Hospital Inventory Resilience Using Conformal Prediction and Digital Twin Simulation During Sudden Demand Spikes

Aim: This study aims to develop and evaluate a novel inventory management framework for hospital supply chains that integrates probabilistic demand forecasting using conformal prediction with digital twin-based simulation. The objective is to improve resilience and reduce stockouts of critical medical supplies, such as N95 masks, during sudden demand surges like those experienced during COVID-19. Methods: The proposed model uncertainty-aware demand forecasting using conformal prediction to generate probabilistic demand intervals rather than single-point estimates. It also uses digital twin simulation of hospital inventory dynamics to evaluate alternative inventory control policies under stochastic demand conditions. To demonstrate feasibility, ten synthetic 30-day demand scenarios for N95 masks were generated. These demand curves were tested against two baseline inventory policies using simulation. Performance metrics included stockout frequency and inventory level efficiency. Results: Simulation results indicate that integrating conformal prediction–based demand intervals with digital twin simulation significantly reduces stockout occurrences compared to traditional single-point forecast methods. The proposed approach achieved improved service levels while maintaining inventory at economically sustainable levels. The findings suggest that explicitly modeling uncertainty enhances decision-making robustness under volatile demand conditions. Conclusion: Uncertainty-aware forecasting combined with supply chain simulation provides a more resilient alternative to conventional deterministic inventory planning models. Recommendations: Future research should prioritize validating the proposed framework using real-world hospital demand data to establish external validity and operational relevance. Incorporating empirical datasets would allow for more robust calibration of conformal prediction intervals and improve the fidelity of the digital twin simulation. In addition, the model should be expanded to account for critical operational constraints such as supplier lead times, procurement delays, budget ceilings, and storage capacity limitations, all of which materially influence inventory performance in practice.