Background Non-communicable diseases (NCDs), particularly hypertension and diabetes, impose a growing burden on health systems in low- and middle-income countries like Kenya. Artificial intelligence (AI)-driven Clinical Decision Support Systems (CDSS) may enhance diagnostic accuracy and adherence to clinical guidelines, yet their feasibility and acceptability among frontline clinicians in real-world settings remain underexplored. Methods We conducted a mixed-methods implementation study in 10 health facilities in Kiambu County, Kenya. The evaluation comprised three components. First, a retrospective review of 1,929 patient records established baseline NCD prevalence and care patterns. Second, we assessed the clinical acceptance of the NCDAI platform, an AI-CDSS using a Large Language Model with Retrieval-Augmented Generation, through 300 independent expert physician reviews of its recommendations. Third, we captured clinician perspectives via a cross-sectional Knowledge, Attitudes, and Practices (KAP) survey (n=29) and key-informant interviews (n=11). Results The baseline cohort demonstrated a substantial NCD burden: 72.8% had a history of hypertension and 43.1% had diabetes. Expert validation showed high acceptance of AI-generated recommendations, with 67.0% “Agreed,” 26.3% “Partially Agreed,” and only 6.7% “Disagreed,” yielding 93.3% overall (partial or full) agreement. Most disagreements arose in medication and treatment plan recommendations. Clinicians demonstrated strong digital readiness; 86% reported moderate or good IT proficiency, and 69% were already aware of AI in healthcare. Patient-related factors were the most commonly cited barriers to NCD care (33%). Qualitative findings identified operational challenges particularly duplicative data entry arising from parallel paper-based workflows as the main impediment to NCDAI adoption amid high patient volumes. Conclusions An AI-driven CDSS for NCD management is feasible and highly acceptable to expert physicians and frontline clinicians in Kenya. The key barrier is not reluctance toward AI but workflow friction. Effective scale-up will require investment in digital infrastructure to enable seamless integration and replacement of paper-based systems.
Kamau et al. (2026) studied this question.