Forty-eight experimental studies of non-mechanically assisted relaxation techniques used to control a variety of clinical symptoms were synthesised using meta-analysis. Effect sizes for three types of comparisons, experimental-control, experimental-placebo, and pre-post, ranged from .43 to .66, demonstrating that treatment of any type included in the analysis moved the client from the 50th to the 67th percentile of an untreated group at minimum and from the 50th to the 75th percentile at maximum. All treatments included in the analysis except Benson's relaxation technique demonstrated evidence of effectiveness, particularly for non-surgical samples with chronic problems such as hypertension, headache, and insomnia.
Many clinical problems in nursing, such as nausea, vomiting, and pain, are known to be associated with anxiety and tension, states that have been found to elicit a stress response (Benson, 1975, Jacobson, 1938). In addition to standard medical procedures, several techniques broadly categorised as self-regulating mechanisms have been proposed as useful treatments for stress related symptoms. These include varieties of relaxation exercises, biofeedback, and meditation. Studies of these mechanisms have used a wide variety of relaxation techniques with different populations, symptoms, and outcome measures. These variations have made it impossible to have confidence in the overall effectiveness of such techniques or to isolate or study a number of the features critical to the better understanding of their potential value.
The commonalties among techniques as well as the uniqueness of each are unclear. Moreover, we do not know the efficacy of specific techniques in relationship to specific populations, human states or stages, or for different clinical problems. Therefore, the purposes of this study were to determine: the overall effectiveness of relaxation techniques in relieving clinical symptoms; the effectiveness of some techniques over others, particularly in relation to specific symptoms; whether some outcome measures are more sensitive than others; and, whether other variables related to study design or to characteristics of the subjects studied affected findings.
Sampling: To limit the sample of studies, relaxation training was defined as a non-pharmacologic, non-mechanical assisted technique to facilitate a relaxed state, like Benson's (1975) relaxation technique, Jacobson's (1938) progressive muscle relaxation (PMR), rhythmic breathing (RB), imagery, Lamaze, meditation, autogenic training, hypnosis, transcendental meditation (TM), Yoga and Zen. Biofeedback assisted relaxation studies were included only when an alternative relaxation technique was used as another treatment in the study, and then only the other treatments were examined. Biofeedback treatments were omitted because the investigators wanted to study only treatments relatively easy to use in nursing practice and not those requiring complicated technology. Only studies published after 1970 that investigated the effects of relaxation training on adult subjects not hospitalised for psychiatric reasons were used. In addition, to be included in the meta-analysis, studies had to contain data that permitted calculation of effect size (ES). Three types of comparisons were included : experimental (E) versus control (C) (E/C), E versus placebo (P) (E/P), and pret-test versus post-test (pre-post).
The descriptors relaxation, hypnosis, biofeedback, meditation, Yoga, Zen, imagery, pain, hypertension, and anxiety were used for a computer search of the literature. In addition, abstracts were reviewed from Dissertation Abstracts International, Psychological Abstracts, Sociological Abstracts and Cumulative Index to Nursing and Allied Health Literature, and World Nursing Research were also reviewed. Professional networking and contacts with major institutions likely to initiate relevant studies were used to locate unpublished studies (Schwartz, Moody, Yarandi & Anderson, 1987).
The literature search resulted in a sample of approximately 100 studies, of which 48 fit the criteria for inclusion. A list of these studies can be obtained from the investigators.
Based on the recommendations of Glass, Mc Gaw and Smith (1981), a number of independent variables were assessed for each study. These variables included publication year and source( published versus unpublished) and characteristics of subjects, treatment, study design, and outcome measures. Two variables of particular interest were study validity and reactivity of the outcome measure. Validity of each study was rated Low, medium, or high based on two criteria : randomisation and mortality (both its extent and its equivalence in treated and untreated groups). Reactivity of outcome measure was defined as the extent to which the measure revealed or paralleled the obvious goals of the experimenter, was under the control of the experimenter, or was subject to the client's need and/or ability to alter scores to show more or less change than actually occurred. Reactivity was rated using a 5-point scale from low for physiological measures too high for experimenter rating of symptoms or improvement (Glass et al., 1981, pp. 83, 84).
Initially, a total of 57 quantifiable variables were abstracted from each study. Percents of agreement for four doctoral students in nursing and one of the investigators after 2½ days of training were calculated for 54 categorical items over total of five studies coded immediately following the training. The range of percent of agreement was from 70% to 100%, with a median of 92%; for 90% of the items, agreement was 80% or better.
Reliability's for threats to validity and effect size, computed using analysis of variance (ANOVA) (Rosenthal, 1984), were .88 and .92, respectively. Reactivity of outcome measure could not be assessed using ANOVA due to restricted variance in the ratings. When one study which seemed to have caused considerable difficulty was omitted, 83%, or 36 pairs of ratings either agreed or fell within one point of each other.
The measure of ES used in this study was the standardised mean difference as defined by Glass et al. (1981):
|Xe - Xc|
When the appropriate means and standard deviations were unavailable, ES estimates were derived from other available statistics, such as t values, p values and proportions, using formulas and tables provided Glass et al. When both pre- and post- treatment scores were reported, ES values were adjusted for pre-treatment differences on that variable. The obtained value for ES was assigned a positive sign when the desired outcome was greater for the treatment group, a negative sign when the reverse was true. If data were insufficient to calculate ES, attempts were made to contact the authors of that particular study. These attempts were, by and large unsuccessful.
To study the overall effectiveness of relaxation techniques in relieving symptoms, weighted, unbiased estimates of ES (Wolf, 1986) were also calculated by using procedures which post-date Glass et al. (1981). The un-weighted arithmetic mean was the primary measure of central tendency of effect size for the other research questions.
Data Analysis Techniques: Because studies differed in the number of experimental treatments outcome measures, and time points at which data were collected, they differed as to the number of individual ES values that could be computed. Two analyses were therefore undertaken; one using the individual ES as the unit of analysis, and a second using the average ES per treatment-control (of treatment-placebo) comparison (hereafter referred to as the treatment-type analysis). Sample sizes used in these treatment-type analyses consisted of the total N for the two groups being compared. For purposes of this article, only the treatment-type analyses will be reported, except for the question of whether some outcome measures are more sensitive than others, which could be handled only by the individual ES analysis. Data for overall ES will be presented for all types of analyses.
One of the research questions motivating this study concerned the efficacy of specific techniques for specific problems; therefore two-way ANOVA using treatment type and presenting- problem as the independent variables and ES as the dependent variable was planned.
For such analysis to be meaningful all cells of the design needed to be filled. Because such was not the case, only limited aspects of the treatment-by-symptom question could be addressed, and only for E/C comparisons.
Tests of homogeneity (Wolf, 1986) were done for each analysis. Significant heterogeneity was found only for the individual ES E/C analysis. Thus, each of the other ES samples could be viewed as coming from a single population of ESs.
In interpreting ESs derived from all analyses, Cohen's (1977) criteria for small, medium, and large effects were used. Effects of .20, .50, and .80 were defined as small, medium, and large, respectively.
Analyses were done separately for the three types of comparisons: E versus C (E/C), E versus P (E/P), and pre-post. For the analysis using the treatment type as the unit of analysis, 48 studies resulted in 55 E/G comparisons (6 studies with two comparisons, 5 with three comparisons, and 28 with only one comparison); 8 studies contained 14 E/P comparisons; and 8 studies contained 9 pre-post comparisons. For a more detailed report of the work described here see Hyman, Feldman, Harris, Levin, and Malloy (1988).
Experimental Versus Control Comparisons: Effect sizes for this analysis were fairly evenly distributed over the years 1973 through 1986 except for a relative dearth of studies in 1973, 1981, and 1983. Approximately 60 of the ESs came from published sources. The average age for subjects based on 32 ESs was 40.95 years (SD = 10.47, median = 42.2), and the median percent male was 36.5 (N = 43).
The majority of ESs (80%) in this analysis came from non-surgical groups who suffered largely from chronic continuous (60%) or acute (25.5%) clinical, symptoms. The most common presenting problems were insomnia (20%), acute pain (18.2%), anxiety (16.4%) and hypertension (16.4%). The most prevalent treatment types were PMR(38.2%)}other (29.7%), and combination (20%). The majority of ESs (63.6%) come from studies including more than one treatment. Most were rated medium (45.5%) or high (36.4%) in internal validity. Most studies (64%) employed no-treatment control groups, with 31.6% employing standard treatment control groups. Sample sizes ranged from 10 through 72, with a mean of 33.13 (SD = 15.21, median = 34.0). Reactivity ratings were usually very low, 1, (35.8%), or quite high, 4, (43.1%), reflecting the use of many physiological measures on the one hand and self-report of symptoms on the other.
The average ESs for individual ES and treatment-type analyses appear in Table 1. ESs ranged from - 1.30 to 3.80 in the treatment-type analysis and from -3.26 to 3.62 in the individual ES analysis. The ES calculated using treatment as the unit of analysis was somewhat larger than that using individual ES as the unit of analysis. In the plot of sample size versus effect size, ESs seemed to converge toward the average value as sample size increased, which thus confirmed that all ESs came from a single underlying population (Light & Pillemer, 1984, p.63).
The breakdowns of ES by publication source, type of sample, and treatment type appear in Table 2, along with breakdowns by outcome type and presenting problem.
Because of the presence of many empty cells, a two-way ANOVA with treatment type and presenting problem could not be done. There were, however, sufficient PMR studies to warrant comparing the effectiveness of PMR for various presenting problems. Means and standard deviations for this analysis are presented in Table 3.
Mean, median, and unbiased weighted ESs for the E/G comparisons (Table 1), along with their equivalent percentiles, indicate, depending on the type of analysis and on the particular measure of central tendency, that a person who learned a relaxation technique fell between the 63rd and 72nd percentile as compared to an untreated person, who scored at the 50th percentile. Overall, relaxation techniques were, therefore, effective, with greater effects for non-surgical samples and in published studies (Table 2).
When effects for different types of presenting problems were compared (Table 2), effects were large for headache and hypertension, moderate for insomnia, low or moderate for chronic pain, anxiety, and other, and low for acute pain. Given the limitation of the small numbers of studies in some categories, treatments most effective were meditation, PMR, and combination, with PMR most effective for hypertension, headache, chronic pain, and insomnia, and moderately effective for anxiety. Outcomes most sensitive to relaxation were headache and diastolic blood pressure. There were no relationships of ES with any of the other independent variables examined. Experimental versus Placebo Comparisons; Although publication year for the placebo studies ranged from 1973 to 1986, the majority came from the years 1975 through 1978. There were none in 1974 or from 1979 through 1985. Of the ESs in this analysis, 57.1% were from published, 42.9% from unpublished sources. The average age of subjects based on only six cases was 40.79 years (SD = 12-19, median - 42.0), and the median percent male was 28.0. The majority of ESs (78.6%) came from non-surgical samples who suffered from chronic continuous problems (71.4%), mostly insomnia (42.9%), and acute pain (21.4%). (See Table 2.) Headache was a common problem in the individual ES analysis. Thirteen of the 14 ESs came from studies including more than one treatment, and the primary treatment used was PMR (35.7%). Sample sizes ranged from 18 through 50, with a mean of 35.5 and a median of 37.0. Most studies in this sample were rated either medium (42.9%) or high (35.7%) invalidity. Again, most reactivity ratings were either very low (35.3%) or quite high (54.4%). The most prevalent outcome measure was pain (30.9%).
The average ESs for both types of analyses appear in Table 1. ESs for the treatment type analysis ranged from - 0.14 to 1.52, with those from the individual ES analysis ranging from -0.96 to 1.86. The ES calculated using treatment as the unit of analysis for E/P comparisons was thus quite similar to that computed using the individual ES as the unit of analysis.
In general, ESs for specific outcome types in this analysis were larger than in the E/G analysis, as was the average ES- Particularly striking are the relatively large. ESs for fear/anxiety and systolic blood pressure.
Depending on which measure of average ES was used, effects shown in Table 1 for the E/P comparisons indicate that a person in a relaxation-treated group fell anywhere from the 71st to the 73rd percentile as compared to an untreated person, who fell at the 50th percentile. Relaxation techniques were again shown to be effective, more so for non-surgical samples. Effects were large for combination, Benson's, and RB, and moderate for PMR. Symptoms showing the largest effects were hypertension and headache, with moderate effects for anxiety, insomnia, and acute pain. Outcome types with the largest effects were fear/anxiety and systolic BP. A low negative relationship, r = -.20, was found between ES and reactivity, indicating that the more potentially biased measures produced lower ESs. A small positive relationship between validity rating and ES, r=.36, was also found.
Pie-Post Comparisons: Publication year for these studies ranged from 1974 to 1982, with half of the studies published in 1974. Seven ESs were from published sources, data on this variable missing for the eighth ES. The average age of subjects (n = 6) was 4888 (SD = 5.03, median = 47.65), and the median percent male (u = 6) was 62.0. All subjects were from non-surgical samples and had chronic continuous problems (See Table 2.), Five of these eight ESs came from studies on samples with the symptom of insomnia (n = 2) or hypertension (n=3); treatment type was largely meditation (62.5%), with some autogenic relaxation studies included as well. (These treatments did not appear m the other two analyses). All but one ES came from studies using only one treatment. Outcome variables were more limited than those for the E/C or E/P comparisons to mostly systolic BP and other. Sample sizes ranged from 7 through 51 (M=21.88, SD = 17.42). with a median of 14. Reactivity ratings followed a pattern similar to that discussed previously: they were either very low or quite high.
The pre-post average ESs for both types of analysis appear in Table 1. ESs for the treatment-type analysis ranged from -0.96 to 2.80, with those for the individual ES analysis ranging from 0.06 to 1.21.
Depending on which measure of average ES was used, effects in this analysis put treated subjects anywhere from the 64th to the 76th percentile, compared to untreated subjects, who fell at the 50th percentile. Meditation was shown to be an effective treatment, and symptoms most affected by treatment in general were chronic pain and hypertension, with insomnia and perhaps headache also somewhat affected. Pain was the outcome most affected. Statistically significant correlation's between ES and sample size, validity, and publication year resulted from this analysis but all were suspect for various reasons (see Hyman et al., 1988).
The data from this study indicate that relaxation techniques do affect some clinical symptoms. The uncorrected ESs from both individual ES and treatment analyses ranged from .43 through .66 and demonstrated that treatment of any type included in the analysis moved the client from the 50th to the 67th percentile of the untreated groups at minimum and from the 50th to the 75th percentile at maximum. Comparable figures for the corrected estimates indicate movement to the 66th percentile at minimum or the 76th percentile at maximum. Clearly, the treatments worked better for non-surgical than for surgical subjects.
Given the small numbers of studies in some categories, conclusions regarding efficacy of specific treatments are tentative. All treatments included in the study when examined individually, however, showed some evidence of effectiveness; one possible exception was Benson's, which had a medium ES in one analysis but zero effect in the other two.
The presenting problem most consistently effected was hypertension, with headache and insomnia also showing fairly consistent evidence of treatment effects. Results for chronic pain and anxiety were low to moderate, and the results for acute pain were quite low. The high ES obtained from the E/P comparisons is reminiscent of findings from a previous study (Levin, Malloy, & Hyman, 1987), in which the placebo was associated with an increase in the distress component of pain post-operatively. It would be interesting to know how subjects perceive the experience of being in a placebo group.
The fact that effects were larger for unpublished sources in this study runs counter to many findings in previous meta-analyses. The only relationships of study design variables with ES occurred for the E/P comparisons, where ES was positively related to validity rating and negatively related to reactivity. The finding with regard to reactivity was opposite from what would be expected. More potentially biased measures would be expected to produce larger effects. The fact that studies included in this meta-analysis represent approximately half of those obtained through the literature search constitutes a limitation on general-izability of the study results. The absence of means and/or standard deviations from so many studies is a problem that must be rectified.
The study results suggest that future research be designed to test specifically whether the efficacy of specific relaxation techniques varies depending on symptom to be treated, on acute versus chronic problems and/or in surgical versus non-surgical settings. A meta-analysis of biofeedback studies would also be of value.
|Unit of Analysis||E/C1||E/P||PRE/POST||E/G||E/P||PRE/POST|
1 E/C = experiment versus control; E/P = experiment versus placebo
Note: percentiles given are those at which a subject in a experimental group would be located as compared to an untreated subject who would fail at the 50th percentile.
|SELECTED VARIABLES FOR TREATMENT TYPE ANALYSIS|
|Published||34||0.64||0.70||8||0.61||0.51||100% of cases|
|Unpublished Sample||20||0.46||1.02||6||0.73||0.58||no cases|
|Nonsurgical||44||0.67||0.92||11||0.73||0.55||100% of cases|
|OUTCOME TYPE FOR INDIVIDUAL ES ANALYSIS|
|Lemgth of Stay||3||0.08||0.35||2||0.32||0.16||-||-||-|
|PRESENTING: PROBLEMS FOR TREATMENT-TYPE ANALYSIS|
Note: Where effects from the Individual ES analysis differed substantially (i.e. related in a percentile diference ≥ 7) they are presented in parenthesis.
a data missing for the study
b PMI - progressive muscle resistence: GSR - galvanic skin response; EMG - electramygram.