Using Machine Learning Approaches for Risk Prediction and Stratification in Healthcare

Abstract

In healthcare, ensuring patient safety and delivering high-quality care continue to be significant challenges. These issues have a substantial impact on patient outcomes and healthcare costs. This thesis explores how integrating machine learning (ML) with systems engineering can enhance risk management and quality improvement in healthcare settings. The research focuses on a series of interconnected studies examining Hospital-Acquired Pressure Injuries (HAPIs), hospital no-shows, and inpatient falls. The research methodology begins with systematic reviews and bibliometric analyses of each focus area, identifying key trends, gaps, and contributing factors. Building on these insights, novel predictive models were developed, integrating advanced ML techniques with systems engineering principles, particularly the Systems Engineering Initiative for Patient Safety (SEIPS) model. This innovative approach captures the complex and dynamic nature of healthcare risks, considering patient-related, operational, and environmental factors.

For each focus study, the research process involved collaborative brainstorming sessions with multidisciplinary focus groups, followed by a methodological framework including data engineering, feature selection, SEIPS feature categorization, and SEIPS-informed ML model development. This process yielded probability estimates and identified significant predictors for each outcome of interest. The research then explored causal relationships between predictors through Bayesian Belief Networks (BBNs) and Chain Event Graphs (CEGs). Uplift modeling and average treatment effect (ATE) analyses were conducted to assess the efficacy of proposed interventions, allowing for the quantification of intervention impacts and identification of optimal risk mitigation strategies. Building on these analytical insights, the research progressed to the design and development of decision-support system user interfaces (UIs). The UIs incorporated key design principles and underwent user experience evaluation through surveys, aligning with best practices in clinical decision support system (CDSS) implementation.

Based on the insights gained from the user experience evaluation, modified standard operating procedure (SOP) flow charts were introduced for each focus case. These enhanced SOPs were designed to integrate the new ML-driven approaches into existing clinical workflows, demonstrating superiority against current practice in key operational metrics.

This comprehensive approach not only advances the understanding of risk factors and their interrelationships in healthcare settings but also provides practical, human-centered solutions for enhancing patient safety and care quality. The integration of ML, systems engineering, and user-centered design principles offers a novel framework for addressing persistent challenges in healthcare risk management and quality improvement. This research opens the door for healthcare institutions to adopt and benefit from these advanced predictive and preventive strategies, transforming the landscape of patient safety and healthcare quality management from reactive to proactive.

Date of Award	9 Dec 2024
Original language	American English
Supervisor	Mecit Simsekler (Supervisor)