A search of the English language literature was performed through August 2003 to identify reports of randomized controlled trials of OMT. We searched MEDLINE, OLDMEDLINE, EMBASE, MANTIS, OSTMED, Alt Health Watch, SciSearch, ClinicalTrials.gov, CRISP, and the Cochrane Central Register of Controlled Trials. A detailed description of the search strategy is provided in the Appendix [see Additional file 1]. Additionally, reports were sought from relevant reviews or meta-analyses of spinal manipulation 9151617181920212223242526272829303132 and manual searches of reference citations in the reviewed literature sources.

The search bibliographies and relevant reports were reviewed by the authors to identify randomized controlled trials involving OMT in human subjects. To assess the efficacy of OMT in primary care, eligibility was limited to randomized controlled trials of OMT performed by osteopaths, osteopathic physicians, or osteopathic trainees that included blinded assessment of low back pain in ambulatory settings. Trials that involved manipulation under anesthesia, industrial settings, or hospitalized patients were not included. Because there is potential confusion regarding the type of manipulation performed in some trials 33, the reported methods in each trial were carefully reviewed to assess eligibility for the meta-analysis. Overall, seven studies known or purported to involve OMT for low back pain were reviewed and excluded for not meeting all eligibility criteria 34353637383940. A subsequent source 41 indicated that an osteopathic manipulation technique was used in the Irvine study 42. Although several of the six included OMT trials were identified in multiple bibliographic databases, five 4243444546 were indexed in MEDLINE. The Cleary 47 trial was identified exclusively through the Cochrane Central Register of Controlled Trials.

Each eligible trial was independently reviewed by two of us to abstract data on methodological characteristics, OMT and control treatments, and low back pain outcomes. Conflicting data were resolved by consensus.

As shown in Table 1, the six OMT trials were conducted between 1973 and 2001 in the United Kingdom or the United States 424344454647. Two of the six trials each included two control treatments 4346, thus providing eight OMT vs control treatment comparisons. The trials generally were comparable in their methodology, with the possible exception of the Cleary 47 trial. Twenty contrasts were reported in the six trials (a contrast refers to a within-trial comparison between OMT and a control treatment with respect to a low back pain outcome at a given point in time). Following randomization, nine contrasts were reported within one month (short-term outcomes), another seven contrasts were reported within three months (intermediate-term outcomes), and the remaining four contrasts were reported within 12 months (long-term outcomes).

The methodological quality of four of the OMT trials 42434445 was independently confirmed in a recent systematic review that included a best evidence synthesis incorporating eight explicit quality criteria, including similarity of baseline characteristics of subjects or reporting of adjusted outcomes; concealment of treatment allocation; blinding of subjects; blinding of providers or other control for attention bias; blinded or unbiased outcomes assessment; subject dropouts reported and accounted for in the analysis; missing data reported and accounted for in the analysis; and intention-to-treat analysis or absence of differential co-interventions between groups in studies with full compliance 13. The Cleary 47 trial was not eligible for this review because it did not include a sufficiently large number of subjects. Although the Licciardone 46 trial was not eligible for the review because it was published after the closing date of December 2002, it has been characterized as an innovative and important trial with many rigorous design features 48, and more recently has been identified as an evidence-based supplement relative to the previous review from the Cochrane Library 49.

We used the effect size, computed as Cohen's d statistic, to report all trial results 50. A negative effect size represented a greater decrease in pain among OMT subjects relative to control treatment subjects. Dichotomous pain measures were transformed to effect sizes by first computing the relevant P-value and then determining the effect size and 95% confidence interval (CI) that would obtain under the assumption of a two-tailed t-test for measuring the standardized mean difference between OMT and control treatments in the relevant number of subjects 50. The meta-analysis results were weighted by the inverse variance for each OMT vs control treatment comparison. The Q statistic was used to test the homogeneity of trials included in each analysis 51.

The overall meta-analysis included the eight OMT vs control treatment comparisons. Four of the six trials, involving six of the eight OMT vs control treatment comparisons, each reported three contrasts 42434546 (Table 1). The median contrast, as identified by the intermediate effect size among the three reported pain outcomes for a given OMT vs control treatment comparison, was used to represent the pain outcome for each of these six comparisons. These median contrasts were then combined with the lone contrasts reported in each of the two remaining OMT vs control treatment comparisons 4447. Based on the similarity among trials (Table 1), a fixed-effects model initially was used to perform meta-analysis and the results were then compared with those of a random-effects model.

A series of sensitivity analyses were then performed. First, to address the possibility of bias by using the median contrasts method, analyses were repeated using the best-case and worst-case scenarios for the six relevant OMT vs control treatment comparisons 42434546. Second, to address the possibility of bias by including comparisons involving the same OMT group vs two different control treatment groups in two trials 4346, analyses were repeated using only one OMT vs control treatment comparison for each of these trials. Each of the four possible combinations of contrasts was analyzed. Third, the analysis was repeated after excluding the Cleary 47 trial. Finally, an analysis was performed using all 20 low back pain contrasts. Similar analyses were performed after stratifying trials according to control treatment, country where the trial was performed, and duration of follow-up.

As summarized in Table 2, there were 43 analyses performed, including the overall meta-analysis, seven stratified meta-analyses, and 35 sensitivity analyses. Meta-analysis was performed only when there were at least three contrasts available for data synthesis. Database management and analyses were performed using the Comprehensive Meta-Analysis software package (Version 1.0.23, Biostat, Inc, Englewood, NJ 07631, USA).

The search for reports is summarized in Figure 1. A total of 525 subjects with low back pain were randomized in the eligible trials. The overall results are presented in Figure 2. There was a highly significant reduction in pain associated with OMT (effect size, -0.30; 95% CI, -0.47 – -0.13; P = .001). The Q statistic was non-significant, thus supporting the assumption of homogeneity among trials. The primary sensitivity analyses are presented in Table 3. Using a random-effects model, the results were virtually identical to those observed with a fixed-effects model. There were 729 (3⁶ × 1²) possible combinations of contrasts for analysis based on three contrasts for each of six OMT vs control treatment comparisons 42434546 and one contrast for each of the two remaining OMT vs control treatment comparisons 4447. The efficacy of OMT for low back pain was supported in both the best-case (effect size, -0.37; 95% CI, -0.55 – -0.20; P < .001) and worst-case (effect size, -0.18; 95% CI, -0.35 – 0.00; P = .046) scenarios. Similarly, when each trial was limited to one OMT vs control treatment comparison, OMT was found to be efficacious in each of the four analyses. OMT also demonstrated significantly greater low back pain reduction than control treatment in analyses with the Cleary 47 trial excluded and with all 20 contrasts included.

The results of stratified meta-analyses are presented in Table 4. There was a significant reduction in low back pain associated with OMT in trials vs active treatment or placebo control (effect size, -0.26; 95% CI, -0.48 – -0.05; P = .02), independent of fixed-effects or random-effects model specification. There were 243 (3⁵ × 1¹) possible contrast combinations based on three contrasts for each of five OMT vs control treatment comparisons 42434546 and one contrast for another remaining OMT vs control treatment comparison 47. Both the best-case and worst-case scenarios demonstrated a greater reduction in pain with OMT than active treatment or placebo control, although the worst-case results did not achieve statistical significance. OMT was found to significantly reduce pain in the remaining analyses that limited OMT vs active treatment or placebo control comparisons to one per trial, excluded the Cleary 47 trial, and included all 16 contrasts. The OMT vs no treatment control comparisons were observed in trials in which all subjects received usual low back care in addition to their allocated treatment (ie, OMT and usual care vs only usual care) 4447. For these trials, the all-contrasts model (ie, the only model with sufficient contrasts for data synthesis) demonstrated a highly significant reduction in pain with OMT.

Trials in both the United Kingdom (effect size, -0.29; 95% CI, -0.58 – 0.00; P = .050) and the United States (effect size, -0.31; 95% CI, -0.52 – -0.10; P = .004) demonstrated significant reductions in low back pain associated with OMT. In the sensitivity analyses, effect sizes were generally of comparable magnitude in both countries, although results in American trials consistently achieved statistical significance as a consequence of the larger sample sizes in these trials (Table 4).

There were significant reductions in low back pain associated with OMT during the short-term (effect size, -0.28; 95% CI, -0.51 – -0.06; P = .01), intermediate-term (effect size, -0.33; 95% CI, -0.51 – -0.15; P < .001), and long-term (effect size, -0.40; 95% CI, -0.74 – -0.05; P = .03) follow-up periods. Sensitivity analyses for temporal outcomes demonstrated that intermediate-term results consistently achieved statistical significance, generally because of the greater number of subjects in these analyses (Table 4). The results of the meta-analyses and sensitivity analyses are further summarized in Figure 3.

The overall results clearly demonstrate a statistically significant reduction in low back pain with OMT (Figure 2). Further, the meta-analysis results are quite robust as indicated by the comprehensive sensitivity analyses (Figure 3). Stratified meta-analyses to control for moderator variables demonstrated that OMT significantly reduced low back pain vs active treatment or placebo control and vs no treatment control. If it is assumed, as shown in a review 52, that the effect size is -0.27 for placebo control vs no treatment in trials involving continuous measures for pain, then the results of our study are highly congruent (ie, effect size for OMT vs no treatment [-0.53] = effect size for OMT vs active treatment or placebo control [-0.26] + effect size for placebo control vs no treatment [-0.27]).

It has been suggested that the therapeutic benefits of spinal manipulation are largely due to placebo effects 53. A preponderance of results from our sensitivity analyses supports the efficacy of OMT vs active treatment or placebo control and therefore indicates that low back pain reduction with OMT is attributable to the manipulation techniques, not merely placebo effects. Also, as indicated above, OMT vs no treatment control demonstrated pain reductions twice as great as previously observed in clinical trials of placebo vs no treatment control 52. Thus, OMT may eliminate or reduce the need for drugs that can have serious adverse effects 44.

Because osteopathic physicians provide OMT to complement conventional treatment for low back pain, they tend to avoid substantial additional costs that would otherwise be incurred by referring patients to chiropractors or other practitioners 54. With respect to back pain, osteopathic physicians make fewer referrals to other physicians and admit a lower percentage of patients to hospitals than allopathic physicians 1, while also treating back pain episodes with substantially fewer visits than chiropractors 55. Although osteopathic family physicians are less likely to order radiographs or prescribe nonsteroidal anti-inflammatory drugs, aspirin, muscle relaxants, sedatives, and narcotic analgesics for low back pain than their allopathic counterparts, osteopathic physicians have a substantially higher proportion of patients returning for follow-up back care than allopathic physicians 56. In the United Kingdom, where general practitioners may refer patients with spinal pain to osteopaths for manipulation, it has been shown that OMT improved physical and psychological outcomes at little extra cost 57.

In our study, the effect sizes for OMT in the United Kingdom, where osteopaths are not licensed physicians, were generally comparable to those in the United States, where OMT is provided by licensed physicians. This consistency suggests that the results truly reflect the effects of OMT itself, and not other elements of low back care. It is not surprising that osteopaths in the United Kingdom achieved pain reduction with OMT similar to that of their physician counterparts in the United States. The training of osteopaths in the United Kingdom is highly focused on OMT, whereas osteopathic physicians undertake a medical curriculum that necessarily relegates OMT to one of many therapeutic approaches, albeit a fundamental one for osteopathic practitioners. Regardless of the career training path of the provider, it appears that OMT achieves clinically important reductions in low back pain.

There are several potential limitations of this study that should be addressed. First, as with any meta-analysis, the individual trials varied somewhat with respect to methodology, including trial setting, subject characteristics, OMT and control treatment interventions, and pain measures (Table 1). Such heterogeneity has been commonly observed in previous meta-analyses of spinal manipulation, including a recent meta-analysis performed in collaboration with the Cochrane Back Review Group 31. The latter study addressed potential heterogeneity by presenting stratified results according to chronicity of low back pain, type of control group, and duration of follow-up. This approach is analogous to the methods used in our study. Further, it should be noted that the assumption of homogeneity among trials was not rejected statistically in any of our eight overall or stratified median contrasts meta-analyses.

Second, because five trials each included repeated pain measures and two trials each included two control treatments, there was no unique set of independent outcomes for meta-analysis. Such interdependencies were noted to be a problem in an early meta-analysis of spinal manipulation 15. We used the median contrasts method to address this problem because the median outcome represents an observed outcome that is easy to compute and is less vulnerable to extreme observations than other measures of central tendency. Further, sensitivity analysis was used to assess the range of possible combinations of outcomes. Thus, for the overall meta-analysis, there were 729 potential contrast combinations. Of these, both the best-case and worst-case scenarios demonstrated statistically significant results favoring OMT, thereby providing unequivocal evidence for the efficacy of OMT. Robust findings were also observed for trials performed in the United States and for intermediate-term outcomes.

Third, because there were a relatively small number of eligible trials, there were not sufficient contrasts for certain analyses and some results were imprecise. The latter phenomenon likely obviated the statistical significance of some results. Nevertheless, it is important to note that the direction of results favored OMT in each of the 43 meta-analyses and sensitivity analyses presented herein (Figure 3).

Fourth, there exists the possibility that the results of unpublished trials of OMT for low back pain may have altered significantly the conclusions of this study. To address this issue, we performed file drawer analysis by computing the fail-safe N 58. This represents the number of unpublished trials of OMT for low back pain that would have met our inclusion criteria, and that also would have demonstrated an effect size averaging ≥ -0.10, which is assumed to reflect clinically insignificant levels of pain reduction. A total of 16 unpublished trials (assuming one control group per trial) with, in the aggregate, clinically insignificant pain reduction outcomes would have been needed to obviate the significance of our results. Only recently has government funding for research in the area of complementary and alternative medicine become more widely available, in response to the public's interest in such treatments. Historically, it is highly unlikely that 16 trials of OMT for low back pain would have been sponsored, conducted, and subsequently not published.

Finally, this study focused only on the efficacy of OMT with respect to pain outcomes. Generic health status, back-specific function, work disability, and back-specific patient satisfaction are other recommended outcome domains 59 that were not assessed because the included OMT trials did not consistently report these data.