Much of evidence‑based clinical medicine—and the clinical trials that inform it—rests on a classical assumption: that comparative clinical outcomes are determined primarily by biological treatment effects, holding broader social and environmental determinants of health constant. This assumption holds when biological contrasts between treatments are large, but it breaks down in many common clinical decisions in which treatments are biologically similar yet outcomes diverge. Although patient choice has long been acknowledged in such settings, it has been treated primarily as an ethical consideration rather than as a contributing cause of physical outcomes. What has changed is not the presence of patient choice in clinical trials, but what can be identified, interpreted, and acted upon. Across successive generations of trial designs, patient preference has shifted from an implicitly acknowledged feature of care to a measurable contributor to outcomes. Recent reanalysis of the Spine Patient Outcomes Research Trial (SPORT) illustrates this shift by isolating a preference‑mediated component of outcome (β) on the same clinical scale as biologically mediated treatment effects (α). Under near biological equipoise, biological differences were clinically negligible, whereas preference‑mediated effects were large and stable across analyses. Behaviors such as crossover and nonadherence are therefore better understood as structured causal responses rather than as analytic bias. Taken together, these findings define distinct regimes in which outcomes are governed by biology and those in which choice is decisive. Recognizing this distinction constrains both interpretation and action: evidence must identify biological and preference‑mediated effects jointly, decision systems must defer when biology cannot decide, and patients must be supported—through medical literacy—to enact informed choice. Treating patient choice as causal does not weaken evidence‑based medicine; it clarifies its domain and renders preference‑sensitive care coherent and rule‑governed.

Randomized clinical trials in preference‑sensitive care often yield attenuated or equivocal results, particularly in the presence of crossover and nonadherence. These features are typically treated as bias or noise, obscuring whether modest or null effects reflect ineffective treatments, insufficient power, or misaligned interpretation. We argue that the core problem is estimand mismatch, with direct consequences for trial design, power, and interpretation. In many contemporary trials, physical outcomes arise jointly from biological treatment effects (α) and preference‑mediated patient choice effects (β), yet most designs and interpretive conventions remain α‑centric. We present a principle‑based framework that formalizes this dual causation once β is empirically identified on physical outcome scales. We state a minimal set of axioms and design constraints, define canonical co‑estimands (α, Δα, β, Δβ) on a common clinical scale, and specify prespecified interpretive rules that route inference, design, and action by the joint magnitude of Δα and β relative to clinical materiality. These rules yield four empirically distinct regimes—α‑dominant, mixed, β‑dominant (equipoise), and decision‑limited ambiguity—each governed by distinct rules of interpretation and trial architecture. The framework shows that β is not incidental to crossover, but a causal quantity that can be identified when randomized assignment is paired with structured observation of treatment choice on a common outcome scale. From this result, we derive a small set of regime‑routed β‑identifying trial designs that render preference effects empirically legible rather than residual. As aging, multimorbidity, and therapeutic multiplicity compress biological contrasts while amplifying preference‑mediated effects, many trials predictably shift toward β‑dominant regimes. In these settings, β‑identifying evidence is not optional but structurally necessary, providing a principled basis for preference‑concordant decisions and N‑of‑1 inference when biological contrasts are non‑decisive.

AI decision-support systems are increasingly deployed in clinical settings where biological treatment effects are small and outcomes depend materially on patient choice. In prior work, we showed that such preference-sensitive decisions reveal a correctness failure of prediction-centric AI: systems trained to rank options by average biological effects act illegitimately under near equipoise unless they detect preference-sensitive regimes, defer premature ranking, and elicit patient preferences neutrally. That β-aware framework secures non-maleficence by enforcing restraint where biological superiority cannot decide. This paper addresses the downstream question: once safety is assured, what forms of optimization are causally, technically, and ethically admissible? We introduce β-optimization, a constrained framework for AI decision support in preference-sensitive care. Rather than inferring or shaping preferences, β-optimization treats explicitly elicited patient values and feasibility constraints as measured inputs and seeks improvement by maximizing concordance between evidence and what patients value and can sustain. We formalize concordance-based objectives appropriate to observer-dependent decision regimes, specify architectural constraints that preserve neutrality, deferral, and causal separation, and show how large language model–based systems can condition recommendations on elicited preferences without exercising illegitimate authority. We further propose concordance-first evaluation metrics—epistemic, practical, and decisional—for settings where prediction accuracy is ill-posed. Together, β-awareness and β-optimization define a regime-aware theory of AI decision support for preference-sensitive domains: first, do not decide when biology cannot decide; then, once preferences are measured, help decisions succeed—without shaping them.

Large language models (LLMs) increasingly support clinical decisions by synthesizing population-level evidence and estimating average biological treatment effects (α). However, they are not designed to represent the causal contribution of patient choice (β)-a limitation that becomes clinically decisive in preference-sensitive care, where biological differences between options are small. Using the Spine Patient Outcomes Research Trial (SPORT) as an orienting case-a uniquely designed trial in which randomization and patient choice coexisted-we show that preference-mediated effects were clinically meaningful under conditions of near biological equipoise. In such settings, α-centric decision-support systems systematically misrank options because they lack access to β. We formalize this limitation as "α-bias" and "β-blindness," and propose a regime-routed architecture-Detect → Elicit → Recommend → Learn-with explicit deferral rules, neutrality constraints, provenance, and auditable guardrails. When outcomes hinge on choice rather than biological superiority, elicitation is a precondition for recommendation, not an optional refinement. These principles extend beyond medicine to any domain in which outcomes depend on decision-contingent preferences rather than fixed parameters. When outcomes depend on choice rather than biology, improving choices is improving outcomes. A companion paper addresses how β-aware systems may permissibly improve concordance once these safety conditions are satisfied and thus AI decision support moves to βoptimization.

Education is classified as a social determinant of health, yet its monolithic placement hides a functional triad. If determinants act through streams, and education through three, it cannot remain one. With set-theoretic reasoning in plain prose, we show that education divides into general education (GE), health literacy (HL), and medical literacy (ML). GE shapes social opportunity; HL guides prevention; ML enables recovery. A simple healthy ↔ morbid flow makes the HL-ML divide explicit: HL dampens entry into illness, and ML hastens return to health. Beyond mechanism, HL and ML differ in time scale and life-course leverage: HL operates upstream with diffuse, long-latency effects that accrue early in life, whereas ML acts peri-morbidly, yielding nearterm, attributable effects on recovery that dominate later in life. Medical literacy is not merely conceptual: it is empirically instantiated in shared decision-making, patient decision aids, decision-quality measurement, patient activation, and risk-communication practices that produce observable, near-term effects during clinical recovery. Conflating these distinct temporal roles obscures both causal structure and practical leverage, further substantiating and motivating a tripartite classification of education determinants.

Public health policy often treats education as a single social determinant, obscuring the distinct mechanisms through which its components act. The tripartite model-General Education (GE), Health Literacy (HL), and Medical Literacy (ML)-maps each component to a different element of a simple two-state public-health structure: GE strengthens the healthy state (H); HL reduces preventable transitions from health to morbidity (H → M); and ML accelerates recovery from morbidity to health (M →  H) by improving comprehension, navigation, and decision quality. We apply this framework to policy design, measurement, and allocation, distinguishing effectiveness changes that strengthen states or flows, level changes that expand GE, HL, or ML supply, and cross-elasticities through which GE amplifies both literacies. Age-structured and time-latency differences shape policy returns: HL produces long-horizon preventive gains that accrue slowly and are difficult to attribute, whereas ML yields short-horizon, attributable recovery gains aligned with clinical and budget cycles. Mechanism-aligned metrics, certified decision aids, comprehension checks, and digitally supported ML infrastructure reduce unwarranted variation, advance equity, and close persistent gaps in understanding, decision quality, and recovery. Education is not one determinant but three-and policy must follow mechanism. 

A document by Ogan Gurel. Click on the document to view its contents.