IT and Data Governance Constraints in Healthcare Analytics

Abstract

Healthcare organizations increasingly apply artificial intelligence and machine learning to data generated by operational information systems, despite frequent mismatches between data quality and the requirements of predictive analytics. This IT-focused case study examines the feasibility of applying machine learning to non-standardized operational healthcare data, using pressure ulcer documentation as an illustrative example. A Random Forest classifier was applied to a multi-hospital dataset extracted from a bed management system, achieving moderate overall accuracy but limited performance in predicting the minority class due to data heterogeneity, class imbalance, and limited feature availability. Feature importance analysis nevertheless identified anatomically plausible patterns, demonstrating that noisy operational data can still support exploratory analytics. The study highlights practical implications for healthcare IT leaders regarding data governance, system design, and realistic expectations of AI-driven initiatives.