Background: Predicting psychiatric outcomes in individuals indirectly affected by natural disasters remains a significant challenge. While direct victims receive immediate attention, indirect victims—including family and friends of those affected—can also experience considerable psychological distress.
Objective: This study explores the tendency to report decisional uncertainty (selecting “not sure” responses) regarding emotional and academic states, which may be associated with depressive symptoms. We operationalized this tendency using a pragmatic proxy: the sum of “not sure” responses across four self-report domains, rather than a validated decisional uncertainty scale.
Methods: Medical students (n = 129), both indirectly affected and unaffected by the Kahramanmaras earthquake through family exposure, were recruited. The proxy measure for decisional uncertainty was calculated from responses concerning hobbies, nutrition, occupational satisfaction, and academic success. Associations with depression, anxiety, and earthquake exposure were analyzed using correlation analyses. Exploratory machine learning (ML) models—including random forest (RF), gradient boosting machines, and support vector machines—were employed to explore the feasibility of classifying this decisional uncertainty proxy.
Results: A weak positive correlation was found between the decisional uncertainty proxy and depression scores (r = 0.247, p = 0.005). However, a multivariable model identified academic and nutritional status as stronger independent predictors of the proxy than depression or anxiety scores. In exploratory ML analysis, models such as RF demonstrated the feasibility of classifying decisional uncertainty, achieving above-chance accuracy.
Conclusion: Preliminary findings suggest a weak association between a simple proxy for decisional uncertainty and depressive symptoms. Exploratory ML approaches show initial promise in identifying this proxy status. These results highlight a novel methodological pathway; however, they are preliminary and underscore the necessity for future validation and calibration against a dedicated, clinically validated decisional uncertainty instrument before any substantive clinical or predictive claims can be made.
Machine learning-driven prediction of decisional uncertainty among medical students post-Kahramanmaras earthquake
