A cross-field review of seven psychiatric research areas finds big data yields mechanistic insight and transdisciplinary collaboration, but sample size alone does not guarantee more precise estimates and a translational gap remains

The review argues that big data is not only about large sample sizes but also about data diversity and complexity, a response to the reproducibility crisis and the democratization of science, links to precision psychiatry, and the need to dissect intricate interacting mechanisms. Whereas prior discussions often equate big data with large samples, this paper explicitly places multi-site, multimodal, multivariate and high-dimensional data, along with open data and pre-registration practices, within a single framework. Organized as a cross-field literature review with tables (Tables 1–3) listing databases, promises and pitfalls for each area, it is a narrative synthesis rather than new empirical data.

The review states that big data research has provided insights into mechanisms underlying mental disorders, while emphasizing the intricacy, heterogeneity and variability of these mechanisms and arguing for triangulation between large-scale and small-scale research. It unifies findings such as the many small-effect polygenic contributions and cross-disorder genetic overlap in genomics, and the low case-control effect sizes in imaging, as evidence of mechanistic complexity rather than a single mechanistic model. Draws on representative studies and consortia across fields (e.g., Psychiatric Genomics Consortium, ENIGMA, NESDA, WMHS) as support, constituting an integrative appraisal of existing evidence.

The review finds a translational gap from big data to clinical application, noting that polygenic risk scores currently account for quite small amounts of variance and that few predictive models have been clinically validated, so diagnoses should be held lightly and explanations offered humbly. It elevates the question of clinical relevance from a debate within individual fields to a cross-field conclusion, and on that basis proposes that clinical training should cover big data methods and rigorous evaluation of AI tools. Based on a review-level judgment across multiple fields of clinical translation evidence; the text explicitly states it relied on existing systematic reviews and meta-analyses rather than conducting such work itself.

The review argues that beyond continuing to build databases, advances in conceptual models and asking the right questions are equally valuable, and that big data can serve as a source for both hypothesis generation and hypothesis testing. While affirming the value of data-driven analyses, it places theory-driven analyses on equal footing and points to needed conceptual advances in neuronal development and degeneration, integration of molecular and social determinants, and evolutionary aspects of psychiatry. A directional argument presented in the discussion and outlook sections based on the authors' cross-field review, without new empirical testing.

The paper is aimed at researchers, funders and clinical educators in psychiatric big data, and applies to seven research settings that rely largely on publicly accessible or consortium-available data: community and register-based surveys, cohort and biobank studies, electronic health records, digital phenotyping, brain imaging, genomics and other -omics, and randomized controlled trials. Its conclusions can guide priorities for data quality and methodological rigor, strategies for cross-field harmonization and triangulation, and discussion of embedding clinical questions in big data project design.

Readers should still watch that data quality metrics and harmonization standards continue to evolve across fields, and that in areas such as digital phenotyping, missingness and heterogeneity in processing pipelines may affect the stability of conclusions; the clinical utility of polygenic risk scores and predictive models remains to be validated; and this is a narrative review that did not conduct new systematic reviews or meta-analyses, with the authors noting they covered only a subset of large databases and omitted several big data directions, so depth of coverage for specific fields is limited.