Can We Make Early Predictions to Identify Which Medical Interns are at Risk of Failing a Clinical Posting?

Passing a medical internship is a crucial professional milestone. We conducted a prospective study to assess the feasibility of early prediction in identifying interns at risk of failing a clinical posting. We surveyed 496 newly enrolled interns across 26 Malaysian hospitals from January to April 2020, using validated instruments to evaluate various factors related to personal attributes, place of study, and place of practice. After one year, we followed up with the participants to identify those who had failed a clinical posting. Significant predictors were determined using the supervised machine learning (ML) framework to linear discriminant analysis (LDA), with the prediction performance validated through split train-test and crossvalidation. The LDA identified a higher risk of clinical posting failure among interns from 8 specific hospitals and 13 medical schools, as well as those with poor interpersonal skills, an avoidant coping style, greater preparedness in information technology, married status, and a shorter gap between graduation and internship commencement. The model achieved 100% sensitivity, 85.7% specificity, 86.7% accuracy, and an area under the curve (AUC) of 0.969 for the training dataset. Cross-validation showed 92.0% sensitivity, 82.9% specificity, 87.4% accuracy, and an AUC of 0.905, while the validation dataset achieved a sensitivity of 75.0%, specificity of 81.0%, accuracy of 80.7%, and an AUC of 0.861. This study shows that surveys of newly enrolled medical interns can effectively predict those at risk of failing a clinical posting within a year. We recommend replicating this study in various countries and practice regions to improve generalisability. Additionally, qualitative studies could provide deeper insights.