Abstract

Study at a Glance

Study Snapshot: Key Statistics

Study Facts

What Researchers Actually Did

Trott and colleagues conducted a preregistered systematic review and meta-analysis, searching PubMed, EMBASE, PsycINFO, Scopus, and CENTRAL on 6 December 2025. Their inclusion criteria were narrow and explicit: placebo-controlled RCTs of semaglutide or tirzepatide in adults with a formal SSD diagnosis, with no restriction on dosage, duration, or outcome. Two independent reviewers handled screening, full-text assessment, and data extraction at every stage; agreement was 100% throughout, and no arbitration was required. Risk of bias was assessed with the Cochrane RoB2 tool. Three trials met criteria, all examining semaglutide: Ganeshalingam et al. (n = 154, 1 mg, 30 weeks, Denmark), Sass et al. (n = 73, 1 mg, 26 weeks, Denmark), and Siskind et al. (n = 31, 2 mg, 36 weeks, Australia). No tirzepatide trials were identified.

Statistical pooling used a random-effects framework with restricted maximum likelihood estimation; heterogeneity was quantified with τ², and prediction intervals were calculated alongside point estimates. Continuous outcomes required at least three contributing studies to enter the meta-analysis; fewer than three were narratively synthesised. Adverse events were pooled as risk ratios after log-transformation, using the same random-effects method. Evidence certainty was graded with GRADE. Publication bias was not formally assessed because the pool contained fewer than ten studies.

Key Findings: Primary Outcomes

• Body weight: Semaglutide reduced weight by a mean of 11.32 kg (95% CI −15.35 to −7.29 kg; p <0.001; τ² = 10.15; prediction interval −18.75 to −3.89 kg; GRADE high certainty).
• BMI: Mean reduction of 3.58 kg/m² (95% CI −4.86 to −2.30; p <0.001; τ² = 1.00; prediction interval −5.93 to −1.24 kg/m²; GRADE high certainty).
• HbA1c: Mean reduction of 0.37% (95% CI −0.51 to −0.22; p <0.001; τ² = 0.01; prediction interval −0.62 to −0.11%; GRADE high certainty).
• Fasting glucose: Mean reduction of 0.54 mmol/L (95% CI −0.94 to −0.13; p = 0.01; τ² = 0.09; prediction interval −1.26 to 0.19 mmol/L; GRADE moderate certainty).
• The authors note that the 11% reduction in body weight was achieved at semaglutide doses of 1.0 to 2.0 mg over 26 to 36 weeks, compared with the 2.4 mg dose used over 68 weeks in the Wilding et al. STEP 1 general-population trial, which yielded approximately 15% weight loss.

Key Findings: Secondary Outcomes

No outcomes other than those listed above reached statistical significance in the pooled meta-analysis. Several narratively synthesised outcomes showed directionally consistent but non-significant patterns:

• Body composition: Significant reductions in body fat percentage, waist circumference, and visceral fat were reported across the two trials measuring these outcomes; hip circumference decreased significantly in Ganeshalingam et al. Waist-to-hip ratio and bone mineral content showed no significant change in any trial.
• Insulin resistance: Two studies reported consistent decreases, but neither reached statistical significance.
• Lipids: Consistent non-significant reductions in triglycerides and heart rate; non-significant and conflicting effects for LDL and total cholesterol.
• Blood pressure: Systolic BP: −2.34 mmHg (95% CI −5.53 to 0.84; not significant; low certainty). Diastolic BP: −0.39 mmHg (not significant; low certainty).
• Psychiatric symptom severity: PANSS total, PANSS subscales, CGI-S, and most cognitive domains (BACS) showed non-significant or conflicting results. No consistent signal of clinical deterioration was observed across any trial.
• Clozapine and norclozapine concentrations: No significant or consistent changes reported by Siskind et al., suggesting absence of pharmacokinetic interaction in that sample.
• Nicotine dependence: A significant reduction on the Fagerström Test for Nicotine Dependence was reported by Sass et al. (single study; not meta-analysed).
• BACS verbal memory learning: Siskind et al. reported a significant increase, but this was an isolated finding from a single small trial and should be interpreted with considerable caution.

No formal subgroup analyses by antipsychotic class, baseline metabolic status, or demographic variables were reported or feasible given the three-trial evidence base.

Adverse Events and Safety Profile

Six adverse event categories contained sufficient data for meta-analysis. Three showed statistically significant elevations in the semaglutide group:

• Abdominal pain: RR 2.93 (95% CI 1.13 to 7.60; p = 0.027; τ² = 0; prediction interval 1.13 to 7.60; GRADE high certainty)
• Vomiting: RR 2.57 (95% CI 1.39 to 4.77; p = 0.003; τ² = 0; prediction interval 1.39 to 4.77; GRADE high certainty)
• Constipation: RR 3.23 (95% CI 1.14 to 9.18; p = 0.028; τ² = 0.38; prediction interval 0.65 to 16.03; GRADE moderate certainty)

Three categories did not differ significantly between groups:

• Diarrhoea: RR 1.50 (95% CI 0.69 to 3.24; not significant; low certainty)
• Nausea: RR 1.44 (95% CI 0.87 to 2.39; not significant; low certainty)
• Serious adverse events: RR 0.81 (95% CI 0.52 to 1.26; not significant; low certainty)

Non-meta-analysed adverse events, including psychiatric admissions, self-harm, suicidal behaviour, neurological symptoms (dizziness, fatigue, headache), and infections, were infrequently and inconsistently reported across trials, with no consistent pattern of increased risk. Rates of any adverse event and adverse events leading to discontinuation were broadly comparable between groups, with risk ratios close to unity where reported. The timing of adverse events relative to dose escalation was not reported in any component study, a gap the authors identify as clinically important.

Statistical Approach and Rigor

The analytic strategy was appropriate for the available data. Random-effects pooling with restricted maximum likelihood estimation is the correct default when between-study heterogeneity is expected, and the authors provided prediction intervals alongside point estimates, which is a more informative and honest representation of effect variability than confidence intervals alone. The weight prediction interval of −18.75 to −3.89 kg, for example, captures the plausible range of effects across settings, not merely the uncertainty around the average. GRADE was applied systematically to rate certainty. Risk-of-bias assessment used the Cochrane RoB2 tool with dual independent reviewers and 100% agreement. All three component trials were rated low risk of bias.

Two methodological constraints bear emphasis. First, with only three contributing studies, heterogeneity estimates (τ²) are themselves imprecise; the pooled τ² for body weight of 10.15 suggests meaningful between-study variability, and the prediction interval should temper confidence in the point estimate for any specific clinical context. Second, publication bias could not be formally assessed with fewer than ten studies. Given that three ongoing trials were identified but had not yet published results, selective reporting of positive findings cannot be excluded. The authors correctly note this constraint.

Why This Matters Clinically

People with schizophrenia spectrum disorders die 10 to 20 years earlier than the general population, and the majority of those premature deaths are attributable to cardiovascular and metabolic disease, not to the psychiatric disorder itself. Antipsychotic medications, particularly clozapine and olanzapine, accelerate metabolic risk through mechanisms that include weight gain, dyslipidaemia, and dysglycaemia. The prior standard pharmacological adjunct, metformin, produces roughly 3 kg of weight loss relative to placebo in this population. Semaglutide, at lower doses and shorter durations than general-population obesity trials, appears to generate three to four times that effect. That magnitude, if durable and translatable to cardiovascular outcomes, has the potential to materially alter life expectancy trajectories in a population historically underserved by metabolic medicine. The simultaneous absence of psychosis worsening and pharmacokinetic interaction with clozapine (at least in one trial) removes what had been a reasonable theoretical concern about GLP-1 agonism in a dopaminergically vulnerable population.