A comparative study of patient and staff safety evaluation using tree-based machine learning algorithms

Highlights: • Medical errors harm patients and staff in dynamic and complex healthcare systems. • Machine learning algorithms (RF and GB) were developed to evaluate medical errors. • Using hospital-level survey data, RF and GB provided similar prediction accuracy. • Health and wellbeing and work-related stress were leading factors affecting errors. Medical errors constitute a significant challenge affecting patient and staff safety in complex and dynamic healthcare systems. While various organizational factors may contribute to such errors, limited studies have addressed patient and staff safety issues simultaneously in the same study setting. To evaluate this, we conduct an exploratory analysis using two types of tree-based machine learning algorithms, random forests and gradient boosting, and the hospital-level aggregate staff experience survey data from UK hospitals. Based on staff views and priorities, the results from both algorithms suggest that "health and wellbeing" is the leading theme associated with the number of reported errors and near misses harming patient and staff safety. Specifically, "work-related stress" is the most important survey item associated with safety outcomes. With respect to prediction accuracy, both algorithms provide similar results with comparable values in error metrics. Based on the analytical results, healthcare risk managers and decision-makers can develop and implement policies and practices that address staff experience and prioritize resources effectively to improve patient and staff safety.