Competition, Gatekeeping, and Health Care Access

Competition, Gatekeeping, and Health Care Access Abstract We study gatekeeping physicians’referrals of patients to specialty care. We derive two theoretical results when competition in the physician market intensi es. First, physicians refer patients to specialty care more often due to competitive pressure. Second, physicians earn more by treat-ing patients themselves, so refer patients to specialty care less often. We assess empirically the overall e¤ect of competition with data from a Norwegian survey in 2008-9 and Statistics Norway. From the data we construct three measures of competition: the number of open primary physician practices with and without population adjustment,and the Herndahl-Hirschman Index. The em-pirical results suggest that competition has negligible or small positive e¤ects on referrals overall. Our results do not support the policy claim that increasing the number of primary care physicians reduces secondary care. 1 Introduction Primary health care has generated many policy discussions. In many European countries, each inhabitant must be enrolled with a primary care physician to receive national health services. In the United States, Title V of the A¤ordable Health Care Act provides subsidies for the training of primary care physicians and allied health professionals (see http://www.healthcare.gov/law/full/index.html). Furthermore, Title IV of the Act promotes prevention, and it is expected that preventive care will be provided by primary care physicians. Primary health care is less expensive than secondary and specialty care, so the emphasis on primary care for cost control is understandable. Perhaps, the most explicit cost-control perspective is the primary care physician’s gatekeeping function. In many health plans in the U.S. and European countries, a patient can only obtain specialty care upon a referral made by his primary care doctor, also referred to as a gatekeeper. In this paper, we model the primary care physician’s referral decision, and empirically assess the relationship between physician market conditions and gatekeeping. A referral decision by a primary care physician or general practitioner (GP) likely depends on many factors such as medical conditions, current medical practice guidelines, availability of secondary care, the GP’s service capacity, and …nancial incentives. The current policy recommendation of increasing the number of GPs adds one more dimension to the complex referral decision. Given a population of patients, more GPs will ultimately mean a more competitive market for the doctors. This paper studies the relationship between competition in the GP market and a GP’s referrals of patients to specialist care. Such a study faces a number of di¢ culties. First, the number of GPs in any market changes slowly, even under any policy intervention. For example, subsidies in the U.S. A¤ordable Care Act are for physician training. This “natural” experiment will generate data only after many years, or perhaps even a decade. Similarly, in an experiment of a long duration, confounding factors a¤ecting referral decisions will change over time. These changes may be di¢ cult to track or unobservable to the analyst. Second, in a multi-payer system such as the U.S., di¤erent health plans use di¤erent incentive contracts. Referral decisions likely will be in‡uenced by these incentives. However, information on physician payment contracts is seldom available. Our strategy is to use a cross-sectional data set, which can be interpreted as a snapshot that 1 captures long-run changes, because di¤erent locations have had unique experiences for some time. (For example, di¤erences between two countries at a given point in time result from long-term cultural developments.) We use data from a 2008-2009 survey in Norway as our primary source, and supplement them with register data from Statistics Norway, and from the National Health Insurance Administration. Because data are collected over a one-year period, time-varying confounding factors are irrelevant. The details of the survey and data are in the next section. Here, we would like to point out that i) all self-employed Norwegian GPs are paid by the same …nancial contract, ii) 95% of all Norwegians GPs are self-employed, and iii) each Norwegian should be listed with a GP who is a gatekeeper for secondary care. In sum, problematic selection issues in multi-payer systems are avoided. In Norway, a GP either lets his or her medical practice be open or closed to new patients. We use the number of open practices (with and without population density adjustments) as a measure of competition intensity in the GP market. The GP market is more competitive when there are more open practices because consumers have more options and each GP faces a more elastic demand. We also use the more conventional Her…ndahl-Hirschman index as an alternative measure. Our use of the number of open practices in a geographical area as a measure of competition is unique. We are unaware of any other data set that contains similar information. Our empirical work seeks to explain specialty referral by competition intensity. We start with a model of GPs’referral decisions. As in much of the literature, we assume that a GP is guided by a pro…t motive and a concern for the patient. A GP’s practice style is how he values a patient’s potential bene…ts from specialty care and pro…ts from providing services himself. Practice styles are assumed to be a¤ected by market conditions. When the GP market becomes more competitive, the patient has more options. A GP who wants to retain a patient should adopt a practice style that values the patient’s bene…ts more. Competition may have a second e¤ect. As the GP market becomes more competitive, a GP has less patients. Therefore, the GP incurs less disutility when he treats the patient himself. For a …xed reimbursement rate the net pro…t from providing services increases. Competition in the GP market has two opposing e¤ects on referrals. More competition encourages a GP to show more concern for the patient, and therefore increases specialty referral. More competition also raises a GP’s net pro…t for providing service himself, and therefore decreases specialty referral. Our model o¤ers this new perspective, and we are able to assess empirically the 2 overall e¤ects of competition on specialty referrals. The data sets allow us to control for patients’socioeconomic status, age and gender, as well as self-assessed health and chronic illness conditions. We also control for general and specialty health care access at the market level. Our (logit and negative binomial) regressions also account for clustering at the municipality levels. We …nd that competition either has insigni…cant or positive e¤ects on GPs’ referrals for patients to specialty care. In other words, we …nd no evidence that more competition among GPs will reduce their specialty referrals. Our results do not lend support to the secondary-care-reduction e¤ect envisioned by a policy that promotes primary care. Our data do not let us estimate separately the two opposing e¤ects derived from our theoretical model. This, however, does not make our results less relevant. Our model of referral does capture the multi-faceted e¤ects of competition on referrals, and an increase in primary care physician density results in more than a single change. This is an important aspect of the complexity in physician-patient interaction. The literature on the primary and secondary health services is huge, whether that literature refers to health economics, health services research, or medicine. The health economics literature on the relations between primary and secondary care is smaller but growing. In any case, the interest in primary care and health cost is topical. Using U.S. data, Baicker and Chandra (2004), and Chernew et al. (2009) …nd that the percentage of primary care physicians in a market is negatively associated with Medicare’s reimbursement per bene…ciary. Chernew et al. (2009), however, …nd no correlation between the percentage of primary care physicians and the growth in Medicare spending; thus Medicare policies that seek to reduce spending levels, but not growth rates, will ultimately fail to address cost issues. Bradley Wright and Ricketts III (2010) use area-level data to show that within a location, a higher density of primary care physicians is associated with less inpatient admission and emergency room visits. Fortney et al. (2005) present results from a natural experiment at the U.S. Department for Veterans A¤airs, in which the number of primary care facilities were increased in some districts but not in others. Using a di¤erence-in-di¤erence analysis of longitudinal data and instrumental variables for potential endogeneity problems, they …nd that an increase in primary care encounters is associated with a decrease in specialty medical encounters. Fortney et al. conclude that primary care is a substitute for specialist health care. With survey data at the individual-patient level, Atella and Deb (2008) study whether primary care physicians and secondary specialists are substitutes 3 or complements. They estimate a structural simultaneous equation model where visits to di¤erent types of physicians are endogeneous. When unobserved heterogeneity is appropriately accounted for, they …nd that primary care physicians and specialists are substitutes. We model primary care physicians’referral decisions. The theoretical literature on referrals is quite rich. Barros and Olivella (2005) study cream skimming due to physicians working in public service self-referring patients to their own private practice. Biglaiser and Ma (2007) examine the welfare e¤ects of allowing dual practice and self-referrals. In our model the physician does not self-refer. Also, the referred specialists can reject referrals; this option has not been considered by existing papers in the literature. Allard et al. (2011) consider how referral to secondary care is a¤ected by incentive contracts for primary care physicians. Jeolvac (2003) compares optimal payment schemes with and without gatekeeping. Brekke et al. (2007) study the e¤ect of GP gatekeeping on equilibrium quality in an imperfectly competitive secondary care market. Gonzales (2010) investigates the interactions between patients and GPs when some patients are informed about whether specialty care is appropriate. Our paper does not deal with the issues in these four papers. Our model is parsimonious, and focuses on competition in the GP market. Yet, it derives a set of predictions that we have taken to data. The literature on competition in the health market is extensive; Gaynor and Town (2011) provide the latest review. It is fair to say that studies of competition have mainly focused on prices, qualities, costs, and health outcomes, and studies that use U.S. data outnumber those that use non-U.S. data. We are not aware of another paper that addresses the e¤ect of competition in the primary care physician market on secondary care referral. Our paper therefore is the …rst to o¤er some evidence on this issue. The common measures of competition in the literature are the number of providers (hospitals, physicians, nursing homes, etc.) within a geographical area, the n-…rm concentration ratio, and the Her…ndahl-Hirschman Index (see for example Wong et al. (2005)). In a patient-list system such as Norway, these measures do not capture the fact that patients can switch to another GP only if they can …nd open practices. We are unaware of any other study that uses the number of open GP practices as a measure of competition, with one exception Iversen and Ma (2011) show that more intense competition, measured either by GP open practices or GP’s desired practice sizes, signi…cantly leads to more diagnostic radiology referrals. 4 The rest of the paper is organized as follows. The next Section describes the Study Setting. A model of GP referral to either private or public specialist is set up in Section 3. Then we present our data set and descriptives in Section 4. The estimation results are in Section 5 while concluding remarks follow in Section 6. Finally, an Appendix contains the proofs of propositions. 2 Study setting Norway has a three-tier government structure. At the top is the state, the next tier consists of 18 county councils, and the bottom tier consists of 430 municipalities. Norwegians’ health care is covered by a decentralized national health services system. Municipalities are responsible for primary health care. Since 2002, secondary health care is the responsibility of the state. Each Norwegian is listed with a primary care physician, or a General Practitioner (GP). The Norwegian government is concerned about geographical distribution of GPs, and strictly regulates the entry of new GPs into municipalities. Licenses for GPs to practice at municipalities are allocated by the Directorate of Health. Patients may switch GPs twice a year, and in a year about 3% of the patients do so. Patterns of patients switching physicians vary considerably, and depend on physician characteristics (Iversen and Lurås, 2011). Almost all GPs (95%) are self-employed. A GP typically contracts with his resident municipality. A GP’s revenue from practicing medicine can be divided into three roughly equal parts. First, he receives a capitation fee from the contracted municipality for each listed patient. Second, he receives fee-for-service reimbursements from the National Insurance Scheme (NIS, a public scheme that is an integral part of the state account) according to a …xed schedule negotiated between the state and the Norwegian Medical Association. Third, a GP receives copayments from patients for o¢ ce consultations and tests. All fees and copayments are set at the national level, without any geographical variations. Hence, to GPS (and specialists) the payment system is entirely given. Almost all hospitals in Norway are public. Four Regional Health Authorities (RHAs) manage public hospitals. A number of private hospitals, both for-pro…t and not-for-pro…t, operate in Norway. Major private not-for-pro…t hospitals contract with the RHAs, and provide acute and elective care on the same terms as public hospitals. Private for-pro…t hospitals are very few and are geographically concentrated around the capital Oslo. The RHAs buy some services from private for-pro…t hospitals. The remaining services are paid for by private health insurance or by patients 5 directly. Patients receive specialty outpatient consultations at public hospitals or at private o¢ ces. Specialists working at public hospitals receive salaries. Most private specialists contract with RHAs. Such a contract gives a private specialist an annual practice allowance from an RHA and feefor-service reimbursements from the NIS. Private specialists are mainly located in urban areas. Approximately one third of all outpatient consultations are given by private specialists. A patient pays the same copayment whether the consultation is at a public hospital or a private o¢ ce. A GP is a gatekeeper. A GP must grant a referral before a patient receives specialty care at a public hospital or at a private o¢ ce operated by a specialist under an RHA contract. A referral allows a patient to go to several visits for a de…ned medical condition within a year. In 2009, the Norwegian government presented a major proposal for reforming the health care sector (Report No. 47 (2008–2009)). The coordination reform was to solve problems due to low care quality for patients with chronic diseases, and high care costs. A part of the proposal suggested a substantial increase in GPs and related resources. However, due to lack of evidence, the Norwegian parliament rejected this part of the reform. Our research, therefore, provides evidence for this policy discussion. 3 A model of referral A patient is under the care of a primary care physician, or a General Practitioner (GP). The GP has to decide between treating the patient himself and referring the patient to secondary care. A referral can be made to a Public Specialist or a Private Specialist, who work, respectively, in the public and private sectors. 3.1 Patient, GP, and Specialists The patient is fully insured, and delegates treatment decisions to physicians. Let u denote the patient’s bene…t from the GP’s treatment. This bene…t depends on a patient’s health status, and may take any value in an interval [0; L]. The GP observes this bene…t u before making the referral decision. The GP does not know how much bene…t the patient may obtain from secondary care, but believes that this is a random variable v. To simplify notation, we also let v vary on [0; L], and it has a distribution F , and a density f . We assume that the distribution of v is independent of u; 6 if they were correlated, we would simply replace the distribution F by a conditional one. Upon seeing the patient, a Specialist learns the value of v, as well as the value of u. Our interpretation is that the GP sends along the patient’s medical information to the Specialist, who can infer the bene…t u from primary care. There is a delay when a patient is referred to the Public Specialist, so the bene…t becomes v, 0 < < 1, if the patient is treated by the Public Specialist. Most public systems use waiting time as a rationing mechanism. This is true in Norway, and motivates our delay assumption. There is no delay when the patient is referred to the Private Specialist, so if the Private Specialist provides treatment, the patient’s bene…t is v. A private physician working in the private sector is paid according to a fee-for-service contract with a national insurance system. The private GP has a fee-for-service rate p, while the Private Specialist has a rate q. We interpret p and q as unit pro…ts, net of service costs. Very often we have p < q, so a Specialist receives a higher rate than a GP (although we do not use this inequality). The Public Specialist receives a salary. This di¤erence in payments implies di¤erent incentives for service provisions. Physicians behave as if their preferences are weighted average of pro…ts and patient’s (expected) bene…ts. Physicians practice medicine according to professional protocols, but also care about pro…ts. Alternatively, we can regard medical services as implicit or explicit bargaining outcomes between physicians and patients. Such outcomes (such as the Nash bargaining solution) are often a weighted average of a physician’s pro…t and a patient’s bene…t. In the literature a physician’s tradeo¤ between patient bene…t and pro…t is said to constitute his practice style. It is well-known that physician practice styles vary widely. De…nitive medical guidlines and protocols are not always available, and doctors operate under a grey area of medicine (Iversen and Lurås, 2000, and Chandra et al.,2011). It is well documented that the same medical condition may be treated di¤erently across locations. Further, patients’ preferences may have an e¤ect on how physicians select treatments (Skinner, 2011). We further hypothesize that practice-style variations are in‡uenced by GP market conditions. When GPs compete vigorously for patients, they may adjust their practice styles to put on more weight on patient bene…t. Alternatively, a more competitive GP market may endow patients with better outside options. This implies a better bargaining outcome for them. 7 3.2 The referral process and physician utilities We allow a Specialist the option to reject a referral and send the patient back to the GP for primary care services; this option is often ignored in the literature. The referral process is modelled as follows: Stage 1: The GP observes the patient’s bene…t value u from his treatment, and decides between treating the patient, referring the patient to the Private Specialist, and referring the patient to the Public Specialist. Stage 2: Upon a referral, the Private or Public Specialist gets to learn both u and v, the latter having being drawn according to distribution F . The Specialist decides between treating the patient and sending the patient back to the GP (who then has to treat the patient). There will be a delay if the referral has been to the Public Specialist. The GP’s utility is p + u if he treats the patient; he values pro…t p from fee-for-service (net) revenue and the patient’s bene…t at u, where the strictly positive practice-style parameter measures the importance of the patient’s bene…t. If the GP refers the patient to the Private Specialist, and the referral is accepted, his utility is v. In this case, the GP no longer receives the fee-for-service payment p, but his concern for the patient remains. We continue to use the practice-style parameter to measure how the GP values the patient’s bene…t v from specialty care. If the GP refers the patient to the Public Specialist, the patient experiences a delay, so the GP’s utility is discounted by a factor of to v. In the Norwegian system, the GP also receives a capitation payment for each patient under his care. At the time of referral, the GP has already received the capitation payment, so we do not write it down explicitly. The Private Specialist’s utility is similarly de…ned as q + v if he accepts the referral, where the practice-style parameter > 0 is the Specialist’s weight on the patient’s bene…t. If he rejects the referral and sends the patient back to the GP, his utility is u. Again, the Specialist values the patient’s utility at the practice-style parameter even when the patient is referred back to the GP. The Public Specialist receives a …xed salary, so we let his payo¤ from treating the patient derive entirely from his concern for the patient. We normalize the Public Specialist’s salary to 0, and his practice-style parameter to 1. When the Public Specialist sees the patient, the delay is already a sunk cost, so we write his utility from treating the patient as v, and his utility from sending the 8 patient back to the GP as u. 3.3 Equilibrium decisions by Specialists The Public Specialist receives a salary and acts in the patient’s best interest, so he treats the patient if v > u, and sends the patient back to the GP otherwise. The Private Specialist, however, may not act in the patient’s best interest. If the Private Specialist accepts the referral, his payo¤ is q + v. If he rejects the referral, the patient receives treatment from the GP, so the Private Specialist’s payo¤ is u. The Private Specialist accepts the referral if and only if q + v u. Even when v < u, he may not redirect the patient back to the GP because he earns a monetary pro…t q by providing treatment. He will send the patient back to the GP only if the monetary payo¤, q, is less than the incremental utility, (u 3.4 v), or equivalently v < u q= . GP’s utilities from treating and referring the patient We now consider the GP’s expected utilities from his three options in Stage 1. He takes into consideration the best responses of both Specialists. First, if the GP treats the patient, his utility is p + u: (1) Second, if the GP refers the patient to the Public Specialist, the referral will be accepted if and only if v u. The GP’s expected utility from this referral is Z [p + u] dF (v) + v u) corresponds to the Public Specialist accepting the referral, so the GP’s payo¤ is v, and again this happens after a delay. Third, if the GP refers the patient to the Private Specialist, the referral will be accepted if and only if q + v is u. The GP’s expected utility from referring the patient to the Private Specialist Z [p + u] dF (v) + v > > [p + u] dF (v) + v dF (v); > > < v > > [p + u] dF (v) + v dF (v) > > : v > > > > = > > > > > ; ; (4) the two terms on the right-hand side of the inequality being the expected utilities from referrals to the Public and Private Specialists, respectively. The derivative of (1) with respect to Z is u. The derivative of (2) with respect to Z u dF (v) + v pF (b u), we have u b< 8 > : v : v = v dF (v) : > ; 9 > = v dF (v) : > ; (7) This inequality says that at u b, the derivative of (1) with respect to , namely u b, is strictly smaller than (5), the derivative of (2) with respect to . At u b, as increases, the GP’s expected utility from referring the patient to the Public Specialist increases faster than the utility from treating the patient. Because of (7), we conclude that at u b the GP refers the patient to the Public Specialist when increases. Next, suppose that at some u, say u e, we have Z [p + u e] dF (v) + p+ u e= Z v dF (v) v u e q= v pF (e u 8 > < q= ) + > : Z v < u e< > : Z u e dF (v) + v u e q= u e dF (v) + Z v u e q= v = v dF (v) : > ; (8) 9 > = v dF (v) : > ; This inequality says that at u e, the derivative of (1) with respect to , namely u e, is strictly smaller than (6), the derivative of (3) with respect to . At u e, as increases, the GP’s expected utility from referring the patient to the Private Specialist increases faster than the utility from treating the patient. Because of (8), we conclude that at u e, the GP refers the patient to the Private Specialist when increases. Proof of Proposition 2: With respect to p, the derivative of the GP’s utility from treating the patient in (1) is 1. The derivative of the GP’s utility from referring the patient to the Public Specialist (2) is Z dF (v) < 1: v > > [p + u] dF (v) + v dF (v); > > < v > > [p + u] dF (v) + v dF (v) > > : v > > > > = > > > > > ; : (9) As p increases, the utility term on the left-hand side of (9) increases at a rate of 1, but each utility term on the right-hand side of (9) increases at a rate less than 1. Hence, the inequality in (9) remains valid as p increases. 21