Moral distress risk profiling in ICU nurses: A cross-sectional study

Background Moral distress is common among ICU nurses and has been linked to burnout, diminished care quality, and turnover. Which factors matter most – and how they combine to shape individual risk – remains poorly characterised. Research objective To identify factors associated with moral distress in ICU nurses and develop a parsimonious, exploratory risk-profiling model. Research design Multicentre cross-sectional survey with machine learning analysis. Participants and research context A total of 318 registered nurses were recruited from ICUs of multiple tertiary hospitals across China between January and June 2025. Participants completed the Chinese-validated Moral Distress Assessment in Professional Practice Scale, the Connor-Davidson Resilience Scale, and the Pittsburgh Sleep Quality Index. High moral distress was present in 28.6% of participants. Data were split into training ( n = 223) and test ( n = 95) sets. LASSO regression selected 12 predictors from 29 variables, and 11 machine learning algorithms were compared. SHapley Additive exPlanations (SHAP) analysis guided feature reduction. Ethical considerations Approved by the Ethics Committee of Jinzhou Medical University (No. JZMULL2025136). All participants gave written informed consent. Findings Gradient boosting machine (GBM) achieved the best balanced performance among all models (F1 = 0.767). SHAP-guided reduction to six features preserved full discriminative accuracy (AUC = 0.907, 95% CI: 0.834–0.980; DeLong p = 0.969 vs the 12-feature model). The six predictors, in order of importance, were monthly night shifts, financial responsibility role, psychological resilience, sleep quality, nurse-to-patient ratio, and weekly working hours. The model was well-calibrated (Hosmer–Lemeshow p = 0.220; Brier score = 0.097) and demonstrated net clinical benefit in decision curve analysis. Night shift frequency exhibited a nonlinear relationship with moral distress risk. Conclusions Six modifiable or assessable factors accounted for most of the predictive signal. These exploratory findings suggest directions for future research and potential intervention – night-shift arrangements, financial strain, and resilience- and sleep-focused support – but the model requires external validation before any clinical application.