Comparative Effectiveness of Spinal Manipulation for Low Back Pain: A Bayesian Hierarchical Analysis Incorporating Natural History

Background Spinal manipulation therapy (SMT) is widely used for low back pain (LBP), yet comparative effectiveness findings across trials remain inconsistent. One potential explanation is incomplete separation of treatment-specific effects from natural recovery and placebo response, which may differ between acute and chronic presentations. Methods We conducted a Bayesian hierarchical analysis of 120 randomized controlled trials (298 arms; 19,059 patients) evaluating SMT, exercise, medical management, physical modalities, self-care, and control conditions. Natural history (NH) trajectories were derived from independent observational cohorts and incorporated as external model components. Treatment effects were estimated as deviations from NH, and sham-corrected effects as deviations from control conditions. Results In acute LBP, NH trajectories predicted substantial improvement by 26 weeks. Estimated treatment effects relative to NH were negative across all categories, and sham-corrected effects were near zero, with credible intervals overlapping zero. In chronic LBP, NH trajectories predicted more gradual improvement. Control conditions exceeded NH, consistent with a substantial placebo response. Exercise and SMT demonstrated positive sham-corrected effects with credible intervals excluding zero; however, estimated magnitudes were modest and did not consistently exceed commonly cited thresholds for clinical importance. Higher study quality was associated with smaller estimated effects in chronic trials. Conclusions Modeling natural history as a structural component of treatment response alters estimates of comparative effectiveness in LBP. Improvements observed in acute LBP appear largely consistent with underlying recovery processes, whereas in chronic LBP, contextual and placebo-related effects contribute substantially to observed outcomes. Explicit separation of these components may help reconcile inconsistencies in the literature and inform interpretation of treatment effects.