Robust strategic planning for mobile medical units with steerable and unsteerable demands

TL;DR

This paper develops a robust optimization framework for strategic planning of mobile medical units, addressing demand variability and demand types to improve healthcare access in rural areas.

Contribution

It introduces a novel integer linear programming model for the SPMMU, incorporating demand steerability and robustness against demand uncertainty.

Findings

The model effectively handles demand variability and uncertainty.

Solution methods outperform traditional approaches in computational tests.

Application to German rural healthcare demonstrates practical benefits.

Abstract

Mobile medical units (MMUs) are customized vehicles fitted with medical equipment that are used to provide primary care in rural environments. As MMUs can be easily relocated, they enable a demand-oriented, flexible, and local provision of health services. In this paper, we investigate the strategic planning of an MMU service by deciding where MMU operation sites should be set up and how often these should be serviced. To that end, we study the strategic planning problem for MMUs (SPMMU) $-$ a capacitated set covering problem that includes existing practices and two types of patient demands: (i) steerable demands representing patients who seek health services through a centralized appointment system and can be steered to any treatment facility within a given consideration set and (ii) unsteerable demands representing walk-in patients who always visit the closest available treatment facility. We propose an integer linear program for the SPMMU that can be solved via Benders decomposition and constraint generation. Starting from this formulation, we focus on the uncertain version of the problem in which steerable and unsteerable demands are modeled as random variables that may vary within a given interval. Using methods from robust optimization and duality theory, we devise exact constraint generation methods to solve the robust counterparts for interval and budgeted uncertainty sets. All our results transfer to the session-specific SPMMU and we evaluate our models in a computational study based on a set of instances generated from a rural primary care system in Germany.