Drone pre-positioning for cardiac arrest response : A feature-driven distributionally robust chance constrained programming

Cardiac arrest is a leading cause of mortality worldwide, with survival critically dependent on treatment within the strict 4–6 min golden time window. Conventional emergency medical service (EMS) relying on ground transport often faces delays from traffic congestion and geographic barriers. To address this challenge, we propose a feature-driven distributionally robust chance-constrained program (DRCCP) that determines drone base locations and assignments to minimize worst-case response time. A feature-driven Wasserstein ambiguity set is constructed to capture travel time uncertainty arising from wind speed and direction. To ensure reliable response within the golden window, we further develop two target-oriented DRCCPs, including a risk-oriented model that minimizes the violation probability and a robustness-oriented model that maximizes the admissible level of distributional ambiguity for a prespecified travel-time target. We solve these models by deriving tractable reformulations and designing a bisection-search algorithm. Numerical experiments demonstrate that the proposed DRCCP consistently outperforms benchmark models, yielding lower expected response times, improved tail-risk performance, and a reduced probability of exceeding critical thresholds. The two target-oriented DRCCP models achieve not only lower violation probabilities within the golden time window, but also smaller violation magnitudes when the target is violated. Moreover, our proposed algorithms significantly outperform state-of-the-art commercial solvers in computational efficiency.