From Data to Decisions in Ambulance Redeployment
Prof. Shane Henderson (Cornell University)
NICTA SEMINARDATE: 2011-04-01
TIME: 11:30:00 - 12:30:00
LOCATION: NICTA - 7 London Circuit
CONTACT: JavaScript must be enabled to display this email address.
ABSTRACT:
Emergency medical service (EMS) providers attempt to provide quick ambulance response to calls for medical attention, along with transport to hospital if necessary. Budgets are invariably tight, so there is great pressure to make the most of EMS resources. Many EMS providers now use some form of ambulance redeployment, which is also known as system-status management or move-up. Redeployment involves moving vehicles in real time in response to real-time system information, in an attempt to better match available ambulances to future calls. Many existing redeployment implementations are designed in somewhat ad-hoc ways,leading to limited performance gains and crew frustration at what are perceived as pointless moves.
I will describe our efforts in using approximate dynamic programming (ADP) to make these redeployment decisions more carefully. I'll describe how ADP works in our context, why traditional methods for tuning ADP coefficients are not completely satisfactory in our context and perhaps more broadly, our revised tuning methods, and numerical results for a couple of major centres. I''ll also briefly explain the statistical methods we use to obtain parameters for our models from Computer-Aided Dispatch databases, emphasizing how we can obtain travel times on road networks from Global Positioning System (GPS) breadcrumb data through a Bayesian formulation. If time allows, I'll also describe our efforts to obtain an upper bound on achievable performance improvements using redeployment, which is important in knowing when we researchers can stop looking for improvements.
BIO:
Shane Henderson received his BSc in Mathematics from the University of Auckland, New Zealand in 1992, his MS in Statistics in 1995 and Ph.D in Operations Research in 1997 from Stanford University. Henderson joined the OR faculty in 2001.
Henderson's research is concerned with discrete-event simulation, from input analysis (for example, extension of simple input models to capture correlation between inputs) to output analysis (for example, using martingales in simulation to achieve variance reduction). A new emphasis in his work is the interplay between optimization and simulation. He is particularly interested in structured simulation optimization, where the optimization problem enjoys certain properties such as convexity or quasi convexity that can be used to develop algorithms that are robust and fast. Specific applications in this area include radiation treatment planning; call center planning, yacht match racing, ambulance deployment, adaptive Monte Carlo and policy identification in complex networks.
Current work also explores the use of general variance reduction techniques, in particular, looking at the use of martingales to obtain variance reduction in simulations of Markov processes. This involves exploring adaptive methods to tune the variance reduction and extending the methods from the Markov setting to general discrete-event simulations. Additional research interests concern regenerative methods of simulation, dependence structures and input uncertainty.
