Economics has a long tradition of studying asymmetric information between market participants and the inefficiencies that might consequently arise. In the health care context, the interaction between physicians and patients has received special attention. Physicians, the rationale goes, are better informed about the spectrum of treatment alternatives and medical technologies (e.g., “wait and see”, surgery, medication) than the patients (e.g., Gottschalk et al., 2020, Johnson, 2014, McGuire, 2011, McGuire, 2000 for surveys of the literature). In an ideal world, physicians would act as “perfect agents” for their patients and simply assist them to demand the quantities of various types of care that patients would have chosen if they had access to the same information and knowledge as the doctors (Pauly, 1980). In the real world, however, physicians need to balance private interests (e.g., income goals) and patient welfare (Ellis and McGuire, 1986). Reimbursement systems that financially reward the quantity of care might therefore distort treatment decisions as some physicians may be willing to trade-off patient benefits for a higher income. Whether physicians actually engage in such behavior ultimately remains an empirical question.

Previous work typically exploits changes in physician fees to analyze the doctors’ responses to financial incentives (e.g., Clemens and Gottlieb, 2014, Jacobson et al., 2010, Gruber et al., 1999, Yip, 1998, Nguyen and Derrick, 1997). The more recent study by Clemens and Gottlieb (2014), for example, examines the consolidation of geographic regions through Medicare in the late 1990s and provides evidence for the presence of substitution effects. Physicians tend to shift quantities from areas that experience fee reductions to services with fee increases to compensate for income losses. Allowing physicians to earn money by selling drugs through their in-house pharmacy to patients - the setting we study in this paper - is equivalent to a fee increase providing an incentive to increase the supply of drugs. Our setting, however, differs from previous studies as physicians may profit from the fee increase even without changing treatment decisions. In case of pharmaceutical prescribing, physicians often face the choice among several product alternatives for a particular diagnosis that provide identical medical benefits to patients but may differ in the financial reward to the physician. To illustrate, consider the high-market share hypertension drug “Amlodipin” which is sold by seven different drug companies. Although the active ingredient is identical between suppliers, the markup differs not only between the brand-name product and the generics but also between generics. Therefore, physicians may increase their income by prescribing more expensive “brands” without sacrificing patient health at all. Engaging in such behavior is a textbook example of rent-seeking behavior: physicians extract a bigger piece of the pie without increasing medical productivity.

In this paper we analyze whether physicians engage in such rent-seeking behavior by exploiting a recent reform in two large Swiss cities. The reform allowed physicians to sell drugs to their patients (through so-called “physician dispensing”) and opened up an additional income channel to them: while only pharmacies earned a markup on drugs before the reform, these markups are now pocketed by the physicians themselves. So even in the absence of behavioral changes, physicians’ earnings are increased by the substitution of markups from pharmacies to physicians. Besides pure substitution, the changes in the incentive structure might also trigger behavioral changes. For one, physicians might choose to prescribe more drugs, potentially sacrificing patient benefits to some extent. This could be both at the extensive (more patients are medicated) or at the intensive margin (patients already treated with medication receive more drugs). On the other hand, they might prescribe more profitable brands without changing the treatment, which would impose additional costs on the patients and the health care system without generating improvements in the treatment quality.

Unlike previous work on drug dispensing (see below), we are able to exploit the variation created by a policy change which allows physicians to dispense drugs. The reform divides patients into a treatment (having a dispensing physician) and a control group (having a non-dispensing physician). All of these patients had a non-dispensing physician prior to the reform. Since dispensing physicians systematically differ in attributes like practice style, adherence to medical guidelines and costs per patient from their non-dispensing counterparts, we dot not base identification on direct group comparisons. Instead, we make use of a special feature in our data: We observe the interaction between patients and physicians both before and after the reform and see the drugs that doctors prescribe to their patients. Specifically, the data offers information on the number of packages, the markups, the costs and the price category of drugs physicians prescribe to the same patients before and after the introduction of dispensing. This data structure has the advantage that we are able to directly address selection biases as we are able to condition on existing physician-patient relationships. For a selection of high market share hypertension and reflux drugs and an antibiotic, we further observe the substance class, the drug brand (e.g., generics or brand-name), the dosage, and the number of pills on each prescription. As a stylized example, suppose Patient A suffers from hypertension and is regularly prescribed the 4-month package (120 pills) of the generic drug “Amlodipin” in the years before the reform. After the reform, Patient A still receives the same active ingredient, but now in the form of two 2-month packages (60 pills) from “Norvasc”, the brand-name drug. Given the current drug price regulation in Switzerland, such a switch is profitable for self-dispensing physicians. For this subset of drugs, we are able to precisely track such changes in physician prescription decisions.

To estimate the effects of dispensing, we apply a two-step procedure that combines balancing with difference-in-difference estimation. Specifically, we use entropy balancing, a preprocessing method similar to nearest neigbhor and propensity score matching that enables us to estimate sampling weights from the data that balance the covariate and pre-treatment outcome distributions between patients of dispensing physicians and patients of non-dispensing physicians. In contrast to other preprocessing techniques, entropy balancing has the advantage to directly produce weights that equalize a potentially large set of sample moments between the treatment and control group. Therefore, entropy balancing always (at least weakly) improves on the covariate balance achieved by other preprocessing methods (Hainmueller, 2012, Hainmueller and Xu, 2013). In this paper, we preprocess the data to make the common trends assumption more credible. Based on the re-weighted data, we then estimate difference-in-differences specifications that account for physician-patient interaction fixed effects and a rich set of patient characteristics (e.g., age, comorbidities).

Our paper contributes to the small body of empirical literature that analyzes the influence of drug dispensing regimes on prescription behavior for Taiwan (Liu et al., 2009, Chou et al., 2003), Japan (Iizuka, 2012, Iizuka, 2007), and Switzerland (e.g., Burkhard et al., 2019; Kaiser and Schmid, 2016; Trottmann et al., 2016; Rischatsch, 2014). In general, this literature finds that the dispensing physicians are responsive to markup differentials between generic and brand-name drugs and that dispensing significantly affects drug expenditures. Although closely related, this paper differs in three dimensions from prior work. First, the majority of previous studies does not attempt to estimate the causal effect of dispensing and instead reports associations between dispensing and prescription outcomes. Previous studies for Switzerland, for example, report a significant association between dispensing and generic substitution (Trottmann et al., 2016; Rischatsch, 2014; Rischatsch et al., 2013). In addition, because the dispensing status of physicians is often not directly observed and has to be imputed the corresponding findings may be biased due to measurement error. Second, previous studies for Switzerland who ask the causal question impose strong identifying assumptions3 and thus the results might be biased due to differences in unobservables. In contrast, we explicitly address selection issues by exploiting the exogenous variation created by the reform. Third, we extend the analysis by analyzing possible channels through which physicians adapt their prescription behavior in response to dispensing.

Our analysis yields four main results: First, we find that dispensing leads to a significant increase in drug costs of about CHF 20 (＋4%–5%) per patient and year. This finding implies a behavioral response of physicians as we would not observe an effect on total drug costs if physicians simply dispensed the drugs for which they previously wrote prescriptions to be filled at pharmacies. Second, we provide evidence that dispensing results in an increase in physician earnings by approximately CHF 30 per patient. We show that the increase in earnings is composed of a substitution and a rent-seeking component. The latter is the result of a behavioral response and we estimate it to be around one third of the revenue increase. Third, we explore possible channels through which the behavioral response we observe may operate. Our analysis is indicative that two channels drive our main findings: (a) physicians substitute larger with smaller packages because the current price regulation of pharmaceuticals generates higher markups per pill for smaller package sizes. Our analysis is compatible with such a “package size channel” as significantly more packages are prescribed to patients of dispensing physicians, while simultaneously we find no evidence for a dosage change among the same patients; (b) physicians prescribe more profitable drugs to patients (the “cherry-picking channel”) as also documented by Iizuka (2012). Fourth, we provide evidence for considerable effect heterogeneity. We show that dispensing unilaterally affects the behavior of classic fee-for-service physicians, while HMO physicians do not respond to dispensing. Furthermore, our analysis suggests that dispensing physicians “target” less price sensitive patients with low deductibles as we solely find effects in this patient segment (i.e., no effect for higher deductible groups).

Overall, our analysis provides evidence that profit motives indeed influence the prescription decisions of physicians. However, we do not find evidence for physicians trading-off patient benefits for profits, a result regularly found in the previous agency literature. Instead, our results indicate that physicians act in the patients’ best medical interest - they do not unnecessarily expand medication or alter dosage decisions - but at the same time, at least some of them also do not shy away from “playing the system” and earn some extra dollars. Hence, even if physicians do not seem to be willing to trade in patient health for profit, they engage in rent-seeking behavior leading to wasteful resource use and avoidable costs for the health care system.

Besides the economic relevance of the problem, our findings are also relevant for policy makers. Drug expenditures amount to about 15%–20% of total health care spending across the OECD in 2017. Switzerland and the US and have the highest drug expenditures per capita in the world with more than $1000 (OECD, 2018). In light of this, the question arises whether it is a smart idea to expose physicians to additional monetary incentives. This paper gives a clear answer. Efficiency gains can be achieved by eliminating the counterproductive financial incentives introduced by dispensing. Whether these low-hanging fruits are “harvested” lies in the hands of policy makers.