Sixty-Six Patients, Three Verbal Framings, and the Mediator That Was Not Belief

The Study

Zaworski et al. ran a secondary analysis of a single-session randomized trial conducted at a university physiotherapy department in Poland. The original parent trial (NCT06537739) enrolled 66 young adults (mean age in their early 20s) with chronic non-specific low back pain of at least three months' duration, no radiculopathy, no red flags, and no recent spinal manipulation exposure.

Participants were randomized 1:1:1 to a positive, neutral, or negative verbal framing script. The script was delivered by the treating clinician immediately before a single lumbar high-velocity low-amplitude (HVLA) thrust technique. Manipulation technique, contact, set-up, and operator were standardized across groups. The only intentionally varied input was the framing language.

Three outcomes were collected at three time points. The 11-point Numerical Pain Rating Scale (NPRS-11) and lumbar pressure pain threshold (PPT) at L3, L4, and L5 were measured at baseline (t0), immediately post-manipulation (t1), and 24 hours after the visit (t2). Pre-treatment, the participants also completed measures of expectations for treatment success, attitude toward manual therapy, and (at t2) whether their expectations had been met.

The analytic question for this secondary analysis was specific. Holding the manipulation itself constant, which of the candidate cognitive-affective variables (expectations, attitude, expectation fulfillment) explained variation in NPRS-11 and PPT at t1 and t2? And how did those candidates compare to the verbal framing assignment itself when entered into the same model?

What They Found

Two findings, one direct, one by absence.

First, the cognitive-affective candidates that the literature most often invokes (expectations, attitude, expectation fulfillment) produced sparse and inconsistent associations across groups. Expectation fulfillment, measured at the 24-hour follow-up, was not consistently linked to either NPRS-11 or PPT at t1 or t2. Some scattered correlations appeared inside subgroups. In the positive framing group specifically, higher pre-treatment expectations correlated with higher immediate PPT at t1 L4 (r = 0.47) and L5 (r = 0.43), and a more favorable attitude toward manual therapy correlated with lower NPRS-11 at t2 (r = -0.45). Outside the positive group, the pattern dissolved.

Second, when the authors entered the candidate variables together with the framing assignment in a multiple regression predicting NPRS-11 at 24 hours, framing was the only retained predictor.

Stated in one sentence: the message the patient heard immediately before the thrust explained more of their next-day pain than any pre-treatment belief the patient brought into the room.

What This Means Clinically

In clinical practice this finding lands harder than the abstract suggests, because the field has spent two decades operating on a different mental model. The standard biopsychosocial framing places patient beliefs (expectations, catastrophizing, fear avoidance, locus of control) upstream of outcome. The clinician's job, in that model, is to assess and address those beliefs across the course of care. The Zaworski analysis does not refute that model. It does suggest that, at the level of a single visit, what the operator says in the seconds before contact is doing more work than the beliefs the patient walked in with.

For me, this maps cleanly to what happens at the adjusting bench every day. A patient comes in tight, guarded, partway through a flare, and the temptation is to compress the encounter. Position, set-up, thrust. The framing language compresses with it. The patient hears something perfunctory ("alright, breathe out") instead of something specific ("this is going to feel like a release through the side of your low back, and most patients tell me the next morning is when they notice the change"). Zaworski et al. are quantifying the cost of that compression. Roughly a quarter of the variance in next-day pain may be sitting in the thirty seconds before the contact.

That is a high return on a small clinical investment.

The Negative-Framing Asymmetry

Notice the size of the effects. Positive framing predicted lower pain at β = -0.26. Negative framing predicted higher pain at β = +0.35. The negative effect is larger. This is consistent with the broader nocebo literature, which has repeatedly shown that warnings and negative expectancy cues land harder than reassurance does. The clinical implication is twofold. First, the upside of careful positive framing is real but modest. Second, the downside of careless negative framing (offhand warnings, defensive over-disclosure of post-adjustment soreness, low-confidence delivery of the technique itself) is larger and more reliable. The asymmetry argues for ruthless audit of the language clinicians use without thinking.

Why Belief-State Measures Fall Short Here

The candidate variables that did not survive the regression (pre-treatment expectations, general attitude, expectation fulfillment) are stable, dispositional, and trait-like. They were measured before the patient ever met the framing script. The framing script was acute, situational, and immediately prior to the intervention. State variables outperformed trait variables in this single-encounter design. That is not a surprising result if the question is framed correctly. It is a surprising result if the field has spent years filling intake paperwork with trait-level psychometrics in the assumption that those scores would predict short-term response.

Across a full course of care, trait variables likely matter more than this one study can detect. Across a single visit, this study suggests that the state-level cue from the clinician dominates.

Where This Fits in the Literature

The Zaworski analysis sits at the intersection of three threads in the manual therapy evidence base.

The first thread is the Cochrane systematic review of spinal manipulation for chronic low back pain, which found a small to moderate average effect for SMT with substantial between-trial heterogeneity (de Zoete et al. 2026). The Zaworski finding offers one mechanistic candidate for that heterogeneity. If verbal framing is doing this much work inside a single visit, framing variability across trial operators is a non-trivial source of trial-to-trial noise that the standard meta-analytic apparatus is not equipped to detect.

The second thread is the PACBACK trial, in which biopsychosocial care embedded inside a manipulation arm produced different long-term outcomes than manipulation alone (Bronfort et al. 2026). PACBACK was framed as evidence for adding structured psychological care to the manual therapy encounter. The Zaworski result raises a quieter question. How much of PACBACK's biopsychosocial signal might already be available in the framing dose embedded in the verbal contact, before any formal psychological protocol is added?

The third thread is the methodological critique that twenty-two years of SMT trial evidence may have already saturated certainty on average effect size, with diminishing return on further trials of the same design (Elliott et al. 2026). If Elliott et al. are correct that the field has wrung what it can from average-effect-size questions, the next frontier is the moderator and mediator literature. Zaworski et al. are an early data point in exactly that frontier. Verbal framing is a moderator the field can manipulate cheaply and reproducibly.

Put together, the three threads suggest the same conclusion. The next decade of meaningful progress in manual therapy outcomes research is unlikely to come from another generic SMT-versus-control trial. It is likely to come from quantifying the contextual variables (operator framing, contact ritual, encounter pacing) that determine whether the same physical technique, delivered the same way, produces a useful response or a non-response.