Topic I. The Practice of Meditation

The following questions pertain to the description of the practice of meditation and meditation techniques:

1.

What is known about the practice of meditation?

a.

What are the main components of the various meditation practices (e.g., breathing, chanting, mantras, and relaxation)? Which components are universal and which ones are supplemental?

b.

How is breathing incorporated in these practices? Are there specific breathing patterns that are integral elements of meditation? Is breathing passive or directed?

c.

For each type of meditation practice, where is the attention directed during meditation (e.g., mantra, breath, image, nothing)?

d.

To what extent is spirituality a part of meditation? To what extent is belief a part of meditation?

e.

What are the training requirements for the various meditation practices (e.g., the range of training periods, frequency of training, individual and group approaches)

f.

How is the success of the meditation practice determined (i.e., was it practiced properly)? What criteria are used to determine successful meditation practice?

Topic II. State of Research on the Therapeutic Use of Meditation Practices in Healthcare

The following key questions pertain to the scope of research on meditation in healthcare:

2.

What meditation practices have been examined in clinical trials and observational studies? What control groups are used?

3.

Can these practices be separated by the diseases, conditions, and populations for which they have been examined?

4.

What outcome measures are used? Are psychosocial outcomes included in these studies? If so, what types?

Topic III. Evidence on the Efficacy and Effectiveness of Meditation Practices

The following key questions pertain to the potential benefits and harms of meditation:

5.

What is the evidence that meditation practices are efficacious for the three most studied diseases or conditions identified in question 2 above?

6.

If more than one form of meditation has been studied for a particular disease or condition (as identified in question 2 above), does the efficacy of these practices differ?

7.

For specific disease populations, are meditation practices that are used as a complement to conventional therapy more effective than either the conventional therapy or meditation therapy alone?

Topic IV. Evidence on the Role of Effect Modifiers for Meditation Practices

The following key questions pertain to specific elements of the meditation practice, population and practitioner that may influence the outcomes:

8.

What dose of meditation is necessary before successful health outcomes are realized? That is, is the duration of meditation important for outcomes?

9.

Does the direction of attention during meditation affect outcomes?

10.

To what extent is a rhythmic aspect (i.e., mantra, controlled breathing, or other ordered, recurrent sound or motion) critical to the practice of meditation and to health outcomes? Do such approaches to meditation that rely on these rhythmic behaviors demonstrate consistent effectiveness versus nonrhythmic approaches to meditation? More broadly, do the number and types of components that make up the various meditation practices influence the outcomes?

11.

Do individual difference variables (age, gender, race, education, income, other) predict success in the process of meditation (i.e., adherence, acceptance), as well as predicting health outcomes?

Topic V. Evidence on the Physiological and Neuropsychological Effects of Meditation Practices

The following key questions pertain to the physiological and neuropsychological effects of meditation practices:

12.

What is known regarding cardiopulmonary, endocrine, immunologic, metabolic, and autonomic changes seen during meditation practices?

13.

What is known regarding the effects of meditation practices on brain function (e.g., brain imaging, electroencephalogram (EEG), neuropsychological and cognitive functions)?

Figure 1 presents the analytic framework for the review. We used two main methodological approaches to address the research topics discussed in this report.

Figure

Figure 1. Analytic framework for evidence report on the state of research on meditation practices in healthcare.

For topic I, the practice of meditation, the steps involved in the development of the descriptive overview included:

• development of an operational definition of meditation
• literature search
• study selection
• data extraction
• qualitative synthesis of information

For topic II on the state of research for the therapeutic use of meditation practices, topic III on the efficacy and effectiveness of meditation practices, topic IV on the role of effect modifiers of meditation practices, and topic V on the physiological and neuropsychological effects of meditation practices, a number of steps were involved in conducting the literature review and synthesis of the evidence:

• literature search and retrieval
• study selection
• assessment of study quality
• data extraction
• data analysis and synthesis

Development of Operational Parameters to Define Meditation Practices

There is no consensus on a definition of meditation in the scientific literature. For the purposes of this report, a set of parameters to describe meditation practices was developed using a modified Delphi methodology.^36,37 The systematic process used to reach consensus on the operational definition of meditation was documented and is described briefly below (Appendix B).¹

A first-round questionnaire was distributed to TEP members to solicit their opinion on a set of parameters extracted from the scientific literature to describe meditation. Participants independently rated the importance of each parameter to characterize a practice as meditation. They were also asked to suggest other parameters not included in the questionnaire that they considered important. A feedback summary from the first-round responses was sent to TEP members along with a second-round questionnaire. In light of round-one group responses, participants were asked if they would reconsider their first-round responses. The process stopped when consensus among participants was reached. Responses to questions were analyzed and categorized by frequency of endorsement. Consensus was defined as agreement on a value or category by 80 percent of the Delphi participants.³⁷ If consensus was not reached by the Delphi technique, the TEP convened and group consensus techniques were used in a teleconference.

Literature Search and Retrieval

Databases and search terms. The research librarian worked closely with the TEP to refine search strategies for all questions of the review. Comprehensive searches were conducted of the electronic databases listed in Table 1 for the time periods specified. The order of the databases in Table 1 is the sequence in which the databases were searched (Appendix C).^*

Table 1

Databases searched for relevant studies.

In addition to the electronic databases, the following journals and collections were hand searched: International Journal of Behavioral Medicine (1994-2005), Scientific Research on The Transcendental Meditation^® Program: Collected Papers (Volumes 1 to 4), Journal of Bodywork & Movement Therapies (1996-2005), Journal for Meditation and Meditation Research (2001-2003), International Journal of Yoga Therapy (1997-2005), and Explore: The Journal of Science and Healing (2005).

The reference lists of relevant studies (e.g., included studies, other systematic or narrative reviews) were reviewed to identify potentially relevant studies. Gray literature was searched to identify unpublished studies and works in progress. Scientific abstracts from the Society of Behavioral Medicine (2005) and the American Psychosomatic Society (1999-2005) annual scientific meetings were reviewed. The National Research Register from the National Health Service was searched for ongoing trials. Primary authors of potentially eligible ongoing studies were contacted if this was necessary to clarify whether those studies did indeed meet the inclusion criteria. TEP members were also requested to provide additional information about potentially relevant studies.

Criteria for Selection of Studies

A set of strict eligibility criteria was used to determine the inclusion and exclusion of studies for the report. The inclusion criteria for topic I are documented in Table 2. It is important to emphasize that the review on Topic I does not constitute a manual for any meditation practice. A more detailed explanation of any specific meditation practice described in this report should be sought in specialized texts or from master practitioners.

Table 2

Inclusion criteria for topic I.

Information from primary studies and other original research identified for topics II to V was considered for topic I if it provided a detailed description of the meditation practice under study according to the parameters defined by the Delphi process.

Our inclusion criteria for topics II to V are documented in Table 3. Some criteria are common to all of these topics, but some criteria were specifically developed for inclusion of studies in topics III and IV only. We sought to match the type of evidence required to the nature of the questions and to identify the highest quality of evidence appropriate to answer each group of questions. For topics III and IV on the efficacy and effectiveness of meditation practices, and on the role of effect modifiers for meditation practices, we looked for evidence from randomized controlled clinical trials (RCTs) and nonrandomized controlled clinical trials (NRCTs). No restrictions were applied for setting or geographical location of the studies. Only studies published in the English language were eligible according to the scope outlined by NCCAM for this review.

Table 3

Inclusion criteria for topics II to V.

Study Selection Process

Screening of titles and abstracts. We developed a predefined set of broad criteria to apply to the results of the literature searches to ensure that potentially relevant articles were not excluded early in the selection process (Appendix D).² Four independent reviewers evaluated the title and abstract of each study to select references potentially relevant to the topics of the report. The full-text of studies meeting the criteria was retrieved as was the full-text of those that reported insufficient information to determine eligibility.

Identification of studies eligible for the review. Two independent reviewers appraised the full-text of potentially relevant articles using a standard form that outlined the inclusion and exclusion criteria for each research topic (Appendix D).* Decisions regarding inclusion and exclusion and the reasons for exclusion were documented.

The level of agreement among reviewers at all stages of the selection process was evaluated using the Kappa (κ) statistic.³⁸ A κ score in the range from 0.0 to 0.40 was considered poor agreement; 0.41 to 0.60 moderate agreement; and 0.61 to 0.80 substantial agreement.³⁹ Disagreements about the inclusion or exclusion of studies were resolved by consensus. When consensus was not reached, a decision was made in consultation with the TEP.

Evaluating the Methodological Quality of Studies

Rating the quality of individual articles. Studies included in the descriptive overview on the practice of meditation (topic I) were not assessed for methodological quality; therefore, the following methods for quality assessment apply to studies meeting eligibility criteria for topics II to V only.

Quality of intervention studies (RCTs, NRCTs, and before-and-after studies). The methodological quality of RCTs was assessed using the criteria for concealment of allocation^40,41 and the Jadad scale.⁴² The former is based on the evidence of a strong relationship between the potential for bias in the results and allocation concealment: failure to conceal the process of treatment allocation can undermine randomization and, consequently, a selection bias may occur.⁴⁰ The Jadad scale is a validated scale that includes questions related to bias reduction: randomization, double-blinding and description of dropouts and withdrawals. This tool scores quality from 0 to 5. Studies scoring less than 3 points are usually considered to be of low quality.⁴² The psychometric properties of the Jadad scale have been thoroughly tested, providing rigorous evidence to support its use.^42,43 We used individual components of the Jadad scale to create a 3-point scale based on blinding and participant attrition to assess the methodological quality of NRCTs.

No completely or partially validated instruments are available to assess the methodological quality of uncontrolled before-and-after studies. Quality of reporting of uncontrolled before-and-after studies included in topics II and V was evaluated with four questions assessing whether the study participants were representative of the target population, the method of outcome assessment was the same for the pre- and postintervention periods for all participants, outcome assessors were blind to the intervention and the purpose of study, and the number of and reasons for study withdrawals were reported.

Quality of observational analytical studies. The methodological quality of observational analytical studies (i.e., prospective and retrospective observational studies, case-control studies, and cross-sectional studies with controls) was assessed using the Newcastle-Ottawa Scales (NOS).⁴⁴ These are eight-item instruments that use a star system to assess methodological quality across three categories: the methods of selecting the study groups, their comparability, and the ascertainment of the outcome of interest. Scores range from 0 to 9 stars. The NOS scales have been recommended by the Cochrane Nonrandomized Studies Methods Working Group, and studies on their psychometric properties are in progress.⁴⁴

The assessment of quality of observational studies is more difficult than the assessment of RCTs and NRCTs. Empirical research has shown that numeric scores based on arbitrary weights given to each item in a scale are unreliable and difficult to interpret.⁴⁵ Therefore, we decided to describe the methodological quality of observational analytical studies using the individual components of the NOS scales.

Finally, information regarding the source of funding was collected for all the included studies.⁴⁶

Two reviewers assessed the methodological quality of studies independently. Disagreements were resolved by consensus or, when no consensus could be reached, a senior methods expert adjudicated (Appendix D).^*

Data Collection

For topic I on the practice of meditation, a single reviewer extracted information that was organized according to narrative categories (e.g., components of the meditation practices, role of breathing and spirituality, training requirements, and criteria for success) to allow for a systematic description of the meditation practices considered in this report.

For topics II to V, trained research staff at the UAEPC extracted the information. A comprehensive and pretested data extraction form and guidelines explaining the extraction criteria were developed (Appendix D)^*. Information regarding the study design and methods, the characteristics of participants, interventions, comparison groups, and outcomes of interest were extracted. Data collection on study design and methods included information on the country and year of publication, type of publication, objective of the study, study design, duration, number of centers, and source of funding. Data on characteristics of the participants included setting of the study, type of primary health problem or health condition of study participants, and diagnostic criteria (as reported by the authors of the studies). Data on characteristics of the intervention (i.e., meditation practices) included a description of the practice in terms of components, content and format, frequency, and intensity. Likewise, data on the characteristics of the control group included a description of the components, content, and format. Finally, information was extracted on the type of outcomes and on the units or instruments of measurement for each outcome. A single reviewer extracted the data from the primary studies and another independent reviewer verified the accuracy and completeness of the data. Any discrepancies in data extraction were solved by consensus between the data extractor and the data verifier. During this process, the reviewers consulted with TEP members both for content and methodological advice as needed.

Study selection, methodological quality assessment, and data extraction were managed with the Systematic Review Software™ (SRS), version 3.0 (TrialStat!; Ottawa, ON). Graph extraction was performed using Corel Draw^®, version 9.0 (Vector Capital, San Francisco, CA). Extracted data were exported into Microsoft Excel™ (Microsoft Corporation, Redmond, WA) spreadsheets.

Data Analysis and Synthesis

Classification of the meditation practices. The first step in synthesizing the data for topics I to V was to create categories of analysis for the meditation practices described in the scientific literature. Based on data from the Delphi study, input from the TEP members, and a review of the literature, a set of seven categories was constructed to classify the meditation practices. Two independent researchers coded each study according to this classification scheme. Coding was discussed between researchers on a study-by-study basis. Coding discrepancies were resolved by consulting the original research study.

The following seven categories were used for data synthesis for topics II to V:

Mantra meditation. This category comprises meditation practices in which a main element of practice is mantra: the Relaxation Response technique (Relaxation Response or RR), the Transcendental Meditation^® technique (hereafter, simply “Transcendental Meditation^®” or “TM^®”), Clinically Standardized Meditation (CSM), Acem meditation, Ananda Marga, and other concentrative practices that involve the use of a mantra such as Rosary prayer, and the Cayce method.

Mindfulness meditation. Though described slightly differently by Eastern and Western interpreters, this category refers generally to meditation practices that cultivate awareness, acceptance, nonjudgment, and require paying attention to the present moment.^47–49 This category includes Mindfulness-Based Stress Reduction (MBSR), Mindfulness-Based Cognitive Therapy (MBCT), Vipassana meditation, Zen Buddhist meditation, and other mindfulness meditation practices not further described.

Qi Gong. This category refers to an ancient practice from traditional Chinese medicine that combines the coordination of different breathing patterns with various physical postures, bodily movements, and meditation. External Qi Gong, in which a trained practitioner directs his or her own qi outward, with the intention of helping patients clear blockages, remove negative qi and balance the flow of qi in the body, to help the body rid itself of certain diseases is not a form of meditation according to the working definition developed for this report.

Tai Chi. This category describes a Chinese martial art characterized by soft, slow, flowing movements that emphasize force and complete relaxation. It has been also called “meditation in motion.”

Yoga. This category includes a broad group of techniques rooted in yogic tradition that incorporate postures, breath control, and meditation. It includes practices such as Hatha yoga, Kundalini yoga, and individual components of Yoga such as pranayama (breath control exercises).

Miscellaneous meditation practices. This category describes techniques that combine different approaches to meditation in a single intervention, without giving prominence to one. It includes combined practices such as Yoga plus RR, TM^® and Buddhist Meditation, and RR plus Mindfulness meditation. The category was also used to describe meditation practices that do not fall within any of the other categories (e.g., coloring mandalas).

Meditation practices (not described). This category refers to meditation practices that were not described in sufficient detail to allow them to be assigned to a more specific category, including techniques that were described by vague terminology such as “meditation,” “movement meditation,” and “concentrative meditation.”

Topic I. Data for topic I on the practice of meditation were synthesized qualitatively. Information was presented in a structured format, with narrative categories of interest for the different practices of meditation identified in the scientific literature. Once categorized, the similarities and differences among the various meditation practices could be appraised. Categories of analysis include the main components of the meditation practices, the role of breathing, attention, and spirituality, the training requirements, and the criteria of success for the various meditation practices.

Topic II. Data collected for topic II on the state of research for the therapeutic use of meditation practices were summarized using descriptive statistics (e.g., proportions and percentages for categorical data, means with standard deviations [SD], or medians with interquartile ranges [IQR], for continuous data). Evidence and summary tables were constructed to summarize relevant characteristics of the included studies. Data from the included studies were synthesized qualitatively. We used the systematic approach of the Cochrane Collaboration for the synthesis of the evidence.⁵⁰ The basic conceptual framework of the qualitative synthesis for topic II focused on the types of meditation practices that have been examined in intervention studies (RCTs, NRCTs, and uncontrolled before-and-after studies) and observational analytical studies (cohort studies, case-control studies, and cross-sectional studies), the types of control groups, the populations, and the types of outcome measures that have been examined in the included studies.

Topics III and IV. Based on the results of topic II describing the populations that have been examined, RCTs and NRCTs assessing the effects of meditation practices for the three most studied clinical conditions were included in the analyses of efficacy and effectiveness of meditation practices (topic III) and the role of effect modifiers for meditation practices (topic IV). The first step in synthesizing the data for topics III and IV was to construct evidence tables that included information on each article's source, study design, study population (e.g., sample size, age, and gender), treatment groups, and outcomes. The evidence tables also included summaries of study quality and comments to help interpret the outcomes.

Meta-analyses were planned as part of the data analysis to derive pooled estimates from individual studies to support inferences regarding the magnitude and direction of the effect of the meditation practices. If studies evaluating specific meditation practices were sufficiently similar, effect sizes were combined and weighted using the standard inverse variance method⁵¹ to produce an overall effect size for a given outcome. Meta-analyses used a random effects model. In this method, study means are averaged, weighting by a combination of inverse variance augmented by heterogeneity.

The types of summary statistics considered were risk ratios (RR) or odds ratios (OR) with 95 percent confidence intervals (95% CI) for dichotomous outcomes and weighted or standardized mean differences (WMD and SMD, respectively) with 95% CI for continuous outcomes.⁵² WMD was chosen as the default method, with SMD being used only when units for the outcome were different among the studies being compared (i.e., stress measured on different scales).^50,53

Hedges adjusted g was used as the SD estimate when the SMD was used.⁵⁴ If the means were not reported, they were either imputed from medians or discarded from meta-analysis if neither mean nor median was available. Occasionally studies did not report SDs of their estimates. In these cases, we determined the SD exactly from confidence intervals or exact p-values; estimated the SD from upper-bound p-values, interquartile ranges, ranges, or exact nonparametric p-values; or imputed from other studies reporting similar outcomes in a similar population. All the meta-analyses used endpoint data or change from baseline to endpoint data instead of using the average of separate mean changes calculated at different intervals of time. Forest plots were used to display the individual and pooled results.

Since some common outcomes were reported for many interventions, indirect comparisons⁵⁵ were made of these active interventions. This type of comparison involves taking the differences between the differences derived from separate meta-analyses. For example, by taking the difference between the derived meta-analysis of A versus B, and the derived meta-analysis of A versus C an estimate of the comparison of B versus C can be obtained. For some outcomes, when more than four interventions could be compared indirectly, a mixed treatment comparison was conducted. Indirect comparisons are a valid approach to meta-analysis when there is insufficient direct evidence from randomized trials reporting head-to-head comparisons between interventions.^55,56

In this method, a Bayesian formulation of the data is employed. The differences between each intervention and a baseline intervention (in this case, “no treatment” was chosen as the baseline) are modeled by choosing a prior distribution for the effect and combining this prior value with the data from the studies to arrive at a posterior estimate and 95% credible interval. Such an estimate was obtained for all pairwise comparisons of interventions as well as the comparisons to the baseline intervention. Since the resulting posterior distributions are too complex for direct computation, a Markov Chain Monte Carlo simulation⁵⁷ was used to obtain the posterior estimates. This procedure involved simulating the unconditional, unknown posterior distribution by sampling many times from the conditional distribution and averaging the results. We used a sample of 20,000 burn-in iterations followed by 200,000 samples and noninformative normal (point estimate) and uniform (variance estimate) priors to obtain the distributions. We also computed a statistic to estimate the probability that each intervention was the best (e.g., lowered blood pressure the most) by recording the best intervention at each iteration. This simulation was performed using the WinBUGS software, version 1.4 (MRC Biostatistics Unit, Cambridge, United Kingdom).

We tested for statistical heterogeneity using the chi-square test⁵¹ and quantified it using the I² statistic.⁵⁸ When there was evidence of clinical or statistical heterogeneity among studies, effect size estimates with corresponding 95% CI were presented separately for each study.^59,60 Sources of heterogeneity were explored qualitatively. They may be methodological (differences in design or quality), or clinical (differences in key characteristics of participants, interventions, or outcome measures).⁶¹ Where appropriate, subgroup analysis based on patient, intervention, and study characteristics were conducted and sensitivity analysis based on study quality (Jadad score of greater than and equal to 3 points or less than 3 points) were conducted to assess the effect of quality on precision of the pooled estimates if the number of studies per comparison allowed it.⁶²

Two analytic strategies were considered for topic IV on the effect modifiers of meditation practices. First, a meta-regression analysis using RCT-level covariates was planned to explore whether certain characteristics of the participants (e.g., age, gender, ethnicity, education, and income) or the interventions (e.g., dose, frequency, and duration) were associated with increased benefits of meditation practices. The outcome (or dependent) variable in the meta-regression analyses would be the pooled effect size (log OR for binary outcomes, or WMD or SMD for continuous outcomes). If a meta-regression was not feasible due to a small number of trials, or limited data from primary studies, subgroup analyses would be conducted based on participant or intervention characteristics.

Topic V. Based on the types of outcomes identified in topic II, RCTs, NRCTs, and uncontrolled before-and-after studies (i.e., without a parallel control group) were included in the analysis of the physiological and cognitive/neuropsychological effects of meditation practices. Evidence tables were constructed to summarize each article's source, study design, study population (e.g., setting, sample size, age, and gender), treatment groups, and outcomes. The evidence tables also included summaries of the strength of the evidence, study quality, and comments to help interpret the outcomes.

Meta-analyses of RCTs and NRCTs using the methods described above for topic III were also planned for topic V. For studies with pre- and post-measures, data on change from baseline were used if available; otherwise, endpoint data were used. If meta-analytic methods were not feasible, effect size estimates with corresponding 95% CIs were presented separately for each study.^59,60 Data from uncontrolled before-and-after studies were analyzed separately, and, if appropriate, the individual estimates of the treatment effect were pooled using the generic inverse variance method. Sensitivity analyses were conducted to assess the robustness of the findings when necessary. Data were displayed using forest plots.

Publication bias. Publication bias, or the selective publication of research depending on the results, was assessed using funnel plots, and the trim and fill method⁶³ if enough data were available from the meta-analyses. Funnel plots of effect sizes (axis X) against the SD (axis Y) for each meta-analysis were examined to identify gaps suggesting publication bias. Finally, the trim and fill method provided estimates of the number of studies potentially missing from a meta-analysis and the effect these omissions might have had on its outcome.

All analyses were performed using SAS/STAT^® software version 9.1 (SAS Institute Inc., Cary, NC), Statistical Package for the Social Sciences^® for Windows^® (SPSS^® version 14.1, SPSS Inc., Chicago, IL), and RevMan version 4.1 (Cochrane Collaboration, Oxford, UK).

Potential limitations, conclusions, and implications for future practice and research were discussed. The results were interpreted in light of the heterogeneity of the individual studies (e.g., differences in design, study populations, interventions or exposures, and outcome measures) and any evidence of publication bias, if present. Recommendations for practitioners and researchers were based solely on the evidence available.