Comprehensive Summary
Mental health issues are an escalating concern among healthcare professionals, with suicide risk significantly higher than the general population as the suicide rate among physicians in the U.S., for instance, is more than twice the national average. Medical students face similar challenges, often driven by academic pressure and career uncertainty. Psychological resilience, the capacity to recover or adaptively regulate psychological functioning amid adversity, is crucial for these students. Enhancing resilience helps medical students cope with stress, reduce burnout, and develop professional identity, ultimately preparing them for future crises and complex work environments. This study aimed to identify high-risk individuals and explore the mechanisms influencing their psychological resilience. Researchers surveyed 843 medical students across various majors (e.g., Clinical Medicine, Dentistry, Nursing) at one university using scales like the Chinese version of The Resilience Scale, as well as the Smartphone Addiction Scale and Pittsburgh Sleep Quality Index. An Extreme Gradient Boosting (XGBoost) model was developed to predict psychological resilience, integrating basic demographic data (like gender, socioeconomic status, grade level, etc.) with daily health behaviors, including sleep disturbance, smartphone addiction, and perceived social support. The XGBoost model was compared against other machine learning models such as Random Forest and Logistic Regression for predictive effectiveness. The XGBoost model demonstrated the best predictive performance among the compared models, achieving an accuracy of 0.822. Key findings, revealed by Shapley additive explanations and mediation analysis, were that perceived social support was positively correlated with psychological resilience, and smartphone addiction and sleep disturbance were negatively correlated. Smartphone addiction and sleep disturbances act as independent or chain mediators between perceived social support and psychological resilience, meaning low social support indirectly lowers resilience via these factors. A total of 255 medical students were identified with low psychological resilience, representing a detection rate of 30.2% in the sample. The study’s primary limitations are that the dataset originated from a single institution and the model has not yet been validated on an external dataset, which restricts the generalizability of the findings. Also, the limited sensitivity of the model suggests that some important predictive factors for psychological resilience may still have been overlooked.
Outcomes and Implications
Traditional mental health screenings in universities typically rely on questionnaires, which are insufficient for real-time monitoring, and are susceptible to inauthentic responses due to the sensitive nature of the topics. This study addresses these limitations by constructing a machine learning psychological resilience risk prediction model (resilience being a key predictor of these adverse symptoms) which combines basic demographic data with objective daily behavioral features. This approach offers an effective, low-cost, and non-stigmatizing screening tool for the early identification of individuals with low psychological resilience, creating a foundation for developing personalized interventions and cultivating a more resilient future healthcare workforce. The results of this study particularly emphasize the complex mediational roles of smartphone addiction and sleep disturbance. Specifically, smartphone addiction mediates the relationship between perceived social support and resilience, as students lacking real-world support may seek validation in the virtual realm, increasing dependency and ultimately lowering resilience. Further, sleep disturbance also acts as a mediator as lower perceived social support is associated with a higher risk of poor sleep which itself is a known risk factor for anxiety, depression, and suicide. The most significant discovery is the chain mediating effect, where a lack of perceived social support increases the risk of smartphone addiction, which then heightens the risk of sleep disturbance, leading to a significant reduction in psychological resilience. Some limitations, however, are the model's relatively limited sensitivity and the use of a single-center dataset. Therefore, future research should focus on validating the model across diverse populations using transfer learning, designing longitudinal studies to explore long-term causal mechanisms, and incorporating more interpretable machine learning methods to enhance the transparency and applicability of the predictive framework.