Toward precision medicine: can neuroimaging prospectively predict early treatment outcomes in schizophrenia spectrum disorders? A systematic review and meta-analysis.
Authors
Affiliations (5)
Affiliations (5)
- Azienda Sanitaria Locale Napoli 3 Sud, U.O.S.M. 55-57 Torre del Greco, Ercolano, Via Guglielmo Marconi, 66, 80059, Torre del Greco, Italy. [email protected].
- Department of Advanced Biomedical Sciences, University of Naples "Federico II", Naples, Italy.
- Department of Advanced Medical and Surgical Sciences, University of Campania "Luigi Vanvitelli", Naples, Italy.
- Azienda Sanitaria Locale Napoli 1 Centro, Unità Operativa Complessa Salute Mentale 32 e 33, Via Walt Disney, 6, Naples, Italy.
- Department of Psychiatry, University of Campania "Luigi Vanvitelli", Naples, Italy.
Abstract
Variability in antipsychotic response results from interaction of illness-related, treatment-related, environmental factors, and intrinsic interindividual differences that have not yet been addressed. Since early non-response may predict later non-response, biomarkers capable of identifying individuals at higher risk of treatment failure may have important implications for management personalization. This study aimed to evaluate whether baseline neuroimaging can predict subsequent treatment outcomes in Schizophrenia-spectrum Disorders. A systematic search of PubMed/EMBASE/IEEE Xplore was conducted until 04/20/2026, in accordance with PRISMA-DTA guidelines and a pre-registered protocol. Prospective studies using Magnetic Resonance Imaging/Positron Emission Tomography with predictive models for subsequent treatment outcomes were included. A random-effects multi-level meta-analysis was performed to pool areas under the curve (AUC). Hierarchical summary receiver operating characteristic (HSROC) curves estimated sensitivity/specificity. QUADAS-2 assessed Risk-of-bias. Sixteen studies were included. The overall pooled discriminatory performance of multi-level hierarchical models was good (AUC = 0.75,95%C.I.[0.70-0.80],τ²=0.071,I<sup>2</sup> = 28.1%,k = 50,n = 14), best-AUC model showed similar, slightly better results (AUC = 0.81,95%C.I.[0.76-0.85],τ²=0.053,I²=21%,k = 11). Functional connectivity-based models showed the highest performance, significantly higher than activity/metabolism-based, structural models, at the Q-test statistics (p = 0.03). Baseline drug status and clinical stage did not significantly moderate estimates. The HSROC model yielded a sensitivity = 0.75,95%C.I.[0.68-0.79], specificity = 0.75,95%C.I.[0.67-0.81]. Baseline neuroimaging may provide relevant prognostic information in Schizophrenia Spectrum, especially based on functional connectivity, supporting that treatment response may be related to disturbances in large-scale network organization than to isolated structural abnormalities. These findings support the potential of prognostic neuroimaging in precision psychiatry and early-risk stratification. However, methodological standardization, external validation, and geographically-diverse samples remain necessary before a full clinical translation.