Evidence/estimand critique

### PDF page 14 contested strands, and the central asymmetry rests on the discipline’s own highest-quality evidence; the review supports only the weak, defensible claim that the burden of proof for compulsion is unmet, not the stronger claim that net harm from medication is proven. Conclusion The central finding is not merely that the evidence for sustained compulsory antipsychotic treatment is weak; it is that the wrong evidence has been used to justify a misclassified intervention. Placebo-controlled trials test drug exposure in consenting participants; compulsory maintenance imposes a different intervention — drug exposure together with legal force, surveillance, threat of recall, loss of autonomy and a transformed therapeutic relationship — and evidence for the former cannot validate the latter. When a society removes a person’s right to refuse treatment it assumes one of the highest evidentiary burdens in medicine. Across the six- condition standard that burden is not discharged, and it fails most clearly at the independently dispositive condition: legal force has not been shown to improve long-term patient-centred outcomes beyond supported voluntary care — a gap measured against the discipline’s own randomised trials and its drugs’ own regulatory labels. The move from “the burden is unmet” to “the practice should be deimplemented as a default” rests on one explicit normative premise: where a competent refusal is overridden, the default is non-coercion unless the overriding element is affirmatively justified, because consent is protective and its removal is what requires warrant. This is a premise about the direction of the burden of proof, not a claim that net harm has been demonstrated. On that premise, the defensible response is not minor procedural reform: sustained compulsory maintenance should no longer be treated as evidence-based default care unless direct evidence shows that legal force improves long-term, patient-centred outcomes beyond voluntary, supported, less- restrictive care. Its routine use should be wound back accordingly; coercion should be confined to genuinely time-bounded emergencies; and resources should be redirected toward the supports a person could rationally choose without force. This entails neither the abolition of psychiatric care nor a bar on time-limited emergency intervention. A society cannot justify removing consent by citing evidence for an intervention that was never the one being imposed.

Design. Structured critical review built around six proposed conditions for justified coercion — valid measurement (M), clinically meaningful benefit (B), incremental benefit of legal force over voluntary care (F), individual predictability of risk (R), proportionate harms (H), and exhaustion of less-restrictive alternatives (L) — with F identified as the load-bearing condition. Data sources. MEDLINE, Embase, PsycINFO and the Cochrane Library; regulatory prescribing information; and Australian government statistics, to May 2026. Results. The argument is carried by one condition, F. Randomised evidence on compulsory

applicable and are documented in online supplemental file 1; meta-analytic pooling was not appropriate because the included evidence spans measurement validity, efficacy, coercion- specific trials, risk prediction, harms and alternatives, which are not commensurable as a single effect size. The standard, and its external warrant. This review proposes that sustained coercive antipsychotic treatment is justified only if six conditions hold: (M) the disorder and any improvement can be validly measured; (B) the drug confers a clinically meaningful benefit; (F) adding legal force improves outcomes over voluntary, supported care; (R) the serious risk said to justify compulsion is individually predictable; (H) the harms are proportionate to the benefit; and (L) no less-restrictive alternative would be adequate. These are advanced as the author’s framework, not as a pre-existing consensus checklist. Each, however, reflects a principle already established in evidence-based medicine, law or ethics — valid measurement underlies admissibility standards for scientific evidence; clinically meaningful benefit is a standard EBM

cannot be proportionate, and the availability of less-restrictive options not shown to have been exhausted (L) defeats the necessity defence. Conditions M (measurement), B (benefit) and R (risk prediction) are corroborating: they show that the intervention being compelled is not even cleanly measurable, large, or targetable, but each contains links that are individually contestable, and the conclusion does not depend on any of them. Presenting the framework this way concentrates the argument on the ground that cannot be conceded without abandoning the discipline’s own randomised and cross-jurisdictional evidence, and it makes explicit that a

therapeutic margin maintenance is supposed to provide narrows substantially once the withdrawal confound is acknowledged. Condition R: individual risk is not reliably predictable Where compulsion is justified by predicting that a particular person will, without forced treatment, commit serious violence, the instruments do not supply the individual-level certainty required. The most comprehensive synthesis — a systematic review of 73 samples and nearly 25,000 people — found that violence-risk instruments discriminate only moderately (areas under the curve roughly 0.66–0.74) and perform far better at identifying who is low-risk than at confirming who is high-risk.38 The decisive quantity is not discrimination but positive predictive value: because serious violence is comparatively rare in this population, even a moderately discriminating instrument flags many more false positives than true ones, so a positive result cannot supply the individualised certainty that overriding autonomy is supposed to rest upon. Compulsion is also justified on other statutory grounds — self-harm, deterioration, inability to self-care — whose individual-level predictability is likewise unestablished; the violence case is treated here because it is the one for which formal predictive evidence exists, and it fails on its own terms. Rational refusal and the preference-sensitive threshold A further implication follows from the benefit–harm balance. Sustained antipsychotic maintenance is not a binary, life-saving intervention like antibiotics for sepsis; it is high-risk and physical and regulatory measures (H); less-restrictive alternatives are not shown to be exhausted before compulsion (L); and the corroborating measurement, benefit and risk- prediction conditions converge. The asymmetry is the heart of the matter: coercive treatment imposes harms measurable in direct physical units to secure a benefit that the discipline’s own instruments cannot cleanly separate from sedation. Murray — Compulsory antipsychotic treatment and the evidentiary standard for coercion 10

required Status Is risk individually predictable enough to override consent? Group-level AUCs High positive predictive value for serious harm in the individual Not shown Are less-restrictive

authorises treatment without consent on diagnosis and risk alone. The evidentiary and legal critiques are moving in the same direction. What would refute this argument. The thesis is falsifiable. It would be defeated by: diagnostic reliability shown to be high (kappa ≥ 0.6) under blinded naturalistic assessment; a pooled drug– placebo difference exceeding the field’s minimal-clinically-important threshold once sedation is controlled and unpublished trials included; evidence that compulsory maintenance produces superior long-term functional outcomes to supported voluntary care; or a validated risk instrument able to predict serious harm at the individual level with positive predictive value sufficient to justify detention. The evidence required need not be an unethical randomised trial of coercion itself: target-trial emulation, natural experiments and stepped-wedge service reforms could all isolate the effect of legal force on patient-centred outcomes. The demand is not for the impossible, but for evidence capable of isolating the coercive component before that component