Dissecting the primary to specialist referrals using graph neural networks: exploring the relationship between physician referral patterns, primary care access and healthcare spending

Abstract

The healthcare referral system affects all points of the healthcare ecosystem – access to care, patient satisfaction, physician utilization and healthcare costs. The state of these variables plays a critical role in determining healthcare efficiency. In this paper we dissect the medical referrals from primary to secondary care in Florida in 2015 and tackle them from three perspectives – influence of physician experience in referral choice, relationship between physician referral choice and Medicare spending, and pattern detection given different referral windows. To accomplish our goal of identifying patterns in primary to secondary referral mechanisms, we use Graph Neural Networks (GNN) unsupervised model to learn the vectoral representation of our physician nodes and their properties in the network. This work provides new discoveries on factors that influence the referral patterns and can be used to make better decisions when aiming to improve the efficiency of referrals.