Treatment (α) and Patient Choice (β) Effects in Medicine: Quantifying the Causal Effect of Preference on Physical Outcomes

Randomized trials are designed to elicit biological treatment effects, and patient choice-treated as a potential confounder-is therefore typically suppressed.When preference-mediated selection is permitted within a randomized framework-as in the singular case of the Spine Patient Outcomes Research Trial (SPORT)-the resulting divergence between randomized and observational estimates, under substantial bidirectional crossover, can be interpreted not as noise but as a measurable component of outcome.Using published contrasts, instrumentalvariable analysis with randomized assignment as the instrument estimated biological effects among compliers (α).We defined a preference-mediated effect (β) as the residual between observational contrasts and causal (IV/CACE) estimates under coexistence of randomization and choice.Across SF-36 Bodily Pain, Physical Function, and the Oswestry Disability Index, the aggregate biological effect was substantial (αG ≈ 15.7), but the net biological difference between surgical and nonoperative care was small (Δα ≈ 0.65).By contrast, the preference-mediated effect was clinically meaningful (βG ≈ 7.4) and exceeded Δα by more than an order of magnitude, placing SPORT in a β-dominant regime in which comparative outcomes depend more on preference alignment than on treatment superiority.To our knowledge, this is the first quantification of a preference-mediated effect on physical outcomes.Unlike patient-preference trials-which document preference without identifying a causal effect on outcomes-our framework separates β from α, reframes crossover as informative behavior, and generalizes to conditions of clinical equipoise with biologically similar outcomes.More broadly, Δα and β define orthogonal axes of clinical decision-making-a 2×2 regime map spanning conditions and settings.Accordingly, trials in preference-sensitive domains should be β-identifying by design, and decision support β-aware, eliciting and incorporating patient preferences rather than optimizing only average (α-centric) effects.