JMIR Publications

ABSTRACT

Healthcare artificial intelligence (AI) research is expanding rapidly, but its growing volume risks contributing to a new form of research waste. Traditional definitions of research waste emphasize flawed methodology, redundancy, and misaligned research priorities. Healthcare AI introduces an additional pathway: the proliferation of technically sophisticated yet clinically unimplemented models. Across the healthcare AI literature, concern is mounting that some studies prioritize predictive performance, novelty, and publication output over clinical relevance, external validity, workflow integration, governance, equity, and post-deployment monitoring. This Viewpoint argues that “algorithmic research waste” arises when AI systems are developed without clear clinical use cases, are trained on nonrepresentative datasets, lack robust external validation, and remain disconnected from real-world care delivery. These limitations undermine both internal and external validity and contribute to an accumulating evidence base that is publishable yet not implementable. Two forms are distinguished: primary waste, in which limited value is foreseeable from the design stage, and contextual waste, in which models may be useful in some settings but are mismatched to others. To address these challenges, this Viewpoint proposes a shift from model-centered evaluation to implementation-ready evidence. It introduces the VALUE-AI framework, which emphasizes Valid clinical questions, Appropriate data and population fit, Local workflow integration, Usability and safety with clear accountability, Equity-aware design and external validation, and continuous auditability and impact monitoring. Local workflow integration does not require every study to optimize an AI system for all settings; rather, researchers should design for adaptability, while local customization remains the responsibility of deploying health systems. VALUE-AI is not intended to replace existing reporting, validation, or implementation science frameworks, but to provide an AI-specific metascience lens for identifying where research value is lost across the evidence lifecycle, from question selection to post-deployment monitoring. Reducing research waste in healthcare AI requires coordinated action from researchers, reviewers, editors, funders, health systems, regulators, patients, public representatives, and industry partners to prioritize clinically meaningful, transparent, reproducible, equitable, and implementation-ready evidence. These patterns are not merely errors in individual researchers’ judgment. They are reinforced by academic incentives that reward novelty, publication volume, and benchmark performance more than external validation, replication, implementation planning, negative results, or post-deployment learning. AI research should ultimately be judged not by its ability to predict, but by its capacity to responsibly and equitably improve patient care in the settings where care is delivered.