Physicians’ role in the health care industry is to bring expert information to the doctor-patient relationship. However, surveys suggest that they typically lack knowledge on a key type of information: the cost of care (Shulkin, 1988, Tierney et al., 1990, Reichert et al., 2000, Allan and Innes, 2004, Allan et al., 2007, Allan and Lexchin, 2008, Okike et al., 2014). Acting as agents for patients, doctors heavily influence health care consumption choices, so their dearth of knowledge on the costs of these choices could result in inefficient allocation of health care resources. Despite this, there have been few efforts to provide physicians with information on costs. The lack of highly credible research to guide such efforts may be one reason for this.

This paper presents a field experiment investigating whether providing specialists’ cost information to primary care physicians (PCPs) affects their referrals to those same specialists. Collaborating with a group of medical practices, I created a report on average relative costs for ophthalmology practices and distributed it to randomly selected PCP groups. I then measured the effect of this intervention using administrative referral and claims data, comparing the subsequent ophthalmology referrals of the PCPs that received the report to those that did not.

The experiment took place in 2014 in a western American state among a group of medical practices organized together as an Independent Practice Association (IPA). The IPA provides health services to customers of health insurance companies and, informally speaking, serves as a middleman. It receives payment from insurers, then distributes those payments to the physicians who are the actual service providers. When a patient is referred to a specialty, the IPA pays the specialist (who bills the IPA) in accordance with the patient’s insurance coverage. The costs included on the report distributed to PCPs in this experiment were averages of the total charges to the IPA from the ophthalmologists after receiving new ophthalmology referrals. As I discuss further below and in Section 3.1, differences in the costs reflect variation in treatment intensity across ophthalmologists, not differences in unit prices for specific services.

Changes in referral patterns are measured using IPA administrative data, which identifies when a referral is made, who sent it, and to whom it was made. I estimate effects for two different types of patients. For the first, the PCPs have a financial incentive to refer to less costly specialists. For these patients, I find that the treatment group PCPs increased the share of their referrals sent to less expensive ophthalmology practices by 4.6 percentage points for each reduction in costliness rank during the first two months after treatment. This effect dissipated over the following four months. In contrast, I find little evidence of a response to the treatment for the second type of patient. For these patients, the PCPs’ financial incentive to refer to cheaper specialists is muted because the IPA reimburses ophthalmologists for treating these patients primarily on a flat-rate basis. This heterogeneous result is consistent with cost reduction motivating the PCPs’ responses. Ultimately, the change in referral patterns for the first type of patients (where financial incentives were stronger) resulted in a reduction in average treatment cost for a referred patient by roughly $80 in the first two months after distribution of the cost information. This represents a decrease of about 45% of the pre-intervention referral cost.

My work in this article is most closely related to Ho and Pakes (2014), which retrospectively studied how the allocation of women giving birth across hospitals was affected by the hospitals’ relative costs. The authors found a negative effect for cost on the likelihood a woman would deliver her baby at a given hospital. The effect grew in magnitude for women whose PCPs had a financial incentive to send their patients to lower cost hospitals. I advance their work along three dimensions. First, while their research design relied on cost variation across patient-condition-by-hospital cells and did not have a control group, I rely on randomized assignment of cost information to physician subjects across treatment and control groups. Second, their analysis relied on claims data to infer physician referrals, but I use the IPA’s administrative data that recorded actual physician referrals. Finally, Ho and Pakes do not explain how physicians would obtain a detailed understanding of expected hospital costs at the patient-condition level, an important question in the face of the surveys cited above that show doctors being unaware of costs. The distribution of the report on costs to the PCPs in my project provides a clear answer to this concern. Despite these differences, my results are consistent with theirs: PCPs refer more to lower cost providers when incentivized to do so. The combined studies, therefore, provide strong evidence that physicians respond to cost information in making referrals.

This paper also adds to a literature on physician price transparency. In these studies, physicians are provided information on prices, typically those paid by patients or their insurance companies, for diagnostic lab tests, imaging, or medications. Two articles are particularly relevant as they are based on randomized assignment of price information at the physician level (the same level of randomization this paper uses). Tierney et al. (1990) studied orders for diagnostic tests and found price transparency reduced test volume and cost of testing. Monsen et al. (2019) investigated prescriptions, finding price information led to doctors reducing their prescribing of three of four high-cost medicines studied.1 Relative to these studies, this paper makes three main contributions.

First, there is an important difference in the type of information being provided to physicians between this study and those above. In those studies, physicians were given unit prices for individual services or products, meaning the cost of buying (for example) one blood test or medication. Variation in cost, therefore, came from different mark-ups across providers or because the services or products were themselves different and not necessarily substitutes (such as blood tests that diagnose different conditions). In contrast, the ophthalmology practices in this study all provide a similar set of services, and because this project took place within the IPA, for any given patient the unit price of a service is the same for all specialists. Moreover, the costs were also risk adjusted for differences in underlying patient populations before being presented to the PCPs. Thus, costs vary across specialists because of variation in the types of treatments they use or recommend for patients. This study, therefore, adds to the price transparency literature by showing that physicians will respond to costs when they convey information about treatment intensity.

Second, this study provides new evidence that cost transparency affects physicians’ referral behavior. Referrals have an important place in modern health care. Their use by physicians has increased significantly, and recent work has shown that referrals influence where health care is received (Barnett et al., 2012b, Ho and Pakes, 2014, Baker et al., 2016, Chernew et al., 2021). Moreover, since a referral is an interaction between two humans, there are social scientific aspects of referrals that are not present in doctors’ decisions to order tests or prescriptions. For example, PCPs may have social relationships with the specialists to whom they refer and consider them friends or colleagues. They may consider how changes in their referral patterns would affect the specialists’ welfare. These types of relationships, therefore, may make PCPs less likely to change their referrals in response to cost information. This paper performs the first test of physician price transparency in the context of such social factors.

Third, the authors of the transparency studies above use concern over high healthcare spending to motivate their work, but they do not discuss mechanisms, such as physician incentives, as to how cost transparency might affect spending. This leaves ambiguous the underlying process these interventions are supposed to be testing, and it is unclear whether their findings could be extrapolated. In my analysis, I clarify the doctors’ financial incentives and provide evidence they influenced the PCPs’ responses to the cost information. This provides a coherent interpretation for the PCPs’ behavior and suggests a manner in which cost transparency could be used in other contexts to help reduce health care spending.

This paper also contributes to a more general literature on how physicians make referrals. Conceptually, a referral could be broken into two steps. First, a PCP must decide to make a referral to a specialty. Surveys of physicians have indicated that, at this stage, referrals are most often made for advice on diagnosis or treatment, they depend on patient symptoms or conditions, and they are more likely for uncommon medical conditions (Donohoe et al., 1999, Forrest et al., 1999, Forrest and Reid, 2001, Forrest et al., 2002, Forrest et al., 2006). Second, having decided to make a referral, a PCP must decide to which physician within the specialty to refer. Evidence suggests at this stage that PCPs weigh their previous experiences with specialists, appointment availability, specialist communication quality and relevant skills, previous patient experiences, and PCPs’ own pre-dispositions towards particular specialists (Forrest et al., 2002, Starfield et al., 2002, Anthony, 2003, Kinchen et al., 2004, Barnett et al., 2012a, Hackl et al., 2015). My investigation here shows that specialist cost information can play a role in the referral process at this second stage, serving as one of several key considerations in PCPs’ decision making.

Lastly, this paper also relates to recent work studying fee-for-service reimbursement versus capitation (e.g., Shafrin, 2010, Hennig-Schmidt et al., 2011, Ho and Pakes, 2014, Johnson and Rehavi, 2016, Brosig-Koch et al., 2017, Chalkley and Listl, 2018), referral fees (Waibel and Wiesen, 2021), and vertical integration of physician practices and referrals (e.g., Baker et al., 2016, Whaley et al., 2021, Richards et al., 2022, Whaley and Zhao, 2024). This last literature on vertical integration is particularly relevant because it tends to find that vertically-integrated physicians refer more often to higher-priced providers within their integrated systems. These results are typically interpreted as physicians enriching themselves or their systems via these referrals. However, it could be the case that higher cost providers are higher quality along some unobserved dimension. Or, it could be, at least in principle, that physicians merely refer more to higher priced providers as a rule, irrespective of their personal financial incentives. My paper provides evidence inconsistent with these interpretations, since I show PCPs sending referrals to lower priced providers when it is in their financial interest. This shows PCPs do not merely to refer to higher priced providers as a rule. Moreover, it also implies that they do not think higher cost means better quality. If they did, the estimates to this experiment would have had the opposite sign. This leaves enrichment as a more likely explanation for the findings in the vertical-integration literature.