The Natural History Trajectory of Acute and Chronic Low Back Pain: A Bayesian Meta-Analysis

Background: Improvement observed during treatment is not the same as improvement caused by treatment. For low back pain (LBP), interpretation of treatment-associated change therefore requires an independent estimate of what would have happened anyway. We treated natural history as an uncertain empirical counterfactual and estimated it by Bayesian meta-analysis. Methods: Natural-history and prognosis cohorts were identified iteratively from bibliographic databases, library resources, and citation searching and were synthesized separately for acute and chronic LBP. The acute analysis included eight independent cohorts contributing 26 observations and estimated a combined, weighted trajectory of recovery over time. The chronic analysis treated improvement as a time-varying favorable clinical course rather than permanent cure and synthesized 12 independent cohorts (15 observations) spanning months to years. Study-level heterogeneity and differences in outcome family were represented hierarchically. Each supported chronic observation contributed at its reported follow-up time; 54 weeks was a reporting horizon, not an eligibility cutoff. Posterior distributions, rather than single benchmark values, were carried forward for later treatment comparisons. Results: Acute LBP showed a strong natural recovery trajectory, with an estimated 75.2% recovery by 26 weeks without the treatments evaluated in the companion analysis (posterior mean; 95% credible interval 49.8% to 94.3%). Chronic LBP showed a substantially lower probability of a favorable course, 18.1% at 54 weeks (posterior mean; 95% credible interval 8.8% to 31.9%), and much greater uncertainty. The chronic posterior was asymmetric: most of its density lay at low probabilities of improvement, with a longer tail toward higher values. Together, the acute and chronic posterior distributions quantify both the expected course and the uncertainty in that expectation. Conclusions: Natural history should be estimated rather than assumed, and should be the baseline against which LBP treatments are judged. Acute and chronic LBP require different clinical estimands and different interpretations of improvement, because they are substantially different conditions. This Bayesian meta-analysis provides a phase-specific counterfactual for the companion comparative-treatment analysis. Data: https://doi.org/10.5281/zenodo.23820882Code: https://doi.org/10.5281/zenodo.23655011