BACKGROUND AND OBJECTIVE: Nurses are essential for safeguarding public health, and their physical condition directly affects care quality and patient safety. Work-related musculoskeletal disorders (WMSDs) are highly prevalent in this workforce and may be amplified by circadian disruption. We therefore integrated individual and environmental risk factors, with special attention to night-shift characteristics, to identify and rank determinants of WMSDs among shift-working nurses. Seven machine-learning algorithms were compared to generate a comprehensive, validated prediction tool that enables managers and nurses to implement targeted, proactive interventions and reduce occupational injury.
METHODS: This study is a cross-sectional study. The general information, lifestyle, psychosocial data, working environment and shift characteristics of shift nurses were collected, and the influencing factors of WMSDs were analyzed. The mean square error increase and residual sum of squares are calculated, and the importance of influencing factors is sorted respectively. The independent influencing factors of WMSDs in shift nurses were included. After screening variables again by Lasso regression, seven prediction models of LDA, PLS, RDA, GLM, RF, SVM-Radial and SVM-Linear were established by machine learning. The AUC, accuracy and specificity median were used to evaluate the prediction efficiency, and the best prediction model was obtained and the accuracy of the prediction factors was verified.
RESULTS: Among 1 080 shift-working nurses, the WMSD prevalence at any body site was 85.19%. The top-ranked determinants were perceived control, perceived social support, Pittsburgh Sleep Quality Index (PSQI) score, chronotype, frequent bending over, night-shift nap duration, friend support, work support, family support, prolonged neck flexion and years in nursing. The LASSO-selected predictor set comprised dairy intake frequency, shift pattern, monthly night shifts, number of nurses on night duty, post-night-shift recovery days, post-night-shift catch-up sleep, frequent trunk flexion, prolonged neck flexion, PSQI, chronotype and perceived control. Random forest achieved the highest predictive performance (median AUC = 0.919).
CONCLUSIONS: Individual characteristics, lifestyle, physical condition, occupational features, shift schedule, biomechanical load and psychosocial factors collectively influence WMSD occurrence in nurses. Random forest outperformed the other algorithms and should be carried out in conjunction with various factors in the model.
