A Judea Pearl Structural Causal Model for a disorder defined by a paradox: people with schizophrenia die 15–20 years prematurely, yet the medications most feared by the public are, in the largest cohorts, protective against all-cause mortality — most strongly the long-acting injectables and clozapine. The model separates the relapse axis (what antipsychotics plainly prevent) from the mortality axis (where the surprising survival signal lives). All effect sizes are real published values; aggregation is an upper-bound ceiling.
Causal question. For schizophrenia (~24 million people), what is the identifiable causal contribution of antipsychotic strategy to (a) relapse prevention and (b) all-cause mortality, after backdoor adjustment — and how do long-acting injectables and clozapine compare?
| Axis | Endpoint | Anchor (no antipsychotic) | Decisive lever |
|---|---|---|---|
| Relapse | 1-year relapse / rehospitalization | ~64% relapse at 1 yr (placebo, Leucht 2012) | Maintenance antipsychotic; LAI best |
| Mortality | All-cause mortality | Reference (no antipsychotic use) | SGA-LAI & clozapine (largest protective effects) |
The 15–20-year life-expectancy gap is driven largely by cardiovascular disease and suicide — not directly by psychosis. Comorbid substance-use disorder is a strong independent multiplier of mortality (RR 1.62).
Antipsychotic strategy acts on two outcomes through partly different paths: it suppresses psychotic relapse directly, and it lowers mortality through relapse/suicide prevention and (for some agents) better engagement with care. The central threat is confounding by indication — clozapine and LAIs are prescribed to, and retained by, particular patients — so the mortality estimates are adjusted associations, bounded by E-values rather than proven do-effects.
| Operation | Method | Implementation here |
|---|---|---|
| Relapse effect | RCT meta-analysis | Maintenance antipsychotic vs placebo relapse RR (Leucht 2012); LAI/depot lower than oral |
| Mortality effect | Cohort meta-analysis | All-cause mortality RR by strategy (Correll 2022): SGA-LAI 0.39, clozapine 0.43, any SGA 0.53 |
| Confounding control | Backdoor + E-value bound | Estimates are adjusted associations; E-value quantifies unmeasured-confounding robustness |
| Independent multiplier | Effect-measure modifier | Comorbid substance-use disorder ×1.62 on all-cause mortality (Correll 2022) |
| Adverse-effect caveat | Out-of-model flag | Metabolic / cardiac / agranulocytosis risks noted but not fully priced in this mortality model |
| Minimum-effective set | Pareto frontier (threshold slider) | Strategies ranked by all-cause mortality reduction (Section IV) |
| Intervention | Effect | Source / grade |
|---|---|---|
| Maintenance antipsychotic vs placebo — relapse | RR ~0.40 (27% vs 64% at 1 yr) | Leucht 2012, Lancet (RCT meta) A |
| Depot / LAI vs oral — relapse | RR 0.31 vs 0.46 | Leucht 2012, Lancet A |
| Any antipsychotic vs none — all-cause mortality | RR 0.71 (0.59–0.84) | Correll 2022, World Psychiatry B |
| Second-generation LAI — all-cause mortality | RR 0.39 (0.27–0.56) | Correll 2022 B |
| Clozapine — all-cause mortality | RR 0.43 (0.34–0.55) | Correll 2022; reduces suicide (InterSePT) B |
| ≥80% antipsychotic use — all-cause mortality | aOR 0.73 (0.60–0.88) | Taipale 2020, FIN20 (n=62,250) B |
| Comorbid substance-use disorder — mortality | RR 1.62 (1.47–1.80) | Correll 2022 B |
| Premature mortality gap | 15–20 years | Correll 2022; driven by CV disease & suicide B |
The counterintuitive core: in the largest cohorts, antipsychotic use — and especially the agents associated with the best adherence (LAIs) or reserved for the sickest (clozapine) — tracks with lower all-cause mortality, not higher. Whether that is fully causal is the analysis's central open question.
Antipsychotic strategies ranked by relative all-cause mortality reduction (1 − RR) versus no antipsychotic use. Drag the threshold to set a minimum: strategies at or above the line form the minimum-effective set. Untreated comorbid substance use appears as a net-harmful (red) bar — a reminder that the mortality gap is multi-causal.
Public narrative treats antipsychotics as life-shortening; the largest cohorts find the opposite for all-cause mortality. The likely mechanism is mundane — preventing relapse, suicide, and disengagement — but it inverts the intuition and is the analysis's most important finding to stress-test.
LAIs show the largest mortality benefit partly because they guarantee delivery. The causal target may be less "which molecule" than "continuity of treatment" — which reframes the lever as a systems problem.
Most of the 15–20-year deficit is cardiovascular, not psychiatric. Antipsychotics that worsen metabolic risk could erode their own survival benefit — a tension this mortality-only model cannot fully resolve.
Reserved for the sickest, burdened by agranulocytosis monitoring, clozapine still shows among the lowest mortality and uniquely reduces suicide — suggesting under-use, not over-use, may cost lives.
| Challenge | Why it threatens the conclusion | Mitigation here |
|---|---|---|
| Confounding by indication | Who receives and stays on clozapine/LAI differs systematically; healthier-adherer bias inflates benefit | Mortality estimates labelled adjusted associations; E-value quantifies the confounding needed to nullify them |
| Survivor / immortal-time bias | Cohort designs can credit treatment with survival time accrued before exposure | Graded B; figures framed as associations, not proven do-effects; RCT relapse data kept on a separate axis |
| Adverse effects under-priced | Metabolic, cardiac, and agranulocytosis harms are real and not fully in a mortality-only model | Flagged explicitly; the model is a ceiling for benefit, not a full risk–benefit ledger |
| Relapse ≠ mortality | The two axes have different evidence quality (RCT vs cohort) | Reported separately; relapse from RCT meta (grade A), mortality from cohorts (grade B) |
| Heterogeneity | High I² across mortality subgroups | Confidence intervals shown; effects presented as ranges, not point certainties |
relapse risk = baseline × relapseRR ; relapseRR: oral 0.46 · LAI 0.31 · clozapine 0.40 · none 1.00
mortality RR (vs no antipsychotic): oral SGA 0.53 · SGA-LAI 0.39 · clozapine 0.43 · none 1.00
comorbid SUD: mortality RR × 1.62
mortality reduction = 1 − RR ; E-value(RR) = RR* + √(RR*·(RR*−1)), RR* = 1/RR (RR<1)
Worked example: SGA-LAI, no SUD, baseline 64% → 1-yr relapse 0.64×0.31 = 20%; mortality RR 0.39 → 61% relative reduction; E-value 2.41 (substantial unmeasured confounding would be needed to explain it away). Add comorbid SUD → mortality RR 0.39×1.62 = 0.63.
1. Leucht S, Tardy M, Komossa K, et al. Antipsychotic drugs versus placebo for relapse prevention in schizophrenia: a systematic review and meta-analysis. Lancet 2012;379:2063–2071. Relapse RR ~0.40 (drug 27% vs placebo 64% at 1 yr); depot/LAI 0.31 vs oral 0.46.
2. Correll CU, Solmi M, Croatto G, et al. Mortality in people with schizophrenia: a systematic review and meta-analysis of relative risk and aggravating or attenuating factors. World Psychiatry 2022;21:248–271. Any antipsychotic vs none RR 0.71; SGA-LAI 0.39; clozapine 0.43; any SGA 0.53; comorbid SUD RR 1.62; 15–20-year premature mortality.
3. Taipale H, Tanskanen A, Mehtälä J, et al. 20-year follow-up of physical morbidity and mortality in relationship to antipsychotic treatment (FIN20). World Psychiatry 2020;19:61–68. ≥80% antipsychotic use → all-cause mortality aOR 0.73 (0.60–0.88).
4. Tiihonen J, Lönnqvist J, Wahlbeck K, et al. 11-year follow-up of mortality in patients with schizophrenia (FIN11). Lancet 2009;374:620–627.
5. Meltzer HY, et al. (InterSePT). Clozapine treatment for suicidality in schizophrenia. Arch Gen Psychiatry 2003;60:82–91. Clozapine reduced suicidal behaviour vs olanzapine.
6. Siskind D, et al. Clozapine v. first- and second-generation antipsychotics in treatment-refractory schizophrenia: systematic review and meta-analysis. Br J Psychiatry 2016;209:385–392.
Longevity Research Initiative — Bayesian Causal Atlas. SCM methodology after Judea Pearl; E-values after VanderWeele & Ding. All effect sizes are published values cited above; none were fabricated. Mortality estimates are adjusted observational associations bounded by E-values, not proven causal effects. Outputs are upper-bound ceilings for research, not clinical guidance. Report generated .