A data-driven safety surveillance framework was developed in a study that incorporated supervised machine learning models and geospatial analytics to forecast and map occupational hazards in the United States healthcare sector. The incident records of 12,347 cases were de-identified in more than 500 facilities in the 30 states (2018-2022). There were four predictive models: logistic regression, support vector machines, random forest, and XGBoost, and the optimal one was XGBoost (AUC-ROC = 0.91, F1 = 0.85). The main predictive variables were the duration of a shift (importance score = 0.178), the type of department (0.154), and PPE compliance (0.139), as well as the ratios of staff to their patients and the time at which the incident occurred. The frequency of injuries was very high in some high-risk departments, i.e., the emergency and ICU, especially under morning and early shifts. Musculoskeletal injuries (31%), needlestick (22%), and infectious exposures (19%) were the most common hazards. Kernel density estimation and Getis-Ord Gi hotspot analysis on spatial mapping revealed the targeted areas of risk concentration within hospitals and long-term care facilities, but only in New York, California, Texas, and Ohio. These results validate the fact that occupational injuries within care are statistically predictable and spatially patterned, and that there should be a transition to real-time and predictive, geographically informed safety governance.
Islam et al. (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: