AN AI-ENHANCED DECISION FRAMEWORK FOR STRATEGIC INVESTMENT IN HEALTHCARE FACILITIES (Kenya, Stanford Clinical Health Resilience Dashboard)

Effective pre-disaster investment in healthcare facility resilience is hindered by a fragmented landscape of complex assessment frameworks and disparate funding requirements, leaving local operators without an actionable tool to build a compelling case for investment. The work presented in this thesis directly meets this need by developing and validating a unified framework that operationalizes research findings into an accessible facility resilience investment decision-support dashboard.

The research was guided by the following objectives:
- Formally test the null correlation between community wealth (proxied by the Atlas AI Asset Wealth Index) and functional facility resilience (RO1).
- Analyze geographic inequities through the lens of distributive justice defining the 'resource' as the fair spatial distribution of operationally resilient facilities (RO2).
- Move beyond structural views to analyze the critical role of Mechanical, Electrical, Plumbing, and Communication (MEP+C) systems in maintaining operational continuity (RO3).
- Address the "problematic profusion" of inconsistent frameworks by developing a granular, standardized assessment instrument (RO4).

The ultimate objective is to synthesize all findings into a powerful decision-support dashboard that moves beyond technical metrics to create a compelling financial argument. Drawing on the principles of Break-Even Analysis and Prospect Theory, the tool is designed to translate the risk of operational downtime into a quantifiable cost of inaction. This empowers operators to reframe resilience not as an expense, but as a crucial investment to avert future losses, culminating in an evidence-based, actionable funding request (RO5).

These objectives were pursued through a mixed-methods analysis of 280 clinics acrossKenya's official six-tier health system. The study integrates facility resilience data from respondent surveys using the Argonne Resilience Measurement Indicator (RMI) framework with community economic data. The findings reject the initial null hypothesis, revealing a highly nuanced relationship where community wealth is not a direct predictor of resilience. Key findings show that lower-tier facilities (e.g., Level 2 Dispensaries) in high-wealth zones can exhibit declining resilience, while factors such as profit-driven ownership models, staff retention, and underinvestment in MEP+C systems emerged as the most critical determinants of resilience.

The five objectives were addressed as follows, first, by establishing a novel, high-resolution dataset essential for testing resilience hypotheses with empirical rigor. With this foundation, the study then aimed to solve the "problematic profusion" of inconsistent assessment tools by developing a validated, standardized resilience metric (the mRMI). This new metric was then used to investigate a central question: is facility resilience driven by aggregate community wealth, or by specific, measurable institutional and system-level factors? The resulting empirical findings prove the latter, identifying a new conceptual model of systemic failure centered on MEP+C instability. Finally, to ensure these insights translate into practice, the research culminated in an evidence-based dashboard, designed as a practical advocacy tool to help local operators secure resilience investments with data-driven evidence.

While this study validates a prototype approach, its limitations define a clear agenda for futureresearch. The reliance on survey data introduces potential respondent/enumerator bias; the analysis was constrained by aggregate community wealth data and clinic-level addresses; and importantly, the study did not include direct patient health outcomes or clinic business metrics. Future work should therefore employ objective data collection (e.g., sensors, drones), integrate granular patient-level and health outcome data, and develop the dashboard’s full predictive capabilities to create a fully dynamic decision-support tool.