Building Supply-Chain Resilience in Healthcare Via AI-Driven Scenario Planning and Multi-Tier Inventory Optimization

Abstract

Healthcare supply chains face unprecedented volatility arising from fluctuating patient demand, regulatory shifts, and global disruptions. This paper presents an integrated framework coupling deep generative scenario planning with a multi‐tier distributionally robust inventory optimization model to enhance resilience across hospital networks. The scenario planning module employs a conditional variational autoencoder augmented by normalizing flow layers to learn complex demand distributions from heterogeneous time‐series and exogenous event streams. Generated latent trajectories are clustered via Wasserstein barycenter methods to form a balanced scenario tree capturing tail risks and regime shifts. The inventory optimization component formulates a two‐stage distributionally robust mixed‐integer program that enforces service‐level chance constraints under ambiguity sets defined by φ‐divergences, and integrates conditional value‐at‐risk measures to bound loss in extreme states. Decision variables span manufacturers, central warehouses, regional hubs, and hospitals, incorporating perishability, capacity, and lead‐time correlations. A custom branch‐and‐price algorithm with Benders‐driven cut generation and dual stabilization solves large‐scale instances within operational horizons. Experiments on a synthetic network with 150 products and 200 hospitals demonstrate a 42 \% reduction in stockouts, a 15 \% decrease in holding costs, and maintenance of 99 \% service levels under simulated pandemic scenarios. Sensitivity analysis over ambiguity radii and risk‐aversion parameters traces a convex trade‐off surface. The framework’s modular architecture supports seamless integration with ERP platforms and streaming analytics, offering a powerful decision support system for proactive resource allocation in volatile healthcare environments.