Twenty-Two Years of Spinal Manipulation Trials Were Already Sufficient. The Field Kept Running Them Anyway.

The Question

For most clinical questions, the standard meta-analysis treats accumulating evidence as a single static dataset. The pool is everything that exists, summarized as a single effect with a confidence interval. What this approach hides is the temporal structure of how the evidence accumulated. Did the first ten trials answer the question, with the next forty just ratifying the answer? Or did the question remain genuinely open through trial sixty?

Trial Sequential Analysis (TSA) was developed to address this hidden structure. It is borrowed from interim-analysis methodology used in single trials, where researchers monitor accumulating data and stop the trial early if the evidence becomes overwhelming for benefit, harm, or futility. TSA applies the same logic to a research field over time. As trials accumulate, the cumulative Z-curve is plotted against monitoring boundaries. When the curve crosses a boundary, the question is statistically answered, and further trials of the same design are unlikely to change the conclusion.

Elliott and colleagues asked one of the most uncomfortable questions in chiropractic research: when, if ever, did the evidence base for SMT in chronic low back pain become sufficient to draw a stable conclusion? And what does that mean for the trials that came afterward?

The Methods

The team used data from the same Cochrane systematic review this issue covered two days ago, the de Zoete et al. 2026 update of CD008112. Two of the lead authors of that review, Sidney Rubinstein and Annemarie de Zoete, are co-authors on this Elliott analysis. This is significant: the Elliott team had access to the cleanest possible version of the dataset, and the analytic framework is internally consistent with the Cochrane review's findings.

TSA was applied to the comparison most clinicians care about: SMT versus other conservative interventions, at one-month and longer follow-up. The primary outcomes were pain and functional status. For each outcome, the team plotted the cumulative Z-curve over the chronological order of trial publication, and identified the point at which the Z-curve crossed either the conventional statistical-significance boundary, the futility boundary, or the required information size.

60 RCTs were eligible, with 10,312 total participants. Note that this is fewer than the 76 RCTs in the de Zoete Cochrane review. The 16-trial difference reflects the Elliott analysis being limited to the SMT-versus-conservative comparison and excluding sham and no-treatment comparisons.

What They Found

The findings are remarkable in their consistency. Across all four primary comparisons, the evidence base reached its informational ceiling in the early 2000s. From that point forward, additional trials of similar design did not change the answer. The futility boundary, which signals that further evidence is unlikely to demonstrate a clinically meaningful effect or non-effect, was crossed at the same time as the information-size boundary.

Stated plainly: by the time most chiropractors practicing today finished their training, the question of whether SMT outperforms other conservative interventions for chronic LBP was already statistically resolved. The answer, established in 2002 to 2004, is the same answer the de Zoete Cochrane review reports today: SMT and other conservative treatments produce similar outcomes, with a small advantage for SMT on functional status, low certainty.

The trials conducted after the boundary was crossed are not invalid in any individual ethical sense. They were designed and approved with the best intentions. What they failed to do is contribute new information. Patients enrolled, time and resources allocated, careers built around findings that ratified what was already known.

What This Means Clinically and Methodologically

For a practicing clinician, the immediate clinical message is identical to the message from our Cochrane SMT commentary: the evidence supports SMT as a reasonable conservative option for chronic LBP, with effects similar in size to other active conservative interventions. The Elliott paper does not change that bottom line. It strengthens it. Twenty-two years of subsequent trials all ratified the same conclusion. The clinical answer is stable.

The harder message is for the research community, the funders, and the practitioners who consume the literature. If Elliott and colleagues are correct, an enormous amount of research effort has been spent producing redundant evidence. That effort could have been directed at the questions the field actually needs answered:

Which patient phenotype responds best to SMT? The current evidence treats "chronic LBP" as a single homogeneous category, when clinicians know it is anything but. Discogenic, facet-mediated, sacroiliac, central-sensitization-driven, fear-avoidance-mediated. These are different patient profiles and likely respond differently to manual therapy.

Which SMT technique produces the best outcomes for which mechanism? Diversified, Gonstead, drop-table, instrument-assisted, mobilization-only. These are categorically different interventions pooled under one label.

What is the optimal dose? Most trials use four to twelve sessions over four to six weeks. Few have systematically varied dose to identify the inflection point.

What predicts non-response? Identifying the patient profile that does NOT benefit from SMT would let clinicians redirect those patients to other interventions earlier, saving them time and the system money.

The authors of the Elliott paper are not anti-SMT researchers. Two of them are themselves co-authors on the foundational Cochrane review. This is internal critique from the people best positioned to make it.

The Methodological Innovation

TSA as a tool deserves attention beyond this specific application. The method makes visible something that has been hidden in conventional meta-analysis: the temporal structure of evidence accumulation. Every clinical field with a long history of redundant RCTs (acupuncture for various conditions, vitamin supplementation, certain surgical interventions) could benefit from a TSA pass to identify when the question was effectively answered. The Elliott team explicitly advocates for this in their significance statement. Reporting standards from the EQUATOR Network already provide much of the methodological scaffolding for trials that ask new questions instead of replicating old ones.

For chiropractic researchers and journal editors specifically, the implication is more direct: a proposed trial that asks the same question as a 2004 trial, with a similar design, should face a higher publication bar. Reviewers should ask: what does this trial add that 60 prior trials have not?

What I Tell Patients

Patients do not need to know about Trial Sequential Analysis. They need to know that the evidence for chiropractic care for chronic low back pain is mature, stable, and broadly favorable, and that the question of effectiveness is settled enough that the research community is now arguing about which subgroups benefit most. That framing serves patients better than the older "the evidence is mixed and evolving" story, which was true in 1998 and is no longer true today.

When patients ask "is there evidence chiropractic works for back pain," the answer is yes. The evidence reached statistical sufficiency in 2002 to 2004 and has been ratified continuously since. The remaining open questions are about precision (which technique, which patient, which dose), not about whether the intervention works at all.