The Impact of Physician Networks on Treatment Decisions

The Network Effect

Physicians can have several options when prescribing medications to treat chronic conditions. How do they choose between a new, name-brand drug or a more familiar alternative? Physician decisions like these impact more than just patient outcomes: According to the Centers for Medicare and Medicaid Services, U.S. prescription drug spending in 2015 reached $324.6 billion.

"We have very large healthcare costs in this country and very rapid growth, and it's not necessarily bad," says Seth Richards-Shubik, assistant professor of economics. "Spending money on health is fundamentally a great thing. It saves lives. So it's a matter of controlling or improving it to the extent that we save as many or more lives, but perhaps we spend the money more efficiently."

Supported by a four-year, $3 million grant from the National Institutes of Health (NIH), Richards-Shubik and co-investigators from the University of Pittsburgh and Harvard University are trying to understand the role of physician networks in how doctors select patient treatment options.

In what Richards-Shubik calls an informational contagion, ideas spread across a network of physicians, who learn from one another about treatments and their benefits and risks. For better or worse, these interactions then influence the treatments those physicians choose and the drugs they prescribe.

Richards-Shubik and his colleagues are looking at five drugs, all introduced in recent years and used to treat cardiovascular disease or diabetes. Using prescription data for nearly all of Pennsylvania's 50,000 physicians as well as Medicare and Medicaid claims data for more than 2 million adults in the state, they are measuring the rate and speed of adoption of these drugs among these physicians. They are also examining the influence of various factors on physician adoption of new medications: physician characteristics, patient characteristics, institutions and different types of professional networks (those formed during training, at the same hospital or within the same medical group, and, in a more novel approach, those who share patients via referrals).

If you see enough common patients between two physicians, Richards-Shubik says, you can infer there is some kind of relationship between them. This establishes what's called a patient-sharing network.

"Once you have these networks, then quantifying the peer influence is a fairly straightforward application of econometric techniques," he explains. "What you're trying to do is take a particular physician's peers at the hospital and those that the physician shares patients with. What proportion of them have started prescribing the new drug? How does that influence the physician's own propensity to prescribe it?"

The team can then measure the strength of that influence and determine, based on that magnitude of influence and on the established network, the more influential players in the network.

"We're finding a big influence from the colleagues that physicians share patients with," says Richards-Shubik. "We're seeing ways where you could identify influential physicians even better using this network information."

Understanding the role of these physician networks can help determine effective interventions like academic detailing, which allows trained health-care professionals to educate prescribers about treatment options, potentially helping them make choices that could improve patient outcomes or reduce costs—or perhaps even both.

"I'm not saying this is an easy slam dunk," says Richards-Shubik. "But to me it's a very important road to try to go down."

This story appears as "The Network Effect" in the 2017 Lehigh Research Review.