The LAI-Oral Antipsychotic Efficacy-Effectiveness Paradox in Schizophrenia

Background and Objectives: Long-acting injectable (LAI) antipsychotics are associated with reduced hospitalization and relapse in real-world observational studies of schizophrenia; however, randomized controlled trials (RCTs) demonstrate no statistically significant advantage over oral antipsychotics for relapse prevention. This efficacy-effectiveness paradox has remained unresolved in the literature for decades and has directly contributed to the persistent underutilization of LAIs in clinical practice. Objective: To examine the methodological sources of the LAI efficacy-effectiveness paradox and to propose four conceptual, precision-medicine-informed methodological solutions, including artificial intelligence (AI)-supported neuroimaging, to guide the design of future comparative effectiveness research. Materials and Methods: This report comprises two components: Part I, a scoping review of existing comparative studies, conducted and reported in accordance with the PRISMA extension for scoping reviews (PRISMA-ScR); and Part II, a conceptual methodological framework proposed by the authors on the basis of gaps identified in Part I, which is not itself a product of systematic scoping-review methodology. For Part I, we searched PubMed, EMBASE, PsycINFO, and the Cochrane Library from January 2000 to March 2026. Search terms included 'long-acting injectable antipsychotics,' 'oral antipsychotics,' 'schizophrenia,' 'medication adherence,' 'confounding by indication,' 'pharmacogenomics,' 'CYP2D6,' 'digital medicine,' 'pragmatic trials,' 'positron emission tomography,' 'functional magnetic resonance imaging,' 'neuroimaging,' 'machine learning,' and 'artificial intelligence. Results: The paradox is primarily explained by confounding by indication in observational studies and inadequate adherence measurement across all study designs (Part I). On this basis, we propose, as a conceptual framework rather than a validated finding, four methodological solutions (Part II): (1) objective digital adherence monitoring via ingestible sensor or equivalent technology, subject to explicit informed consent and comprehensive ethical safeguards; (2) pharmacogenomic stratification using CYP2D6, CYP3A4, and CYP2C19 metaboliser phenotyping to control for pharmacokinetic confounding specific to antipsychotic drug metabolism; (3) pragmatic RCT designs with real-world eligibility criteria and objective primary adherence endpoints; and (4) artificial intelligence-supported neuroimaging positron emission tomography (PET)-derived dopamine receptor occupancy and machine learning analysis of functional magnetic resonance imaging (fMRI) connectivity data as an objective pharmacodynamic and predictive complement to the three adherence- and pharmacokinetics-focused solutions. Conclusions: Current evidence on LAI versus oral antipsychotic superiority cannot be considered definitive. Integration of digital monitoring, pharmacogenomics, pragmatic trial design, and AI-supported PET and fMRI neuroimaging into future research represents a necessary methodological correction that could resolve decades of contradictory findings and provide clinicians with genuinely actionable, patient-centered evidence.