|Home | About | Journals | Submit | Contact Us | Français|
Background. Back pain is a common problem and a major cause of disability and health care utilization. Purpose. To evaluate the efficacy, harms, and costs of the most common CAM treatments (acupuncture, massage, spinal manipulation, and mobilization) for neck/low-back pain. Data Sources. Records without language restriction from various databases up to February 2010. Data Extraction. The efficacy outcomes of interest were pain intensity and disability. Data Synthesis. Reports of 147 randomized trials and 5 nonrandomized studies were included. CAM treatments were more effective in reducing pain and disability compared to no treatment, physical therapy (exercise and/or electrotherapy) or usual care immediately or at short-term follow-up. Trials that applied sham-acupuncture tended towards statistically nonsignificant results. In several studies, acupuncture caused bleeding on the site of application, and manipulation and massage caused pain episodes of mild and transient nature. Conclusions. CAM treatments were significantly more efficacious than no treatment, placebo, physical therapy, or usual care in reducing pain immediately or at short-term after treatment. CAM therapies did not significantly reduce disability compared to sham. None of the CAM treatments was shown systematically as superior to one another. More efforts are needed to improve the conduct and reporting of studies of CAM treatments.
Back pain is a general term that includes neck, thoracic, and lower-back spinal pain. In the majority of cases, the aetiology of back pain is unknown and therefore is considered as “nonspecific back pain”. Back pain is considered “specific” if its aetiology is known (e.g., radiculopathy, discogenic disease). Although back pain is usually self-limited and resolves within a few weeks, approximately 10% of the subjects develop chronic pain, which imposes large burden to the health-care system, absence from work, and lost productivity . In a recent study, the direct costs of back pain related to physician services, medical devices, medications, hospital services, and diagnostic tests were estimated to be US$ 91 billion or US$ 46 per capita . Indirect costs related to employment and household activities were estimated to be between US$ 7 billion and US$ 20 billion, or between US$25 and US$ 71 per capita, respectively [3–5]. One study published in 2007 showed that the 3-month prevalence of back and/or neck pain in USA was 31% (low-back pain: 34 million, neck pain: nine million, both back and neck pain: 19 million) .
The prevalence of back pain and the number of patients seeking care with complementary and alternative medicine (CAM) therapies in the US has increased over the last two decades . The most prevalent CAM therapies for back and neck pain in the US are spinal manipulation, acupuncture, and massage . The exact mechanisms of action of CAM therapies remain unclear. Recently, many randomized controlled trials (RCTs) have been conducted to study the effects of CAM therapies for back pain. The results of many systematic reviews [8–12], meta-analyses , and clinical practice guidelines [14–17] regarding the effectiveness of CAM therapies for back pain relative to no treatment, placebo, or other active treatment(s) in reducing pain and disability have been inconsistent.
The agency for healthcare research and quality (AHRQ) and the national center for complementary and alternative medicine (NCCAM) commissioned the University of Ottawa Evidence-based Practice Center (UO-EPC) to review and evaluate evidence regarding the effectiveness, cost-effectiveness, and safety of the most prevalent CAM therapies (i.e., acupuncture, manipulation, mobilization, and massage) used in the management of back pain. This technical report can be viewed at the AHRQ website (http://www.ahrq.gov/) . The present paper summarizes the evidence from this technical report with a focus on a subset of studies reporting pain, disability, and harms outcomes compared between CAM therapies and other treatment approaches deemed relevant to primary care physicians (i.e., waiting list, placebo, other CAM therapies, pain medication, and physical therapy including exercise, electrotherapy and/or other modalities). The specific aims of this study were to systematically review and compare the efficacy, cost-effectiveness, and safety of acupuncture, manipulation, mobilization, and massage in adults (18 years or older) with neck or low-back pain.
We searched MEDLINE (1966 to February 2010), EMBASE (1980 to week 4 2010), the Cochrane Library (2010 Issue 1), CINAHL (1982 to September 2008), AMED (Allied and Complementary Medicine Database: 1985 to January 2010), Mantis (1880 to October 2008), and EBM Reviews—ACP Journal Club (1991 to August 2008). Two specialized CAM databases, the Index to Chiropractic Literature (ILC; October 2008) and Acubriefs (2008 October) were also searched. We searched using controlled vocabulary and keywords for conditions pertaining to neck pain, back pain, spinal diseases, sciatica, and various CAM interventions including acupuncture, electroacupuncture, needling, acupressure, moxibustion, manipulative medicine, manipulation, chiropractic, and massage. (Appendix A: Complete search strategies for each database). The searches were not restricted by language or date. We also reviewed reference lists of eligible publications.
RCTs reporting efficacy and/or economic data of CAM therapies in comparison with no treatment, placebo, or other active treatments in adults with low-back, neck, or thoracic pain were eligible. Nonrandomized controlled trials and observational studies (e.g., cohort, case-control, cross-sectional) reporting harms were also included. Reports published in English, German, Dutch, Chinese, Japanese, Italian, French, Portuguese, and Spanish were eligible for inclusion. Systematic and narrative reviews, case reports, editorials, commentaries or letters to the editor were excluded.
Two independent reviewers screened the titles and abstracts and later reviewed the full-text reports of potentially eligible records. Discrepancies were resolved by consensus.
Two independent reviewers extracted data on study and population characteristics, treatment, study outcomes, and duration of posttreatment followup. The abstracted data were verified and conflicts were resolved by consensus.
Treatment efficacy outcomes were pain intensity (e.g., Visual Analog Scale-VAS, McGill Pain Questionnaire-MPQ) and disability (e.g., Roland-Morris Disability Questionnaire-RMDQ, Northwick Park Neck Pain Questionnaire-NPQ, Pain Disability Index-PDI, Oswestry Disability Index). The timing of posttreatment followup for outcomes was ascertained and categorized into four groups: immediate, short- (<3 months), intermediate- (3 to 12 months), and long-term (>12 months) posttreatment followup. Harms (e.g., any adverse event, withdrawals due to adverse events, specific adverse events) were extracted as proportions of patients with an event.
For cost-effectiveness analysis, data was extracted on: (a) costs to the health care sector, (b) costs of production loss, (c) costs in other sectors, (d) patient and family costs, and (e) total costs.
The risk of bias for RCTs was assessed using the 13-item criteria list (item rating: Yes, No, Unclear) recommended in the Updated Method Guidelines for Systematic Reviews in the Cochrane Collaboration Back Review Group . The risk of bias for each RCT was classified into three groups: good (score: 4), fair (score: 2-3), and poor (score: 0-1) depending on the number of “Yes” ratings (score range: 0–4) across the four domains (treatment allocation concealment, balance in baseline characteristics, blinding, and number/reasons for dropouts). Assessment of quality of reporting in observational studies was done by using the modified 27-item tool of Downs and Black . Methodological quality of economic studies was determined using the 19-item Consensus Health Economic Criteria .
The overall strength (i.e., quality) of evidence was assessed using the grading system outlined in the Methods guide prepared for the AHRQ Evidence-based Practice Center (EPC) program . The grading was based on four domains: overall risk of bias, consistency, directness, and precision (applied to pooled results only). The overall risk of bias (high, medium, and low) was derived by averaging the risk of bias (good, fair, and poor) across individual trials. If evidence consisted of only one study (or multiple studies of the same risk of bias score), then the risk of bias for individual study corresponded to the overall risk of bias for this evidence as follows: “poor” (score: 0 or 1) = risk of bias (high), “fair” (score: 2 or 3) = risk of bias (medium), and “good” (score: 4) = risk of bias (low). In case of evidence consisting of multiple studies with different risk of bias scores (studies that scored “poor”, “fair”, and “good” mixed together), the mean risk of bias score (i.e., mean number of “Yes”) was calculated and the overall risk of bias was defined as “high” (mean score < 2), “medium” (2 ≤ mean score < 4), and “low” (mean score = 4). Consistency was judged based on qualitative assessment of forest plots of meta-analyses (direction and 95% confidence intervals of the effects in individual trials). Results were considered consistent when statistically significant or nonsignificant effects in the same direction were observed across trials. When pooling was not possible, consistency was judged based on qualitative summary of the trial results. The pooled estimate with relatively narrow 95% CIs leading to clinically uniform conclusions was considered as “precise evidence”. Relevant health outcomes (pain, disability) were defined as “direct” as opposed to intermediate or surrogate outcomes (“indirect”). The grade of the evidence for a given outcome was classified into four groups: high, moderate, low, or insufficient (no evidence). The initial “high” grade was reduced by one level (from high to moderate) for each of the domains not met (i.e., overall risk of bias, consistency, directness, precision) and by two levels in case of high risk of bias (e.g., from high to low grade).
The results were grouped according to the type of experimental intervention (e.g., acupuncture, manipulation, mobilization, massage), pain location in spinal region (low-back, neck, head, thorax), duration of pain (acute/subacute, chronic, mixed, unknown), and cause of pain (specific, nonspecific). Study, treatment, population, and outcome characteristics were summarized in text and summary tables.
We meta-analyzed RCTs with similar populations (demographics, cause, location, and duration of spinal pain), same types of experimental and controls treatments, and outcomes measured with the same instruments (and scale) at similar posttreatment followup time points. The meta-analyses of pain were based on the Visual Analogue Scale (VAS; score range: 1–10). The random-effects models of DerSimonian and Laird were used to generate pooled estimates of weighted end point mean difference (WMDs) with 95 percent confidence intervals (95% CIs). Statistical heterogeneity was evaluated using the Chi-square test and the I2 statistic (low: 25.0%; moderate: 50.0%; high: 75.0%). Subgroup (e.g., patients' age, gender) and sensitivity (e.g., trial quality) analyses were planned to investigate the sources of heterogeneity.
The degree of clinical importance for the observed differences in pain scores between the treatment groups was specified according to the Updated Method Guidelines of Cochrane Collaboration Back Review Group : small (WMD < 10% of the VAS scale), medium (10% ≤ WMD < 20% of the VAS scale), and large (WMD ≥ 20% of the VAS scale).
Publication bias was examined through visual inspection of funnel plot asymmetry and the Egger's regression-based method .
This topic was nominated by NCCAM and selected by AHRQ. A representative from AHRQ served as a Task Order Officer and provided technical assistance during the conduct of the full evidence report and comments on draft versions of the full evidence report. AHRQ did not directly participate in the literature search, determination of study eligibility criteria, data analysis or interpretation, preparation, review, or approval of the paper for publication.
Our literature search identified 152 unique studies: 147 RCTs and 5 nonrandomized studies (1 controlled trial and 4 observational) were included in the review (Figure 1). One hundred and fifteen RCTs reported data on efficacy (pain and disability) and/or harms. Additionally, 23 RCTs that did not report pain and disability outcomes provided data on harms. Five nonrandomized studies reported harms. Ten RCTs reported on cost-effectiveness (one of the 10 RCTs also reported efficacy).
The included studies were published between 1978 and 2009. The studies were published in English (74.5%), Chinese (3.3%; all acupuncture) [24–28], German (<1.0%; massage of lumbar region) , Japanese (2.6%; all acupuncture) [30–33], and one in Spanish (spinal mobilization) . All 10 reports of economic evaluation of CAM treatments were published in English [35–44].
The majority of trials (>90%) included adult men and women aged 18–65 years. Six trials included adults aged 55 years or older [45–50]. In total, 61% of all studies included subjects with nonspecific pain. About 85%, 14%, and 12% of acupuncture, spinal manipulation/mobilization, and massage trials, respectively, enrolled subjects with nonspecific cause of back pain. The remaining trials enrolled subjects with specific causes of back pain (e.g., disc perturbation, whiplash, myofascial pain, cervicogenic headache, or underlying neurological causes).
A large variety of methods of acupuncture treatments were used to compare the effect of acupuncture and control treatments. The control treatments in these trials included active (i.e., physical modalities and exercise) or inactive treatments (i.e., placebo, no treatment). The treatment providers were trained or licensed acupuncturists, general practitioners or physicians with especial training in acupuncture, neuropathy physicians, general practitioners, and trained physiotherapists. In the majority of Chinese trials, the treatment provider was referred as “therapist”.
Interventions were provided by experienced and licensed chiropractors, physical therapists, general practitioners, licensed or qualified manual therapy practitioners, nonspecified clinicians, neurologists or rheumatologists, folk healers, and osteopaths.
Treatment providers were licensed or experienced massage therapists, physical therapists, reflexologists, acupressure therapists, folk healers, general practitioners, manual therapists, experienced bone setters, and chiropractic students.
The risk of bias was assessed for 131 RCTs. Overall, the methodological quality of the RCTs was poor (median score = 6/13; inter-quartile range: 4, 7). Only 71 (54%) of the studies scored 6 or higher based on the 13 items of risk of bias tool. An adequate method of randomization was described in 57 (43.5%) studies. The remaining 74 studies either did not report the method used for randomization (n = 8; 6.0%) or the method used was not clearly described (n = 66; 50.0%). Concealment of treatment allocation was judged as adequate for 41 (31.3%) of RCTs and inadequate for 20 (15.3%) of RCTs (Table 1 and Figure 2).
Of the 10 studies reporting cost-effectiveness data, 3 studies collected costs appropriate to their chosen perspective. Two studies did not state the perspective adopted for the economic evaluation. Most studies measured costs using diaries, questionnaires, practice/insurance records, and valued costs appropriately using published sources. Most studies conducted an incremental cost-effectiveness analysis. The length of followup across the studies was at least one year. In one study with a length of followup of more than one year, discounting was undertaken .
The objectives and the main outcome (an adverse event) of the 4 studies were well described. The studies had a large sample size ranging from 68 to 3982 subjects, providing sufficient power to detect clinically important effects.
This section included 33 trials (Table 2 for efficacy results and evidence grading (Appendix B)) [26, 30–33, 35, 36, 41, 45, 47–49, 51–73]. One study  was published in Chinese and four studies were published in Japanese [30–33]. The trials were conducted in China (37%), Europe (United Kingdom, Germany, Ireland, and Sweden; 35%), and USA (28%).
One meta-analysis (Figure 3) showed that subjects with chronic nonspecific LBP receiving acupuncture had statistically significantly better short-term posttreatment pain intensity (3 trials; pooled VAS: −1.19, 95% CI: −2.17, −0.21) [48, 52, 74] and less immediate-term functional disability (1 trial)  compared to subjects receiving no treatment.
Trials comparing acupuncture to placebo yielded inconsistent results with respect to pain intensity. For subjects with acute/subacute nonspecific LBP, acupuncture did not significantly differ from placebo on pain or disability outcomes [31, 53]. In a meta-analysis (Figure 4) of subjects with chronic nonspecific LBP, acupuncture compared to placebo led to statistically significantly lower pain intensity, but only for the immediate-posttreatment followup (10 trials; pooled VAS: −0.59, 95% CI: −0.93, −0.25) [51, 55, 56, 58, 59, 61–65, 67]. The mean pain intensity scores in the acupuncture and placebo groups were not significantly different at short- [51, 55, 56, 58] intermediate-[51, 54, 58], and long-term [51, 54, 63, 67] followups. Acupuncture did not significantly differ from placebo in disability [62, 67]. Trials using sham-TENS, sham-laser, or placebo medication tended to produce results in favor of acupuncture in relation to pain intensity and disability compared to trials using sham-acupuncture.
Two meta-analyses showed that acupuncture did not significantly differ from pain medication in reducing immediate posttreatment pain (4 trials; VAS score) [49, 69–71] or disability (2 trials; Oswestry score) [69, 70] in patients with chronic nonspecific low-back pain (Data is not presented in Figures).
Another meta-analysis (Figure 5), based on subjects with chronic nonspecific low-back pain, indicated that manipulation was significantly better than acupuncture in reducing pain immediately after the treatment (2 trials; VAS score: 3.70, 95% CI: 1.50, 5.80) [69, 70].
One trial showed that subjects receiving acupuncture had significantly better immediate posttreatment pain and disability than subjects receiving a combination of physical modalities (the light, electricity, heat) .
Massage was significantly better than acupuncture in reducing pain intensity and disability at immediate- or long-term followups for subjects with chronic nonspecific LBP .
Subjects with chronic nonspecific LBP receiving acupuncture compared with those receiving usual care (analgesics, anti-inflammatory drugs, primary care, recommendation for physical therapy visits) had significantly better short-/intermediate-term posttreatment pain intensity (2 trials; VAS score) [47, 67] and disability (2 trials; RMDQ score) [47, 67]. However, in subjects with acute nonspecific LBP, posttreatment disability (RMDQ) was not significantly different between the acupuncture plus usual care (limited bed rest, education, and nonsteroidal anti-inflammatory drugs, activity alterations) and usual care alone groups (1 trial) .
This section included 24 trials (Table 3 for efficacy results and evidence grading (Appendix B)) [24, 27, 28, 69, 70, 72, 75–94]. About 38% of studies were conducted in Europe (Germany, Spain, Sweden, Turkey, United Kingdom), 17% in Australia, 8% in Japan, and 8% in the USA. The remaining 29% of trials were conducted in Brazil, South Korea, and Taiwan. All studies in this section were published in English language.
In one trial of subjects with unknown duration of myofascial neck pain , acupuncture was significantly better than no treatment in reducing pain intensity (McGill pain questionnaire) shortly after the end of treatment (mean change from baseline: −15.2 ± 13.3 versus −5.3 ± 8.7, P = 0.043). There was no evidence comparing acupuncture to no treatment in subjects with neck pain of acute/subacute, chronic, and mixed duration.
Two meta-analyses (Figure 6) indicated no significant difference between acupuncture and sham-acupuncture in subjects with chronic-specific (two trials; VAS score: 0.27, 95% CI: −0.60, 1.13) [77, 78] or nonspecific pain (three trials; VAS score: − 0.24, 95% CI: −1.20, 0.73) [80–82] for immediate posttreatment pain intensity. Similarly, one trial of subjects with mixed specific pain showed no significant difference between acupuncture and placebo in reducing pain intensity (VAS score) or improving disability immediately after treatment . There was no evidence comparing acupuncture to placebo in subjects with acute/subacute duration of neck pain.
There were inconsistent results for immediate- or short-term posttreatment pain intensity between acupuncture and pain medication in subjects with chronic and unknown duration of pain (8 trials) [28, 69, 70, 87, 89–92]. For subjects with chronic nonspecific pain, acupuncture was significantly better in reducing pain than NSAIDs immediately after treatment . Similarly, in two trials, acupuncture was significantly more effective than injection of Lidocaine in short-term followup for treatment of unknown nonspecific neck pain [28, 92]. In other five trials, there was no significant difference between acupuncture and pain medication [69, 70, 87, 89, 90].
There were inconsistent results for immediate- or short-term posttreatment pain intensity between acupuncture and spinal manipulation for chronic pain (3 trials) [24, 70, 72]. Immediate/short-term posttreatment disability score (NDI) was better in manipulation than acupuncture groups of subjects with chronic nonspecific pain (2 trials) [69, 70].
In one trial , acupuncture was significantly better than massage in reducing pain intensity at short-term posttreatment followup (mean VAS score change from baseline: 24.22 versus 7.89, P = 0.005).
This section included 13 studies using manipulation alone [69, 70, 72, 97–108]. (Table 4 for efficacy results and evidence grading (Appendix B)). About 62% of studies were conducted in North America (USA and Canada), 15% in Australia, and the remaining 23% in Europe (United Kingdom, Italy), and (Egypt).
In subjects with acute/subacute [97, 99–101, 109] and mixed duration [98, 104] nonspecific LBP, manipulation was significantly more effective than placebo [97, 99–101, 104, 109] or no treatment [97, 98] in reducing pain intensity immediately or in the short-term following treatment. There was no significant difference between manipulation and placebo in posttreatment pain disability. In subjects with chronic nonspecific LBP, manipulation was significantly more effective than placebo in reducing pain intensity (VAS score) immediately or short-term after the end of treatment [100, 102, 103].
Manipulation was significantly better (in immediate posttreatment pain) or no different (in intermediate-term posttreatment pain) than pain medication in improving pain intensity [69, 70]. Manipulation did not differ from pain medication in reducing pain intensity at short- and intermediate-term followup after treatment .
In older subjects with mixed LBP duration, spinal manipulation was significantly better than medical care (exercise, bed rest, analgesics) in improving immediate and short-term posttreatment disability (Oswestry), although no significant difference could be found in pain intensity .
In two large trials [110, 111], subjects receiving combination of manipulation and exercise or manipulation and best care by general practitioner (analgesics or muscle relaxants) improved in pain and disability compared to subjects with no spinal manipulation treatment.
This section included 12 trials (Table 5 for efficacy results and evidence grading (Appendix B)) [69, 70, 72, 112–121]. About half of the studies were conducted in North America (USA and Canada), 16% in Europe (Germany, Spain) and the remaining 34% of the studies in Australia.
There was no significant difference in reducing pain intensity between manipulation and “no treatment” groups in immediate-term posttreatment in subjects with unknown nonspecific pain (1 trial) .
Subjects with acute, subacute, chronic or unknown neck pain receiving manipulation had significantly better posttreatment pain (4 trials) [113–116] and disability (1 trial)  compared to those taking placebo.
In two trials [69, 70], manipulation was significantly better than medication (e.g., NSAIDs, Celebrex, Vioxx, Paracetamol) in reducing pain intensity and improving disability score at immediate/short-term followup.
In subjects with acute/subacute nonspecific pain there was no statistically significant difference between manipulation and mobilization immediately after treatment (1 trial) . In subjects with mixed duration nonspecific neck pain, manipulation was statistically significantly more effective than mobilization in reducing pain immediately after treatment (2 trials) [119, 120]. In one trail , there were no clinically or statistically significant differences between manipulation and mobilization in reducing pain or improving disability at intermediate term followup .
This section included 13 trials (Table 6 for efficacy results and evidence grading (Appendix B)) [25, 34, 122–134]. About 30% of the trials were conducted in the US, 54% in Europe (Finland, United Kingdom, Sweden, Spain), and 16% in Australia, Thailand, and China. Two studies were published in either Spanish  or Chinese .
Subjects with acute/subacute  and chronic nonspecific LBP  receiving mobilization experienced significantly improved pain intensity VAS, MPQ  compared to subjects not receiving any treatment, immediately posttreatment [34, 122]. Results regarding disability (RMDQ, Oswestry) were inconsistent, showing either a significant difference in favour of mobilization  or no difference  between mobilization and no treatment. In one trial of subjects with mixed duration of LBP, there was no significant difference in pain intensity immediately posttreatment compared to no treatment .
In subjects with acute/subacute specific (pelvic joint dysfunction) [125, 126] and nonspecific mixed duration LBP  there were no significant differences in pain intensity (VAS) between mobilization and placebo groups immediately [125, 126] and in the short-term  after treatment.
In two meta-analyses, subjects with chronic nonspecific LBP receiving mobilization (traditional bone setting) compared to physiotherapy (massage, stretching, trunk exercise) had significantly lower pain intensity (pooled VAS score: −0.50, 95% CI: −0.70, −0.30) [128–130] and disability (pooled Oswestry score: −4.93, 95% CI: −5.91, −3.96) [128–130] immediately posttreatment.
In one trial, the manipulation group had a significantly better disability score compared to the mobilization group immediately posttreatment . In two trials, mobilization was shown either significantly worse than  or no different  from massage in reducing short-term posttreatment pain intensity amongst subjects with chronic nonspecific  or unknown duration of LBP .
The immediate- posttreatment pain intensity (VAS)  and disability (Oswestry)  did not significantly differ between mobilization and exercise in trials with mixed duration of LBP (2 trials) [131, 134]. In a trial including subjects with nonspecific pain of mixed duration, mobilization was significantly superior to exercise in reducing disability (Oswestry) at intermediate- and long-term posttreatment followup .
In two trials [135, 136], subjects with chronic or mixed nonspecific pain receiving mobilization had significantly lower pain intensity compared to no treatment. Mobilization was significantly better than placebo in subjects with acute/subacute nonspecific pain (1 trial) , but did not differ from placebo in subjects with chronic nonspecific pain (1 trial) .
Mobilization was significantly better than massage  or physiotherapy (massage, stretching and exercise) [137, 138] in improving pain (VAS score) and disability (NDI) in subjects with chronic and mixed nonspecific pain at intermediate-term posttreatment followup (2 trials) [137, 138]. Subjects with nonspecific pain of mixed duration in the mobilization and continued general practitioner care (analgesics, counselling, and education) groups had similar posttreatment pain intensity (VAS) and disability (NDI) .
This section included 10 trials (Table 8 for efficacy results and evidence grading (Appendix B)) [29, 139–147]. About half of the studies were conducted in Europe (Belgium, Germany, United Kingdom), 30% in North America (USA and Canada), and 20% in Taiwan. One study was published in German Language .
Subjects with acute/subacute nonspecific LBP receiving massage had significantly better pain intensity (VAS, MPQ) and disability (Oswestry) compared to no treatment (1 trial)  or placebo (2 trials) [139, 141] immediately or short-term after the end of treatment. In subjects with chronic nonspecific LBP, massage did not significantly differ from no treatment  or placebo  in improving immediate or intermediate-term posttreatment pain intensity (SF-36 pain scale, MPQ; 2 trials) [140, 142] or disability (Oswestry, RMDQ; 2 trials) [140, 142].
In two meta-analyses, massage was significantly better in reducing pain compared to relaxation (2 trials, pooled VAS score: −1.27, 95% CI: −2.46, −0.08) [145, 146] or physical therapy (2 trials; pooled VAS score: −2.11, 95% CI: −3.15, −1.07) [143, 144] immediately after treatment of subjects with chronic nonspecific LBP.
In subjects with chronic nonspecific LBP, there was no significant difference between receiving massage and usual care (advice and exercise) in improving pain (VAS score) or disability (RMDQ) intermediate-term after the end of treatment (1 trial) .
This section included 6 trails (Table 9 for efficacy results and evidence grading (Appendix B)) [76, 148–152]. Four trials were conducted in Europe (Finland, Germany, the Netherlands) and two trials in North America (USA and Canada).
Massage compared to no treatment significantly improved pain intensity (NPQ, VAS scores) in subjects with chronic or unknown duration of nonspecific pain, immediately after the end of treatment (2 trials) [148, 150]. Subjects with acute/subacute, chronic, or unknown duration of nonspecific pain receiving massage had significant improvement in pain intensity (≥2-point decrease on NRS-11, VAS) compared to subjects receiving placebo (2 trials), immediately after treatment [76, 151].
In subjects with chronic nonspecific pain, massage compared to exercise significantly improved disability (NPQ) immediately after the treatment (1 trial) .
This section included 5 trials (Table 10 for efficacy results and evidence grading (Appendix B)) [153–158]. The studies were conducted in Europe (the Netherlands, United Kingdom, and Norway), Australia, and the USA.
Subjects with acute/subacute nonspecific LBP receiving manipulation plus mobilization were not significantly better than subjects who received a double placebo (sham manipulation and placebo analgesic) (1 trial) .
Manipulation plus mobilization was significantly better in reducing pain than physiotherapy (exercise, massage, heat, electrotherapy, ultrasound) in subjects with mixed duration of LBP (1 trial) , better than hospital outpatient treatment in subjects with nonspecific LBP of unknown duration (1 trial) , and better than exercise for pain (VAS) and disability (RMDQ) in subjects with chronic nonspecific LBP (1 trial) . However, there was no difference between manipulation plus mobilization and usual care (analgesics, muscle relaxants, instruction in proper back care, life-style recommendations, and exercise) in subjects with mixed duration of nonspecific LBP (1 trial) .
In one trial, in subjects with chronic nonspecific pain, spinal manipulation plus mobilization was significantly better in reducing pain intensity and the frequency of headache than no treatment (P < 0.001) [160, 162].
In one trial , spinal manipulation plus mobilization did not differ from exercise alone in reducing headache frequency (number per week), intensity (VAS score: 0–10) and neck pain (percentage of patients who improved ≥50% on a 10-point MPQ scale). However, the combination was significantly better than physiotherapy (exercise, massage, heat, electrotherapy, ultrasound, shortwave diathermy) in reducing pain intensity (1 trial) [155, 161].
Visual inspection of the funnel plot (Figure 7) for the acupuncture trials comparing immediate posttreatment mean VAS scores between acupuncture and placebo treatment groups suggested some degree of asymmetry. Specifically, there was a relative lack of trials with negative results (i.e., fewer trials in areas of statistical nonsignificance), indicating a potential for publication bias. The Egger's regression-based analysis  yielded a statistically significant result (P = 0.03).
This section included results from 10 studies of full economic evaluations of acupuncture (low-back pain: 2 studies, neck pain: 1 study), spinal manipulation (low-back pain: 4 studies, neck pain: 2 studies), and massage (1 study) for low-back [35, 37–40, 43, 44] and neck pain [163–165].
Two economic evaluations showed that acupuncture was cost-effective compared to usual care and compared to no treatment in patients with chronic low-back pain [35, 39]. However, in both studies health gains were small and one study used no treatment control group and had only 3 months of followup.
One study  showed that in subjects with chronic neck pain acupuncture use was associated with significantly higher total costs compared to usual care ($1,565 versus $1,496).
There were no differences in costs between manual therapy, general practitioner care (rest, sick leave, direct prescription, advice about posture, and information about nature of the pain), and intensive therapy for acute LBP . Costs were higher for manipulation compared with medical care (analgesics or muscle relaxants) without producing better clinical outcomes for patients with mixed duration of LBP . This was associated with significantly more visits to chiropractic care than medical care. Spinal manipulation in addition to general practitioner care (active management; back book) was relatively cost-effective compared to general practitioner care alone for patients with subacute and chronic LBP . In chronic LBP patients, there were no differences in costs between physician consultation, spinal manipulation plus stabilizing exercises, and physician consultation alone . Results are difficult to compare due to differences in health care systems, perspectives, interventions, populations, and methods used.
One study  in subjects with neck pain found that pulsed short-wave diathermy was less cost-effective compared with manual therapy or exercise/advise. In another study , manual therapy was less costly and more effective than physiotherapy (functional, active and postural or relaxation exercises, and stretching) or general practitioner care (advice and exercise).
One study  reported an economic evaluation of therapeutic massage, exercise, Alexander technique, and usual general practitioner care (counselling, education, and pain medication) in patients with chronic low-back pain showing that massage was more costly and less effective than usual care by the general practitioner.
Reports of 57 trials provided data on harms. The reporting of harms was poor across the studies (e.g., lack of consistency, not detailed, not comparable). No definitions were presented. Therefore, rates of adverse events between the different interventions could not be meaningfully compared.
The reported events in RCTs [35, 36, 41, 45, 47, 49–51, 55, 56, 61–63, 63, 67, 69, 73, 76, 78, 80, 81, 83, 84, 86, 89, 92, 166–179] were mostly of moderate and transient nature. Most commonly reported events were soreness/pain at the site of needling, bruising light headedness, minor bleeding, dizziness, or headache. The proportion of subjects with any adverse event did not reportedly differ in acupuncture versus TENS or usual care groups.
In one nonrandomized controlled trial , discomfort or soreness in the acupuncture, chiropractic therapy, and massage groups were 5.0%, 8.0%, and 7.0%, respectively.
The reported events in RCTs were mostly moderate in severity and of transient nature (e.g., increased pain) [69, 98, 106, 118, 121, 180–184]. In one RCT [121, 185], after 2 weeks of treatment, patients with neck pain receiving manipulation were not at significantly increased risk for having an adverse event compared to patients receiving mobilization (OR = 1.44, 95% CI: 0.83, 2.49). In another RCT , the proportion of patients with neck pain having adverse events was similar in manipulation versus Diazepam groups (9.5% versus 11.1%).
In two case control studies [186, 187], subjects younger than 45 years of age with vertebro-basilar artery (VBA) stroke were more likely to visit a chiropractic or primary care physician than subjects without VBA stroke. This association was not observed in older subject visiting the chiropractic clinic. In the first case-control study , the excess risk of vascular accident was observed for both, subjects undergoing chiropractic care and subjects undergoing primary care treatments. In the second case-control study , subjects with cervical artery dissection were more likely to have had spinal manipulation within 30 days (OR = 6.62, 95% CI: 1.4, 30.0). In one cohort study, rate of complications did not differ between subjects with low-back pain receiving manipulation plus mobilization versus no treatment . In another prospective cohort study of 68 subjects with chronic LBP , treatment with medication-assisted manipulation or spinal manipulation alone for at least 4 weeks did not lead to any complications requiring institutional review board notification.
In a few RCTs [76, 142, 147, 190–192], subjects receiving massage experienced worsening of back/neck pain or soreness of mild and transient nature. One study reported allergic reactions (rashes and pimples) in 5 subjects due to massage oil. In one RCT , the proportion of patients with neck pain having adverse events in massage group was lower (7.0%) compared to acupuncture (33.0%) or placebo-laser (21.0%).
This paper identified a large amount of evidence on comparative effectiveness of single mode CAM interventions for management of low-back and neck pain in subjects with a wide spectrum of causes of pain.
The benefits of CAM therapies were limited mostly to immediate and short-term posttreatment periods when compared to inactive treatments (no treatment or placebo). The observed benefits were more consistent for the measures of pain intensity than disability. Trials that applied sham-acupuncture tended to produce negative results (i.e., statistically nonsignificant) compared to trials that applied other types of placebo (e.g., TENS, medication, laser) between acupuncture and placebo groups. One explanation for the beneficial effect of sham acupuncture is the diffuse noxious inhibitory controls (DNIC) where neurons in the dorsal horn of the spinal cord are strongly inhibited when a nociceptive stimulus is applied to any part of the body, distinct from their excitatory receptive fields . Another explanation could be the nonspecific effects of attention and beliefs in a potentially beneficial treatment.
The results were less consistent regarding comparison of CAM therapies to other active treatments (e.g., other CAM therapy, physiotherapy, pain medication, usual care). The degree of clinical importance for the differences in pooled pain intensity observed between the treatment groups for low-back pain was small (acupuncture versus placebo; mobilization versus physical modalities), medium (acupuncture versus no treatment; massage versus relaxation), or large (acupuncture versus manipulation, in favour of manipulation; massage versus physical modalities).
Due to the small number of economic evaluations, inconsistent standards of comparison, and substantial heterogeneity as well as different healthcare payment systems used in the countries these trials were conducted, it was not possible to apply these findings globally or to reach clear conclusions about the cost-effectiveness of any of these CAM treatments. Acupuncture was cost-effective relative to usual care or no treatment in subjects with back pain. Evidence for massage and mobilization was insufficient.
We identified 4 systematic reviews of acupuncture: one for LBP  and 3 for neck pain [195–198]. The LBP review found either acupuncture being superior (1 trial) or no different from sham acupuncture (3 trials). Although the present paper included a much wider range of trials, its results for neck pain were consistent with those of the three reviews [195, 197, 198] in finding acupuncture moderately more beneficial compared to no treatment or placebo immediately or the short-term after treatment. There were 2 reviews that evaluated manipulation and/or mobilization for acute, subacute, or chronic LBP [199, 200]. The first review  found manipulation more beneficial than sham but similar to general practitioner care, physical therapy, or exercise. The other review , indicated that manipulation did not differ from NSAIDs but was more beneficial than mobilization, general practitioner care, detuned diathermy, or physical therapy.
The results are similar across the three systematic reviews with respect to the superiority of manipulation and mobilization compared to no treatment of placebo for the various duration of LBP. The discrepancies lie when comparing manipulation or mobilization to other treatments. One review  concludes that manipulation or mobilization is equally effective compared to all other treatments, while the other  generally finds manipulation more effective than most other forms of therapy, but mostly in the short-term. In our paper, manipulation and mobilization effectiveness is variable depending on symptom duration, outcome, comparator, whether there is exercise or general practitioner care and followup period. Although this variability can be considered as “inconsistent findings”, the overall evidence suggests that manipulation and mobilization are an effective treatment modality compared to other therapies. The three systematic reviews also differ significantly on definition of SMT: the review by Assendelft et al.  lumps spinal manipulation and mobilization together and also allows for cointerventions). The synthesis methods were different, one has more language restrictions  and uses best evidence methodology, while the other uses meta-analysis for all included trials and includes patients with leg pain . In addition, they only included RCT published prior to 2002. All these reasons can explain differences in the findings and conclusions.
The findings of this paper regarding the effects of manipulation on neck pain were consistent with those of other reviews [9, 201–203]. While some differences in results between this and other two reviews can be explained by the inclusion criteria and grading of trials, the major results in findings were similar. Two other reviews [204, 205] assessed multimodal interventions (mobilization and manipulation combined with other interventions) and therefore were outside the scope of this review. One Cochrane review  found massage to be more beneficial than placebo or no treatment for chronic nonspecific LBP at short or long-term followup.
One of the strengths of this paper is that it identified a large amount of relevant evidence. The reviewers used systematic, comprehensive, and independent strategies to minimize the risk of bias in searching, identifying, retrieving, screening, abstracting, and appraising the primary studies. The search strategy, not restricted by the language or year of publication, was applied to multiple electronic sources. Further strength of this paper is the inclusion of only those trials from which an effect of a single CAM therapy could be isolated. Moreover, the results of individual trials were stratified by spine region (e.g., low-back, neck), duration of pain (acute, subacute, chronic, mixed, and unknown), and cause of pain (specific or nonspecific).
This paper has its limitations. The reviewed evidence was of low to moderate grade and inconsistent due to substantial methodological and/or clinical diversity, as well as small sample size of many trials, thereby rendering some between-treatment comparisons inconclusive. The differences in the therapy provider's experience, training, and approaches (e.g., deep or superficial massage, choice of trigger points, needling techniques) may have additionally contributed to heterogeneous results. Evidence for acute, subacute, and mixed specific pain was sparse relative to that for chronic nonspecific pain. Quantitative subgroup analyses exploring the effects of age, gender, race, type of treatment provider, or dose of treatment could not be performed due to lack or insufficient data. Poorly and scarcely reported harms data limited our ability to meaningfully compare rates of adverse events between the treatments. This paper focused on manipulation or mobilization to estimate the efficacy. Results from these studies may not be readily applicable to various combinations of interventions used in today's practice. However, the assessment of a single intervention is the first step in teasing out which therapeutic item is more effective in reducing pain and improving function.
This paper assessed the extent of publication bias using a visual inspection of the funnel plot and the Egger's regression-based technique . Although the visual inspection method is not very reliable, it conveys some general idea as to how symmetrical the dispersion of individual trial effect estimates is around more precise effect . The funnel plot of acupuncture placebo-controlled trials showed some degree of asymmetry which may have arisen due to publication bias. Publication bias, if present, may have led to overestimation of the treatment effect of acupuncture compared to placebo in reducing pain intensity.
In future, results from long-term large head-to-head trials reporting clinically relevant and validated outcomes are warranted to draw more definitive conclusions regarding benefits and safety of CAM treatments relative to each other or to other active treatments. More research is needed to determine which characteristics of CAM therapies (e.g., mode of administration, length of treatments, number of sessions, and choice of spinal region/points) are useful for what conditions. Future studies should also examine the influence treatment-, care provider-, and population-specific variables on treatment effect estimates. It is clear that strong efforts are needed to improve quality of reporting of primary studies of CAM therapies.
This project was funded under Contract no. HHSA290-2007-10059-I (EPCIII) from the Agency for Healthcare Research and Quality, USA Department of Health and Human Services. The authors of this paper are responsible for its content. Statements in the paper should not be construed as endorsement by the agency for healthcare research and quality, the national center for complementary and alternative medicine, national institute of health or the USA department of health and human services.
The authors declare no conflict of interests.
Agency for Healthcare Research and Quality.
The following filters were applied and overlap removed:
The following filters were applied and overlap removed:
The following filters were applied and overlap removed:
The following filters were applied and overlap removed:
The following filters were applied and overlap removed:
(((((“BACK PAIN” or “NECK PAIN”) or “SPINAL DISEASES”) or “BACK INJURIES”) or “SPINAL INJURIES”) or “NECK INJURIES”) or “SCIATICA” [Descritor de assunto] and acupuncture or electroacupuncture or acupressure or massage or manipulation or chiropractic or osteopathic [Palavras]
Excluded PubMed refs, ACP Jnl Club, Cochrane, ClinicalTrials.gov, animal studies