JMIR Publications

ABSTRACT

Background:

Background:

Mental health and wellbeing services generate growing volumes of operational and client-level data. Business intelligence, data visualisation, and machine learning are increasingly proposed to translate these data into insight that can improve service delivery and decision making. The review-level evidence base for these technologies in mental health and wellbeing services is fragmented across methodologies, populations, and technology categories, and has not been integrated.

Objective:

Objective:

To synthesise review-level evidence on how business intelligence (BI), data visualisation, and machine learning (ML) are used within mental health and wellbeing services; to characterise the service contexts, methods, applications, benefits, and barriers reported; and to identify evidence gaps and future priorities.

Methods:

Methods:

An umbrella review was conducted per Joanna Briggs Institute (JBI) methodology and reported following PRISMA 2020. Scopus, Web of Science, IEEE Xplore, ACM Digital Library, and PsycINFO were searched on January 27, 2026 for peer-reviewed English-language reviews published between 2021 and 2026. Two reviewers screened records, with third-reviewer resolution of conflicts. Quality was appraised with the JBI critical appraisal checklist. The protocol was registered on the Open Science Framework, and findings were synthesised narratively around five research questions.

Results:

Results:

Eight reviews were included (two systematic, three scoping, one umbrella, one narrative, one mini-review). Across the set, the reviews drew on several hundred primary studies and source documents spanning mental health, broader healthcare, and digital health. Reported methods were dominated by AI and ML (conversational agents, NLP, deep learning, large language models), while BI and data visualisation were thinly represented. Most reviews addressed broad healthcare contexts in which mental health was one application domain among several; only one focused specifically on mental health helpline services. Applications clustered around clinical decision support, screening and triage, conversational delivery, and patient-feedback analytics. Service-level benefits (efficiency, accessibility, satisfaction) were reported more consistently than clinical benefits, and short-term effects more reliably than sustained ones. Notably, several reviews reported no significant clinical benefit at three- to six-month follow-up, one reported chatbot use associated with increased depressive symptoms, and two flagged large language model bias or hallucination, so positive service-level signals are not matched by durable clinical effect. Cross-cutting barriers included data privacy, technology immaturity, poor integration with clinical workflows, alert fatigue, and absent design-to-evaluation frameworks. Two reviews were judged low quality on JBI criteria and retained with their limitations explicitly flagged.

Conclusions:

Conclusions:

Review-level evidence on data-driven methods in mental health and wellbeing services has grown rapidly, but the field is uneven. AI and ML methods are well covered, while BI and operational data visualisation are not. Service-level outcomes are reported more reliably than clinical outcomes, and helpline-specific evidence is concentrated in a single review. Priorities for future work include service-delivery-focused evaluation, longer follow-up, helpline-specific implementation evidence, and explicit treatment of safety, equity, and ethics.