Perceived impact of generative artificial intelligence on clinical learning and decision-making among internal medicine residents in Ethiopia: a mixed- methods study
Background Generative artificial intelligence (GenAI) is increasingly used in medical training, but evidence from African residency programs is scarce. We assessed how internal medicine residents in Ethiopia use GenAI and how they perceive its impact on clinical learning and clinical decision-making. Methods We conducted an explanatory sequential mixed-methods study consisting of a national cross-sectional online survey of internal medicine residents (June to August 2026), followed by two focus group discussions (FGDs) with 10 residents at two universities (September 2026) to explain the survey findings. Perceived impact was measured with five-point Likert items and two composite scores (learning, five items; decision-making, eight items). Associations with frequency of use were tested with chi-square or Fisher's exact tests. FGD data were organized deductively into categories based on the survey topics. Results Of 252 respondents from 18 institutions, 98.8% had used GenAI and 75.4% used it daily; ChatGPT was used by 91.6% of users. The mean composite score for perceived impact on clinical learning was 3.83 out of 5 (70.7% high) and for clinical decision-making 3.86 out of 5 (67.7% high). Agreement was highest for improved differential diagnosis (90.8%), learning efficiency (81.9%) and support for self-directed learning (79.5%), and lowest for ward-round performance (64.3%) and confidence in complex cases (71.4%). Daily use was more frequent among residents with more than three hours of daily internet use (82.3% versus 66.7% to 67.4%; p = 0.024) and was not associated with age, sex, marital status, year of residency, device access or formal training. In the FGDs, residents described GenAI as a fast, supportive tool for learning and clinical work, as an aid to reasoning and not a replacement for physician judgment, and as a possible threat to independent learning. Conclusions Residents perceive GenAI as helpful for self-directed learning and diagnostic reasoning while keeping final responsibility for clinical decisions. These findings reflect perceptions and not measured outcomes, so objective evaluation of its effects on learning and clinical performance is needed.
Authors
- Tamirat Godebo Woyimo (ORCID: https://orcid.org/0009-0006-5943-944X)
- Hayatu Awel Abdela (ORCID: https://orcid.org/0009-0005-8946-3021)
- Kenzudine Assfa Mossa
Institutions
- Jimma University (ET)
- Wolkite University (ET)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-18
- DOI
- https://doi.org/10.21203/rs.3.rs-11541959/v1
- Primary Topic
- Artificial Intelligence in Healthcare and Education
- Type
- preprint