Integrating machine learning and experimental design to analyze unstructured healthcare decisions: A mixed-methods study at Kawsar Hospital.

Objective: This study investigates decision-making processes among healthcare managers at Kawsar Hospital in Sanandaj, Iran, using a mixed-methods approach that integrates qualitative insights with data-driven analysis. Methods: Semistructured interviews with 15 managers were analyzed through the Critical Incident Technique and the NatCen Framework, revealing distinct patterns in unstructured, multilevel decisions. Results: Organizational knowledge played a central role in the initial stages of decision making, while external research became more influential in later phases. A dataset derived from these interviews was examined using Random Forest and K-means clustering. The machine learning results highlighted time pressure as the most influential predictor of decision quality, whereas managerial experience contributed only marginally. Clustering analysis further revealed heterogeneous managerial profiles, reflecting differences in experience, decision speed, and contextual stressors. Additionally, Response Surface Methodology (DOE/RSM) underscored the importance of information sources in optimizing decisions under uncertainty. All participants expressed satisfaction with their decisions, emphasizing the adaptive and naturalistic nature of their decision-making styles. Conclusion: These findings, while exploratory given the limited sample size, provide a practical framework for enhancing healthcare management by integrating qualitative evidence with machine learning and experimental design. Future studies with larger datasets and expanded coding (e.g. a decision types table) are recommended to strengthen the generalizability and applications of this research.