Brown, C. (2018). Recognition of family physicians as experts rather than gatekeepers requires “cultural shift”. Canadian Medical Association Journal, 190(17), E550-E551. Available in the Trident Online Library.
Dumontet, M., Buchmueller, T., Dourgnon, P., Jusot, F., & Wittwer, J. (2017). Gatekeeping and the utilization of physician services in France: Evidence on the médecin traitant reform. Health Policy, 121(6), 675. doi:10.1016/j.healthpol.2017.04.006
What was the intended purpose of the “gatekeeper” system? Based on what you’ve read, do you believe it has achieved its purpose? Why or why not? If not, what could be changed that would result in it meeting its intended goal? Be sure to read and comment upon a few of your fellow students’ remarks on this subject.
Health Policy 121 (2017) 675–682
Contents lists available at ScienceDirect
j our na l ho me page: www.elsev ier .com/ locate /hea l thpol
atekeeping and the utilization of physician services in France: vidence on the Médecin traitant reform
agali Dumonteta,e, Thomas Buchmuellerb,c, Paul Dourgnona, Florence Jusota,d, érôme Wittwerb,∗
Institut de Recherche et Documentation en Economie de la santé, Paris, France Université de Bordeaux, Inserm U1219, France University of Michigan, Ann Arbor, MI, USA Université Paris Dauphine, PSL, LEDa-LEGOS, France LIRAES (EA 4470) & Endowed Chair AGEINOMIX, Univ. Paris Descartes, SPC, Paris
r t i c l e i n f o
rticle history: eceived 15 July 2016 eceived in revised form 13 March 2017 ccepted 23 April 2017
eywords: atekeeping hysician services utilization eform evaluation
a b s t r a c t
In 2005, France implemented a gatekeeping reform designed to improve care coordination and to reduce utilization of specialists’ services. Under this policy, patients designate a médecin traitant, typically a general practitioner, who will be their first point of contact during an episode of care and who will provide referrals to specialists. A key element of the policy is that patients who self-refer to a specialist face higher cost sharing than if they received a referral from their médecin traitant. We consider the effect of this policy on the utilization of physician services. Our analysis of administrative claims data spanning the years 2000–2008 indicates that visits to specialists, which were increasing in the years prior to the implementation of the reform, fell after the policy was in place. Additional evidence from
the administrative claims as well as survey data suggest that this decline arose from a reduction in self- referrals, which is consistent with the objectives of the policy. Visits fell significantly both for specialties targeted by the policy and specialties for which self-referrals are still allowed for certain treatments. This apparent spillover effect may suggest that, at least initially, patients did not understand the subtleties of the policy.. Introduction
While the French healthcare system is well regarded for provid- ng universal access to high quality care (WHO ), policy makers truggle to address the same problems as in other countries. From 000 to 2015, health expenditures increased from 7.9% to 8.9% ercent of GDP (DREES ). Rising expenditures represent a partic- larly important challenge, especially given the weak performance f the French economy in recent years. The French public health nsurance system has been in deficit since 2002, a situation that is xpected to persist until at least 2020 (Commission des Comptes e la Sécurtié Sociale ).
One factor that may contribute to higher expenditures is that
rench patients have traditionally faced no restrictions on their hoice of physician, including the ability to self-refer to special- sts. Most specialists in France are in private practice, where they∗ Corresponding author at: Université de Bordeaux, ISPED, 146 rue Léo Saignat, 3076 Bordeaux, France.
E-mail address: email@example.com (J. Wittwer).
ttp://dx.doi.org/10.1016/j.healthpol.2017.04.006 168-8510/© 2017 Elsevier B.V. All rights reserved.
© 2017 Elsevier B.V. All rights reserved.
are paid on a fee-for-service basis and enjoy considerable clinical autonomy. Although the lack of constraints on both patients and providers may be one reason that the French report high levels of satisfaction with their health care system (Rodwin , Or et al. ) it makes it difficult for health authorities to control expenditures. Moreover, the freedom afforded to both patients and providers may lead to care that is not well coordinated, which in turn may lead to suboptimal patient outcomes.
In 2005 France implemented a health reform designed to alter the use of specialty care with the goals of better controlling expen- ditures and improving the coordination of care by introducing a form of gate keeping. Under this policy, patients are encouraged through financial incentives to select a physician—their médecin traitant—who will be their first point of contact for any episode of care. That physician can either provide the necessary care or refer the patient for specialty care. Thus, the médecin traitant plays a role similar to that of a designated primary care provider in a US
managed care plan.Although the médecin traitant represents a significant reform to the French health system, there is surprisingly little research on the policy. In particular, no prior peer-reviewed study has examined
6 lth Pol
t i i c v G t o e (
i t t d d c a s
m d c d c t o s
p t b G t f G a i c h
c a c p o p a m i k
i 1 m t p
76 M. Dumontet et al. / Hea
he effect of the policy on the utilization of physician services. Even f the reform took place in 2005, this subject is still relevant because n January 2016, the médecin traitant scheme was expanded to cover hildren. Given important similarities between the physician ser- ices sector in France and in other European systems (Or et al. ; roenewegen et al. ), an analysis of this reform has the potential
o inform health policy more broadly. However, outside of studies n managed care systems in the U.S., there is very little research vidence on the effect of health policies related to gatekeeping Garrido et al. ).
This paper fills this gap in the literature by investigating changes n the utilization of physician services in France before and after he implementation of the reform. Our analysis is based on mul- iple years of survey and administrative claims data. The survey ata provides self-reported information on whether or not patients eclared a médecin traitant and whether they self-referred to a spe- ialist in the past year. The administrative data provides detailed nd accurate information on visits to GPs and different types of pecialists.
We find that compliance with the policy was very high, with ore than 9 out of 10 survey respondents saying that they had
esignated a médecin traitant. The administrative claims data indi- ate declines in the number of specialist visits and the number of ifferent GPs seen in a year. There was also a decline in the per- entage of patients who saw only a specialist during the course of he year. These changes coincided with a decline in the percentage f survey respondents saying that they self-referred to their last pecialist visit.
. Policy background
.1. The French system
In France, ambulatory care relies mainly on a fee-for-service ayment system where self-employed physicians who are required o accept regulated fees (sector 1) and physicians allowed to extra- ill (sector 2) coexist and offer the same services. In 2004, 85% of Ps and 61% of specialists were in sector 1. Regardless of the sector,
he public health insurance system reimburses 70% of the regulated ee, which in 2004 (the midpoint of our analysis period) was D 20 for P visits and D 25 for specialist visits. The difference between that mount and the full fee—which in the case of sector 2 physicians ncludes extra-billing—is paid by the patient directly or through omplementary health insurance. Ninety-two percent of patient ave a complementary health insurance.
.2. The Médecin traitant reform
In France, patients have long been free to choose their physi- ians, including specialists as well as GPs. This freedom of choice long with the fee-for-service reimbursement model had histori- ally received strong support from most organizations representing hysicians (Hassenteufel ). Yet, starting in the 1980s, the idea f containing health expenditure growth through innovations in rimary care, including the use of gate-keeping, began to gain cceptance, not only among health policy analysts and policy akers, but also among GPs. Only organizations representing the
nterests of specialists were opposed to the introduction of gate- eeping.
The médecin traitant reform was passed into law in 2004 and put nto effect in January 2005. According to the policy, all patients age
6 and older are encouraged to designate a physician to be their édecin traitant. The policy does not strictly require individualso make this declaration, though survey results suggest that many eople believe this to be the case (Dourgnon et al. ). Similarly, the
icy 121 (2017) 675–682
policy does not require that patients designate a GP as their médecin traitant, though almost all do (Dourgnon et al. ). Indeed, only 5% of patients have declared a specialist as a médecin traitant.
The policy uses financial incentives to encourage patients to des- ignate a médecin traitant. Patients who do not designate a médecin traitant or who directly seek care from another physician with- out a referral receive a lower level of reimbursement from the public health insurance system. Initially, the reimbursement rate is reduced by 10% for patients outside the authorized pathway meaning that rather than covering 70% of the regulated fee, the public system paid only 60%. For a specialist visit with a standard fee of D 25, this corresponds to a penalty of D 2.5. Reimburse- ment for care falling outside the authorized pathway fell to 50% in September 2007 and to 30% in January 2009, corresponding to a D 10 financial penalty. (As 2017, the 30% figure still applies.) The law included provisions that discouraged providers of complementary insurance from covering these additional copayments. In practice, all insurance companies offer ”responsible contracts” that do not cover penalties for patients who go outside the preferred pathway, though the exact details vary across policies. A second financial penalty for seeking care without an authorized referral comes in the form of extra billing. The law allows specialists in sector 1, who are normally not allowed to extra-bill, to charge up to 17.5% more than the standard fee to patients who have self-referred.
The policy provides only limited incentives for the gatekeeper physicians themselves. They are eligible to receive annual pay- ments of D 40 per patient for coordinating the care of patients with certain costly and severe long lasting conditions (cancer, dia- betes, cardiac insufficiencies, . . .). For these patients, the médecin traitant writes a protocol in tandem with the patient and other doc- tors who are involved in the patient’s care. It includes information on the treatment and physicians treating the patient. Importantly, in contrast to systems often used in U.S. managed care plans, the médecin traitant has little or no incentive to not to write a referral and essentially no ability to enforce the policy’s financial penal- ties. It is the specialist who must report whether or not a patient is respecting the preferred pathway and the public insurance system has no way to verify that this reporting is accurate. Although the ability to extra bill may provide an incentive to report patients who have self-referred, physicians also have an interest in keeping their patients happy. Thus, we might expect some specialists to claim that a patient was referred, even if that were not the case.
Prior to the reform, self-referrals varied considerably across medical specialties. The three specialties with the highest rates of self-referral were dermatology, ophthalmology and gynecology; otolaryngology head and neck surgery (OHNS) and rheumatology also had higher than average rates of self- referral (Le Fur et al. ). Under the reform, patients are considered to be in the authorized pathway if they self-refer to ophthalmologists to get a prescription for eyeglasses, but not for acute conditions, such as conjunctivi- tis. Similarly, women do not need a referral to see a gynecologist regarding contraception, prenatal care and cancer screening, but they do for other types of care. The policy also exempts self-referrals for psychiatric care for patients under the age of 25. According to a report by the High Commission on the Future of the Health Insur- ance System (HCAAM ), when the policy first came into effect, these rules concerning when one could and could not self-refer to these “direct access” specialists were not well understood by patients.
3. Empirical predictions
A key challenge in evaluating the impact of the reform is that because the policy was implemented nationally, there is no plausi- ble “control group” of patients who were not subject to the policy.
G d d
i t n G t H s o o
t o v t n p v r o b a i
v i T w r t p r t t s a c
i t v e f o o a p a b m
t b t o w t v
h v t a
M. Dumontet et al. / Hea
iven this limitation, it is especially important to have clear pre- ictions about how the reform might or might not have affected ifferent outcomes.
Regarding GP visits, the strongest prediction we can make is that f the policy was successful in reducing “medical nomadism”—the endency to receive care from multiple providers in an uncoordi- ated fashion—we should see a decline in the number of different Ps that a patient sees in a given year. Such an effect might, in
urn, correspond to a reduction in the total number of GP visits. owever, if patients who would have otherwise self-referred to a
pecialist now see their médecin traitant either to obtain a referral r to receive treatment, GP visits will increase. Therefore, the effect f the policy on GP visits is ambiguous.
We can make stronger predictions regarding specialists. Because he policy made it costlier to see a specialist—in terms of time, r money or both—we would expect to see a decline in specialist isits. If patients have a clear understanding of the policy, the reduc- ions should be largest to those specialties for which a referral is ow required, with less change for gynecology, ophthalmology, and sychiatry (for patients under age 25). However, as noted, not all isits to these “direct access” specialists are exempt from the refer- al requirements and there is reason to believe that the subtleties f the policy are not well understood. Therefore, it is possible that y emphasizing the importance of beginning episodes of care with
visit to a médecin traitant, the policy may have led to a reduction n all types of specialist visits.
To the extent that the reform did cause a reduction in specialist isits, the effect should be driven by a decline in self-referred vis- ts. We examine both direct and indirect measures of self-referrals. he survey data we use has questions that provide information on hether or not patients who had a specialist visit received a refer-
al from another physician. Referrals are not observed directly in he administrative data, but we can identify certain patterns that rovide indirect evidence on whether or not a patient received a eferral. If we observe a patient who saw at least one specialist in he year and did not also see a GP, it is likely that she self-referred. If he policy was effective, we should see a decline in this outcome. We hould note, however, that we are not able to determine whether
decline in specialist self-referrals led to improvements in care oordination, which was a goal of the policy.
In addition, given the pre/post nature of our research design, t is important to be aware of other changes occurring around the ime of the reform that could also explain the decline in specialist isits. During the decade of the 2000s, the per capita number of self- mployed physicians in France was slightly decreasing. We control or this by including in our regressions measures of the number f physicians per 100,000 persons measured annually at the level f the département, a relatively small geographic unit. Addition- lly, the proportion of specialists in sector 2 increased during the eriod—from 37.6% in 2000 to 40.6% in 2008—and the average mount of extra-billing increased as well . We account for this y including in our models the proportion of physicians in sector 2 easured annually at the département level. Another supply side factor that may have affected the evolu-
ion of physician visits is the establishment in 2004 of a national reast cancer screening program through which women between he ages of 50–74 are invited to receive free mammograms every ther year (Buchmueller and Goldzahl ). Because previously omen would have needed a prescription from a doctor—often
heir gynecologist—this policy may have reduced the number of isits to gynecologists.
On the demand side, it is well established that complementary
ealth insurance coverage is positively associated with physician isits (Buchmueller et al. ). During the period of our analysis, he share of the French population with complementary insur- nce increased slightly 89% in 2000–94% in 2008 (HCAAM ).icy 121 (2017) 675–682 677
Private complementary health insurance is not observed in the administrative claims data that we use for our main analyses. How- ever, a subset of these data can be matched to survey data, which includes information on insurance coverage and other individual level covariates. Conducting the analyses on this subsample yields results that are qualitatively the same as those from the full sam- ple, albeit less precisely estimated. Importantly, the results are not sensitive to whether or not we control for insurance coverage or other covariates.
Another potentially important demand side factor is the income shock associated with the financial crisis. In many countries, the crisis led to a decrease in public spending including spending on health care. However, a comprehensive survey of European coun- tries suggests that the impact on health spending was less in France than in other countries (Thomson et al. ). Furthermore, the fact that the impact of the crisis was felt after 2008, which is the last year of our data, limits its implication for our results. Lusardi et al.  report results from a 2009 survey that asked respondents whether they changed their use of routine health care because of the financial crisis. The percentage of French respondents saying that they reduced utilization was greater than the percent reporting decreases, though over 80 reported no change.
4. Data sources
Our evaluation is based on both survey and administrative claims data, each of which has important strengths as well as cer- tain limitations.
4.1. Survey data
The survey is the Enquête santé et Protection Sociale (ESPS), which is administered biennially to a nationally representative sample of French households. In the years after the reform was in effect, the ESPS includes a direct question on whether or not the person had declared a médecin traitant. Thus it is possible to document the rate of compliance with the policy and examine how compliance varies with individual characteristics. We use a sample of 24, 386 respon- dents to the médecin traitant question available in 2006 (n = 11,716) and 2008 (n = 12,670).
Starting in 2004, there are several questions about the respon- dent’s most recent specialist visit, including one about whether they were referred to that specialist by another physician or whether they self-referred. It is important to note that since these questions were asked only about the most recent visit to a specialist, we cannot estimate the change in the total number of self-referred visits. Nonetheless, a binary based on this question does provide evidence on whether the tendency of patients to self-referral to specialists declined after the médecin traitant policy was in effect. For this analysis, our sample contains 29,726 individuals.
4.2. Administrative claims data
We analyze changes in physician visits using administrative claims data for a random sample of individuals covered by France’s main sickness fund. This file, the echantillon permanent des assures sociaux (EPAS), provides information on all outpatient claims for the even-numbered years from 2000 to 2008. We form an unbalanced panel (each person is not observed every year) of 84,831 individuals age 16 and older, giving us 293,022 observations in total (Table 1).
A clear strength of the EPAS is that it provides precisely detailed information on all health care utilization. We observe not only the
number of physician visits in a year, but also the number of visits by specialty. Because each claim record includes a unique provider identifier, we also observe the number of different physicians seen during the year, which is important given the goal of the policy.
678 M. Dumontet et al. / Health Policy 121 (2017) 675–682
Table 1 Descriptive statistics of EPAS database.
2000 2002 2004 2006 2008
I. Outcomes: Number of visits Number of GP visits 5.35 5.25 5.19 5.26 5.14
Number of specialist visits all specialists 2.80 2.85 2.92 2.81 2.82 “direct access” specialists 0.82 0.82 0.82 0.81 0.79 gynecologists visits (among women) 0.75 0.75 0.72 0.71 0.68 ophtalmologists visits 0.39 0.39 0.40 0.40 0.40 “non-direct access” specialists 1.98 2.03 2.10 2.00 2.03 OHNS, rheumatologists dermatologists 0.49 0.49 0.49 0.44 0.44
II. Outcomes patterns of utilization Prob(saw more than one GP|saw a GP) 46.27% 45.67% 45.30% 41.16% 41.49% Number of different GP|saw a GP 1.74 1.72 1.72 1.62 1.63 Prob(Saw 1+ specialist, with seeing a GP) 6.87% 6.90% 6.91% 5.65% 5.76%
III. Covariates Female 54.2% 54.80% 55.00% 54.90% 55.00% Age 45.9 46.4 46.5 46.8 47.2 Has free complementary (Cmu-c) 7.80% 6.10% 5.90% 5.60% 4.80% Has chronic disease 14.30% 15.90% 16.90% 17.80% 18.80% Density of GPs per 100,000inhabitants 100.5 100.3 100 100.2 100.2 Density of specialists per 100,000inhabitants 89.3 88.1 86.7 86.9 86.8
W r o
t a t H g s o p o m t e o s u s
w h t C 9 E f t h r p w o o
Proportion of sector 2 GPs 14.8 Proportion of sector 2 specialists 33 N 56 471
ith such a large sample, we can precisely estimate changes in elatively rare events, such as the number of visits to certain types f specialists.
The main weakness of the database is that the sample is limited o individuals who had at least one medical claim during the year. As
result, we cannot distinguish individuals who used no care from hose who left the sample, say by moving to another sickness fund. owever, potential concerns about sample selection bias are miti- ated by the fact that in a given year over 90% of all individuals has ome type of outpatient utilization (Aligon et al. ). Moreover, ur key outcome variables—for example, the number of different hysicians seen and the mix of GP and specialist visits—are defined nly for people with positive utilization. As noted, we are able to atch a subsample of the EPAS data to the ESPS survey. In addition
o allowing us to test the sensitivity of our results to the inclusion or xclusion of covariates, it is also possible in the matched data set to bserve “true” zeroes. The fact that regressions using the matched ample yield essentially the same results as the full sample, gives s additional assurance that our results are not sensitive to sample election.
.1. Compliance with the reform
We begin the analysis by examining self-reported compliance ith the policy. According to official figures from the French public ealth insurance fund, the proportion of patient declaring a médecin raitant was 80% in 2006, increasing to 85.5% in 2008 (Cour des omptes ). In the ESPS, the rate was higher: 91% in 2006 and 8% in 2008. This discrepancy may be explained by the fact that the SPS rate is based on self-reports. Some respondents may have a amily doctor who they consider their médecin traitant, but whom hey have not yet formally declared as such, perhaps because they ad not visited that doctor since the reform was passed. Indeed, esults from probit regressions (not reported) indicate that the
robability of declaring a médécin traitant is positively correlated ith factors that themselves are associated with greater utilizationf care. Women, older adults, individuals with a chronic condition r poorer self-reported health, and those with private complemen-
14.3 13.7 13.3 12.6 33.4 34.2 35.2 36.3 57 292 59 121 59 585 60 568
tary health insurance are significantly more likely to have declared a médecin traitant. This pattern suggests that many people who did not immediately declare were not rejecting the policy, but rather did not yet have a need to declare because they had no reason to see a physician.
5.2. Physician visits
Even before the reform it was very common for French patients to have a regular GP, who would be the most likely physician they would designate as their médecin traitant. Thus, while a high rate of compliance is a necessary condition for the policy to have an effect on the utilization of outpatient services, it is not sufficient. To consider whether, in fact, the reform changed the way patients accessed care, we turn to the administrative claims data.
Using the EPAS sample, we estimate random effects nega- tive binomial regressions to examine how the number of GP and specialist visits has evolved over time, controlling for the lim- ited set of individual-level covariates available in the EPAS (Age, gender, coverage by public complementary insurance) plus the département-level measure of physician density and the share of physicians in sector 2. Results from these regressions are presented in Table 2.
Evidence on the effect of the médecin traitant reform comes from the coefficients on the year variables. The reference year is 2004, giving us two period before the reform (2000 and 2002) and two periods after (2006 and 2008). As noted, there is no clear hypothesis regarding GP visits, but there is reason to expect a negative effect on specialist visits. The results are consistent with these predic- tions. For the number of GP visits, we see a diminution before the reform but after the reform the trend is unclear. The mean num- ber of GP visits increased in 2006, but in 2008 was lower than in 2004. For specialist visits, we see a different story: specialist visits were increasing in the years leading up to the implementation of the policy and then decreased by between 4% and 5% after the pol-
icy was in place. The decline in visits was especially large for three specialties, dermatology, rheumatology and OHNS, for which self- referrals were very common before the reform and which are now subject to the new rules. For this set of specialties, a comparison
Table 2 Change in number of visits by type of physicians.
Direct access specialties Specialties without direct access
OHNS, GPs’ visits Specialists
Visits All Gyneco- logists’
visits Ophthal- mologists’ visits
All rheum- atologists derma- tologists visits
Female 0.270*** 0.543*** 1.117*** 0.264*** 0.327*** 0.297*** (0.005) (0.006) (0.009) (0.010) (0.007) (0.011)
Age 0.015*** 0.047*** 0.043*** 0.134*** 0.034*** 0.050*** 0.023*** (0.001) (0.001) (0.001) (0.002) (0.001) (0.001) (0.001)
Age2 −1.1 × 10−5* −3.8 × 10−4*** −3.7 × 10−4*** −0.002*** −1.4 × 10−4*** −3.9 × 10−4*** −1.5 × 10−4*** (5.7 × 10−6) (7.3 × 10−6) (1.1 × 10−5) (2.1 × 10−5) (1.1 × 10−5) (8.3 × 10−6) (1.3 × 10−5)
Having free complementary (Cmu-c) (Ref. No) 0.200*** 0.045*** −0.098*** −0.197*** −0.076*** 0.119*** −0.071*** (0.007) (0.010) (0.015) (0.022) (0.019) (0.011) (0.021)
Chronic disease (Ref. No) 0.444*** 0.378*** 0.073*** −0.130*** 0.097*** 0.486*** 0.055*** (0.005) (0.007) (0.011) (0.020) (0.012) (0.008) (0.014)
Density of GPs (log) 0.273*** −0.017 −0.057 −0.120*** −0.045 0.023 −0.217*** (0.023) (0.029) (0.037) (0.043) (0.034) (0.033) (0.053)
Density of gynecologists (log) 0.241*** (0.018)
Density of ophthal- mologists(log) 0.186*** (0.014)
Density of specialists (log) −0.120*** 0.221*** 0.251*** −0.160*** (0.012) (0.015) (0.017) (0.056)
Density of direct access specialties (log) 0.156*** (0.015)
Density of 3 specialties with high self-referral (log) 0.643*** (0.050)
Proportion of Sector 2 GPs (%) −0.006*** (0.000)
Proportion of sector 2 specialists (%) 0.001*** 0.002*** (0.000) (0.000)
Proportion of sector 2 gynecologists (%) 0.002*** (0.000)
Proportion of sector 2 ophtalmologists (%) 2.4 × 10−4 (0.000)
Proportion of sector2 for direct access specialties (%) 0.002*** (0.000)
Proportion of sector 2 for 3 specialties with high self-referral (%) 2.2 × 10−4 (3.5 × 10−4)
2000 0.072*** −0.029*** 0.034*** 0.082*** −0.005 −0.053*** −0.014 (0.004) (0.006) (0.008) (0.013) (0.010) (0.007) (0.012)
2002 0.029*** −0.022*** 0.017** 0.065*** −0.012 −0.037*** −0.008 (0.004) (0.006) (0.008) (0.012) (0.010) (0.007) (0.012)
Ref. 2004 2006 0.011*** −0.045*** −0.026*** −0.034*** −0.004 −0.058*** −0.114***
(0.004) (0.006) (0.008) (0.012) (0.010) (0.007) (0.012) 2008 −0.022*** −0.050*** −0.057*** −0.081*** −0.024** −0.051*** −0.093***
(0.004) (0.006) (0.008) (0.013) (0.010) (0.007) (0.012) Constant 0.048 −1.955*** −1.379*** −2.226*** −0.107 −2.689*** −1.093***
(0.076) (0.096) (0.147) (0.188) (0.151) (0.108) (0.182) LR-test (Pooled versus RE negbin) 78286*** 48363*** 32742*** 16361*** 18922*** 40283*** 19465*** LR-test (Poison RE versus Negbin RE) 69382*** 131613*** 22425*** 17674*** 1933*** 128531*** 36118*** Observations 293022 293022 293022 161108 293022 293022 293022 N 84831 84831 84831 45860 84831 84831 84831
Sources: EPAS 2000, 2002, 2004, 2006, 2008,* p < 0.10, ** p <0.05, *** p <0.001
6 lth Pol
i r c d b c o g “ w
w r E d T m f r a e 2 c p s p b a
E o i v t t c w o t a a a i
c r f v s b w o t a
80 M. Dumontet et al. / Hea
f the two later years to 2004 implies a reduction between 9% and 1% relative to 2004.
At the same time, there was also a statistically significant decline n visits to “direct access” specialists, for whom referrals are only equired for certain services. When we examine each of these spe- ialties individually, we see that this decline is mostly due to a iminution in visits to gynecologists. As noted, that change may e related to the implementation of France’s national breast can- er screening program, which meant that women between the ages f 50 and 74 no longer needed a prescription (often provided by a ynecologist) to obtain a mammogram. Alternatively, this apparent spillover” effect may reflect patients’ imperfect understanding of hen the penalties for self-referral apply.
.3. Survey-based evidence on self-referrals
If the observed decline in specialist visits is a result of the policy, e should also see a reduction in the tendency of patients to self-
efer. Direct evidence on self-referrals to specialists comes from the SPS. These results are reported in Table 3. In 2004, 37.2% of respon- ents said that they had self-referred to their most recent visit. he rate fell to 35.3% in 2006 and then to 31.4% in 2008. The esti- ated decreases remain statistically significant when we control
or individual characteristics (gender, age, education, income, self- eported health, a département level measure of physician density nd indicators for chronic conditions, complementary insurance, mployment status, and smoking) except in column 1 where the 006 effect was negative but non-significant. Given that the per- entage of people with specialist visits was declining over the eriod, it is also useful to calculate this percentage over the full ample, treating observations with no specialist visits as zeroes. The robit models estimated on the full sample imply that the proba- ility of self-referring fell by 2.3% points between 2004 and 2006 nd by 4.0% points between 2004 and 2008.
.4. Inferring self-Referrals from claims data
In addition to this evidence from survey responses, we use the PAS data to define two outcomes that provide indirect evidence f self-referral behavior. One is a binary variable that equals one f a patient had at least one specialist visit in a year and no GP isit. Given that the vast majority of individuals declared a GP as heir médecin traitant, we can assume that most patients for whom his variable equals one have self-referred. The results for this out- ome, which are reported in column (3) of Table 4, are consistent ith a policy effect. Whereas we see no trend in the percentage
f patients seeing only specialists during the pre-reform period, here is a small, but statistically significant decline of between 0.5 nd 0.6% points after 2004. Although this variable will not capture ll self-referrals (patients can see a GP for one medical condition nd self-refer to a specialist for another), we can interpret a decline n this outcome as a reduction in self-referrals.
.5. Tendency to receive care from multiple GPs
The reform should have not only reduced self-referrals to spe- ialists, but should also have reduced the tendency of patients to eceive care from multiple GPs. The last two columns test (Table 4) or such an effect. Since the number of GPs seen in a year is a count ariable, we analyze it using a negative binomial model. However, ince the modal number of GPs seen is 1, most of the variation is etween 1 and 2. Therefore, we also estimate a probit model in
hich the dependent variable is 1 for patients who saw more thanne GP in a year, and 0 for those who saw only one. Consistent with he intended effect of the policy, both of these outcomes declined fter 2004. The fact that we see no significant trend in the pre-
icy 121 (2017) 675–682
reform period supports the interpretation that this is a true effect of the reform.
By emphasizing the central role of the primary care physician in coordinating care and incentivizing patients to follow a particular pathway, the médecin traitant reform represents a small step in the direction of managed care. Because most French patients already had a “family doctor,” it is not surprising that the vast majority quickly complied with the policy and designated a médecin traitant. The manner in which the policy built on pre-existing relationships combined with the fact that patients were not obligated to respect the médecin traitant pathway, may explain why the reform did not produce the type of “managed care backlash” that resulted from the imposition of gatekeeping strategies in the U.S.
At the same time, in light of the modest, incremental nature of the reform, it is not clear that it should have had much effect on the utilization of physician care. Yet, our results indicate that visits to specialists, which were increasing in the years leading up to the reform, declined significantly after the policy was in place. Consistent with the goals of the policy, we also find small, but sig- nificant decreases in the percentage of patients saying that they self-referred to a specialist for care, the number of different physi- cians seen and the probability of seeing only specialists during a year. The relatively small magnitude of these effects is not surpris- ing for two reasons. First, the financial incentives associated with the policy were themselves quite small. Second, even before the reform most patients had a “family doctor,” who provided referrals to specialists.
However, not all of the results are easy to interpret. Visits fell both for specialties where penalties for self- referral applied after the reform and those for which some self-referrals were still allowed. It may be that patients did not fully understand the sub- tleties of the policy and thought that financial penalties applied to all self-referrals. Or, perhaps the primary mechanism was not the financial incentives per se, but by establishing the preferred pathway as normative behavior.
Indeed, evidence from other settings suggests that when a pol- icy is well publicized, its effect may exceed what would be expected based on financial incentives alone. For example, consider the case of “Lifetime Health Cover,” enacted in Australia in 2000 to increase the take-up of private health insurance. Under the policy con- sumers face financial penalties for each year after age 30 that they delay purchasing private insurance. The policy was introduced by a major publicity campaign, including television and print ads. One empirical study concludes that the Lifetime Health Cover initia- tive had an effect on insurance coverage that was much larger than could be reasonably explained by the financial incentives alone (Ellis and Savage ). Similarly, in the U.S., there has been some debate among researchers about the impact of the financial penalties for not having insurance that were established by the Affordable Care Act. In projecting coverage effects, some modelers assumed that because the penalties are fairly modest, their effect would be limited; others hypothesized stronger effects due to a general “taste for compliance” (Frean, Gruber and Sommers ).
A key question concerns the effect of the médecin traitant policy on health expenditures. Previous research on the effect on gate- keeping finds a wide range of estimated effects (Garrido et al. ). A full evaluation of the effect of the reform on health expendi- tures is beyond the scope of this study, though, all else equal, our
results would suggest a slight reduction in ambulatory care spend- ing. Consistent with this, data from the public health insurance fund indicates that health expenditures associated with special- ist consultations decreased by approximately 2% between 2004
M. Dumontet et al. / Health Policy 121 (2017) 675–682 681
Table 3 Probability of self-referral behavior.
Having a specialist visit Full sample
Marginal Effect Standard error Marginal Effect Standard error
Ref: 2004 2006 −0,011 (0.008) −0,023*** (0.005) 2008 −0,038*** (0.008) −0,04*** (0.006) N 18364 29726 Predicted probability 0.34 0.21
Sources: ESPS 2004, 2006, 2008 pooled. Marginal effects (probit estimations), same individual controls we use to estimate probability of having a visit (Table 3). *p < 0.10, ** p < 0.05, *** p < 0.001.
Table 4 Changes in patterns of utilization among utilizers.
See more than one GP|seeing a GP Number of different GP|seeing a GP See only specialist (no GP)
Marginal Effect Standard error Coefficient Standard error Marginal Effect Standard error
Female 0.1137*** (0.0029) 0.13*** (0.003) −0.0013** (0.0006) Age −0.0002 (0.0004) −0.0005 (0.0004) 0.0008*** (0.0001) Age2 −1.3 × 10−5*** (3.58 × 10−6) −1.1 × 10−5*** (0.000004) −1.2 × 10−5*** (8.8 × 10−7) CMU-c (Ref. No) 0.125*** (0.0053) 0.214*** (0.007) −0.0153*** (0.0012) Chronic disease (Ref. No) 0.084*** (0.004) 0.097*** (0.005) −0.0278*** (0.0011) Density of GPs (log) −0.0301** (0.015) −0.039** (0.017) −0.0247*** (0.0031) Density of specialists (log) 0.0515*** (0.008) 0.079*** (0.009) 0.0205*** (0.0017) Proportion of sector 2 GPs 0.0006** (0.0003) 0.001** (0.0003) 0.0004*** (0.0001) Proportion of sector2 specialists −0.0007*** (0.0002) −0.001*** (0.0002) −0.00013*** (0.00004) Dummies years variables 2000 0.0107*** (0.0035) 0.004 (0.003) −0.0013* (0.0007) 2002 0.0052 (0.0033) 0.0001 (0.003) −0.0007 (0.0007) Ref: 2004 2006 −0.0478*** (0.0033) −0.055*** (0.003) −0.006*** (0.0007) 2008 −0.0406*** (0.0034) −0.05*** (0.003) −0.0047*** (0.0007) Constant 0.343*** (0.056) Predicted probability 0.445 0.0165 Model used Probit with random effect Poisson random effect Probit with random effect LR-test (Pooled vesus RE) 11072*** 251*** 5789*** Obs 254,580 254,580 293,022 N 79,003 79,003 84,831
a g c t e e
r i a t p s S t m i p i [ c
r s i l c
ources: EPAS 2000, 2002, 2004, 2006, 2008 *p < 0.10, ** p < 0.05, *** p < 0.001
nd 2006. To make the reform more palatable to specialists, the overnment also increased fees for certain providers in certain ircumstances, which works in the opposite direction. These addi- ional payments, along with the impact of population growth, likely xplain the lower decrease in expenditure compared to our 5% stimated decrease in number of specialist visits.
As difficult as it is to determine whether the médecin traitant eform had a causal effect on the utilization of physician services, t is even more difficult to evaluate its impact on care coordination nd the quality of care received by French patients. Research from he U.S. and Canada suggests that reducing the number of different hysicians treating a patient can improve health outcomes, pre- umably through more effective coordination (Tamblyn et al. ; kinner et al. ). This suggests that a strategy like the médecin raitant reform can enhance quality, particularly if it makes patients
ore aware of the preferred care pathway. On the other hand, an mportant result from the Rand Health Insurance Experiment is that atients responded to higher cost sharing by reducing the use of
neffective and effective care in roughly equal measure (Lohr et al. 25]). Thus, it is difficult to draw normative conclusions regarding hanges in utilization (Baicker ).
Our study suggests some interesting avenues for further esearch. Developing interviews could be a way to better under-
tand patients view on self-referral and their understanding of the ncentives of the reform. It would also be interesting to observe the ong-term effects of the reform and to analyze the impact on health are quality and population health. Finally, it could be interestingto compare the effects between various socioeconomic groups and by types of diseases.
In summary, our results indicate that visits to specialists fell slightly but significantly after the médecin traitant reforms were put in place. While the pre/post nature of the evidence precludes strong statements about causality, the pattern of the changes is consistent with the intended effect of the policy. Patients were less likely to self-refer to specialists or to see multiple GPs in a given year.
Although it is important to assess the initial impact of this reform, it is also important to recognize that the policy represents another step in the gradual introduction of managed care prin- ciples into the French health care system. The same 2004 health law laid the groundwork for improvements in health information, which could be an essential element of a more fully developed strat- egy emphasizing coordination among providers. Additionally, the patient lists created by the requirement that patients designate
a médecin traitant provide a framework for pay-for-performance schemes and other approaches to improving population health. That may be in the long run the most durable impact of the médecin traitant reform on the French healthcare system.
6 lth Pol
[ case of myocardial infarction. Health Aff 2006;25:w34–7.
 Lohr KN, Brook RH, Kamberg CJ, Goldgerg GA, Leibowitz A, Keesey L, et al. Use of medical care in the rand insurance experiment. Rand Corporation; 2006.
 Baicker K, Mullainathan S, Schwartzstein J. Behavioral hazard in health insur-
82 M. Dumontet et al. / Hea
This work was supported by the Health Chair—a joint initiative y PSL, Université Paris-Dauphine, ENSAE, MGEN and ISTYA under he aegis of the Fondation du Risque (FDR)
 World Health Organization. Health Systems: Improving Performance. World Health Report. World Health Organization; 2000.
 Direction de la Recherche des Etudes de l’Evaluation et des Statistiques. Les dépenses de santé en 2015-Résultats des comptes de la santé 2016.
 Commission des Comptes de la Sécurité Sociale. Les Comptes de La Sécurité sociale. Résultats 2014, Prévisions 2015 et 2016. France 2015.
 Rodwin VG. The health care system under french national health insur- ance: lessons for health reform in the United States. Am J Public Health 2003;93(1):31–7.
 Or Z, Cases C, Lisac M, Vrangbaek K, Winblad U, Bevan G. Are health problems systemic? Politics of access and choice under beveridge and Bismarck systems. Health Econ Policy Law 2010;5(3):269–93.
 Groenewegen P, Dourgnon P, Stefan Greß S, Jurgutis A, Willems S. Strengthen- ing weak primary care systems: steps towards stronger primary care in selected western and eastern european countries. Health Policy 2013;113(1–2):170–9.
 Garrido VM, Zentner A, Busse R. The effects of gatekeeping: a systematic review of the literature. Scand J Prim Health Care 2011;29(1):28–38.
 Hassenteufel P. Les médecins face à l’Etat. Presses de Sciences Po; 2007.  Dourgnon P, Naiditch M. The preferred doctor scheme: a political reading of a
French experiment of gatekeeping. Health Policy 2010;94:129–34. 10] Dourgnon P, Guillaume S, Naïditch M, Ordonneau C. Les assurés et le médecin
traitant: premier bilan après la réforme. Questions d’économie de la santé
2007:124.11] Le Fur P, Lengagne P. Modes D’accès Aux Spécialistes: état Des Lieux Avant La Mise En Place Du Parcours Coordonnés. Questions d’économie de la santé 2006:106.
12] Haut Conseil Pour l’Avenir de l’Assurance Maladie. Rapportannuel 2006.
icy 121 (2017) 675–682
13] http://www.ecosante.fr/index2.php?base=DEPA&langh=FRA&langs=FRA. 14] Buchmueller T, Goldzahl L. The effect of a national breast cancer screening
program on mammography use: evidence from France. Unpublished working paper. University of Michigan; 2016.
15] Buchmueller T, Agnes C, Grignon M, Perronnin M. Access to physician services: does supplemental insurance matter? Health Econ 2004;13(7):669–87.
16] Haut Conseil Pour l’Avenir de l’Assurance Maladie. La généralisation de la cou- verture complémentaire en santé. Rapport 2013.
17] Thomson S, Figueras J, Evetovits T, Jowett M, Mladovsky P, Maresso A, et al. Economic crisis health systems and health in Europe: impact and implication for policy. WHO; 2015. WHO policy summary 2015.
18] A Lusardi, DJ Schneider, P Tufano, The economic crisis and medical care usage. National bureau of economic research, workingpaper 2010 march.
19] A Aligon, L Com-Ruelle, P Dourgnon, S Dumesnil, M Grignon, A Retaileau, La con- sommation médicale en 1997 selon les caractéristiques individuelles – Données de l’appariement EPAS-ESPS. Rapport CREDES 2001 mai.
20] Cour des Comptes. Le Médecin Traitant et Le Parcours de Soins Coordonnés. France; Rapport publicannuel 2013.
21] Ellis R, Savage E. The impact of premiums, threats and deadlines on private health insurance in Australia. Int J Health Care Finance Econ 2008;8:357–77.
22] M Frean, J Gruber, B Sommers, Premium Subsidies, the Mandate, and Medicaid Expansion: Coverage Effects of the Affordable Care Act. NBER Working Paper 2016, n◦22213 (April).
23] Tamblyn RM, McLeod PJ, Abrahamowicz M, Laprise R. Do too many cooks spoil the broth? Multiple physician involvment in medical management of elderly patients and potentially inappropriate drug combinations. Can Med Assoc J 1996;154(8).
24] Skinner JS, Staiger DO, Fisher ES. Is technological change always worth it? The
ance. Q J Econ 2015;130(4):1623–67.
- Gatekeeping and the utilization of physician services in France: Evidence on the Médecin traitant reform
- 1 Introduction
- 2 Policy background
- 2.1 The French system
- 2.2 The Médecin traitant reform
- 3 Empirical predictions
- 4 Data sources
- 4.1 Survey data
- 4.2 Administrative claims data
- 5 Results
- 5.1 Compliance with the reform
- 5.2 Physician visits
- 5.3 Survey-based evidence on self-referrals
- 5.4 Inferring self-Referrals from claims data
- 5.5 Tendency to receive care from multiple GPs
- 6 Discussion
- 7 Conclusions