Data driven approaches to improve operational efficiency of emergency medical services

We study data-driven approaches to maximize the service level of Emergency Medical Services (EMS) in emerging economies. These systems usually operate under heavy resource constraints and face significant operational challenges, making them structurally and operationally different from systems in developed countries. In this thesis we study two specific issues - (i) modeling human behavior, and (ii) accounting for risk metrics due to tail behavior. First, we address the issue of ambulance abandonment that occurs when a patient's willingness to wait is less than the ambulance response time resulting in the vehicle not being utilized. We present a maximum likelihood estimation approach to estimate willingness to wait for different types of patients. We then use the estimate of waiting times in a greedy simulation based optimization model to redesign the EMS network to maximize the number of patients served within their waiting time thresholds. Computational experiments using data from an Indian metropolitan city show that our proposed resource allocation model reduces abandonment by approximately 2 percentage points with the current ambulance fleet size, 5 percentage points by doubling the fleet size and 6 percentage points by tripling the fleet size. Next, we present a risk-based optimization approach to make the EMS network robust to unexpected changes in demand patterns. This is motivated by the fact that when few parts of the network face heavy-tailed demand patterns, the demand for entire network under the resource constrained setting behaves in a heavy-tailed manner. To achieve a robust location strategy we include risk metrics, specifically the Conditional Value at Risk, that focus on tail behavior in addition to average case performance metrics. Computational experiments show that planning with a view of minimizing risk leads to solutions that perform well in heavy-tailed settings