaboration with the TEP (Technical Expert Panel), developed and implemented search strategies designed to identify relevant evidence for key questions of the review. A systematic search of 21 electronic databases was conducted. Table 1 outlines the electronic databases that were searched. Table 2 outlines the subject headings and keywords that were used in the search. Appendix A ^♦ contains details of the search strategy. Most of the searches were limited to humans, and no age restrictions were applied to any of the searches.

Table 1

Databases searched.

Table 2

Subject headings and keywords used in searches.

For Question 1, which relates to the definition, classification, diagnosis, and etiology of chronic insomnia in adults, we searched for narrative and systematic reviews, book chapters, diagnostic manuals and standards of practice parameters, and applied English-language restrictions. For Question 2, which relates to the prevalence, natural history, incidence, and risk factors for chronic insomnia in adults, and Question 3, which relates to the consequences, morbidities, co-morbidities and public health burden associated with chronic insomnia in adults, we searched for observational studies, encompassing a range of designs including cross-sectional, case-control, and cohort studies, and applied English-language restrictions. For Question 4, which relates to the treatments for chronic insomnia in adults, and the evidence regarding their safety, efficacy, and effectiveness, we searched for randomized controlled trials, and no language restrictions were applied. We did not apply language restrictions to searches for Question 4, since a portion of this question involves a review of complementary and alternative medicine (CAM), and there is evidence to suggest that studies of some CAM topics are often initially published in non-English languages, and many of these are not published in English.²¹ We searched electronic resources that specialize in CAM, including AMED (Allied and Complementary Medicine), Alt HealthWatch, and Cochrane Complementary Medicine Field Registry. In order to systematically search for the different types of studies required for each question, it was useful to refer to: the highly sensitive search strategy for identifying reports of randomized controlled trials in MEDLINE^® from the Cochrane Reviewer's Handbook (Appendix 5b)²²; search strategies for diagnosis, etiology, natural history and morbidities from PDQ Evidence-based Principles and Practice²³; and the search strategy for systematic reviews in MEDLINE^® from the Alberta Research Centre for Child Health Evidence.²⁴ Searches were also conducted in databases that index grey literature, including SIGLE (System for Information on Grey Literature in Europe), OCLC Proceedings First, Dissertation Abstracts and the NLM Gateway (searched specifically for meeting abstracts).

No hand searching was conducted for this review, given that the key journals pertaining to chronic insomnia, such as Sleep and Sleep Medicine Reviews, are indexed in MEDLINE.

Development of Inclusion Criteria

We did not develop formal inclusion criteria for the question pertaining to the definition, classification, diagnosis and etiology of chronic insomnia, nor for the question pertaining to the future direction of insomnia-related research. The former question was answered by providing an overview of the literature, and the latter question was answered by assessing the limitations in the evidence for the other questions of the review.

Inclusion criteria were developed for three questions of the review. Question-specific inclusion criteria appear below. The questions have been numbered according the numbering system outlined in the Introduction of this report. In the interest of clarity, questions 2 and 3 will be referred to as the questions on manifestations of chronic insomnia, while question 4 will be referred to as the question on management of chronic insomnia.

1.

What are the prevalence, natural history, incidence and risk factors for chronic insomnia? Specific risk factors of interest include age, gender, race/ethnicity, psychiatric illness and psychological problems, medical disease, socioeconomic status and shift-work.

A study was considered to be relevant to the portion of Question 2 pertaining to the prevalence, natural history and incidence of chronic insomnia, if it met the following criteria:

• the report was written in English
• participants were at least 15 years old
• it examined chronic insomnia
• it had a cross-sectional or cohort design
• it assessed the prevalence, natural history or incidence of chronic insomnia

A study was considered to be relevant to the portion of Question 2 pertaining to risk factors for chronic insomnia, if it met the following criteria:

• the report was written in English
• participants were at least 15 years old
• it examined chronic insomnia
• it had a cohort, case-control, or cross-sectional design
• it assessed one of the risk factors of interest

2.

What are the consequences, morbidities, co-morbidities, and public health burden associated with chronic insomnia? Specific outcomes of interest include healthcare utilization, psychiatric illness, absenteeism, work performance, accidents, falls in the elderly, quality of life and social relationships, memory, cognitive function, mood, direct and indirect costs.

A study was considered to be relevant to this question of the review, if it met the following criteria:

• the report was written in English
• participants were at least 15 years old
• it examined chronic insomnia
• it had a cohort or cross-sectional design
• it assessed one of the consequences of interest

For Questions 2 and 3, a study was considered to examine chronic insomnia if this condition was defined as a sleep disturbance of four weeks or more, or the report explicitly mentioned that chronic sleep disturbance was examined.

3.

What treatments are used for the management of chronic insomnia and what is the evidence regarding their safety, efficacy, and effectiveness? Specific treatments of interest include prescription medication, over the counter medication, alcohol, behavioral therapy, combination therapy and complementary and alternative care.

A study was considered to be relevant to this question of the review, if it met the following criteria.

• the report was written in English
• participants were at least 15 years old, and the majority were at least 18 years old
• participants suffered from chronic insomnia
• participants were randomized to intervention or placebo
• participants and assessors were blind to treatment received
• it assessed at least one of the following outcomes, listed in order of importance in deriving conclusions of the review:
  • sleep onset latency
  • wakefulness after sleep onset
  • sleep efficiency
  • total sleep time
  • sleep quality
  • quality of life

Sleep onset latency was defined as the amount of time between the participant laying down to sleep and the onset of sleep; wakefulness after sleep onset was defined as the amount of time spent awake in bed following the attainment of sleep; sleep efficiency was defined as the amount of time spent asleep as a percentage of the total time spent in bed; and total sleep time was defined as the total time spent asleep while in bed. We used broad definitions of sleep outcomes in this review. For example, sleep onset latency could be defined as time to sleep, time to stage 1 sleep, time to stage 2 sleep or latency to persistent sleep. We believe that it was acceptable to combine studies with differing definitions of sleep onset latency in the analysis, since differences in the magnitude of estimations across definitions would be accounted for by subtraction of placebo effects from treatment effects. Although it could be argued that these definitions are significantly different, the optimal definition of sleep onset latency has not yet been determined. Nonetheless, differences between polysomnography, sleep diary and actigraphy definitions of sleep onset latency were explored indirectly through sub-group analyses.

Sleep onset latency and wakefulness after sleep onset were given the highest priority in deriving conclusions of the review, since they were considered the best indices of sleep initiation and sleep maintenance, respectively. However, sub-group analyses were conducted only on data relevant to sleep onset latency, since this outcome was the most highly reported outcome across studies.

If the majority of participants met one of the following criteria, the study population was considered to suffer from chronic insomnia:

• participants suffered from a sleep disturbance of 4 weeks or more
• participants were described as having a chronic/long-standing/persistent sleep disturbance
• participants were selected from a sleep disorders clinic

The 4-week cut-point for chronic insomnia was considered long enough to eliminate studies involving transient insomnia, and short enough to include studies involving persistent insomnia.

In the case of combination therapy, the combined treatment could be compared to either placebo or single treatment.

We acknowledged the fact that double-blinding is often not feasible in studies of psychological treatments by not requiring double-blinding in these studies for inclusion in the review. The placebo treatment for relaxation therapy and cognitive/behavioral therapy was minimal treatment, such as sleep hygiene recommendations or minimal instruction. We required a placebo control and randomization of participants to intervention groups in order to account for potential confounders in the analysis. That is, we wanted to control for potential improvements in insomnia symptoms that may occur during the natural course of observation, irrespective of treatment effects, and for systematic differences in the experimental and control groups.

Given that placebo for psychological treatment is variable and not standardized across studies, we restricted our analysis to a particular type of placebo such that our results could be put in some context i.e. the efficacy of psychological treatment could be judged against a particular type of comparator. We required that the placebo resemble the intervention of the study except that it was known to produce either no effect or only a minimal effect. Thus, component controls or attention-placebo were considered appropriate if they were thought to have at most a minimal effect. A waiting-list or measurement control was considered inadequate because no intervention was provided. A pill-placebo was considered inadequate because it did not resemble the experimental intervention, which did not involve administration of a pill.

Study Selection

The research librarian provided three databases containing the titles and abstracts of potentially relevant articles of the review; one database was relevant to the question on the definition and etiology of chronic insomnia, another database was relevant to the questions on manifestations of chronic insomnia, and another database was relevant to the question on management of chronic insomnia. In the first stage of study selection, two reviewers screened the titles and abstracts of all potentially relevant articles, independently. Each reviewer noted the titles and abstracts that were potentially relevant to the review, and these articles were retrieved. In the second stage of study selection, two reviewers appraised the potentially relevant articles, independently, using pre-determined, question-specific, inclusion criteria. Disagreements between reviewers were resolved by discussion and consensus. The rate of disagreement between reviewers and the primary reason for exclusion of potentially relevant articles were noted.

Assessment of Study Quality

The quality of studies relevant to the questions on manifestations of chronic insomnia was assessed using one of three instruments; studies on prevalence and incidence were assessed using a scale designed specifically for this purpose.²⁵ This scale assesses bias in sample selection, sampling frame, sample size, outcomes and their assessment, response rate, confidence intervals and sub-group analysis, and sample description. The maximum score is eight. A priori, it was established that a score of zero to two would be considered low quality, a score of three to five would be considered moderate quality and a score of six to eight would be considered high quality. All other studies relevant to manifestations of chronic insomnia were assessed using one of two Newcastle-Ottawa scales (unpublished), each scale specific to either cohort or case-control studies. The scale specific to cohort studies assesses bias in the selection of exposed and non-exposed cohorts, ascertainment of exposure, presence of outcomes at the start of the study, comparability of cohorts based on design or analysis, outcome assessment, and length and adequacy of follow-up. The scale specific to case-control studies assesses bias in the definition, selection, comparability, ascertainment of exposure, and non-response rate for both cases and controls, and how these groups compare on these items. The maximum score for the Newcastle-Ottawa scales is nine. A priori, it was established that a score of zero to two would be considered low quality, a score of three to five would be considered moderate quality and a score of six to nine would be considered high quality.

The quality of studies relevant to management of chronic insomnia was assessed using the Jadad scale.²⁶ This scale assesses bias in sample selection, outcome assessment, data analysis, and appropriateness of randomization and blinding methods. The maximum score is five. A priori, it was established that a score of zero to one would be considered low quality, a score of two to three would be considered moderate quality and a score of four to five would be considered high quality. The concealment of allocation of participants to treatment groups was also assessed.²⁷ Allocation was considered adequate, inadequate or unclear.

Appendix B contains the quality assessment tools used in this review.

Data Extraction

The following data were extracted for studies relevant to manifestations of chronic insomnia, as applicable: first author and year of publication, site, objectives, design, time-frame, intended sample size, response and follow-up rates, type of participants, definition of comparison groups, participants' gender, age, and ethnicity, and participants' co-morbid conditions at entry. For the question on prevalence, incidence, natural history and risk factors for chronic insomnia, additional data extracted included setting, sampling frame and method of sampling, data collection method, prevalence, incidence and natural history parameters. We did not identify studies with designs that would support the categorization of outcomes as either risk factors or consequences of chronic insomnia; therefore, data relevant to potential risk factors and potential consequences of chronic insomnia were extracted, and these outcomes were referred to as associated factors of chronic insomnia.

The following data were extracted for studies relevant to management of chronic insomnia: first author and year of publication, funding source and role of funding organization, design, whether an intent-to-treat analysis was conducted, number of participants enrolled and their distribution by gender, participants' age, number of withdrawals and reasons for withdrawal, duration of insomnia, participants' co-morbid conditions at entry, methods used to assess outcomes, details of the intervention, such as frequency and duration of treatment and timing and route of delivery, number of participants allocated to treatment groups and number analyzed in each group, length of follow-up, patient preference, and data relevant to sleep onset latency, wakefulness after sleep onset, sleep efficiency, total sleep time, sleep quality, quality of life and adverse events. A trained reviewer extracted relevant data, and a second reviewer verified the data extracted for accuracy and completeness.

Appendix B contains data extraction forms for the questions on manifestations and management of chronic insomnia.

The information gathered by data extraction was used to generate Evidence Tables. Appendix C contains these tables.

Data Analysis

Data relevant to manifestations of chronic insomnia were analyzed qualitatively, while data relevant to management of chronic insomnia were analyzed quantitatively.

Manifestations of chronic insomnia. For the questions on prevalence, natural history, incidence, risk factors and consequences of chronic insomnia, data relevant to each variable were analyzed separately, except for data relevant to potential risk factors and potential consequences of chronic insomnia, which were analyzed together as associated factors of chronic insomnia. The key features of all studies providing information on prevalence, natural history, incidence or associated factors of chronic insomnia were summarized in tables, such that data relevant to each variable appeared in a separate table. The information on prevalence was divided into three tables, one for prevalence in the general population, one for prevalence in outpatients of general practice and one for prevalence in clinical populations.

The following information was included in the tables on prevalence: first author and year of publication, study quality, study design, sampling frame, sampling method, response/follow-up rate, method of data collection, type of participants, duration of sleep complaints and definition of cases and comparison groups, gender distribution of sample, age distribution of sample, and prevalence estimates. The following information was included in the table on natural history: first author and year of publication, study quality, study design, time frame for the study, response/follow-up rate, type of participants, duration of sleep complaints, gender distribution of sample, age distribution of sample, and natural history estimates. The following information was included in the table on associated factors for chronic insomnia: author and year of publication, study quality, study design, type of participants, duration of sleep complaints, gender distribution of sample, age distribution of sample, response/follow-up rate, and a qualitative summary of the findings of the study. The qualitative summary of results was derived by consolidating information available in the results and conclusions of relevant studies. We did not identify information relevant to the incidence of chronic insomnia.

The data provided in the tables were synthesized to provide a description of the methods and results of the studies relevant to a given variable. In the analysis of prevalence of chronic insomnia, a range, median and interquartile range were provided for each population (general, outpatient and clinical), separately for high and moderate quality studies, where appropriate. In the analysis of associated factors of chronic insomnia, the qualitative summary of findings were summarized in terms of the studies that did or did not find an association between chronic insomnia and the various factors of interest.

Management of chronic insomnia. A priori, the drug interventions were categorized according to drug class i.e. benzodiazepines, non-benzodiazepines and antidepressants. It was considered acceptable to combine different drugs of the same category in a meta-analysis, based on similar mechanisms of action. For psychological interventions, it was considered acceptable to combine predominantly cognitive approaches in a meta-analysis, and also to combine predominantly relaxation approaches in a meta-analysis; however, it was considered unacceptable to combine these two types of psychological approaches in a meta-analysis, since they were considered too different in their modes of action. Relaxation techniques address somatized tension, and different forms of this type of therapy (progressive relaxation and group relaxation) were considered similar enough to be pooled. However, cognitive therapy addressing the cognitive aspects of insomnia was not thought to be equivalent to relaxation therapy because it targets different aspects of insomnia, and was considered separately. A few interventions (e.g. L-tryptophan, melatonin and valerian) were categorized under the heading of “complementary and alternative care”; however, separate meta-analyses were presented for these interventions. We did not combine these interventions in a meta-analysis, given their distinct modes of action.

For continuous outcomes (i.e. sleep onset latency, sleep efficiency), studies were combined using a Mean Difference (MD), with the exception of sleep quality and quality of life, where studies were combined using a Standardized Mean Difference (SMD). Dichotomous outcomes (i.e. safety outcomes) were combined using a Risk Difference. A number needed to harm (NNH) was also reported for any safety outcomes that were found to be statistically significant. The Inverse Variance Method²⁸ was used to weight the studies. An efficacy estimate, with corresponding 95% Confidence Interval (CI), was computed for each outcome. For interpreting estimates calculated using the SMD, we used the generalization of 0.2 as small, 0.5 as moderate, and 0.8 as large.²⁹

We were usually able to calculate the efficacy estimates for each study exactly (i.e. mean difference, standardized mean difference, risk difference), but occasionally, estimates had to be made by extracting from graphs or using medians. Standard errors of the differences were calculated exactly from available data (i.e. individual patient data or exact P-values), whenever possible. For studies with a parallel design, this calculation was usually accomplished with the standard formula for variance of difference of independent variables: var(A-B) = var(A) + var(B). For studies with a crossover design, the standard error was estimated using the formula for variance of difference of dependant variables: var(A-B) = var(A) + var(B) -2ρ(var(A)var(B))^½ and using a correlation estimate of 0.5. In cases where exact values could not be obtained, standard errors were estimated using conservative P-values (i.e. p < 0.05), ranges, inter-quartile ranges, and extracting from graphs. As a last resort, an average of standard deviations of other studies was used to impute standard deviations of a study.

For studies with a parallel design, change from baseline data were used if available, otherwise final data were used. For studies with a crossover design, final data were always used. When continuous data were presented for multiple conditions, which we wished to combine, a new mean and standard deviation were computed.

All meta-analyses were performed using a Random Effects Model. Bailey³⁰ suggests that the Random Effects Model is more appropriate when making recommendations for management and treatment of the next given patient.

For some outcomes (sleep onset latency and number of adverse events), treatment categories were compared indirectly, via their relationship to placebo. Differences of differences with 95% CI were computed. Indirect comparisons were not made between pharmacological and psychological treatments for the following reasons (1) although our inclusion criteria required blinding for drug and complementary and alternative care, this criteria was omitted for psychological treatments (2) the placebo intervention was considered to have no effect for drug and complementary and alternative treatments, while it may have had minimal effect for psychological treatments (3) the pool of participants for psychological interventions was much smaller than for either the benzodiazepines, non-benzodiazepines or antidepressants. Thus, only indirect comparisons between non-psychological intervention categories and between psychological intervention categories were made.

All estimates of efficacy were assessed for heterogeneity using the I-squared statistic.³¹ Based on this statistic, heterogeneity for each outcome was classified as negligible (I² = 0 percent), minimal (I² < 20 percent), moderate (20 percent < I² < 50 percent), or substantial (I² > 50 percent). This measure of heterogeneity describes the degree of variation in the efficacy estimates among studies. For our primary outcome (sleep onset latency), heterogeneity was explored in sub-group analyses using a number of variables. The following variables were targeted a priori and explored in sub-group analyses: treatment sub-group (i.e. type of drug or therapy), presence or absence of psychiatric illness (as defined in the study inclusion criteria), length of treatment (short-term and long-term, defined as less than or equal to 4 weeks and greater than 4 weeks, respectively), age (adult and elderly defined as the majority of patients 15–65 years or greater than 65 years, respectively) and gender (male and female). Method of measurement of sleep outcomes (polysomnography, sleep diary actigraphy) was analyzed post-hoc in a sub-group analysis based on comments from peer reviewers. Study quality (low, moderate and high quality defined as Jadad scores of 0–1, 2–3 and 4–5, respectively) was also explored in a sensitivity analysis. Deeks' chi-square statistic ³² was used to test for significant heterogeneity reduction in partitioned sub-groups.

Publication bias is the publication of studies based on the nature and direction of results. We tested for publication bias visually using the Funnel Plot³³ and quantitatively using the Rank Correlation Test,³⁴ the Graphical Test,³⁵ and the Trim and Fill Method.³⁶

Footnotes

♦

The Appendices and Evidence Tables cited in this report are provided electronically at http://www​.ahrq.gov/clinic/tp/insomntp​.htm.