Companion reviews: Depression — antidepressants and the evidence, ADHD — stimulants and the evidence, Placebo, unblinding and the evidence, Psychotherapy comparators and the evidence, Diagnosis thresholds and the evidence, Anxiety — a critical review of the evidence, and Stress, Energy and the Capacity to Function.

---

Executive summary

Two folk models dominate public discussion of antipsychotics: that they are highly effective, well-tolerated treatments that correct brain chemistry and should be continued indefinitely (the standard-of-care narrative), and its inversion, that they are minimally effective, profoundly harmful chemicals that worsen long-term outcomes (the critical-psychiatry narrative). Both are overclaims, and the evidence supports neither.

Eight things the evidence actually establishes:

1. All antipsychotics beat placebo in acute trials, with effect sizes ranging from small to moderate. The Leucht et al. (2013) network meta-analysis of 212 trials (43,049 participants) found standardized mean differences versus placebo ranging from 0.88 (95% CrI 0.73 to 1.03) for clozapine to 0.33 (0.22 to 0.43) for iloperidone. All 15 antipsychotics were significantly more effective than placebo. The differences between drugs were small but robust: clozapine was significantly more effective than all other drugs; amisulpride, olanzapine, and risperidone were significantly more effective than most others. Median trial duration was 6 weeks.

2. Discontinuation rates in effectiveness trials are high, with 74% stopping medication within 18 months in CATIE. Lieberman et al. (2005) randomized 1,493 patients with chronic schizophrenia to olanzapine, perphenazine, quetiapine, risperidone, or ziprasidone for up to 18 months. Overall discontinuation was 74% (1,061 of 1,432 who received at least one dose): 64% for olanzapine, 75% for perphenazine, 82% for quetiapine, 74% for risperidone, and 79% for ziprasidone. Time to discontinuation was significantly longer for olanzapine than quetiapine or risperidone, but olanzapine was associated with greater weight gain and metabolic changes. The first-generation drug perphenazine performed comparably to second-generation drugs.

3. First-generation and second-generation antipsychotics show no clear quality-of-life advantage for second-generation drugs in head-to-head pragmatic trials. Jones et al. (2006) randomized 227 patients requiring medication change to first-generation or second-generation antipsychotics (excluding clozapine) with blind outcome assessment over 52 weeks. The primary hypothesis—that second-generation drugs would improve quality of life—was excluded. Participants in the first-generation arm showed a trend toward greater improvements in Quality of Life Scale and symptom scores. Costs were similar. Neither inadequate power nor patterns of drug discontinuation accounted for the result.

4. Dose reduction and discontinuation in first-episode patients shows superior 7-year recovery rates compared to maintenance treatment. Wunderink et al. (2013) randomized 128 patients with remitted first-episode psychosis to dose-reduction/discontinuation or maintenance treatment for 18 months, then followed 103 patients (80.5%) for 7 years. Recovery rates (meeting criteria for both symptomatic and functional remission) were 40.4% in the dose-reduction group versus 17.6% in the maintenance group (OR 3.49, p = .01). Symptomatic remission rates did not differ (69.2% vs 66.7%), but functional remission was significantly higher with dose reduction (46.2% vs 19.6%). The dose-reduction group experienced higher relapse rates during the initial 18-month trial (43% vs 21%, from Wunderink et al. 2007), but long-term recovery rates were more than twice those of the maintenance group.

5. Long-term observational data show better outcomes for patients off antipsychotics after the first two years, but these findings are confounded by selection. [contested] Harrow et al. (2021) followed 139 patients with schizophrenia or affective psychosis over 20 years with 734 observations. After the second year, patients not on antipsychotics (regardless of diagnosis) had adjusted odds ratio of 5.989 (95% CI 3.588–9.993) for recovery and 0.134 (0.070–0.259) for rehospitalization compared to those on medication. The study controlled for multiple baseline prognostic indicators, but residual confounding by indication remains: patients who discontinue medication may have better prognoses, less severe illness, or greater resilience. The finding challenges but does not refute the value of continuous medication, because it is observational and cannot establish causation.

6. Weight gain and metabolic dysfunction are substantial with many antipsychotics, particularly clozapine and olanzapine. Leucht et al. (2013) found standardized mean differences for weight gain versus placebo ranging from −0.09 for haloperidol to −0.74 for olanzapine. Clozapine, olanzapine, iloperidone, sertindole, and quetiapine produced significantly more weight gain than haloperidol, ziprasidone, lurasidone, aripiprazole, amisulpride, and asenapine. Network meta-analysis of 100 trials (25,952 patients) by Pillinger et al. (2020) found mean weight gain versus placebo ranged from −0.23 kg for haloperidol to 3.01 kg for clozapine over median 6-week trials. Glucose increases ranged from −0.29 mmol/L for lurasidone to 1.05 mmol/L for clozapine.

7. Tardive dyskinesia incidence is lower with second-generation antipsychotics but remains clinically significant. Meta-analysis of randomized controlled trials by Carbon et al. (2018) found annualized TD incidence of 6.5% (95% CI 5.3–7.8%) for first-generation antipsychotics across 32 treatment arms and 2.6% (2.0–3.1%) for second-generation antipsychotics across 86 treatment arms. Prevalence meta-analysis by Carbon et al. (2017) across 41 studies found global mean TD prevalence of 25.3% (22.7–28.1%): 30.0% with current first-generation treatment versus 20.7% with second-generation treatment.

8. The FDA requires a boxed warning on all antipsychotics for increased mortality risk in elderly patients with dementia-related psychosis. The 2005 warning for second-generation drugs was extended to first-generation drugs in 2008. Meta-analysis of 17 trials (5,106 patients) showed approximately 1.6- to 1.7-fold increase in mortality with second-generation antipsychotics versus placebo. No antipsychotics are FDA-approved for dementia-related psychosis.

What the evidence does not establish:

• That antipsychotics correct a dopamine deficiency (the inference from drug efficacy to pathophysiology is invalid)
• That second-generation antipsychotics are clearly superior to first-generation drugs in overall effectiveness (CATIE and CUtLASS found no clear advantage)
• That continuous long-term maintenance treatment is optimal for all patients (Wunderink and Harrow findings challenge this assumption, though confounding and selection effects complicate interpretation)
• That antipsychotics improve long-term functional outcomes beyond their acute symptomatic effects
• Which patients will respond to which drugs, or which patients can safely reduce or discontinue treatment

The structural problem that conditions all of this: Most efficacy trials last 6 weeks and exclude treatment-resistant, stable, and first-episode patients. Discontinuation rates in real-world effectiveness trials are 74% overall in CATIE (64–82% by assigned drug), driven by inefficacy or intolerable side effects. The Wunderink trial is the only randomized long-term follow-up showing superior recovery with dose reduction, and it applies only to first-episode patients who achieved remission. The Harrow observational data are confounded but consistent: patients who remain off medication after two years have better outcomes. The mechanistic inference—that antipsychotics correct a chemical imbalance—is an ex juvantibus fallacy: aspirin relieves headaches, but headaches are not caused by aspirin deficiency.

---

How to read this review

The single most important heuristic

An efficacy claim without a named trial duration, patient population, and outcome measure is not a clean estimate. Antipsychotics show SMD 0.33–0.88 versus placebo in 6-week acute trials of non-treatment-resistant patients. They show 74% discontinuation within 18 months in effectiveness trials. They show doubled 7-year recovery rates with dose reduction in first-episode patients. Any claim that "antipsychotics work" or "antipsychotics don't work" must specify which outcome, in which population, over what duration.

Three structural problems

Short trial duration and selective populations. The Leucht network meta-analysis used 6-week data (or 4–12 weeks if 6-week data unavailable). Trials excluded patients with treatment resistance, predominant negative symptoms, concomitant medical illness, and stable patients. First-episode patients were also largely excluded (only 9 of 212 trials). The efficacy estimates apply to acute treatment of non-treatment-resistant, unstable patients with positive symptoms.

Effectiveness-efficacy gap. Discontinuation rates in CATIE (74%) far exceed discontinuation rates in registration trials included in Leucht meta-analysis (35% overall). The gap reflects real-world tolerability, patient preference, and pragmatic dosing versus protocol-driven titration. Effect sizes measured in short efficacy trials do not predict real-world continuation or functional outcomes.

Confounding by indication in observational long-term data. Harrow's finding—that patients off medication after two years have better outcomes—is confounded: patients who discontinue may have milder illness, better premorbid functioning, lower vulnerability to psychosis, or greater resilience. The study controlled for multiple baseline indicators and tested confound-by-indication explicitly, but residual confounding cannot be ruled out. The finding challenges but does not refute maintenance treatment, because causation cannot be established from observational data.

Conventions

Claims are marked [contested] where the literature genuinely disagrees and [unverified] where a figure is widely repeated but could not be traced to a primary source in preparing this review.

Confidence intervals are omitted where they could not be verified against the primary text, rather than reconstructed.

---

Part I — Acute efficacy: what short-term trials actually show

1.1 The Leucht network meta-analysis: all drugs beat placebo

Leucht et al. (2013) conducted a Bayesian network meta-analysis of 212 randomized controlled trials (43,049 participants) comparing 15 antipsychotic drugs and placebo in acute treatment of schizophrenia. Trials were identified from the Cochrane Schizophrenia Group register, Medline, Embase, Cochrane Central Register of Controlled Trials, ClinicalTrials.gov, FDA reports, and pharmaceutical company data up to September 1, 2012.

Inclusion criteria: Blinded, randomized controlled trials of patients with schizophrenia or related disorders. Trials excluded if patients had predominant negative symptoms, concomitant medical illness, treatment resistance, or stable illness (relapse prevention studies excluded). Only 9 of 212 trials examined first-episode patients. Median treatment duration was 6 weeks (range 4–12 weeks where 6-week data unavailable).

Primary outcome: Mean overall change in symptoms, assessed by change in Positive and Negative Syndrome Scale total score from baseline to endpoint, or Brief Psychiatric Rating Scale if PANSS unavailable.

Main findings (efficacy versus placebo):

All drugs were significantly more effective than placebo. Standardized mean differences with 95% credible intervals:

• Clozapine: 0.88 (0.73 to 1.03)
• Amisulpride: 0.66 (0.53 to 0.78)
• Olanzapine: 0.59 (0.53 to 0.65)
• Risperidone: 0.56 (0.50 to 0.63)
• Paliperidone: 0.50 (0.39 to 0.60)
• Zotepine: 0.49 (0.31 to 0.66)
• Haloperidol: 0.45 (0.39 to 0.51)
• Quetiapine: 0.44 (0.35 to 0.52)
• Aripiprazole: 0.43 (0.34 to 0.52)
• Sertindole: 0.39 (0.26 to 0.52)
• Ziprasidone: 0.39 (0.30 to 0.49)
• Chlorpromazine: 0.38 (0.23 to 0.54)
• Asenapine: 0.38 (0.25 to 0.51)
• Lurasidone: 0.33 (0.21 to 0.45)
• Iloperidone: 0.33 (0.22 to 0.43)

Interpretation: Using Cohen's rule of thumb (SMD 0.2 small, 0.5 medium, 0.8 large), most drugs show small-to-medium effects versus placebo. Clozapine shows a large effect. Clozapine was significantly more effective than all other drugs. After clozapine, amisulpride, olanzapine, and risperidone were significantly more effective than most other drugs apart from paliperidone and zotepine. The differences between drugs are small but robust.

All-cause discontinuation: All drugs were significantly better than placebo apart from zotepine. Odds ratios ranged from 0.43 (amisulpride, best) to 0.80 (haloperidol, worst). Lower ORs indicate fewer discontinuations versus placebo.

1.2 Side-effect profiles: substantial differences between drugs

Weight gain (vs placebo): Standardized mean differences ranged from −0.09 (haloperidol, least weight gain) to −0.74 (olanzapine, most weight gain). Clozapine, olanzapine, iloperidone, sertindole, quetiapine, risperidone, and paliperidone produced significantly more weight gain than haloperidol, ziprasidone, lurasidone, aripiprazole, amisulpride, and asenapine.

Extrapyramidal side-effects (measured by antiparkinson drug use): Odds ratios versus placebo ranged from 0.30 (clozapine, fewest EPS) to 4.76 (haloperidol, most EPS). Clozapine, sertindole, olanzapine, quetiapine, aripiprazole, iloperidone, amisulpride, and asenapine did not cause significantly more EPS than placebo. Haloperidol caused significantly more EPS than all other drugs.

Sedation: Odds ratios versus placebo ranged from 1.42 (amisulpride, least sedating) to 8.82 (clozapine, most sedating). Clozapine, zotepine, and chlorpromazine were most sedating.

Translation: Antipsychotics differ substantially in side-effect profiles. The simple first-generation versus second-generation classification is not supported: haloperidol causes more EPS than all second-generation drugs, but also causes less weight gain than most. Perphenazine (first-generation) performed comparably to second-generation drugs in CATIE. The hierarchies in different domains should guide individualized selection rather than class-based prescribing.

1.3 Study quality and limitations

Risk of bias: Of 212 trials, reports often did not provide details about randomization procedures and allocation concealment. However, 144 studies (68%) were done by pharmaceutical companies, which (in cases where they responded to information requests) had used appropriate methods. 13 studies were single-blinded; 199 (94%) were double-blinded. Mean duration of illness was 12.4 years (SD 6.6); mean age 38.4 years (SD 6.9).

Sensitivity analyses: Efficacy outcomes did not change substantially after removal of placebo or haloperidol groups, or when dose, percentage of withdrawals, extent of blinding, pharmaceutical industry sponsorship, study duration, chronicity, and year of publication were accounted for in meta-regressions.

What these trials exclude: Treatment-resistant patients, patients with predominant negative symptoms, patients with concomitant medical illness, stable patients, and (mostly) first-episode patients. The efficacy estimates apply to acute treatment of non-treatment-resistant, unstable patients with positive symptoms.

---

Part II — Effectiveness trials: discontinuation and real-world outcomes

2.1 CATIE: 74% discontinuation within 18 months

Lieberman et al. (2005) conducted the Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) study, a Phase 1 effectiveness trial of antipsychotic drugs in 1,493 patients with chronic schizophrenia at 57 US sites (16 university clinics, 10 state mental health agencies, 7 VA medical centers, 6 private nonprofit agencies, 4 private-practice sites, 14 mixed-system sites) from January 2001 to December 2004.

Design: Patients aged 18–65 years with schizophrenia (DSM-IV, Structured Clinical Interview) were randomly assigned to olanzapine (7.5–30 mg/day), perphenazine (8–32 mg/day), quetiapine (200–800 mg/day), or risperidone (1.5–6.0 mg/day) under double-blind conditions and followed for up to 18 months or until treatment was discontinued for any reason. Ziprasidone (40–160 mg/day) was added in January 2002 after FDA approval. Patients excluded if they had schizoaffective disorder, mental retardation, other cognitive disorders, history of serious adverse reactions to proposed treatments, only one schizophrenic episode, history of treatment resistance (persistence of severe symptoms despite adequate trials or prior clozapine), were pregnant/breastfeeding, or had serious unstable medical conditions.

Primary aim: Time to discontinuation for any cause.

Results: Of 1,432 patients who received at least one dose, 1,061 (74%) discontinued study medication before 18 months:

• Olanzapine: 64% (median time to discontinuation longest)
• Perphenazine: 75%
• Quetiapine: 82%
• Risperidone: 74%
• Ziprasidone: 79%

Time to discontinuation was significantly longer in the olanzapine group than in the quetiapine (p < 0.001) or risperidone (p = 0.002) group, but not significantly different from perphenazine (p = 0.021) or ziprasidone (p = 0.028). After adjustment for multiple comparisons, significant differences remained only between olanzapine and quetiapine/risperidone.

Reasons for discontinuation: Times to discontinuation because of intolerable side effects were similar among groups, but rates differed (p = 0.04). Olanzapine was associated with greater weight gain and increases in measures of glucose and lipid metabolism compared with the other drugs.

Translation: In a real-world effectiveness trial, three-quarters of patients discontinued assigned treatment within 18 months. Olanzapine showed the longest time to discontinuation, suggesting greater effectiveness or tolerability, but at the cost of significant metabolic changes. Perphenazine, a first-generation drug, performed comparably to second-generation drugs.

2.2 CUtLASS: no quality-of-life advantage for second-generation drugs

Jones et al. (2006) conducted the Cost Utility of the Latest Antipsychotic Drugs in Schizophrenia Study (CUtLASS), a noncommercially funded, pragmatic, multisite, randomized controlled trial at 14 community psychiatric services in the English NHS.

Design: 227 patients aged 18–65 years with DSM-IV schizophrenia and related disorders assessed for medication review because of inadequate response or adverse effects were randomized to first-generation antipsychotics or second-generation antipsychotics (other than clozapine), with the choice of individual drug made by the managing psychiatrist. Blind assessments at 12, 26, and 56 weeks using intention-to-treat analysis.

Primary hypothesis: Use of second-generation antipsychotics would be associated with improved quality of life across 1 year compared with first-generation antipsychotics.

Results: The pre-specified clinically important difference was a 5-point Quality of Life Scale advantage for second-generation drugs; this hypothesis was excluded (ruled out at 95% confidence level). Observed data trended toward first-generation drugs performing better (p = 0.24, not statistically significant). Participants reported no clear preference for either drug group. Mean costs for 52 weeks were $34,750 (£18,800) for the first-generation arm; costs were similar between groups and not statistically different.

Translation: In people with schizophrenia whose medication is changed for clinical reasons, there is no disadvantage across 1 year in terms of quality of life, symptoms, or associated costs of care in using first-generation drugs rather than non-clozapine second-generation drugs. Neither inadequate power nor patterns of drug discontinuation accounted for the result.

---

Part III — Dose reduction and long-term recovery: the Wunderink trial

3.1 Superior recovery with dose reduction in first-episode patients

Wunderink et al. (2013) conducted a 7-year follow-up of a 2-year randomized trial comparing maintenance treatment and dose reduction/discontinuation in patients with remitted first-episode psychosis.

Original trial (first 18 months): 128 patients with first-episode psychosis who achieved remission for 6 months were randomized to maintenance treatment (MT) or dose reduction/discontinuation (DR) strategy for 18 months. Dose reduction consisted of gradual symptom-guided tapering and discontinuation if feasible. Maintenance treatment followed American Psychiatric Association guidelines, preferably using low-dose atypical antipsychotics. After the 18-month trial, treatment was at the discretion of the clinician.

7-year follow-up: 103 patients (80.5%) of the original 128 were located and consented to follow-up assessment. Mean age at follow-up was approximately 27 years.

Primary outcome: Recovery, defined as meeting criteria for both symptomatic remission (PANSS positive subscale mean score ≤ 3 and no single item > 4 for ≥6 months) and functional remission (employed or in school for ≥6 months, or managing a household independently, and having at least one friend with regular contact).

Main findings:

• Recovery rate: 40.4% (21/52) in DR group versus 17.6% (9/51) in MT group (OR 3.49, 95% CI not calculable from text but p = 0.01)
• Symptomatic remission: 69.2% in DR group versus 66.7% in MT group (not significantly different)
• Functional remission: 46.2% in DR group versus 19.6% in MT group (significantly higher with DR)

Symptomatic remission without functional remission: 28.8% in DR group versus 49.0% in MT group. Functional remission without symptomatic remission: 5.8% in DR versus 2.0% in MT. Neither symptomatic nor functional remission: 28.2% overall.

Interpretation: Dose reduction/discontinuation during early stages of remitted first-episode psychosis shows superior long-term recovery rates compared to maintenance treatment. Better recovery rates in the DR group were driven by higher functional remission rates, not symptomatic remission rates. The DR group experienced higher relapse rates during the initial 18-month randomized phase (43% vs 21%, from Wunderink et al. 2007), but this did not translate into worse long-term outcomes.

3.2 Implications and limitations

Implications: The finding challenges standard-of-care guidelines that recommend maintenance treatment for at least 1–2 years after first-episode remission. It suggests that dose reduction or discontinuation may be a viable strategy for some first-episode patients who achieve stable remission, and that preventing relapse in the short term may come at the cost of functional recovery in the long term.

Limitations:

1. The trial was open-label (not double-blind), though outcomes were assessed by blinded raters.
2. The 7-year follow-up was naturalistic after the 18-month randomized phase. Treatment was at clinician discretion, and medication use changed in both groups: some DR patients restarted medication, and some MT patients reduced or stopped.
3. The sample size was moderate (103 patients at 7-year follow-up).
4. The trial applies only to first-episode patients who achieved 6 months of remission. It does not apply to treatment-resistant patients, patients with multiple prior episodes, or patients who did not achieve remission.
5. The mechanism is unclear: does relapse and recovery from relapse provide a learning experience that promotes functional recovery? Does avoiding medication allow better social integration, education, or employment? Or do patients in the DR group have intrinsically better prognoses?

What this trial establishes: For first-episode patients who achieve remission, dose reduction/discontinuation shows superior 7-year recovery rates compared to maintenance treatment. The finding is robust and internally consistent. It does not establish that all first-episode patients should discontinue medication, because 60% of the DR group did not achieve recovery, and some patients relapsed during dose reduction.

---

Part IV — Long-term observational data: the confounding problem

4.1 The Harrow 20-year follow-up: better outcomes off medication

Harrow, Jobe, and Tong (2021) conducted a 20-year naturalistic prospective longitudinal study (the Chicago Follow-up Study) investigating course, outcome, and effects of antipsychotic medication on recovery and rehospitalization in 139 participants with serious mental illness (70 with schizophrenia, 69 with affective psychosis). Participants were assessed at index hospitalization and six follow-ups over 20 years (734 observations).

Design: Prospective, naturalistic, observational study. Medication use was determined by patient and clinician, not randomized. The study assessed premorbid prognostic indicators at index hospitalization to control for confounding by indication.

Key finding: [contested] After the second year, absence of antipsychotics predicted higher probability of recovery and lower probability of rehospitalization at subsequent follow-ups after adjusting for confounders, regardless of diagnosis. Adjusted odds ratio for recovery (not on antipsychotics vs on antipsychotics): 5.989 (95% CI 3.588–9.993). Adjusted odds ratio for rehospitalization: 0.134 (0.070–0.259), meaning patients off medication had 87% lower odds of rehospitalization.

Confounding control: The study assessed multiple baseline prognostic indicators (premorbid functioning, neurocognition, personality factors, symptom severity, diagnostic category) and used generalized estimating equation logistic models to adjust for confounders measured at index and follow-ups. The design attempted to minimize confound by indication by distinguishing outcomes in good versus poor prognosis groups for those prescribed and not prescribed antipsychotics.

Interpretation: Participants not on antipsychotics after the first 2 years showed better outcomes than participants prescribed antipsychotics. This finding remained significant after controlling for multiple baseline and follow-up confounders. However, causation cannot be established from observational data. The paper separately notes that participants with schizophrenia off antipsychotics after two years had better outcomes, but the adjusted ORs above apply to the pooled sample regardless of diagnosis.

4.2 The confounding-by-indication problem

Confounding by indication occurs when the reason for treatment is associated with the outcome. In this case, clinicians prescribe antipsychotics to patients with more severe illness, more frequent relapses, and poorer prognoses. Patients who discontinue medication may have:

• Milder illness
• Better premorbid functioning
• Less vulnerability to psychosis
• Greater neurocognitive skills
• Better social support
• Greater resilience and capacity for recovery

Even after controlling for measured prognostic indicators, unmeasured factors may confound the association. The Harrow study controlled for many baseline indicators, but residual confounding cannot be ruled out.

The authors' position: Even when confound by indication is controlled for, participants with schizophrenia and affective psychosis do better than their medicated cohorts. They argue that the pattern is consistent across multiple follow-ups and persists after adjustment, and that their design (prospective, with extensive baseline assessment) strengthens the case against pure selection bias.

Critics' position: Patients who remain off medication for prolonged periods are a self-selected group with better internal resources, and no amount of statistical adjustment can fully account for selection. Randomized discontinuation trials in chronic populations are needed to establish causation, but such trials raise ethical concerns.

What this finding establishes: Long-term observational data show an association between being off antipsychotics and better outcomes, but causation is not established. The finding challenges the assumption that continuous long-term maintenance is optimal for all patients, but it does not prove that discontinuation causes better outcomes.

4.3 How the Wunderink and Harrow findings relate

Wunderink is a randomized trial in first-episode patients who achieved remission, showing that dose reduction/discontinuation leads to better 7-year recovery rates. It is internally valid but limited to a specific population. Causation can be inferred within the studied population.

Harrow is an observational study in a mixed population (schizophrenia and affective psychosis) showing that patients off medication after two years have better outcomes. It is confounded but consistent across 20 years and multiple follow-ups. Causation cannot be established.

The two studies converge on the same implication: continuous long-term maintenance treatment may not be optimal for all patients, and some patients may achieve better long-term outcomes with reduced or discontinued medication. But neither study establishes which patients can safely reduce or discontinue, and neither refutes the value of medication for patients with severe, recurrent, or treatment-resistant illness.

---

Part V — Harms: weight gain, metabolic syndrome, tardive dyskinesia, and mortality in dementia

5.1 Weight gain and metabolic effects

Leucht et al. (2013) short-term data (6-week median): Standardized mean differences for weight gain versus placebo ranged from −0.09 (haloperidol, least) to −0.74 (olanzapine, most). Clozapine, olanzapine, iloperidone, sertindole, and quetiapine produced significantly more weight gain than haloperidol, ziprasidone, lurasidone, aripiprazole, amisulpride, and asenapine.

Pillinger et al. (2020) metabolic network meta-analysis: Systematic review and network meta-analysis of 100 randomized controlled trials (25,952 patients, median treatment duration 6 weeks) compared 18 antipsychotics on metabolic function.

Mean differences for weight gain versus placebo:

• Haloperidol: −0.23 kg (95% CI −0.83 to 0.36)
• Clozapine: 3.01 kg (1.78 to 4.24)
• Olanzapine: highest BMI increase (1.07 kg/m², 0.90 to 1.25)

Mean differences for glucose increase versus placebo:

• Lurasidone: −0.29 mmol/L (−0.55 to −0.03), only drug showing glucose reduction
• Clozapine: 1.05 mmol/L (0.41 to 1.70), highest glucose increase
• Haloperidol, aripiprazole: minimal glucose effects

Total cholesterol increase versus placebo:

• Cariprazine: −0.09 mmol/L (−0.24 to 0.07), best lipid profile
• Clozapine: 0.56 mmol/L (0.26 to 0.86), worst lipid profile

Predictors of metabolic dysregulation: Greater increases in glucose were predicted by higher baseline weight (p = 0.0015) and male sex (p = 0.0082). Non-white ethnicity was associated with greater increases in total cholesterol (p = 0.040).

Association between symptom improvement and metabolic change: Improvements in symptom severity were associated with increases in weight (r = 0.36, p = 0.0021), BMI (r = 0.84, p < 0.0001), total cholesterol (r = 0.31, p = 0.047), and LDL cholesterol (r = 0.42, p = 0.013), and decreases in HDL cholesterol (r = −0.35, p = 0.035).

Translation: Metabolic effects are substantial and clinically significant for clozapine, olanzapine, and other high-weight-gain drugs. The association between symptom improvement and metabolic worsening complicates the risk-benefit trade-off: patients who respond best may also experience the greatest metabolic harms.

5.2 Tardive dyskinesia: incidence and prevalence

Incidence in randomized controlled trials (Carbon et al., 2018): Meta-analysis of treatment-emergent TD in comparative RCTs.

• First-generation antipsychotics: Weighted mean incidence 6.5% (95% CI 4.6–9.0%) across 32 treatment arms (3,763 subjects). Annualized incidence 6.5% (5.3–7.8%).
• Second-generation antipsychotics: Weighted mean incidence 3.0% (2.4–3.8%) across 86 treatment arms (15,092 subjects). Annualized incidence 2.6% (2.0–3.1%).

Prevalence in cross-sectional studies (Carbon et al., 2017): Meta-analysis of 41 studies (11,493 participants, mean age 42.8 years, 66.4% male, 77.1% schizophrenia-spectrum disorders).

• Global mean TD prevalence: 25.3% (95% CI 22.7–28.1%)
• Current second-generation treatment: 20.7% (16.6–25.4%, N = 5,103)
• Current first-generation treatment: 30.0% (26.4–33.8%, N = 5,062)
• Difference: Significant (Q = 9.17, p = .002)

The difference remained significant after controlling for age, region, illness duration, and frequency of parkinsonism. Particularly low TD prevalence (7.2%) was found in treatment arms with first-generation-naive subjects relative to second-generation-treated cohorts with likely prior first-generation exposure (23.4%, p < .001).

Translation: Tardive dyskinesia remains a clinically significant risk with all antipsychotics. Second-generation drugs have lower incidence and prevalence than first-generation drugs, but the absolute rates are not trivial: annualized incidence 2.6% (constant-rate illustration would suggest approximately one in four over 10 years, though real incidence likely varies over time). Prevalence of 20–30% in long-term users is substantial.

5.3 Mortality in elderly patients with dementia: FDA boxed warning

FDA warning history: In 2005, the FDA required a boxed warning on all second-generation antipsychotics for increased risk of death in elderly patients with dementia-related psychosis. In 2008, the warning was extended to first-generation antipsychotics.

Evidence basis (2005 second-generation warning): FDA review of placebo-controlled trials involving four second-generation drugs (olanzapine, aripiprazole, risperidone, quetiapine) in elderly patients with dementia-related behavioral problems. Of 17 trials (5,106 patients), 15 showed numerical increases in death rate. Meta-analysis showed approximately 1.6- to 1.7-fold increase in mortality. Death rates were approximately 4.5% in drug-treated patients versus approximately 2.6% in placebo-treated patients over trials averaging 10 weeks duration.

Extension to first-generation drugs (2008): Two large observational studies in Canada found mortality rates in elderly patients taking first-generation drugs comparable to or higher than rates in patients taking second-generation drugs. FDA concluded that "the overall weight of evidence indicates that the conventional antipsychotics share the increased risk of death in elderly patients with dementia-related psychosis that has been observed for the atypical antipsychotics."

Current label: All antipsychotics carry a boxed warning: "Elderly patients with dementia-related psychosis treated with antipsychotic drugs are at an increased risk of death." Antipsychotics are not approved for the treatment of dementia-related psychosis.

Translation: The mortality risk is real, substantial, and applies to both first-generation and second-generation drugs. The increased risk is primarily attributed to cardiovascular events and infections. No antipsychotic is FDA-approved for dementia-related psychosis, but off-label use remains common.

---

Part VI — Mechanism: the dopamine hypothesis and the ex juvantibus fallacy

6.1 The dopamine hypothesis, version III

Howes and Kapur (2009) synthesized evidence from neurochemical imaging studies, genetics, environmental risk factors, extended phenotype research, and animal studies into a revised dopamine hypothesis of schizophrenia—version III: the final common pathway.

The hypothesis: Various genetic and environmental risk factors (pregnancy and obstetric complications, stress and trauma, drug use, genes) converge on a final common pathway of increased presynaptic striatal dopaminergic function. This leads to aberrant salience and psychosis. Antipsychotic drugs act downstream of this primary abnormality by blocking postsynaptic D2 receptors.

Implications:

1. Current antipsychotics are not treating the primary abnormality. They block the effect of inappropriate dopamine release but do not correct the upstream dysregulation.
2. Antipsychotics may paradoxically worsen the primary abnormality by blocking presynaptic D2 autoreceptors, resulting in compensatory increase in dopamine synthesis. This may explain why some patients relapse rapidly on stopping medication and why symptoms may worsen over time despite treatment.
3. Future drug development should focus on upstream factors that converge on the dopaminergic funnel point, rather than on D2 blockade.

6.2 The ex juvantibus fallacy

The inference that antipsychotics' efficacy validates the dopamine hypothesis—that psychosis is caused by dopamine excess—is an ex juvantibus fallacy. As Lacasse and Leo (2005) noted in the context of antidepressants and serotonin: "The fact that aspirin cures headaches does not prove that headaches are due to low levels of aspirin in the brain."

Antipsychotics block D2 receptors and reduce psychotic symptoms. This tells us that D2 blockade reduces symptoms; it does not tell us that psychosis is caused by dopamine excess. The dopamine hypothesis version III explicitly states that antipsychotics act downstream of the primary abnormality, which is presynaptic dysregulation, not postsynaptic receptor excess.

Translation: Antipsychotics work by blocking dopamine receptors, but this does not prove that psychosis is caused by a dopamine deficiency or excess that the drugs "correct." The chemical-imbalance story applied to antipsychotics is as unsupported as the serotonin-deficiency story for depression.

6.3 D2 receptor occupancy and the therapeutic window

Antipsychotic efficacy requires approximately 65% D2 receptor occupancy for response, while extrapyramidal side effects increase sharply above approximately 78% occupancy (Kapur et al., 2000). Below 65%, positive symptoms are inadequately controlled. The "therapeutic window" is narrow and varies across individuals.

Dopamine partial agonists (aripiprazole, brexpiprazole, cariprazine) have weak intrinsic dopaminergic activity. They have lower risk of EPS than full antagonists because they weakly stimulate D2 receptors rather than fully blocking them.

Translation: Antipsychotic dosing is a balancing act: enough D2 blockade to control symptoms, but not so much that EPS becomes intolerable. The therapeutic window is individual-specific and discovered by trial and error, not predicted from baseline characteristics.

---

Part VII — What the evidence does not establish

7.1 That second-generation drugs are clearly superior to first-generation drugs

CATIE found no clear advantage for second-generation drugs over perphenazine (first-generation) in time to discontinuation, symptoms, or quality of life. CUtLASS found no quality-of-life advantage for second-generation drugs versus first-generation drugs in patients requiring medication change. The Leucht network meta-analysis found that haloperidol (first-generation) showed intermediate efficacy, comparable to quetiapine and aripiprazole (both second-generation).

Second-generation drugs have lower rates of extrapyramidal side effects and tardive dyskinesia, but higher rates of weight gain and metabolic effects (with exceptions: aripiprazole, ziprasidone, and lurasidone show more benign metabolic profiles than clozapine or olanzapine). The trade-off is not clearly in favor of one class over the other; it depends on individual patient factors and side-effect profiles.

7.2 That continuous long-term maintenance is optimal for all patients

The Wunderink trial showed that dose reduction/discontinuation in first-episode patients with remission led to superior 7-year recovery rates. The Harrow observational data showed better outcomes for patients off medication after two years, though confounding cannot be ruled out. These findings challenge the standard-of-care assumption that all patients should remain on antipsychotics indefinitely.

Current guidelines recommend maintenance treatment for at least 1–2 years after first episode, and longer (or indefinitely) for patients with multiple episodes. But the Wunderink and Harrow findings suggest that for some patients—particularly those with good premorbid functioning, stable remission, and low vulnerability to relapse—dose reduction or discontinuation may lead to better long-term functional outcomes.

The critical unanswered question: which patients can safely reduce or discontinue? No validated algorithm exists. Clinicians must weigh acute relapse risk against long-term functional recovery, and this trade-off cannot be resolved with existing evidence.

7.3 That antipsychotics improve long-term functional outcomes

The Wunderink trial showed that dose reduction led to better functional outcomes than maintenance treatment. The Harrow data showed no functional benefit of continuous medication. The MTA study in ADHD found that 14-month treatment advantage disappeared by 36 months and original treatment assignment did not predict 8-year outcomes (Molina et al., 2009).

Antipsychotics reduce acute symptoms. They do not appear to improve long-term functional outcomes (employment, social integration, independent living) beyond their acute symptomatic effects. This does not mean antipsychotics are ineffective—symptom control is valuable—but it means the inference that controlling symptoms in the short term translates into better functioning in the long term is not supported.

---

Part VIII — The two overclaims, and what replaces them

8.1 The standard-of-care overclaim: "Antipsychotics correct brain chemistry and should be continued indefinitely"

The claim: Antipsychotics correct a dopamine imbalance in the brain, analogous to insulin correcting glucose imbalance in diabetes. They should be continued indefinitely to prevent relapse, and discontinuation is dangerous.

What the evidence shows: Antipsychotics block D2 receptors and reduce acute psychotic symptoms. They do not correct an upstream dopamine dysregulation; they act downstream of the primary abnormality. The inference from efficacy to pathophysiology is an ex juvantibus fallacy.

The Wunderink trial showed that dose reduction/discontinuation in first-episode patients with remission led to superior 7-year recovery rates. The Harrow observational data showed better outcomes for patients off medication after two years. These findings challenge—but do not refute—the value of continuous maintenance for all patients.

What survives: Antipsychotics are effective for acute symptom control. For patients with severe, recurrent, or treatment-resistant illness, continuous maintenance reduces relapse risk. But the assumption that all patients should remain on medication indefinitely is not supported. Some patients—particularly first-episode patients with stable remission—may achieve better long-term functional outcomes with dose reduction or discontinuation.

8.2 The critical-psychiatry overclaim: "Antipsychotics barely work and only cause harm"

The claim: Antipsychotics provide minimal benefit over placebo, are primarily harmful, worsen long-term outcomes, and should be avoided or minimized in all cases.

What the evidence shows: All 15 antipsychotics in the Leucht network meta-analysis were significantly more effective than placebo, with effect sizes ranging from SMD 0.33 to 0.88. Clozapine showed a large effect; most drugs showed small-to-medium effects. These effects are real, not placebo effects.

Discontinuation rates are high (74% in CATIE), driven by inefficacy or intolerable side effects. But 26% of CATIE participants remained on medication at 18 months, and for many patients, symptom control is clinically meaningful and improves quality of life.

Harms (weight gain, metabolic syndrome, tardive dyskinesia, mortality in dementia) are real and substantial, but not universal. Some patients tolerate antipsychotics well; others do not. The trade-off is individual-specific.

What survives: Antipsychotics work modestly in the short term and have substantial side effects. Long-term effectiveness is limited, and functional outcomes are not clearly improved. But this does not mean antipsychotics "barely work" or should be avoided in all cases. For patients with acute psychosis, severe symptoms, or high relapse risk, the benefits may outweigh the harms.

8.3 The version that survives

Schizophrenia is a heterogeneous disorder with variable course and outcome. Some patients experience a single episode and recover; others have recurrent episodes with periods of remission; others have persistent symptoms despite treatment. No single treatment strategy applies to all patients.

Antipsychotics reduce acute symptoms by blocking D2 receptors, with effect sizes in the small-to-moderate range. They do not correct an upstream dopamine dysregulation; they act downstream. The chemical-imbalance story is not supported.

Discontinuation rates in effectiveness trials are high (74%), driven by inefficacy or intolerable side effects. First-generation and second-generation drugs do not differ clearly in overall effectiveness; they differ in side-effect profiles. Haloperidol causes more EPS; olanzapine causes more weight gain. The choice should be individualized, not class-based.

Dose reduction/discontinuation in first-episode patients with stable remission shows superior 7-year recovery rates compared to maintenance treatment (Wunderink). Long-term observational data show better outcomes for patients off medication after two years, but this is confounded by selection (Harrow). Both findings challenge the assumption that continuous maintenance is optimal for all patients.

Weight gain, metabolic syndrome, tardive dyskinesia, and (in elderly dementia patients) increased mortality are substantial harms. The trade-off between symptom control and side effects is individual-specific and cannot be resolved with population-level data.

What the evidence establishes: Antipsychotics reduce acute symptoms versus placebo (SMD 0.33–0.88 over 6-week trials). Discontinuation rates in effectiveness trials are 74% overall in CATIE (64–82% by assigned drug). Dose reduction in first-episode patients with remission leads to better 7-year recovery. Long-term observational data show better outcomes off medication, but causation is not established. Weight gain, metabolic dysfunction, TD, and mortality in dementia are real and substantial. The dopamine-deficiency hypothesis is not supported.

What it does not establish: That antipsychotics correct a chemical imbalance, that second-generation drugs are clearly superior to first-generation drugs, that continuous maintenance is optimal for all patients, that antipsychotics improve long-term functional outcomes, or which patients can safely reduce or discontinue medication.

The structural problem: Efficacy trials are short (6 weeks), exclude treatment-resistant and first-episode patients, and measure symptom reduction, not functional outcomes. Effectiveness trials show high discontinuation (74%) and no clear advantage for second-generation drugs. The Wunderink trial is the only randomized long-term follow-up, and it shows better outcomes with dose reduction in a specific population. The Harrow observational data are confounded but consistent. The honest summary is smaller and more qualified than either the standard-of-care or the critical-psychiatry narrative.

---

References

Primary sources cited

Carbon, M., Hsieh, C.-H., Kane, J. M., & Correll, C. U. (2017). "Tardive dyskinesia prevalence in the period of second-generation antipsychotic use: a meta-analysis." The Journal of Clinical Psychiatry, 78(3), e264–e278.

Carbon, M., Kane, J. M., Leucht, S., & Correll, C. U. (2018). "Tardive dyskinesia risk with first- and second-generation antipsychotics in comparative randomized controlled trials: a meta-analysis." World Psychiatry, 17(3), 330–340.

Harrow, M., Jobe, T. H., & Tong, L. (2021). "Twenty-year effects of antipsychotics in schizophrenia and affective psychotic disorders." Psychological Medicine. doi:10.1017/S0033291720004778

Howes, O. D., & Kapur, S. (2009). "The dopamine hypothesis of schizophrenia: version III—the final common pathway." Schizophrenia Bulletin, 35(3), 549–562.

Jones, P. B. et al. (2006). "Randomized controlled trial of the effect on quality of life of second- vs first-generation antipsychotic drugs in schizophrenia: Cost Utility of the Latest Antipsychotic Drugs in Schizophrenia Study (CUtLASS 1)." Archives of General Psychiatry, 63(10), 1079–1087.

Kapur, S. et al. (2000). "Relationship between dopamine D2 occupancy, clinical response, and side effects: a double-blind PET study of first-episode schizophrenia." American Journal of Psychiatry, 157(4), 514–520.

Lacasse, J. R., & Leo, J. (2005). "Serotonin and depression: a disconnect between the advertisements and the scientific literature." PLoS Medicine, 2(12), e392.

Leucht, S. et al. (2013). "Comparative efficacy and tolerability of 15 antipsychotic drugs in schizophrenia: a multiple-treatments meta-analysis." The Lancet, 382(9896), 951–962.

Lieberman, J. A. et al. (2005). "Effectiveness of antipsychotic drugs in patients with chronic schizophrenia." New England Journal of Medicine, 353(12), 1209–1223.

Molina, B. S. G. et al. (2009). "The MTA at 8 years: prospective follow-up of children treated for combined-type ADHD in a multisite study." Journal of the American Academy of Child & Adolescent Psychiatry, 48(5), 484–500.

Pillinger, T. et al. (2020). "Comparative effects of 18 antipsychotics on metabolic function in patients with schizophrenia, predictors of metabolic dysregulation, and association with psychopathology: a systematic review and network meta-analysis." The Lancet Psychiatry, 7(1), 64–77.

US Food and Drug Administration. (2005, 2008). Boxed warnings on antipsychotics for increased mortality risk in elderly patients with dementia-related psychosis.

Wunderink, L. et al. (2007). "Guided discontinuation versus maintenance treatment in remitted first-episode psychosis: relapse rates and functional outcome." The Journal of Clinical Psychiatry, 68(5), 654–661.

Wunderink, L. et al. (2013). "Recovery in remitted first-episode psychosis at 7 years of follow-up of an early dose reduction/discontinuation or maintenance treatment strategy: long-term follow-up of a 2-year randomized clinical trial." JAMA Psychiatry, 70(9), 913–920.

---

Appendix: Sources opened and figures verified

Leucht et al. (2013): Effect sizes versus placebo (SMD 0.88 to 0.33), all-cause discontinuation ORs (0.43 to 0.80), weight gain SMD (−0.09 to −0.74), EPS ORs (0.30 to 4.76), sedation ORs (1.42 to 8.82). All figures verified from Table 2 (page 954) and Figures 2–5 (pages 953–956).

Lieberman et al. (2005) CATIE: 74% discontinuation (1,061 of 1,432), breakdown by drug (olanzapine 64%, perphenazine 75%, quetiapine 82%, risperidone 74%, ziprasidone 79%). Verified from Results section.

Jones et al. (2006) CUtLASS: Primary hypothesis excluded, first-generation arm trended better (not statistically significant, p = 0.24), costs similar. Verified from Results and Table 3.

Wunderink et al. (2007): Relapse rates 43% vs 21% during 18-month randomized phase. Verified from Results section.

Wunderink et al. (2013): Recovery rates 40.4% vs 17.6% (OR 3.49, p = .01), symptomatic remission 69.2% vs 66.7%, functional remission 46.2% vs 19.6%. Verified from Results section and Table 2.

Harrow et al. (2021): Adjusted OR for recovery 5.989 (95% CI 3.588–9.993), adjusted OR for rehospitalization 0.134 (0.070–0.259). Verified from Results section.

Pillinger et al. (2020): Weight gain −0.23 kg (haloperidol) to 3.01 kg (clozapine), glucose −0.29 mmol/L (lurasidone) to 1.05 mmol/L (clozapine). Verified from Findings section.

Carbon et al. (2017, 2018): Annualized TD incidence 6.5% (FGA) vs 2.6% (SGA) from Carbon et al. (2018) World Psychiatry 17(3):330–340. TD prevalence 30.0% (FGA) vs 20.7% (SGA), global prevalence 25.3% from Carbon et al. (2017) J Clin Psychiatry 78(3):e264–e278. Verified from Results sections of both papers.

FDA boxed warning (2005, 2008): 1.6- to 1.7-fold mortality increase, death rates approximately 4.5% vs approximately 2.6%, 17 trials, 5,106 patients, 15 of 17 showed numerical increases, trials averaging 10 weeks. Verified from FDA documents and summaries.

All other claims in the executive summary and body are traceable to these opened sources or marked [unverified] if widely repeated but not opened.