Abram · Schizophrenia bulletin 2026 · randomized controlled trial · n=58

Metabolic Improvements with a Ketogenic Diet Correlate with Symptom Improvement in Psychosis: A Randomized Controlled Trial.

Level 2 - randomized trial

Individual randomized controlled trial with an uncontrolled open-label extension phase.

PubMed 42415343 · doi:10.1093/schbul/sbag082 · record verified 2026-08-26

What was done

Participants with schizophrenia-spectrum and bipolar-1 disorders were randomized to either a ketogenic diet (KETO; n = 28) or diet-as-usual (DAU; n = 30) for 1 month. An optional extension was offered to both groups, resulting in a subgroup completing 4 months on the ketogenic diet (n = 25). Metabolic health, psychiatric symptoms (positive, negative, depression), and cognition were assessed at 1 month (KETO vs. DAU) and at 4 months (KETO vs. baseline).

What was found

At 1 month, KETO participants exceeded standard ketosis thresholds and showed significant reductions relative to DAU in weight (Padj < .001), HbA1c (Padj = .05), and insulin resistance (Padj = .05); specific numeric values, baseline levels, and effect sizes were not reported in the abstract. In the 4-month extension subgroup, clinical symptoms (positive, negative, depression) and cognitive performance all improved relative to baseline (all P < .001). Increased blood ketone levels correlated with improvements in pre-diabetic markers and depressive symptoms (all P < .001), while weight reduction was not related to metabolic or symptom improvements.

Why it matters

This trial provides randomized evidence that a ketogenic diet is feasible in outpatients with serious mental illness and can rapidly improve metabolic dysregulation linked to psychiatric medications.

Limits

The randomized, controlled comparison lasted only 1 month. Longer-term (4-month) clinical symptom and cognitive outcomes were assessed only in an uncontrolled, open-label extension subgroup compared to baseline, making them vulnerable to expectancy effects, selection bias, and regression to the mean. The sample size was small (n = 58 randomized, n = 25 completed 4 months), and the abstract provides P-values without point estimates, confidence intervals, or effect sizes.