Understanding the Dual Nature of Physician Preferences

Saving Lives or Saving Money? Understanding the Dual Nature of Physician Preferences Abstract This paper develops a model of joint physician and patient decision making. We highlight the interaction between two competing incentives: altruism and imperfect information. Altruism of healthcare providers towards their patients encourages providers to represent the interest of their patients while imperfect information encourages profit‐maximizing behavior. Our model provides an explanation for the observed heterogeneity in pricing dynamics across procedures: for some procedures, increasing Medicare prices increases use, while for other procedures, increasing Medicare prices reduces use. We use our model to derive normative implications for patient welfare when utilizing payment reform to affect quantity. I. Introduction Economists have long emphasized the peculiarities of healthcare markets, compared to other markets for goods and services. Since at least Kenneth Arrow’s pioneering paper on the subject, economists have recognized two features in particular: the altruism of healthcare providers towards their patients, and the reliance of patients on their physicians for information and guidance. Less well recognized is the inherent tension between these two aspects of healthcare provision. Altruism encourages physicians to represent the interests of their patients. For example, an altruistic physician will tend to economize on the use of scarce inputs and attempt to maximize the utility of patients subject to their own resource constraints. However, the informational advantage of physicians creates a classic agency problem that physicians might exploit to pursue their own interests instead of their patients’ interests. (See, for example, Emanual and Emanual, 1992.) Little effort has been devoted to understanding the interaction between these two competing incentives and their implications for healthcare markets. Figure 1 provides some initial insight into the importance of this issue. The figure depicts the histogram of price elasticities within Medicare that experienced an approximately 50% increase in prices – such large and sudden changes are unlikely to reflect changes in demand, but instead reflect movements along a demand curve. Price increases coincided with increased quantity in half the cases depicted, but with decreased quantity in the other half. This pattern is difficult to explain when relying on either agency problems or altruism alone. Physicians exploiting agency problems would drive utilization higher in the event of higher reimbursement. Physicians behaving altruistically would drive utilization lower in the event of higher prices, part of which are borne by patients. What is needed is a unifying framework that considers how altruism and agency problems interact to drive healthcare markets. Significant policy questions are at stake, since price is often viewed as an important lever for influencing behavior. In some market segments, for example, higher physician reimbursements can be expected to curb utilization, while in others, the opposite effect obtains. In this paper, we study how altruism and agency problems compete, and the positive and normative implications for healthcare markets. From a positive standpoint, we show that exogenous price changes may increase or decrease quantity supplied. When higher prices lower quantity, we say dynamics are primarily “patient‐driven,” and when the opposite is true, we say they are primarily “physician driven.” Moreover, the degree to which markets are patient‐driven or physician‐driven endogenously depends on physician incentives. Specifically, pricing is more likely to be patient‐driven when patients are poorer, and when healthcare provision is less profitable. In other words, physician altruism is more likely to win out when the value of behaving altruistically is higher and the cost is lower. From a normative standpoint, consumer welfare is maximized when behavior is patient‐ driven. Thus, we expect less inefficiency, from the consumer’s perspective, when consumers are poorer, patient cost‐sharing is higher, input prices are lower, and profitability is lower. Increases in wealth, therefore, are expected to increase “waste” in healthcare, as are expansions in the availability of insurance. Our paper offers several new insights to the existing literature. First, we build upon the literature examining supplier‐induced demand by offering a patient‐driven counterpart (Dranove, 1988; Pauly, 1980; McGuire 2000). Second, we show how price increases can lead to quantity reductions, even under a fee‐for‐service reimbursement system. There have been some papers analyzing how different reimbursement schemes can lead to quantity reductions. For example, Ellis and McGuire (1986) show that having a prospective‐payment system can lead to too few services being provided if physicians undervalue the benefits of patients relative to hospital profits. Glied and Zivin (2002) discuss how managed care can restrict quantity. Third, our paper relates to work by Dickstein (2014), who quantifies empirically the contributions of patient and physician incentives to prescription drug utilization. We show how physician self‐interest coexists with altruism, and how the interaction of these incentives affects care utilization across patient and procedure types. Understanding how quantity responds among these different dimensions can be important for informing how supply‐side and demand‐side reforms that are effective at containing costs. While policymakers have traditionally focused on amending the method and amount Medicare pays its providers, demand‐side changes to patient cost‐sharing and supplemental insurance are currently being debated (Gruber, 2013; National Commission on Fiscal Responsibility and Reform, 2010; and Zuckerman et al., 2010). The rest of the paper is organized as follows. In Section 2, we propose a theoretical framework for the joint decision‐making between patients and physicians, and we derive the normative implications from our model. In Section 3, we discuss the empirical approach for testing conjectures derived from the model. In Section 4, we present the empirical results. Section 5 concludes. 2. Theoretical framework Due to the nature of joint decision making between patients and physicians, own‐price effects within Medicare may be of any sign, and the same is true of cross‐price effects. We demonstrate this potential ambiguity in a simple theoretical model that traces back to Becker (1957). 2.1. Simple illustration For pedagogical purposes, we first illustrate in a very simple, perfectly competitive model how physician and patient decisions interact. Here, we presume that physicians earn zero economic profits, and patients bear the full cost of healthcare. Suppose health is produced using a good or procedure procedure S, according to , , where 0, 0, and and a substitute good or 0. We also suppose that these two goods are initially health‐improving, but eventually health‐reducing if overused. In this context, suppose for simplicity, that a fully informed representative patient maximizes the value of health net of the cost of production. This results in the following household production function for health: max , , It is straightforward to show in this context that the derived demand for but rising in price is falling in price . Now, however, suppose that the representative patient is not fully informed but instead receives care from a physician, who bears cost , , where 0, 0, and 0. The latter assumption rules out cost complementarities between the procedure and its substitute. The physician maximizes a weighted average of patient well‐being and her own income, as in: max 1 , , The parameter , is an index of altruism. Observe that the physician’s objective function can be rewritten as: max , , 1 , 1 2 This has the following first‐order conditions: Define 1 1 2 0 1 1 2 0 as the determinant of the Hessian matrix for this maximization problem. This allows us to write the comparative statics of the problem as: 2 1 1 2 1 1 If the problem is strictly concave at the optimum, then 0. As a result, if , the own‐ price elasticity is negative and the cross‐price elasticity is positive, because the physician is sufficiently altruistic to weight patient preferences enough that her decision problem resembles that of the fully informed patient. We call these “patient‐driven pricing dynamics.” If, on the other hand, , the opposite dynamics prevail: the own‐price elasticity is positive, and the cross‐price elasticity is negative. We call these “physician‐ driven pricing dynamics.” 2.2 General model The derivation above assumed physicians and patients are risk‐neutral over consumption. It also abstracted from the existence of health insurance. More generally, suppose the representative patient derives utility where is income, , is an ex ante insurance premium, and , ; , ), represents patient out‐of‐ pocket expenditures. Here and elsewhere, we abstract from effects of physician decision‐ making on the insurance premium. This assumption sacrifices little generality in a public insurance scheme or when studying a relatively small set of procedures. Now suppose physicians derive utility from a weighted average of patient utility, and their own utility over consumption, ⋅ , where and are weakly concave utility functions. Assume the utility functions satisfy the assumptions of monotonicity, risk‐ 0, aversion, and weak prudence, as in 0, and 0 (Felder & Mayrhofer, 2011). The generalized physician objective function can then be written as: max 1 , , , , ; , The first‐order conditions now become: ∗ 1 ∗ 0 ∗ 1 ∗ 0 The optimality conditions are weighted averages of physician profit‐maximization and patient utility‐maximization. To simplify the analysis, we follow the convention adopted in much of the insurance literature and abstract from the direct income effects associated with patient out‐of‐pocket payments (Lakdawalla & Sood, 2013). This amounts to holding ′ fixed when prices change. Define , , and as the respective second derivatives of the physician’s objective function, all three of which are negative. The comparative statics now become: 1 1 1 1 ′ These comparative statics suggest a simple empirical test for the presence of physician altruism. Absent altruism, own‐price elasticities will always be positive. To see this, observe from the physician’s optimality condition that in the absence of physician altruism. Therefore, without altruism, it follows that 0. If price increases lead to quantity reductions in real‐world data, this necessarily signals the presence of altruism. To develop further the implications of the comparative statics, note that pricing dynamics are patient‐driven if 1 ′ 1 . We now show that the right‐hand side of the inequality is always positive and decreasing in physician at the consumption. To do so, we require the relatively weak assumption that optimum. That is, patients will not accept an equilibrium in which their own marginal out‐ of‐pocket cost exceeds their own private marginal benefit. The asymmetry of information means this is not a trivial assumption, but the empirical evidence for moral hazard justifies its use. This assumption, coupled with the first‐order condition for , implies that Thus, 1 has (weakly) the same sign as 1 with the prudence assumption ( term 1 ′ 1 ′. This result, coupled 0), implies that increases in income will reduce the . With this condition in place, we can conclude that pricing dynamics are more likely to be patient‐driven if: 1. Physician altruism is higher – i.e., is higher; 2. Patient income is lower – i.e., is lower and ′ is higher; 3. Patient health care spending is higher – i.e., is higher, and ′ higher; 4. Physician income is higher – which implies that ′ and 5. The physician’s price‐cost margin, are lower; , is lower. Intuitively, pricing is more likely to be patient‐driven if: physicians care more about their patients (#1); patients are more sensitive to spending growth (#2 and #3); physicians are richer and willing to pay more to purchase patient welfare (#4); and the opportunity cost to physicians of boosting utilization is lower (#5). . 2.3. Normative implications The simple model above has several implications for patient welfare. Even in the presence of physician altruism, Pareto‐efficiency continues to be characterized by the standard input . Thus, we can characterize the degree of inefficient overuse efficiency conditions, by quantifying . By inspecting the first‐order conditions for physician decisionmaking, we can derive: 1 1 1 There are two sources of input overuse. The first is moral hazard, which obtains when social marginal costs exceed out‐of‐pocket costs. The second is over‐reimbursement, which obtains when physician reimbursements exceed the marginal value of care to patients. (If marginal value of care exceeds pricing, this tends toward underuse instead.) The overall degree of input inefficiency is the weighted average of these two sources, with the weights given by the relative importance of patient versus physician consumption. If physicians are perfectly altruistic, the over‐reimbursement effect vanishes. On the other hand, if they are perfectly self‐interested, the moral hazard effect vanishes. In addition, note that increases in physician consumption levels place more weight on the moral hazard effect, because richer physicians place more value on their patients’ consumption than their own. The relative importance of physician versus patient consumption has implications for which policy levers are most efficient at reducing distortions. If the degree of altruism is high, containing moral hazard by shrinking is relatively more important. If low, on the other hand, aligning physician reimbursements with the value of health improvement becomes more important. More formally, holding all patient and physician incentives constant, reimbursement reforms that compress when 1 will contribute less to efficiency ′, and vice‐versa. With a bit more structure, we can link the normative analysis to the positive predictions of the model more directly. Assume that physician income effects associated with price changes in a procedure are small. This would be true if net income from the procedure, , is relatively small. It would also be true if physicians are close to risk‐neutral, as would be the case if they were employed by large health systems or provider organizations. If any of these conditions obtain, we can ignore the term and derive that pricing is patient‐driven if and only if: 1 1 Holding constant patient cost‐sharing ( , this expression implies that is higher when pricing is patient‐driven than when it is physician‐driven. Since the degree of inefficient overuse is increasing in , we have shown that inefficient overuse is reduced when HCPCS tend to follow a patient‐driven pricing behavior. From this analysis, we can conclude that any given physician reimbursement reform will have a smaller effect in market segments where pricing is patient‐driven than in segments where they are physician‐driven. Here, physician reimbursement reforms refer to changes that move reimbursements closer to social value. By the same logic, patient cost‐sharing reforms will have a smaller effect in market segments where pricing is physician‐driven than elsewhere. 3. Empirical Analysis The theoretical analysis generates at least five testable implications which we test: 1. Both the size and even the sign of own‐price and cross‐price elasticities may vary when there is joint decision making between patients and physicians. 2. When patient income is lower, price elasticities are more likely to reflect patient‐driven pricing behavior. 3. When patient cost‐sharing is higher, price elasticities are more likely to reflect patient‐ driven pricing behavior. 4. When the physician’s price‐cost margin, , is lower, price elasticities are more likely to reflect patient‐driven pricing behavior. 5. Physician payment reforms have a larger effect in market segments where pricing is physician‐driven than elsewhere. 3.1. Data We rely on the Center for Medicaid and Medicare (CMS) Medicare Carrier Claims File (CCF) and the Medicare Current Beneficiary Survey (MCBS). The CCF data contains the fee‐for‐ service Physician/Supplier Part B claims for a random sampling of Medicare enrollees. When examining data from 2003‐2011, we use the 20% claims data. Because the 20% sample is unavailable prior to 1998, we use the 5% sample when examining data from 1993‐2005. All prices are converted to $2010 using the Current Price Index for medical expenditures. We restrict the CCF sample to procedures with an “identifiable” substitute. For each Healthcare Common Procedure Coding System (HCPCS) code assigned to a procedure, we identify a substitutes using CMS’ description of the HCPCS. These descriptions consist of one to four words, and they break the universe of our data into categories ranging from one HCPCS per category to sixteen HCPCS per category. Among the 30% of HCPCS with at least one identifiable substitute, we calculate a quantity‐weighted average price across all substitutes. While the CMS descriptors provide a systematic way of identifying similar HCPCS, it is possible that our definition insufficiently captures close substitutes or groups together HCPCS which are not substitutable.1 Defining a market by the Dartmouth Atlas’ hospital referral region (HRR), we collapse the data to the HCPCS‐year‐HRR level and focus on a balanced panel to avoid variation from the deletion and addition of HCPCS over time. The MCBS data consists of a smaller, but still nationally representative, sample of 12,100 Medicare beneficiaries. By combining patient surveys with administrative payment files, the MCBS data provides a richer set of socioeconomic and demographic covariates. This feature allows us to consider additional patient‐related characteristics, such as income and 1 Type‐I error is occurs when for example, HCPCS 11400 , described as “exc tr‐ext b9+marg 0.5