AI-driven healthcare analytics for early disease detection

A scoping review of clinical applications, validation, and translational challenges

Abstract

Artificial intelligence is increasingly used to support earlier disease detection across imaging, electronic health records, physiologic signals, wearable devices, and biomarker-based platforms. However, reported performance is highly variable across diseases, data environments, and validation settings, and the translational maturity of the field remains uneven. Objective: To map the breadth and characteristics of empirical studies on AI-driven healthcare analytics for early disease detection and to identify patterns in model application, validation, implementation relevance, and evidence gaps. Methods: A scoping review was conducted using the Arksey and O'Malley framework with further refinements and reported in line with PRISMA-ScR. PubMed and Google Scholar were searched for English-language peer-reviewed studies published from 2018 to 2025. Eligibility was defined using the Population, Concept, and Context framework. Data were charted on study characteristics, data modality, target disease, analytic method, reference standard, performance metrics, and implementation relevance, and synthesized narratively. Results: Twenty-nine studies were included. The evidence covered sepsis and septic shock, diabetic retinopathy, breast cancer, skin cancer, colorectal polyps, pulmonary nodules and lung cancer, tuberculosis, pancreatic cancer, atrial fibrillation, and chronic kidney disease. Imaging dominated visual detection tasks, whereas EHR, physiologic, wearable, and lipidomic data were more common in temporal or multimodal prediction tasks. Most studies reported moderate-to-high discriminatory performance, but the strength of evidence depended heavily on external, temporal, or prospective validation. The most consistent gains were earlier case identification, reduced missed lesions, improved prioritization, and workflow support rather than autonomous replacement of clinicians. Conclusion: AI shows substantial promise as a clinically embedded decision-support tool for earlier and more targeted disease detection, but widespread adoption still depends on robust external validation, local adaptation, and stronger real-world implementation evidence.