Introduction
According to International Association for the Study of Pain, pain is defined as an unpleasant sensory and emotional experience associated with, or resembling that associated with, actual or potential tissue damage [
1]. Pain is generally classified into two categories. Acute pain serves as a warning signal and can be defined as a reaction to tissue damage triggered by aversive external stimuli or endogenous processes [
2,
3]. As far as adequate treatment of the cause is possible, acute pain is reversible [
4]. In contrast, chronic pain either persists after the injury is healed or it is associated with a chronic illness [
5]. In pain research, chronic pain includes both persistent and recurring pain with at least 3- to 6-month duration [
6].
Pain is a significant physical strain for the individual, which is associated with decreased quality of life, reduced job productivity, and increased absence from work [
7]. The direct and indirect health care costs of chronic pain disorders in European member states are estimated at 2–3% of gross domestic product across the EU [
8,
9]. For 2016, this was approximately 441 billion euros [
10]. Chronic pain carries a significant burden for employees, employers, and society, and the adverse consequences of chronic pain with its substantial negative impact on work-related outcomes are often underestimated [
11]. According to a large-scale internet-based survey on prevalence and attributes of pain experiences in the UK, France, Spain, Germany, and Italy, one in five respondents had experienced pain in the past few months [
7]. For various pain types, high prevalence rates are reported for 19 European countries. At a pan‐European level, back/neck pain was the most prevalent with 40% of survey participants experiencing pain [
12].
Particularly in the current Covid-19 pandemic, delaying or discontinuing treatment for individuals suffering from severe chronic pain has negative consequences for them, such as an increase in pain [
13]. With the limited availability of multimodal therapeutic approaches in the pandemic, physicians must prescribe pain medications until adequate treatment is available [
14]. In summary, because of their high prevalence, relating costs and worsened chronic pain symptoms due to COVID-19 chronic pain has not only high personal, but also clinical and economic relevance for patients, practitioners, and payers.
Oral analgesics are one of the primary treatments for different types of pain because they can be a quick, cheap, and effective solution to the problem of pain [
15]. However, the use of pharmaceuticals for pain reduction can lead to various unwanted side effects. In addition to physical side effects such as kidney, liver, or cardiovascular problems, these also include drug dependency as well as sedation and tolerance effects [
16]. Non-pharmacological psychological interventions aim at modifying factors that are important in the genesis and maintenance of pain [
17]. Meanwhile, these interventions are essential in multimodal pain therapy. In Germany, almost every facility that offers multimodal pain treatment applies at least one relaxation procedure, such as progressive muscle relaxation, biofeedback, and autogenic training (AT), as routine part of pain treatment [
18].
In the following study, we focus on AT only, because in comparison to progressive muscle relaxation, the patient is not forced to build up additional muscle tension of the painful muscle sections but can achieve an improvement by directing his/her attention to certain relaxation reactions of the body (for more details, see below). Moreover, AT can be performed without medical supervision. In cases where the patient’s health could be affected using the relaxation exercise such as PMR for pain, physical illness, disabilities, or injury, medical supervision is recommended [
19]. Compared to biofeedback, AT is superior because it can be used in everyday life in any situation, whereas with biofeedback, there is a dependence on experts and equipment [
19].
AT is a self-relaxation procedure applying passive concentration on certain combinations of psychophysiologically adapted stimuli, developed by Schultz almost 100 years ago [
20]. Within AT, participants are trained in auto-suggestive techniques to influence their physical condition [
21‐
23]. In its classic form, AT uses six standard exercises that are trained in individual or group settings over a period of 6 to 8 weeks [
24,
25]. Participants sit or lie in a quiet, undisturbed setting and focus on different areas of the body, which are addressed using six suggestive formulas aiming at increasing relaxation and balance between sympathetic and parasympathetic control [
26]. Relaxation is suggested to affect pain by reducing tissue oxygen requirement and degrading lactic acid, by relieving skeletal muscle tension and anxiety, and by releasing endorphins [
27,
28].
In the past decades, several systematic reviews summarized the evidence on AT for various clinical indications including pain. Within the extensive review of Grawe et al. [
29] including more than a thousand psychotherapy studies published up to 1983/84, only 14 trials were controlled AT studies. Based on their results, the authors concluded that the effectiveness of AT has not yet been sufficiently validated compared to other relaxation techniques. At the same time, the first meta-analysis of AT was released, including 24 controlled studies published from 1952 to 1993 [
26]. However, pain was not explicitly considered primary outcome, but was included in the aggregate of ‘behavioral and psychological outcomes.’ AT was associated with medium-sized pre-post effect sizes in migraine and tension headache, but this estimation was based on two or five studies, respectively. Stetter and Kupper [
25] updated this review in 2002 examining 60 clinical studies published between 1952 and 1999, including 35 randomized controlled trials. Outcomes were grouped as either ‘physiological’ or ‘behavioral and psychological’. Eleven randomized controlled trials were included examining the effects of AT in individuals with tension headache/migraine, providing a significantly positive, medium effect size of
d = 0.59 (four studies, 251 participants) on all reported outcomes when AT was compared with passive control groups. In comparison to other psychological interventions, there were significantly negative effects of
d = − 0.25 showing that other psychological treatments performed better than AT in reducing headache/migraine pain (seven studies, 871 participants). Kanji et al. [
30] published a systematic review on the effectiveness of AT in individuals with tension headache including five randomized controlled trials and two non-randomized controlled trials. AT was comparable to other types of interventions, and only some studies revealed inferior effects in contrast to biofeedback. The authors concluded that AT is an effective relaxation technique for individuals with pain; however, this was based on a narrative summary only.
Altogether, the efficacy of AT in individuals suffering from pain has been investigated in numerous randomized controlled trials. However, a previous meta-analysis is about 20 years old and has so far only summarized the existing evidence without specifically considering pain as outcome [
25]. Hence, the aim of this meta-analysis is to investigate the efficacy of AT in individuals with chronic pain on the primary outcome pain in comparison to waiting list control groups, attention control groups, or control groups that received other psychological interventions. In addition to pain as primary outcome, mental distress and health-related functioning are considered secondary outcomes.
Methods
Protocol and Registration
The review was registered at PROSPERO International Prospective Register of Systematic Reviews (CRD42020141812).
Eligibility Criteria
Randomized controlled trials published in English or German language without restrictions of publication date were included. Eligible studies involved individuals with chronic pain and evaluated the efficacy of AT. AT had to be applied for therapeutic purposes, had to be the only or at least the primary therapeutic method, and could be performed individually or in a group. ‘No treatment’, ‘attention control’, or ‘another treatment’ was considered eligible control groups. Attention control groups were defined as delivering a comparable amount of time and attention without specific therapeutic components. Another treatment included standard care or another type of intervention referred to as relaxation intervention. Primary outcome was pain including measures of, e.g., pain intensity, frequency, and duration. Mental distress (including measures of, e.g., anxiety, depression, well-being, relaxation, comfort) and health-related functioning were considered secondary outcomes. Outcomes reflecting quality of life (pre-specified as secondary outcomes in the review protocol) were classified as mental distress or health-related functioning, depending on the subscales of the quality of life measures. Deviating from the review protocol, we excluded studies on somatoform and acute pain and limited study inclusion to a more homogeneous population of individuals with chronic pain.
A systematic literature search was performed using the following electronic databases: Medline, Web of Science, PsycInfo, and PubPsych (date last searched: April 7, 2021; see
Supplementary material for details of the search strategy). The search strategy specified terms referring to the patient population (e.g., pain*), treatment (e.g., autogenic training, autogenic*, autosuggest*), and study design (e.g., random*, control*). The search strategy was developed with consideration of validated search strategies for retrieving randomized controlled trials [
31]. Additionally, relevant treatment guidelines and references of recent reviews, meta-analyses, and primary studies were checked to identify further studies. In order to detect unpublished studies, the ProQuest Dissertations and Theses Full Text database, DART-Europe E-theses Portal, Networked Digital Library of Theses and Dissertations (NDLTD), and the Theses Database of the German National Library were searched.
Study Selection
Title and abstract of studies identified in the literature search were first screened for eligibility by the first author. In a second step, full texts of the preselected studies were examined in detail for eligibility by two independent researchers (AK, JR). Disagreements were resolved by consensus.
The following data were extracted from the included studies: information on publication (e.g., authors, publication year, country of origin), sample characteristics (e.g., sample size, gender, age, type of pain), characteristics of the intervention group (e.g., treatment format, treatment modality, number of sessions, length of sessions, total duration), characteristics of the control group (e.g., type of control group), information on outcomes (e.g., type of outcome category, measure, time point), and statistical data. Descriptive information was coded by the first author. Two authors (AK, JR) extracted information on outcomes and statistics needed for effect size estimation with disagreements resolved by consensus.
Risk of Bias in Individual Studies
To evaluate various indicators of bias, the current version of the Cochrane Risk of Bias Tool for Randomized Trials (ROB2—revised version from August 2019) was used [
32]. Risk of bias was assessed in five distinct domains. Within each domain, one or more signaling questions were answered. Based on defined algorithms, judgments of ‘low risk of bias,’ ‘some concerns,’ or ‘high risk of bias’ were proposed for each domain. The judgments within each domain finally resulted in an overall risk-of-bias judgment per study. Risk of bias was assessed for each study and domain independently by two authors (AK, KW). Disagreement in judgments was resolved either via discussion or another author (JR) was called to adjudicate the final judgment.
Summary Measures
Between-group effect sizes (Hedges’
g) were computed for each comparison, assessment time point, and outcome of interest. Hedges’
g represents the standardized mean difference calculated by subtracting the posttreatment mean of the intervention group from the posttreatment mean of the control group, dividing the result by the pooled standard deviation, and multiplied by a small-sample bias correction factor [
33]. If means and standard deviations were not reported, other statistics (
F,
t, or
p value) were used to calculate effect sizes. For dichotomous outcomes, log odds ratios were calculated and converted to Hedges’
g in order to pool across different effect size formats [
34]. The magnitude of Hedges’ g was interpreted within the same framework as Cohen’s
d, regarding 0.20, 0.50, and 0.80 as small, medium, and large effects between two contrasted groups, respectively [
35]. Positive effect sizes indicate that AT was superior to the comparison treatment, whereas negative effect sizes suggest superiority of the comparison treatment. All summary measures are reported with a 95% confidence interval (CI). The software Comprehensive Meta-Analysis (CMA, Biostat. Inc. Version 3) was used for computing effect sizes and performing data analyses.
Data Synthesis
Outcome data were meta-analyzed using a random-effects approach. The generic inverse variance method with heterogeneity estimated using the DerSimonian-Laird method was applied [
36]. In case of multiple comparisons within one study (two control groups were compared against one shared intervention group), each pairwise comparison was included separately in the meta-analysis as proposed by Higgins et al. [
37]: for dichotomous outcomes, both the number of events and the total number of patients in the shared intervention group were divided evenly among the pairwise comparisons. For continuous outcomes, only the total number of patients was divided and statistical parameters were left unchanged. If multiple outcomes were reported within one outcome domain (e.g., two measures of pain), effect sizes were aggregated within domains for each unit of analysis and correlations between outcomes were set at 0.50 [
38]. All pooled effect sizes Hedges’
g were additionally transformed into numbers needed to treat (
NNT) [
39], representing the number of patients one would need to treat with the intervention in order to have one more patient to have an outcome better than a randomly selected one in the control group (note that negative
NNT values refer to the number needed to harm).
Statistical heterogeneity across studies was assessed with χ
2 heterogeneity tests (Cochrane’s
Q) and
I2 statistic [
40].
I2 describes the percentage of the variability in effect estimates that is due to heterogeneity rather than chance, with values from 0 to 40% indicating no important heterogeneity, 30 to 60% moderate, 50 to 90% substantial, and 75 to 100% considerable heterogeneity, respectively [
41]. Additionally, 95% prediction intervals were computed representing the potential effect in a future study that is similar to the studies in the meta-analysis [
41,
42].
Risk of Bias Across Studies
In order to address potential publication bias, funnel plots were visually inspected and Egger’s regression test for funnel plot asymmetry was performed [
43]. Duval and Tweedie’s trim and fill procedure was used to determine whether small studies with non-significant effects were underrepresented in the meta-analysis [
44]. Possible missing studies were imputed and the effect size estimate was recalculated. Additionally, Classic fail-safe
N [
45] was computed estimating the number of studies with a null effect that would be needed to increase the
p value for the meta-analysis to above 0.05.
Additional Analysis
Subgroup analyses and meta-regression analyses were planned for various pre-specified variables, given a sufficient number of available studies (per group). To test the robustness of effect size estimates, sensitivity analyses were performed by excluding studies with children, approximated effect sizes, and studies with high risk of bias in any domain.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.