AI-enhanced mobile support for skin diseases in primary care in Kenya: usability and accuracy
C Carrion, A Fuster-Casanovas, E Quilter, E Kinyeru, Q Miró, A Pérez-Navarro, M Aymerich, R Nyangacha, D Atiendo, J Ruiz-PostigoAbstract
Background
Skin neglected tropical diseases (skin NTDs) and common dermatoses are major public health problems in resource-limited endemic settings. The WHO SkinNTDs app supports frontline health workers (FHWs) by improving knowledge and decision-making on 13 skin NTDs and 24 common skin diseases. This study evaluates the usability and the performance of a beta version enhanced with non-generative artificial intelligence in routine primary care in Kenya.
Methods
A cross-sectional study was conducted in 2024 with 36 FHWs from five Kenyan counties using the app in routine clinical practice. Training was provided in an initial workshop (April 2024), followed by a second workshop at study completion (November 2024). Usability and user experience were assessed using the uMARS questionnaire, four focus groups and 16 semi-structured interviews. To evaluate AI performance, 605 clinical images were collected. AI outputs were compared with consensus diagnoses from three expert dermatologists in skin NTDs and common dermatoses in Kenya.
Results
The WHO SkinNTDs app showed high usability, with a mean overall uMARS score of 4.11/5. Qualitative findings indicated that the app empowered FHWs, increased self-confidence and trust from colleagues and patients, and contributed to time savings and fewer referrals to specialists. Regarding AI performance, of the 605 images collected, only those with with full agreement among the three dermatologists were included in the analysis (524). For skin NTDs, overall AI accuracy was 67%, with higher accuracy for endemic diseases in Kenya, including lymphatic filariasis (82%; 95%CI: 78-85) and mycetoma (83%; 95%CI 80-86). For common skin conditions, AI accuracy reached 96%.
Conclusions
The beta version of the WHO SkinNTDs app is a usable tool with acceptable performance and high perceived impact among FHWs in Kenya. It may strengthen frontline recognition and management of skin diseases in low-resource settings.
Key messages
• The AI-enhanced WHO SkinNTDs app showed high usability and acceptable diagnostic performance in routine primary care in Kenya.
• The app strengthened frontline health workers’ confidence, saved time and may reduce unnecessary specialist referrals in low-resource settings.