Health Policy 119 (2015) 456–463

Contents lists available at ScienceDirect

Health Policy journal homepage: www.elsevier.com/locate/healthpol

Prescribing behavior of General Practitioners: Competition matters Catherine Schaumans ∗ TILEC, CentER, Tilburg University, Netherlands

a r t i c l e

i n f o

Article history: Received 17 July 2014 Received in revised form 18 November 2014 Accepted 27 December 2014 Keywords: Competition General Practitioners Prescription Drugs Quality JEL classification: D22 I10 I11 L11 L15

a b s t r a c t Background: General Practitioners (GP) have limited means to compete. As quality is hard to observe by patients, GPs have incentives to signal quality by using instruments patients perceive as quality. Objectives: I investigate whether GPs prescribe more units when confronted with more competition. As there is no monetary benefit in doing so, this type of (perceived) quality competition originates from GPs satisfying patients’ expectations. Method: Market level data on per capita and per contact number of items prescribed by GPs is studied for the Belgian market of General Practitioners. I hypothesize that GP competition has a positive impact on the prescribed volume, after controlling for medical needs and GP characteristics. Properly controlling for medical needs implies the use of a two-stage linear regression model. Findings: The analysis indicates that a higher number of GPs per capita results in a higher number of units prescribed by GPs, both per capita and per contact. This is consistent with quality competition in the GP market, while inconsistent with alternatives explanations (GP scarcity, GP inducement and GP dispersing prescription in time). Conclusion: GPs prescribe more units when there is more competition to satisfy patients’ expectations. The paper thus presents empirical evidence of (perceived) quality competition. © 2015 Elsevier Ireland Ltd. All rights reserved.

1. Introduction In social health insurance systems, the price for primary care, delivered by General Practitioners (GPs) is typically regulated, be it in a fee-for-service or capitation way. Furthermore, it is common that GPs are not allowed to advertise their services. As a result, GPs can only to compete in quality. It however remains an open question whether this quality effect of competition occurs in primary care markets. Theory predicts that competition

∗ Correspondence to: PO Box 90153, 5000 LE Tilburg, Netherlands. Tel.: +31 13 466 2044. E-mail address: [email protected] http://dx.doi.org/10.1016/j.healthpol.2014.12.018 0168-8510/© 2015 Elsevier Ireland Ltd. All rights reserved.

improves quality only in case quality elasticity is large compared to price elasticity [1,2]. But even in fixed price settings, empirical studies are not consistent in their finding of an impact of competition on quality of care [3–5]. This is not surprising considering the severe information asymmetry which makes it virtually impossible to judge whether a GP is good in diagnosing, proposing or performing treatment. Furthermore, while GPs have incentives to communicate and signal quality, measures of quality in the GP market are scarce and unreliable. As such, quality competition would entail GPs focusing on specific actions that are measurable, visible to patients and perceived as indicating good quality by patients (i.e. perceived quality). Looking through the literature, quality competition for GPs concerns opening hours and availability of appointments,

C. Schaumans / Health Policy 119 (2015) 456–463

access to the practice by phone or internet, facilities, the degree to which GPs involve patients in their decisions and referral behavior [5–7]. An unvisited topic concerns the prescribing of medication in response to competition. In most countries, GPs do not benefit from their prescriptions. That is, the profession of pharmacist and GP are strictly separated (with the exception of some dispensing GPs in population scarce areas). Therefore, whether a GP prescribes few or many units of medication should not depend on the economic environment in which a GP operates. Due to the information asymmetry in the market however, the GP does have some discretion over whether and how much to prescribe. Walley and Williams [8] lists non-medical factors influencing the prescribing behavior of physicians. While it is documented that physicians overestimate the demand for prescriptions, legitimizing the patient’s illness and perceived patient pressure play an important role. In other words, writing a prescription at the end of a contact satisfies the expectation of a prescription by patients, confirms their health concern and indicates that the GP is taking it seriously. Both Frantsve and Kerns [9] and Zgierska et al. [10] confirm that fulfillment of patient expectations on prescriptions usually results in a more satisfied patient in the context of pain relief (patient satisfaction hinges crucially on communication with the patient and is only limitedly driven by met/unmet expectations [11–13]). When the competitive pressure in the market is higher, it is likely that a physician is more sensitive to patients’ expectations and satisfying patients to retain them. This argument is analog to Iversen and Ma [6], where a GP satisfies patients’ requests for referrals to compete for patients (although there is a financial consequence to this). Furthermore, putting on paper the exact names and quantities of medication to take, even though it concerns over-thecounter (OTC) drugs, can be seen as an extra service toward patients and as clear communication. This paper hypothesizes that more intense competition between GPs leads to an increase in the number of prescriptions, as prescribing behavior can be perceived (by patients) as a quality signal. I test this hypothesis for the Belgian GP market. Previous work on Belgian GPs indicates that GPs do behave differently in the face of competition. More precisely, Schaumans [14] finds evidence of supplier inducement in GP dense areas and indicates that GPs use consultations (as opposed to home visits) to do so. Whereas these findings indicate that Belgian GPs do respond to competitive pressure to increase their income level, I now look at behavior that has no direct monetary impact on GP income. As such, the focus is on pure quality or perceived quality competition. Note already that I investigate the number of units of medication, both OTC- and prescription drugs, prescribed by GPs. The paper thus focus solely on explaining the prescribing behavior of GPs, rather than investigating total consumption of drugs. 2. Institutional details: Belgian market of General Practitioners The General Practitioners (GPs) I study are active in a fee-for-service system, based on the number of contacts, in

457

combination with a third-payer system with copayments. While GPs get a higher fee for home visits and visits during the weekend, the content of the contact nor the GP’s testing or prescribing behavior adds to her income. GPs are free to locate wherever they want, with the exception of physical and business separation from pharmacies. Moreover, GPs are joined in a system of night and weekend duty organized on local market level (LKO’s). GPs can differentiate themselves (and increase their fee) by getting an accreditation, which is conditional on registration and sharing requirements of patient contacts and quality management. As 80% of Belgian GPs are accredited and as this accreditation is not systematically communicated to patients, its signaling function is limited. GPs typically prescribe all products patients need from the pharmacy. That is, a GP would write a prescription for both over-the-counter drugs (OTCs) and prescription drugs. Note that Belgian pharmacies have the professional monopoly to sell not only prescription drugs but also OTCdrugs. Since 1996, the prescribing behavior of individual GPs has been recorded carefully and is made available to GPs for self-evaluation (Farmanet). Starting 2004, institutional incentives concerning GP prescribing behavior came into place. Initial incentives aiming to decrease the consumption of antibiotics quickly made place for cost reduction policy: GPs are given incentives to prescribe more generics since 2006. Apart from that, there is neither institutional incentive nor monetary benefit with respect to the frequency of prescribing and to the number of units prescribed.

3. Explaining the volume of GP prescriptions I study the extent to which GP prescribing behavior is influenced by market characteristics. The volume of prescriptions is clearly driven by medical needs: population characteristics and use of medical care define in essence the inherent demand for medication, which is at least partly correlated to the number of GP prescriptions. Also GP characteristics are likely to impact the volume of prescriptions, as GPs learn (and adjust) how and what to prescribe during their education and on the job, through peer influence (through e.g. LKO’s) and from pharmaceutical representatives and drug samples [8,15]. For example, Gouni-Berthold and Berthold [16] summarizes studies which indicate that the gender of the GP influences the medical therapy suggested. The main hypothesis of this paper is that, after controlling for the needs for medical care and GP characteristics, the level of competition for patients between GPs influences the volume of prescriptions, as GPs compete in quality. To test this, I combine three independent data sets to get information at market level on the prescribing behavior of GPs, the number and characteristics of GPs and the demand for care. I use the postal code as the relevant market level, as in Schaumans and Verboven [17]: in general, studies indicate that patients typically do not travel far for GP care. The resulting market level dataset is far from ideal – a dataset on patient and/or physician level would allow better identification, but is however not accessible at this

458

C. Schaumans / Health Policy 119 (2015) 456–463

time – but has the benefit of covering all active physicians in Belgium. First, for every Belgian postal code with at least one active GP, I have information on the number of prescribing GPs and the number of units prescribed by these GPs (packages) for the year 2003 (IMSHealth Belgium). On average, a local market holds 13.46 prescribing GPs which on average prescribe 121,235 units in 2003. Note that none of the programs to control GP prescribing behavior were yet in place in 2003. Second, I retrieved postal code level data on the gender and experience composition of active GPs and on the number of contacts with patients, including the percentage of home visits and visits performed during the weekend (RIZIV/INAMI). Finally, I add information on population characteristics of the Belgian postal codes (NIS, Ecodata and RSZ). Our empirical strategy consists of estimating how the volume of prescriptions by GPs is affected by GP competition, controlling for GP characteristics and the need for medical care.

• Volume of prescriptions (pres): To correct for market size differences, per capita number of GP prescriptions is used as volume indicator. In our dataset, on average 11.21 units (packages) per capita were prescribed by a GP and subsequently filled at Belgian pharmacies during the year 2003. To control for alternative explanations, I also consider the number of prescription per patient contact. The data indicates that the average GP contact ends with a prescription of 2.43 units of drugs. • GP competition (comp): The degree to which a GP experiences competition from other GPs is captured by the GP density in the market [3,18]. That is, it is not the mere number of GPs in the market, but the resulting number of patients per GP that impacts whether GPs need to put in effort to maintain a certain income level. As the division of patients across GPs can impact the prescribing behavior of GPs, I furthermore include a measure of market concentration (see [7]). I use the Herfindahl–Hirschman Index (HHI) of market concentration: the closer HHI to 1, the higher the concentration of patients with few GPs. Traditionally, a low HHI-value is interpreted as a higher level of competition. But furthermore, conditional on the number of GPs in the market, a higher HHI also indicates a higher diversity of GP practices in terms of size. I therefore also include the absolute number of GPs in the market. • GP characteristics (GP char): Different profiles of GPs can result in different prescribing habits. Firstly, I control for the gender and experience composition of GPs in the market. In the average market 31% of GPs is female, 27% has less than 10 years of experience and 40% has more than 20 years of experience. Second, I account for the extent to which GPs are accredited, as these GPs are constantly trained and confronted by performance of their peers: on average 81% of GPs has an additional accreditation. Finally, I have information on the percentage contacts that take place at patients’ homes. As there is little regulation on this, this is essentially a choice variable of the GP. Note that patients consider this an extra service from

the GP, but as this is costly for the GP, it is mostly reserved for children and elderly. • Medical needs: A direct measure of medical needs, such as subjective health, is not available on a disaggregate level. Therefore, I use the consumption of medical care (cons) as a proxy for medical needs: on average, there are about 5 contacts with GPs per capita in 2003. This variable is however the result of both demand and supply side effects and therefore not a perfect measure of medical needs. Note though that as 80% of GP contacts are at the patient’s initiative (WIV [16]) and as the underlying motive influencing the use of medical care are monetary, consumption of medical care is in principle exogenous to the prescribing decision of GPs. Second, patient characteristics (pat char) have been shown to be good indicators of subjective health: the variation in subjective health indicators almost completely disappears once controlled for variables such as gender, income and age composition [19]. Furthermore, patient characteristics are also good proxies of the use of primary care [20] and of medication [21]. I include the population size and density, the average income and the unemployment rate, the gender, age and nationality composition of the population and the geographic location of the postal code. Third, the percentage contacts that fall under weekend and night duty (duty, performed by the GP on call) gives an indication to which extent the local population needs urgent care. And finally, the number of pharmacies per capita (phar) in the local market is an imperfect proxy the need for medication in the market. That is, while pharmacy density is related to the medical needs of the local population, regulation has put a cap on the number of pharmacies per municipality based on population numbers. The main challenge of the data is to properly control for the medical needs in the local market. This is crucial for the identification of the competition effect: if medical needs are not controlled for, GP density can be seen as a proxy for medical needs. Hence, no distinction can be made between the impact on prescription volume of the demand for prescriptions and of competition in the GP market. Therefore, I put forward a two-stage approach. In a first stage, I estimate the determinants of the number of GP contacts per capita, and include, amongst others, the level of GP density as an explanatory variable (note that I am here not interested in causal effects). The resulting predicted values are a measure of health needs in the markets. In the second stage, I use these predicted health needs as explanatory variable for the prescribing behavior of GPs. Since the first stage corrected for the extent to which GP density affects medical needs, the coefficient of GP density in the second stage equation can be interpreted as the competition effect. The empirical approach can be summarized as follows: Cons = f (comp, GP char, pat char, duty, phar) ons, GP char) Pres = f (comp, c The Appendix (Table A1) presents the full set of descriptive statistics of the 785 local markets used in the analysis: due to the specific nature of GPs related to hospitals and

C. Schaumans / Health Policy 119 (2015) 456–463

459

Table 1 Empirical results for competition measures.a By

Impact on units prescribed per capita

Impact on units prescribed per contact

GP density HHI Health needs ( cons) Adjusted R2

7.72*** (0.43) 0.71 (0.80) 0.75*** (0.17) 0.56

1.84*** (0.13) 0.99*** (0.24) −0.44*** (0.05) 0.26

*** a

Indicates significance at 1%. Controlled for the number of GPs in the market.

Table 2 Empirical results for GP characteristics.a By

Impact on units prescribed per capita

Impact on units prescribed per contact

GP gender Exp < 10 yrs Exp > 20 yrs Accr Home

−1.83* (0.96) −1.06 (0.92) −0.92 (0.88) −0.29 (0.74) 9.93*** (1.62)

−0.20 (0.30) 0.26 (0.29) −0.16 (0.27) −0.53** (0.23) 2.13*** (0.50)

*

Indicates significance at 10%. ** Indicates significance at 5%. *** Indicates significance at 1% Controlled for competition indicators and medical needs (results not shown here. Additionally controlled for the number of GPs in the market. Adjusted RR2 is 0.59 and 0.28 respectively). a

the access to emergence rooms, local markets containing a hospital are dropped from the analysis. 4. Does competition determine the prescribing behavior of GPs? Table 1 summarizes the empirical findings on the relation between competition measures and the volume of prescriptions, controlling for medical needs in the market. I refer to the appendix for the results of the first stage equation (Table A2). Note that all findings on quality competition are robust against a finer or broader selection of local markets and differing definitions of the number of GPs (see Table A3). Our primary finding is that competition matters. In all specifications used, the higher the number of GPs per capita, the higher both per capita and per contact number of units prescribed. The evidence is consistent with the hypothesis of perceived quality competition. That is, to satisfy the expectation of patients and to give as complete information as possible, GPs are in the face of competition inclined to prescribe more units at each patient contact. As such, they provide an extra service and attempt to signal their quality as a physician. There are several alternative explanations for the finding of a positive impact of GP density on the volume prescribed. However, further empirical analysis dismisses these alternatives. I discuss the most important alternatives below: • GP scarcity: In GP scarce areas, there can be an undercapacity of GP care reflected in long waiting times and lower levels of service (e.g. shorter contacts, less home visits). Patients therefore on average visit GPs less often (i.e. fewer opportunities to prescribe). Furthermore, they do so in a later stage of the disease. An increase in GP density then (partly) resolves the scarcity. This ‘GP scarcity’ hypothesis implies a positive impact of GP density on per capita units prescribed, as more patients visit a GP.

However, it would also imply a negative impact of GP density on per contact units prescribed, since the average patient would be less ill. This contradicts the findings. I additionally tested for the impact on per contact prescriptions on a subsample of markets with a low GP density and find no sign reversal. I thus dismiss the GP scarcity hypothesis. • Supplier inducement: Schaumans [14] indicates the presence of supplier inducement as a response to higher competitive pressure: the number of per capita GP contacts in GP dense areas is higher (as in our first-stage equation). Whereas the inherent need of prescribing is low for induced visits, GPs might want to signal the necessity of the visit by prescribing some medication (but less than for a genuine health concern). Per capita prescriptions would thus go up in the face of more competition, but per contact prescriptions would lie lower. Again, the latter is inconsistent with the empirical results. • Dispersion in time: The prescribing behavior of a GP can also be a tool to induce demand. That is, a significant part of GP contacts concerns beginning illness, which is often self-resolving in time with rest and the need for antibiotics is limited to severe, persisting cases. The GP can ask the patient to return in case the condition does not improve and at that time prescribe antibiotics. Alternatively, she can prescribe antibiotics at the first contact, for the patient to fill only in case the condition does not improve over the next few days. The same goes for patients with chronic conditions or systematic use of a specific drug (think of birth control pills or insulin for diabetics): either the GP prescribes multiple packages at a single contact or she follows the patient more closely and has the patient come back multiple times for a same prescription. If GPs are dispersing their prescriptions in time in the face of competition, the total amount of prescribed and filled medication (per capita) is however independent of the dispersing in time. That is, the GP would prescribe more when she is not dispersing in time, but these prescriptions would not be filled and do

460

C. Schaumans / Health Policy 119 (2015) 456–463

therefore not materialize nor appear in our data. The empirical results indicate that Belgian GPs are not dispersing their prescriptions in time due to competition. If anything, competition makes GPs cluster prescriptions in a single contact, which can be consistent with the perceived quality argument to the extent that it would not concern patients with comorbidities that require close follow-up. It would then be considered a service and a sign of trust that the patient does not have to return in the short or medium run for a repeat-prescription. • Medical needs: As indicated above, medical needs cannot perfectly be controlled for in this market-level dataset. I however use many proxies of medical needs, which a study on Belgian health has shown to be explaining most of the variation in subjective health [19], and employ a two-stage approach to filter out any contamination of this in explaining the prescribing behavior of GPs. Robustness checks are reassuring: different specification of the first stage and the use of competition measures of an adjacent year do not alter the results on the competition effect qualitatively. I refer to Table A3 in the appendix for further details on this. In conclusion though, the robustness of the findings suggest that it is not solely the medical needs in the market that drive the effect of GP density on the volume prescribed by GPs. Our other measure of competition – the Herfindahl– Hirschman Index – also impacts the prescribing behavior of GPs. That is, a higher HHI results in a higher number of prescriptions per contact. As the number of GPs in the market is controlled for, I interpret the HHI as a measure of size differences across GP practices (i.e. diversity). I thus find that when the market is served by same-sized GP practices (low HHI), there is a lower number of prescriptions per contact. A possible explanation is that GPs are more inclined to use prescriptions to signal quality when they are small and attempt to attract patients from a larger practice. As such, this type of quality competition would occur less when GPs are more similar in terms of number of patients. Alternatively, when a market is served by same-sized GPs, the GPs might experience more competitive pressure. That is, the results from the first stage analysis suggest that there are more contacts per capita in such market. It might be the case that GPs are more likely to induce demand and/or disperse prescriptions in time in a market with same-sized GPs. Note finally that health needs have a positive significant impact on the number of prescriptions per capita. At the same time, the higher the health needs, the lower is the number of prescriptions per contact. This is consistent with higher health needs resulting in more GP contacts. 5. Do GP characteristics influence prescribing behavior? The literature suggests that GP characteristics play a role in the prescribing behavior [16]. Even though the analysis here is on the market level (and there is no information on type of prescription), I find a significant joint impact of GP characteristics on GP prescribing behavior (Chi2 = 48.46). That is, as indicated in Table 2, the percentage of home visits

has a clear positive significant impact on the prescribed volume. A GP thus prescribes more when she also decides to perform more home visits. Note that medical needs are – at least partly – controlled for: there are more contacts with a GP when the GPs in the market are performing more home visits. The percentage female GPs has a negative impact on per capita volume. As the results also indicate that females do not prescribe more per contact, this effect can be ascribed to the lower level of total contacts with GPs. No directly statements on the role of gender in the volume of prescriptions can therefore be made. With respect to the accreditation level, the results indicate a lower number of prescription per contact. As our level of aggregation reduces the variability in the data, it would be interesting to see whether precise effects of GP characteristics on the prescribing volume can be found when there is data on GP level and preferably differentiated according to type of prescriptions (OTC or prescription drug, type of drug). 6. Conclusion In this paper, I study the prescribing behavior of General Practitioners (GP) in Belgium in 2003. More precisely, I investigate whether the volume of prescriptions systematically differs across markets and test whether the extent of GP competition plays a significant role. Our evidence indicates that the higher the number of GPs per capita, the higher the volume of medication prescribed, both per capita and per contact. Although the control for medical needs is not optimal, the robustness of the results indicates that the competitive pressure GPs experience is responsible for this. This is somewhat surprising as there is no direct monetary benefit for the GP in prescribing more medication. The mechanism responsible for this finding is that GPs use prescriptions as a signal of quality in order to compete with other GPs in the market. This type of quality competition stems from GPs having little means to compete, from patients’ expectation to receive a prescription at the end of a GP contact and from the inability of patients to measure or observe real GP quality. Furthermore, GPs seem to use this quality signal more, the more different GP practices are in size. While there is a vast literature on describing the impact of GP competition of consumption of primary care, I present evidence of (perceived) quality competition in this market using prescribing behavior. This can help us understand what ‘competition’ means in the context of primary care providers [22]. As patient-satisfaction criteria are becoming more important (especially in hospital setting, but also in online ratings), competition in the GP market can be expected to increase the discrepancy in prescribing behavior of GPs depending on the competitive environment in which they operate. Notice that I am focusing on GP prescribing behavior in this paper and not the consumption of medication. Our data does not allow us to differentiate between OTC-drugs prescriptions and prescriptions related to prescription drugs and the data only reports what GPs prescribed and not the consumption of medication as a whole. Policy

C. Schaumans / Health Policy 119 (2015) 456–463

conclusions are therefore limited and premature. Further and complementary research would be able to clarify whether for example there is a higher consumption of non-prescribed OTC-drugs in regions with a lower GP density. The higher number of GP prescriptions in GP dense markets, might therefore not results in a higher consumption of medication as a whole (patients would use GP contacts to guide their drug consumption). Depending on complementary results, our findings can both support a limitation on the number of GPs per local market (if GP prescribing behavior results in a higher total consumption of medication) and stimulating entry of GPs in markets with fewer GPs (if GP prescribing behavior does not results in a higher total consumption of medication).

Acknowledgements The author thanks IMSHealth Belgium and Rijksinstituut voor Ziekte-en Invaliditeitsverzekering (RIZIV/INAMI) for the support in collecting the data and NWO for financial support (VENI 451-09-021). The author also thanks seminar attendants at TILEC and two anonymous referees for useful input.

Appendix A. Tables A1–A3

461

The interpretation of these results is not subject of this paper: it requires a different econometric approach for precise interpretation of coefficients. Shortly though, consumption of primary care is found to be affected by: • GP competition: The more GPs per capita, the higher the per capita number of GP contacts. Note that there is an endogeneity concern here: the number of contacts is a results of demand and supply and GPs locate there were the medical need is highest. As this equation only serves as first stage instrumentation to get predicted values, I do not elaborate on this here. Furthermore, the less alike the GPs are in terms of size of their patientele, the lower is the number of contacts per capita. • GP characteristics: The higher the percentage of female GPs, the lower number of GP contacts (which is in line with female GPs being more likely to work part-time) and the higher the percentage GPs with a lot of experience (>20 years), the lower the consumption of primary care. Finally, the more GPs perform home visits, the higher is the consumption of care. • Patient characteristics: In areas with more people (personal connection), with more foreigners (habits) and with a higher income (substitution to specialists), the consumption of primary care lies lower. The consumption of care is on the other hand higher the higher the percentage females. The base group of the population between 40 and 59 years old are among the highest consumers: there is lower consumption of primary care for

Table A1 Descriptive statistics of variables (nobs = 785). Variable Prescribing behavior units cap units con GP competition GP cap HHI GP characteristics GPfemale Exp < 10yrs Exp10–20yrs Exp > 20yrs Accr Home Medical needs con cap pop popdens foreign female inc cap unempl age0 9 age10 24 age25 39 age40 59 age60 79 age80plus FLA BHG WAL duty phar cap

Explanation

Mean

Std. dev.

Yearly per capita number of units prescribed by GPs in the postal code Number of units prescribed by GPs per patient contact in the postal code

11.21 2.43

6.74 1.59

Number of GPs per capita (/1000) Herfindahl–Hirschman Index based on number of patients per GP

1.07 0.31

0.54 0.28

Percentage female GPs Percentage GPs with less than 10 years of experience Percentage GPs with 10 to 20 years of experience Percentage GPs with more than 20 years of experience Percentage accredited GPs Percentage home visits (during week days)

0.31 0.27 0.33 0.40 0.81 0.38

0.19 0.20 0.21 0.21 0.22 0.12

Per capita number of contacts with patients Population size (/1000) Population density Percentage foreign population Percentage female population Average per capita income Unemployment rate Percentage population under 10 years old Percentage population between 10 and 25 years old Percentage population between 25 and 40 years old Percentage population between 40 and 60 years old Percentage population between 60 and 80 years old Percentage population over 80 years old Postal code located in Flanders Postal code located in Brussels (Brussels Hoofdstedelijk Gewest) Postal code located in Wallonia Percentage home visits during weekends Number of pharmacies per capita

5.02 7.33 0.46 0.05 0.51 1.16 0.06 0.12 0.19 0.21 0.28 0.18 0.03 0.49 0.01 0.50 0.02 0.40

2.20 6.70 0.68 0.06 0.01 0.20 0.03 0.01 0.02 0.02 0.02 0.02 0.01 0.50 0.11 0.50 0.01 0.27

462

C. Schaumans / Health Policy 119 (2015) 456–463

Table A2 Empirical results for the first stage regression. In this first stage, I regress the per capita number of contacts with a GP on market characteristics and other medical needs indicators. The predicted values of this equation (i.e. predicted medical needs) serve as explanatory variable for the prescribing volume. I use Walloon postal codes, the percentage population in age group 40–59 and the percentage GPs with 10–20 years of experience as our baseline. Variable GP density HHI GPfemale Exp < 10yrs Exp > 20yrs Accr Home pop popdens foreign female

Coeff (Std. err) ***

1.85 (0.13) −1.09*** (0.39) −1.27*** (0.38) 0.38 (0.38) −0.72** (0.35) −0.08 (0.32) 4.50*** (0.68) −0.09*** (0.02) 0.09 (0.14) −5.83*** (1.38) 23.8*** (7.91)

Variable

Coeff. (Std. err.)

inc cap unempl age0 9 age10 24 age25 39 age60 79 age80plus FLA BHG duty phar cap

−1.23** (0.52) −0.62 (3.69) −18.5*** (6.16) −17.4** (7.29) −5.47 (6.76) −19.7*** (5.97) −1.03 (8.39) 1.01*** (0.24) 0.39 (0.74) −14.9** (5.76) −0.17 (0.26)

Additionally controlled for the number of GPs in the market. R2 = 0.38. ** Indicates significance at a 5%. *** Indicates significance at a 1% level. Table A3 Robustness checks. Robustness test

Impact on per capita volume prescriptions of GP density

HHI

***

Linear regression, including all explanatory variables: Pres = f(comp, GP char, cons, pat char, duty, phar) 8.61 0.64 Two-stage regression model, with excluded from the first stage equation *** •Competition and GP characteristics1 8.93 0.44 8.26*** 0.55 •GP characteristics Cons = f(comp, pat char, duty, phar) •GP competitiona Cons = f(GP char, pat char, duty, phar) 8.89*** 0.12 5.96*** 0.62 •Patient characteristicsb Cons = f(GP char, duty, phar) Full two-stage regression model, using alternative definitions of GP density based on the active number of GPs which meet the new performance criteria to remain certifiedc •GP density in 2003 3.86*** 1.17 3.76*** 0.86 •GP density in 2004 •GP density in 2002 4.08*** 0.95 ***

Indicates significance at a 1% level. When I exclude GP competition from the first stage equation, I do not account for the impact of GP density on medical needs. Hence, the estimated coefficient of GP density in the prescribed volume equation captures both the medical needs effect and the effect of GP competition. Therefore, the estimated coefficient is higher compared to the full model. b The coefficient of GP density on per capita volume prescribed is significantly lower when I exclude patient characteristics from the first stage equation. As medical needs are not well controlled for, it is the variation in GP density that explains medical care. As a result, little variation in GP density remain to impact prescribing volumes separately from medical needs as defined in the first stage. Therefore, the estimated coefficient is lower compared to the full model. c This selection criterion is based on “Ministrieel Besluit tot vaststelling van de criteria voor de erkenning van huisartsen (21/02/2006)” and amounts to selecting GPs with at least 500 visits and at least 50 patients in one of the last 5 years. In effect, this excludes physicians connected to companies and government or those that only do guard duty. Using this measure of number of GPs, GP density lies much lower compared to the GP density calculated based on the number of prescribers. While one can argue that this is the correct definition to use, I cannot distinguish between GPs and their performance in the prescriptions-dataset (the number of prescriptions is based on all GPs in the local market). For consistency I thus use the larger definition of GPs. Nevertheless, using this more precise definition of GP density as correct measure of competition, I also find significantly positive impact on the prescribing volume. Finally, using the GP density of the previous or subsequent year in principle improves the estimation as these are not correlation to specific health episodes driving the medical needs and prescribing behavior in 2003. a

kids (substitution to pediatricians) and youngsters (better health and substitution to physiotherapists) and for the age group 60 to 79 years old (no need for medical certificates). Finally, in Flanders there is a significantly higher use of primary care. • Duty: The percentage of home visits outside the typical working hours of a GP surprisingly impacts the total consumption of primary care negatively. References [1] Dranove D, Satterthwaite M. The industrial organization of health care markets. In: Culyer AJ, Newhouse JR, editors. The handbook of health economics. Amsterdam: Elsevier; 2000 [chapter 20].

[2] Gaynor M. Competition and quality in health care markets. Found Trends® in Microeconomics 2006;1(6). [3] Sorensen RJ, Grytten J. Competition and supplier-induced demand in a health care system with fixed fees. Health Economics 1999; 8:6. [4] Biørn E, Godager G. Does quality influence choice of general practitioner? An analysis of matched doctor–patient panel data. Economic Modelling 2010;27:4. [5] Pike C. An empirical analysis of the effects of GP competition, MPRA paper no. 27613; 2010. [6] Iversen T, Ma C. Market conditions and general practitioners’ referrals. International Journal of Health Care Finance and Economics 2011;11:4. [7] Godager G, Iversen T, Ma CA. Competition, gatekeeping, and health care access. Journal of Health Economics, 2015;39. [8] Walley T, Williams R.Psychology and sociology of prescribing. In: Courtenay M, Griffiths M, editors. Independent and supplementary

C. Schaumans / Health Policy 119 (2015) 456–463

[9]

[10]

[11]

[12]

[13]

[14] [15]

prescribing: an essential guide, 2nd ed. Cambridge: Cambridge University Press; 2010. p. 53–64 [chapter 6]. Frantsve LME, Kerns RD. Patient–provider interactions in the management of chronic pain: current findings within the context of shared medical decision making. Pain Medicine 2007;8:1. Zgierska A, Miller M, Rabago D. Patient satisfaction, prescription drug abuse, and potential unintended consequences. The Journal of the American Medical Association 2012;307:13. von Ferber L, Köster I, Prüss U. Patient variables associated with expectations for prescriptions and general practitioners’ prescribing behaviour: an observational study. Pharmacoepidemiology and Drug Safety 2002;11:4. Steinke DT, MacDonald TM, Davey PG. The doctor–patient relationship and prescribing patterns. PharmacoEconomics 1999; 16:6. Himmel W, Lippert-Urbanke E, Kochen MM. Are patients more satisfied when they receive a prescription? The effect of patient expectations in general practice. Scandinavian Journal of Primary Health Care 1997;15. Schaumans, C. What’s up doc? Competition in health care markets revisited, working paper 2014. Gallan AS. Factors that influence physicians’ prescribing of pharmaceuticals: a literature review. Journal of Pharmaceutical Marketing & Management 2004;16:4.

463

[16] Gouni-Berthold I, Berthold HK. Role of physician gender in drug therapy. In: Regitz-Zagrosek, editor. Handbook of experimental pharmacology 214, sex and gender differences in pharmacology. Berlin Heidelberg: Springer-Verlag; 2012. p. 183–210. [17] Schaumans C, Verboven F. Entry and regulation: evidence from health care professionals. The RAND Journal of Economics 2008;39(4). [18] Carlsen F, Grytten J. More physicians: improved availability or induced demand? Health Economics 1998;7. [19] Deboosere P, Demarest S, Lorant V, Miermans P, Portret M, Van Oyen H. Gezondheid en Mantelzorg. In: De gezondheid in kaart, Socio-Economische Enquête 2001, 2006. FOD Economie, K.M.O., Middenstand en Energie; 2011 [hoofdstuk 6]. [20] Wetenschappelijk Instituut Volksgezondheid. In: Bayingana K, Demarest S, Gisle L, Hesse E, Miermans PJ, Tafforeau J, Van der Heyden J, editors. Gezondheidsenquête door middel van Interview, België, 2004. 2006. Depotnummer: D/2006/250/4, IPH/EPI REPORTS N 2006035. [21] RIZIV. Farmaceutische Kengetallen, Farmaceutische verstrekkingen, Ambulante praktijk 2003, Comité voor de evaluatie van de medische praktijk inzake geneesmiddelen; 2006. [22] Gaynor M, Town RJ. Competition in health care markets. In: Pauly MV, McGuire TG, Barros Pedro P, editors. Handbook of health economics, vol. 2. 2011. p. 499–637 [chapter 9].