JMIR Publications

ABSTRACT

Background:

Continuity of care—defined as coherent, connected services across providers—is critical for psychiatric patients with chronic conditions but challenged by fragmentation at primary-secondary-community interfaces, poor data interoperability, and transitions. AI/ML tools (e.g., predictive algorithms for relapse, NLP for data synthesis, chatbots for engagement, digital phenotyping) promise to enhance informational, relational, and managerial continuity by automating monitoring, triaging high-risk cases, and bridging care gaps. Despite growing applications, no rigorous synthesis evaluates their efficacy, typologies, patient experiences, or implementation factors in psychiatry.

Objective:

This PROSPERO-registered systematic review will assess AI/ML interventions' effectiveness vs. standard care for continuity in adult psychiatric care; categorize AI architectures; synthesize patient-reported outcomes (PROs); and identify ethical/implementation barriers/facilitators.

Methods:

Guided by Cochrane Handbook and PRISMA-P, we will search MEDLINE, Embase, Cochrane CENTRAL, CINAHL, PsycINFO (2016–2025), trial registries (ClinicalTrials.gov, ICTRP), and references. PICO: P adults ≥21 years with primary psychiatric diagnoses; I any AI/ML for post-encounter continuity (prediction, NLP, chatbots); C standard care (case management, ACT, non-AI reminders); O readmissions, follow-up latency, adherence, attrition, PROs/satisfaction; designs: RCTs, cohorts, qualitative/mixed-methods (English, peer-reviewed). Dual independent screening/extraction in Covidence (pilot-tested, ≥80% agreement); data extraction for study characteristics, PICO details, effect sizes (RR/HR/means/SD); author contact (3 attempts) for missing data. Risk of bias: RoB 2.0 (RCTs), ROBINS-I (NRSIs), CASP (qualitative). Synthesis: random-effects meta-analysis (RevMan) if homogeneous (I²<50%); narrative/thematic otherwise; GRADE for certainty; funnel plots/Egger's for publication bias (≥10 studies); sensitivity analyses for imputation/heterogeneity.

Results:

We expect AI/ML to reduce readmissions (RR 0.7–0.9) and improve adherence via proactive tools, with barriers like privacy/clinician resistance and facilitators like workflow integration.

Conclusions:

Findings will map multi-level (patient/community/system) applications, guide equitable AI deployment in resource-constrained systems like Singapore's, and inform policy for person-centered psychiatric care. Clinical Trial: PROSPERO CRD420251245352