Pharmacogenomics and Neuroimaging for Predicting Treatment Response and Relapse in Co-Occurring Mental Illness and Substance Use Disorders: A Systematic Evidence Review

Abstract

Background: Co-occurring mental illness and substance use disorders (SUDs) are associated with high clinical complexity, heterogeneous treatment response and recurrent relapse. Pharmacogenomic and neuroimaging biomarkers may complement conventional clinical predictors, but the extent to which evidence directly supports dual-diagnosis populations is uncertain. Objective: To systematically evaluate direct and clinically relevant indirect evidence on pharmacogenomic and neuroimaging biomarkers for treatment response and relapse, with explicit attention to evidence directness, methodological quality and clinical transferability to co-occurring mental illness and SUDs. Methods: A systematic evidence review was reported in accordance with PRISMA 2020. The review record documented searches in PubMed/MEDLINE, PsycINFO, Embase and Web of Science, with evidence considered through 15 July 2026 and supplemented by reference-list screening and targeted citation identification. Evidence was stratified as direct (co-occurring mental illness and SUDs) or indirect but translationally relevant (psychiatric-only or SUD-only populations). Primary trials and prediction studies, systematic reviews, clinical guidelines and focused narrative reviews were considered using design-appropriate domains from RoB 2, PROBAST, AMSTAR 2, AGREE II and SANRA. Heterogeneity precluded quantitative pooling across source types. Results: Seventeen outcome-bearing evidence sources were retained: four primary trials/prediction studies and 13 systematic reviews, guidelines or focused narrative reviews. Direct primary biomarker evidence in prospectively defined dual-diagnosis cohorts was sparse. In major depressive disorder, the GUIDED trial did not significantly improve its primary week-8 symptom-improvement endpoint, although secondary response and remission outcomes favoured guided care; PRIME Care showed fewer predicted drug–gene interactions and a small, nonpersistent remission advantage. CPIC supports prescribing recommendations based on CYP2D6, CYP2C19 and CYP2B6, but not SLC6A4 or HTR2A. In alcohol-use populations, neuroimaging studies reported predictive signals for post-treatment drinking and relapse, including a 46-participant study with balanced accuracy of 79.4%, but small samples and limited external validation constrain clinical use. Conclusion: Pharmacogenomic and neuroimaging biomarkers are promising components of precision psychiatry, but the principal finding for co-occurring mental illness and SUDs is an evidence gap. Current signals are largely extrapolated from single-disorder populations and should not be treated as validated dual-diagnosis prediction tools. Prospective, multi-ancestry, externally validated studies that integrate genomic, neuroimaging and clinical variables in explicitly co-occurring populations are required.