JMIR Publications

ABSTRACT

Background:

Artificial intelligence (AI) promises enhanced diagnostic accuracy, predictive analytics, and operational efficiency across clinical settings. However, despite remarkable technical progress, translation into sustainable clinical use remains inconsistent, with many implementations confined to prototypes or local trials and evidence on long-term value and equity fragmented. The Quadruple Aim (patient experience, population health, cost/value, and provider well-being) offers a framework for assessing whether AI advances value-based care.

Objective:

This systematic review aimed to identify key challenges and opportunities associated with the real-world implementation of AI tools in healthcare and to interpret these findings through the Quadruple Aim framework.

Methods:

In accordance with PRISMA 2020 guidelines, a systematic search was conducted across PubMed, Scopus, and Web of Science for articles published between 2015 and 2026, with final searches executed on February 6th, 2026. Studies were eligible if they reported empirical evidence on the implementation, adoption, or integration of AI tools in healthcare delivery settings. The search yielded 652 records; after removing 277 duplicates, 375 were screened by title and abstract (283 excluded), and 92 underwent full-text assessment (58 excluded for lacking an implementation focus, reporting only algorithm development, being editorials, or non-English). Thirty-four studies met all inclusion criteria. Data were extracted using a structured framework, and methodological quality was appraised with the Mixed Methods Appraisal Tool (MMAT). Findings were synthesized thematically and mapped onto the four dimensions of the Quadruple Aim.

Results:

The 34 included studies, published between 2019 and 2026, spanned 11 clinical domains, 22 countries across six geographic regions, and diverse designs including qualitative (n=9), cross-sectional survey (n=9), mixed methods (n=7), cohort (n=5), and other designs (n=4). The most frequently addressed implementation themes were workflow integration and usability (74%), trust and explainability (71%), data quality and interoperability (68%), clinician knowledge and training (62%), and regulatory uncertainty (53%); percentages denote the proportion of studies in which each theme was salient, whether as a barrier encountered or a factor successfully managed. Key opportunities included predictive analytics and early warning systems (56%), stakeholder engagement and co-design (53%), workflow optimization and automation (50%), and patient engagement and access (44%). Mapping findings to the Quadruple Aim revealed that 23 studies addressed provider well-being, 22 cost/value, 21 patient experience, and 20 population health. Of the 33 empirical studies, MMAT appraisal rated 18 as high quality, 12 as moderate, and 3 as low; the remaining study, a conceptual viewpoint, was not eligible for appraisal.

Conclusions:

While most studies confirm AI's potential to enhance quality, safety, and efficiency, barriers related to governance, transparency, workforce readiness, and equitable design continue to impede scale-up. Evidence limitations include heterogeneous study designs, variable reporting depth, and the absence of longitudinal data. Sustainable success requires aligning AI strategies with the Quadruple Aim through robust data governance, inclusive co-design, investment in digital-health literacy, and rigorous economic evaluation.