A Statistical Learning Approach to Evaluate Factors Associated With Post-Traumatic Stress Symptoms in Physicians: Insights From the COVID-19 Pandemic

Physicians facing the COVID-19 pandemic are likely to experience acute and chronic, and often unpredictable, occupational stressors that can incur post-traumatic stress symptoms (PTSS), prevention of which is of utmost importance to enhance healthcare workforce efficiency. Unlike previous studies, in this paper we developed a generalized data-driven framework to generate insights into the complex, nonlinear associations of cognitive/occupational factors with physicians’ PTSS-risk. Data were collected from practicing physicians in the 18 states with the largest COVID-19 cases by deploying a cross-sectional, anonymous, web-based survey, following the second COVID-19 peak in the US. Analyses revealed that physicians directly treating COVID-19 patients (frontline) were at higher occupational risk of PTSS than those who didn’t (secondline). We implemented a suite of eight statistical learning algorithms to evaluate the associations between cognitive/occupational factors and PTSS in frontline physicians. We found that random forest outperformed all other models, in particular the traditionally-used logistic regression by 6.4% (F1-score) and 9.6% (accuracy) in goodness-of-fit performance, and 4.8% (F1-score) and 4.6% (accuracy) in predictive performance, indicating existence of complex interactions and nonlinearity in associations between the cognitive/occupational factors and PTSS-risk. Our results show that depression, burnout, negative coping, fears of contracting/transmitting COVID-19, perceived stigma, and insufficient resources to treat COVID-19 patients are positively associated with PTSS-risk, while higher resilience and support from employer/friends/family/significant others are negatively associated with PTSS-risk. Insights obtained from this study will help to bring new attention to frontline physicians, allowing for more informed prioritization of their care during future pandemics/epidemics.