Learning and Optimization for Patient-Physician Matching in Specialty Care

Abstract

It is extremely difficult for a patient without medical knowledge and education to select a ‘capable’ physician in specialty care. Motivated by this existing patient-physician matching problem, this paper presents an improved multi-disease pre-diagnosing Bayesian network framework to match patients’ symptoms with physicians’ specialties through pre-diagnosed diseases, which is learnt by a tabu search algorithm embedded with expert knowledge. Specifically, we define a novel physician matching index (PMI) based on the symptom-specialty relationship, which can be further integrated with patient preference. A first-come first-serve rule based greedy heuristic is proposed to recommend physician under patient preference and physician capacity constraints. Its performance is evaluated by its upper bound by solving a 0-1 knapsack model. A case study of patient-physician matching problem in ear, nose and throat (ENT) department is conducted. The experimental results show that the proposed matching framework increases the physician matching accuracy under different patient preferences. Moreover, we analyze the effect of weight of primary disease and comorbidity, the physician’s specialty distribution, and the patient preference. We find that the embedded PMI matching mechanism enhances the diagnosing accuracy and patient satisfaction.

Keywords: Patient physician matching, Bayesian network, Expert knowledge, Patient preference