Validating A “Big Data” Algorithm For Discerning Physician Networks

Main Image: 
The largest connected network of physicians in the survey: (A) shows the relationships among the sampled physician population (including nonrespondents) as measured by the relationships presented in the survey; (B) shows the same network, with the addition of all relationships measured by administrative claims.

The goal of this project was to develop and test a survey to measure referral partners and information exchange partners within a sample of primary care and specialist physicians. To accomplish this, the study team examined the relationship between a measure of the number of shared Medicare patients between two physicians (ascertained using Medicare claims data) and the presence of a referral or information exchange relationship (reported by physicians). Patient sharing was assessed directly using Medicare claims as a form of “big data”; and traditional survey methods were used to assess perceived professional relationships with respect to referrals.

In one publication resulting from their work, the investigators reported that the probability of two physicians having a recognized professional relationship increased with the number of Medicare patients shared, with up to 82% of relationships recognized at a threshold of 9 shared patients, overall representing a diagnostic test with an area under the receiver-operating characteristic curve of 0.73 (95% CI: 0.70–0.75). Primary care physicians were more likely to recognize relationships than medical or surgical specialists.

These findings validated a method that has been used in research to map networks of physicians, namely that using claims and the metric of shared patients is a valid way of discerning ties between physicians. Using this “big data” method is much more complete; it is also cheaper and faster. Methods developed in this pilot project led directly to an additional publication examining the relationship between patient-sharing networks and the cost and intensity of medical care delivery and substantially informed other work in this area, including in a landmark paper in JAMA mapping the network of doctors in the entire USA, all among the first publications on physician patient-sharing networks in the field.

Pilot Project Leader: Michael Barnett, MD, Harvard Medical School