Revealing Physicians Referrals from Health Insurance Claims Data

Abstract

Health insurance companies in Brazil have their data about claims organized having the view only for service providers. In this way, they lose the view of physicians' activity and how physicians share patients. Partnership between physicians can be seen as fruitful, when they team up to help a patient, but could represent an issue as well, when a recommendation to visit another physician occurs only because they work in same clinic. This work took place during a short-term project involving a partnership between our lab and a large health insurance company in Brazil. The goal of the project was to provide insights (with business impact) about physicians' activity from the analysis of the claims database. This work presents one of the outcomes of the project, i.e., a way of modeling the underlying referrals in the social network of physicians resulting from health insurance claims data. The approach considers the flow of patients through the physician–physician network, highlighting connections where referrals between physicians potentially occurred. We present the results from the analysis of a claims database (detailing 18 months of activity) from the health insurance company we partnered with. The main contribution presented in this paper is the model to reveal mutual referrals between physicians. Results show the proposed model reveals underlying characteristics of physicians' activity from real health insurance claims data with multiple business applications.

Introduction

Health insurance costs are a main issue of concern in almost every country in the world as budget constraints impact directly on the quality of the service. As a result, health insurance companies have been extensively trying to reach a trade-off between offered services and costs as a way to meet budget constraints.

One way for health insurance companies to address those issues is to better understand the complex relationships among the diverse participants of the healthcare systems, including patients, physicians, hospitals, and other service providers. To support this quest, healthcare insurance companies and other health service providers have often a wealth of data from their own operations at their disposal, especially transactional data.

In the case of health insurance companies, an important piece of transactional data involves the claims presented by their ecosystem of providers. In the present work, a claim represents a report from a physician or a healthcare service provider to a health insurance company requesting some form of fee related to a patient's consultation with a physician, a clinical exam, or a medical procedure. Even though claims data may vary, it generally contains at least the ID of the healthcare professional involved in the procedure (it may also be a group of professionals), the ID of the patient, the type of procedure, and time information related to the event. It may include other types of information such as location of the service, pre-authorization codes, etc.

Traditionally the analysis of claims data is based on applying statistics and Data Mining methods to the individual elements of the system (physicians, service providers, patients) or to the set of claims. However, healthcare is often provided by collaborative teams of physicians, nurses, and technicians which are connected to each other by often strong professional relationships. Physicians that refer patients to other physicians have clear preferences about who they want to team up with for specific procedures and often are involved in master-apprentice structures. Physicians also have preferences for specific service providers such as hospitals and clinical analysis laboratories. Those recommendations could be good for building patient trust or indicate a fraud when this is not the patient's will. Similarly, patients establish bonds of trust and reliance with specific physicians or group of physicians.

In practice, mining claims is difficult because claims are paid to a wide variety of providers, such as hospitals, clinics, or even physicians registered as small companies. A single patient can consult a physician through all those channels. It could be even difficult to know who exactly is taking care of a given patient, since that there are cases in which the physician ID used in a claim is from a professional registered in the health care provider system, but the one taking care of the patient is an unregistered physician. Moreover, claims contain no information about referrals and often the connections among service provider team members are not recorded explicitly in claims. It is also well known that claims data is riddled with errors and unreliable information. Despite all those difficulties, we show in this paper that meaningful and reliable insights about the flow of patients through the network of physicians and how physicians refer each other can be inferred from claims data.

This study took place during a partnership, in a short-term project, between our lab and a major Brazilian health insurance company involving the analysis of their claims database. The main challenge the insurance company brought to our team was to identify physicians that excel at medicine by using the claims database. After decomposing this challenge considering the health insurance workflow, the following components were identified as key factors for outstanding professionals, defined by the health insurance company: physicians referred by peers, relative importance in the network of physicians, and returning behavior of patients. The need for modeling referrals emerged during interactions with the health insurance company staff as a way of identifying physicians that excel in a certain specialty and are referred by their peers. These interactions occurred weekly, fomenting discussions between our team and subject matter experts, IT specialists, analytics team, and the superintendent of the health insurance company. Hence, this work aims at presenting a way of modeling mutual referrals in a physician–physician network, which connects to the hypothesis studied in this work: H1) It is possible to identify underlying physicians' referrals from claims data.

The main contribution of this work is a way of modeling mutual referral patterns in the physician–physician network. The method can therefore be used by health insurance companies to better manage the physicians they have businesses with, nurturing the experience of registered physicians and inviting unregistered physicians that collaborate with registered ones. It can also be used to support patients to receive more integrated care from a group of physicians and service providers.

This paper is organized as follows: section 2 describes the related work, section 3 details the database analyzed, section 4 presents the proposed model for highlighting mutual referrals, section 5 discusses the obtained results, and section 6 concludes and points to future works.