Printer Friendly

PRIMARY CARE PHYSICIANS AS GATEKEEPERS IN THE GERMAN HEALTHCARE SYSTEM: QUASI-EXPERIMENTAL EVIDENCE ON COORDINATION OF CARE, QUALITY INDICATORS, AND AMBULATORY COSTS.

1. Introduction

In many countries a strong primary care system is considered key to an effective provision of healthcare (EXPH, 2014). In such a system, primary care physicians (PCP) might take the role of gatekeepers with respect to secondary and specialized care (Kringos et al., 2013; Scott, 2000). Gatekeeping is broadly defined as a system where PCPs or general practitioners act as primary contact persons, while patients need their "gatekeeper's" referral in order to access most specialist care. Although policy makers' attitudes towards gatekeeping are mixed, these systems are on the rise in many countries (e.g. Abbasi, 2014). In an international overview of healthcare systems in 18 developed and middle-income countries, Mossialos et al. (2016) report that most of these countries have gatekeeping regulations at least in some regions or under specific care plans. According to this overview, only three out of the 18 countries do not offer gatekeeping programs (i.e. India, Singapore and Sweden in 2015). In the US, PCPs act as gatekeepers within managed care plans (Greenfield et al., 2016; Forrest, 2003). In Europe, gatekeeping exists both in tax-funded systems like the UK and Spain and in social insurance systems like in Switzerland, in the Netherlands and in Germany (Garrido et al., 2011). In Germany, gatekeeping contracts are currently broadly introduced and co-exist with standard non-gatekeeping contracts (Brekke et al., 2007; Garrido et al., 2011).

Even if the recent literature summarizes various ethical, clinical and economic arguments in favor for or against physician-centered gatekeeping (Greenfield et al., 2016), to our knowledge, to date there is limited evidence on the quality and cost impacts of these programs. Existing evaluations of gatekeeping programs are hampered by the fact that patients are solely observed in either the gatekeeping or a non-gatekeeping system. Even if both schemes co-exist within a country, the selection of patients into these schemes needs to be taken into account when analyzing the gatekeeping effects.

In the present paper, we analyze the impact of the German PCP gatekeeping program on coordination of care as well as on quality indicators and ambulatory costs. (1) Our paper adds to the existing literature by drawing on objective quality of service and cost measures from an extensive data set. Based on health insurance claims data of a major public health insurance company, we observe all insured individuals in the Southern German state of Baden-Wurttemberg in the year before (2009) and about three years after (2013) the introduction of gatekeeping contracts in 2010. Thus, we are able to control for patients' characteristics in the year before opting or not opting for a physician-centered gatekeeping contract.

One may doubt whether the previously applied control group approaches manage to control for all factors causing selection to the programs and also being related to the outcome variables of interest. Therefore, in contrast to existing studies summarized below, we draw on multiple identification strategies. We do not solely rely on matching or regression analysis controlling for potential sources of selection bias. First, we build on a difference-indifference logic as we control for pre-treatment individual and physician characteristics. Additionally, we combine this approach with exploiting variation in the regional provision of gatekeeping in an instrumental variable (IV) framework. In this respect, we assume that the regional provision affects individual participation but is not directly related to individuals' health outcomes. In a further empirical strategy, we analyze PCP fixed effects based on the observation of patients opting for one of two different contracts within the same physician's office. Based on this, we are also able to discuss potential spill-over effects of gatekeeping (on non-participants) within the physician's office.

Since we consider various (quality) measures which--besides gatekeeping incentives--respond to differential financial incentives, our findings may provide insight to gatekeeping effects within both fee-for-service and capitation remuneration schemes. This might be valuable, for example with respect to U.S. American health plans which use multiple incentive contracts similarly interacting with gatekeeping effects (Godager et al., 2015).

From a theoretical point of view, PCPs' gatekeeping may improve the process of resource allocation, particularly if PCP-ensured continuity of care counteracts information asymmetries and goes along with improved assessment of patients' health histories and needs (Brekke et al., 2007; Forrest, 2003; Scott, 2000). Gatekeeping is expected to avoid overtreatment and to reduce healthcare costs. (2) However, inefficiencies may also be enforced if the gatekeeping physicians' objectives diverge from patients' interests (e.g. McGuire, 2000; Scott, 2000): PCPs may overtreat due to supplier-induced demand related to profit maximization. On the other hand, undertreatment may occur if contracted PCPs are tempted to reduce effort in the absence of potential competitors (Godager et al., 2014; Vedsted and Olesen, 2011; Marinoso and Jelovac, 2003; Gaynor and Town, 2011 for a general summary of competition effects). Over- and undertreatment behaviors are documented in the absence of competition under capitation as well as fee-for-service schemes (e.g. Brosig-Koch et al., 2016, for recent literature and experimental evidence).

Existing empirical research provides mixed evidence on the impacts of strong primary care in general and of the particular impacts of gatekeeping. Garrido et al. (2011) is a systematic review of the international literature on physician-centered gatekeeping: The authors identify 26 studies with documented effects of gatekeeping on health indicators, measures of healthcare utilization as well as costs. Most of these studies relate to gatekeeping in the US (i.e. 16 out of 26 studies). Further evidence is based on data for Switzerland, Denmark, Germany, Scotland and the Netherlands respectively. The majority of studies suggest that gatekeeping is related to lower utilization of health services as well as lower costs. As indicated by Garrido et al. (2011), one drawback of the existing evidence is that "health- and patient-related outcomes have been studied only exceptionally and are inconclusive" (ibid., p. 36). The provided evidence varies in magnitude and also in the direction of effects, especially when patient-related outcomes are considered. Garrido et al. (2011) also report studies to be based on "limited quality" of research designs (ibid., p. 36). Specifically, most studies suffer from multiple sources of bias and fail to take into account patient characteristics such as clinical or socio-demographic information. Only six studies consider morbidity or self-reported health status.

More recent studies relate to cross-country-analyses. Comparing primary care systems in European countries, correlative evidence suggests that strong primary care is associated with lower rates of unnecessary hospitalizations and to improved population health (Kringos et al., 2013). Hansen et al. (2015) argue that patients suffering from chronic conditions are more likely to be in good self-rated health in countries with a strong primary care structure. In contrast, Vedsted and Olesen (2011) highlight potential adverse effects of gatekeeping on healthcare quality and health outcomes. Comparing cancer survival rates in European countries, they find that the first-year relative cancer survival rate is significantly lower in physician-centered gatekeeping systems. According to Kringos et al. (2013), health expenditures seem to be higher in countries with strong primary care systems. However, Docteur et al. (2003) as well as Gerdtham (1998) document lower expenditures related to strong primary care systems.

A number of studies evaluate aspects of the recently introduced German gatekeeping contracts and address the statistical challenge of initial selection of patients into programs. In Germany, the broad introduction of gatekeeping contracts is one result of the Act on Modernization of the Statutory Health Insurance which was implemented in 2004 and aims at improving coordination of service provision between healthcare sectors. Since 2009, all statutory health insurance companies are obligated to offer gatekeeping contracts as an alternative to the standard contracts.

Most of the previous evaluation studies for Germany apply matching techniques to generate control groups based on observables (AQUA, 2013; Ose et al., 2008; Riens et al., 2011; Steiner et al., 2013). Likewise, further studies use regression analysis (Bocken, 2008; Gerlach and Szecsenyi, 2013a/b, 2014, 2016 (3) ; Schnitzer et al., 2011) controlling for observable characteristics. (4) In summary, findings from these studies are mixed, particularly when considering coordination, medication, cost indicators as well as secondary care indicators. More consistently, some studies suggest that gatekeeping is related to a higher number of visits to PCPs (i.e. the studies by Gerlach and Szecsenyi 2013a/b, 2014; Steiner et al., 2013). Concerning objective measures of care quality based on claims data, Gerlach and Szecsenyi (2014/16) suggest that gatekeeping patients are more often treated according to observed treatment guidelines for specific chronic conditions. Similarly, patients under gatekeeping more often receive preventive treatment (AQUA, 2013; Gerlach and Szecsenyi, 2014/16) and are less often reported to be subject to ambulatory care sensitive hospitalizations as well as severe complications observed within the group of patients suffering from diabetes (Gerlach and Szecsenyi, 2014/16). (5)

Our paper consistently finds that the gatekeeping contract yields a higher coordination of care, improved quality (regarding prevention and avoidance of hospitalization) but higher billed ambulatory costs. Our findings are in line with evidence from Gerlach and Szecsenyi (2013a/b, 2014, 2016), which is the most comprehensive analysis of gatekeeping in Germany to date (both concerning outcome measures and control variables).

Our paper proceeds as follows: Section 2 describes the main features of the German gatekeeping contracts. Section 3 introduces our data source and presents descriptive evidence on patients' self-selection into the gatekeeping program. Section 4 presents our identification strategies and provides further detail on the spatial distribution of gatekeeping contracts. Section 5 presents the results, which are then further discussed and concluded in Section 6.

2. Features of the German Gatekeeping Contracts

Health insurance companies in Germany operate within a largely universal public healthcare system where individuals are generally obligated to contract with one of the statuary or private health insurance companies. The statuary health insurance companies' contracts offer identical mandatory services and only differ slightly with respect to membership rates and special services. (6) Since 2009, all statutory health insurance companies are obligated to offer gatekeeping contracts as an alternative to standard health insurance plans. Participation in these gatekeeping programs is voluntary for both PCPs and the insured. Patients may opt for either a gatekeeping or a standard contract. PCPs participating in the gatekeeping program operate under two coexisting schemes: Some of their patients will, while others probably won't, opt for the gatekeeping contracts. This coexistence allows us to directly compare the gatekeeping contract to the standard contract within the same national healthcare system--and even within the same physician's office.

In this section, we detail three main features of the German gatekeeping contracts which are relevant to our empirical designs. We address (a) the key aspects related to the coordination of care, (b) PCPs' remuneration as well as (c) aspects of the structure dimension of care quality.

Concerning the coordination of care (a), patients are generally free to consult different PCPs as well as specialists under the standard (non-gatekeeping) contract in Germany. Under the gatekeeping contract, PCPs act as gatekeepers to specialist care, but the respective patients still have direct access to specific specialists (i.e. to gynecologists, optometrists, dentists, pediatricians as well as in emergency cases). (7) This is similar to the gatekeeping systems in Denmark, Estonia and Poland (EXPH, 2014). In contrast, for example Croatia, the Netherlands, Spain, Slovenia, the United Kingdom apply stricter referral regulations (EXPH, 2014).

A further key feature of the German gatekeeping scheme is that remuneration for participating PCPs differs from the standard remuneration scheme (b). For patients under the standard contract, PCPs' remuneration is based on a mix of a capitation and a fee-for-service scheme: Services are remunerated separately, while physicians receive a minor basic rate per patient treated within the quarter. There is a de facto cap of the available budget within the so-called morbidity-related compensation (morbiditatsbedingte Gesamtvergutung). (8) This means that--beyond a certain threshold--the respective services are remunerated at lower rates if a PCP provides more services or provides services to a higher number of patients.

In contrast, remuneration for PCP services provided for gatekeeping program participants is generally capitation based; only few services are being paid for separately. At the same time, there is no general cost cap in the gatekeeping system--i.e. treating more patients will always increase the gatekeeper's income. (9)

From a theoretical point of view, capitation systems may generally incentivize PCPs to attract as many patients as possible while minimizing effort (Scott, 2000; Blomqvist and Leger, 2005; Allard et al., 2011). This might counteract the desired quality effects of gatekeeping. Similarly, one may expect that systems with stronger capitation schemes yield lower per patient expenditures compared to fee-for-service schemes (cf. the crosscountry evidence in Gerdtham and Jonsson, 2000). On the other hand, PCPs in the German gatekeeping-capitation scheme receive a rather high fixed rate per patient. (10) The relatively generous baseline remuneration structure is supposed to be an incentive for PCPs to participate in the program (Hausarzteverband, 2015). Therefore, there might be limited scope for direct primary care cost savings of the gatekeeping/capitation scheme.

Fee-for-services for patients under gatekeeping relate to prevention measures such as check-ups and influenza vaccinations, ambulatory operations and treatments of psychosomatic conditions. For these specific services, PCPs income is always (under both contracts) directly linked to service provision, without budget caps relating to the number of treated patients. This might generally incentivize PCPs to provide these treatments. In our empirical analysis we consider outcome variables both related and unrelated to contract-specific incentives. This gives some guidance towards the gatekeeping effect interacting with financial incentives (cf. Section 3 and the discussion in Section 6).

A further feature of the gatekeeping contracts relates to service standards for participating physicians (c). According to the terms of contract, participating PCPs are requested to regularly attend advanced training courses on medication and treatment guidelines, to implement up-to-date IT infrastructure, to offer special services such as consultation hours on weekends, and to comply with cost saving requirements when prescribing drugs. These prerequisites probably yield a positive selection of physicians into the gatekeeping program which needs to be taken into account in our empirical identification strategies.

Similar to PCPs, patients also self-select into gatekeeping. Better coordination and a potential rise in service quality provide incentives to participate (Hausarzteverband, 2015). The self-selection of patients will be examined further in Section 3. Participation rates of both physicians and patients vary from state to state and among health insurance companies due to the differing timing of the gatekeeping programs' introduction. Overall, by the end of 2014, about 3.5 million members of all statutory health insurance companies (about 5%) participated in the gatekeeping program. (11)

3. Data

Our analysis is based on claims data from the billing process of the IKK classic, which is a major statutory health insurance company in Germany. (12) Our study is the first work using the health insurance company's claims data for empirical research beyond use within the company. We observe all insured residents of the federal state of Baden-Wurttemberg with consistent documentation of their health insurance status and contacting a PCP in the years of observation. (13) Information from the billing data of the insurance company is observed in 2009 (limited information as presented in Table 1) and 2013 (full information from claims data). Additionally, we access information on participation in the gatekeeping program throughout the relevant period of 2010-2013.

The health insurance company introduced gatekeeping contracts in October 2010. Thus, the data allows observing individuals before and after introduction of the program. Available individual-level information is related to diagnoses, treatments, operations and medical prescriptions (out-patient as well as in-patient) and ambulatory costs. Furthermore, we observe individual characteristics such as age, sex, zip codes and a proxy for labor force participation. We restrict our sample to the relevant age group of persons aged 18 and older at the time of their decision to participate in the program. Additionally, we limit the sample of program participants to patients opting for the contract until the third quarter of 2011. Persons opting for a gatekeeping contract later on are excluded from the analysis. This implies that all patients under gatekeeping have been consistently participating for at least 18 months by the time of our outcome measurement in 2013.

Outcome measures

The database allows us to analyze key features and objectives of the gatekeeping contract. Feasible outcome measures relate to (a) the utilization of healthcare services, (b) the quality of care and (c) ambulatory healthcare costs. For our analysis, we were granted limited data access, under strict data security restrictions. We had to decide on a manageable quantity of indicators which are detailed in the following paragraphs (potential indicators are discussed in our own previous work, Hofmann and Muhlenweg, 2016).

First of all, with respect to the utilization of services (a) and in order to measure whether the gatekeeping program is doing what it is supposed to do, we consider the number of PCPs visited within a year. Theoretically, under gatekeeping the number of PCPs should always be one. Due to vacation replacement regulations, the average gatekeeping patients' number of PCPs may be slightly higher. Also, to date there are no penalties for patients not complying with the terms of their gatekeeping contracts. Therefore, it is possible that patients deviate from the gatekeeping premise.

As a further measure of the utilization of healthcare services (a), we observe the utilization of specialist treatment with and without reported referral as direct indicators for the coordination of care. Specialist visits are observed as treatment cases by medical specialist within one quarter of a year, whereas a treatment case may imply multiple visits to the same specialist: (14) Since 2010, the claims data do not document separate information on each single visit. Also, referrals tend to be underreported because reporting is upon responsibility of the specialist. However, to our knowledge there is no reason for a systematic difference in the documentation practice for program participants and non-participants. Therefore, the numbers of specialist visits with and without referrals are considered to be valid outcome measures (also see Hofmann and Muhlenweg, 2016).

Additionally, we measure coordination of services based on the observation of patients with chronic conditions participating in disease management programs (DMP). Within the statutory healthcare system, DMPs are offered to patients suffering from diabetes, (15) breast cancer, coronary heart disease (CHD), asthma or chronic obstructive pulmonary disease (COPD). Diagnoses and treatment of all of these conditions except for breast cancer are within the scope of responsibility of the PCP. Therefore, we do not consider the breast cancer DMP. For the other DMPs, we measure registrations of patients diagnosed with the respective chronic condition after potential enrolment in the gatekeeping contract.

With respect to healthcare quality (b), we observe three outcome measures. Two of these measures are related to prevention (16) : Particularly, we observe influenza vaccinations in patients aged 65 and older as well as general preventive health "check-ups" in patients aged 35 and older. Concerning influenza vaccinations in patients aged 65 and older, they are considered to be an important preventive measure (see e.g. the Quality Indicators in Ambulatory Health Care, AQUA, 2009). Influenza vaccinations have been shown to be associated with a significant reduction in mortality and in the risk of respiratory diseases in the elderly (Gross et al., 1995; Nichol et al., 1998; Wilde et al., 1999). As to the second measure, the general health "check-up" offered to all insured persons aged 35 and older, this aims at detecting chronic diseases which may be asymptomatic, such as diabetes mellitus, cardiovascular diseases and renal diseases. Early detection of these chronic diseases is considered to be crucial for successful treatment (Hoebel et al., 2013). We consider both prevention measures as depicting aspects of the process dimension of care quality (following the quality classification according to Donabedian, 1988).

Besides these measures related to the process dimension of quality, we draw on Ambulatory Care Sensitive Conditions (ACSH) as an indicator for the outcome dimension of care quality. ACSH are defined as hospitalizations that may be prevented through adequate primary care (see e.g. Harrison et al., 2014). In our analysis, we rely on 19 sub-indications that are considered to be the most relevant ACSH according to the National Health Service (NHS) in England (see Sundmacher and Kopetsch, 2015). (17)

Concerning the cost dimension (c), we observe ambulatory healthcare costs. This measure comprises all billed costs for primary and secondary care services provided by health professionals outside hospitals and generally covered by the statutory health insurance. The available measure is not necessarily identical to the finally reimbursed costs which may be subject to ex post adjustments of fees provided per service to the physicians in the fee-for-service scheme. Also, costs for medication are not included in this cost measure. (18)

We expect our quality measures to interact with the financial incentives of the German gatekeeping program in different ways (cf. section 2). Particularly, for influenza vaccinations, fee-for-service remunerations for gatekeeping PCPs depend on reaching a vaccination threshold of 55% among their elderly gatekeeping patients. Income-maximizing gatekeepers have a clear incentive to conduct vaccinations among the gatekeeping patients. Therefore, observing higher vaccination rates among program participants might be rather due to financial incentives than to the gatekeeping effect. In contrast, general health check-ups are equally remunerated under the fee-for-service scheme for program participants and non-participants. In this case, one may expect that there is no, or to a lower extent, interaction of the gatekeeping effect with a financial incentive.

There are no specific financial incentives (fee-for-services) related to Ambulatory Care Sensitive Conditions. If PCPs maximize profits, we would expect that they limit treatment quality under the capitation scheme. In this case, the capitation element of the gatekeeping contracts might counteract the potentially improved quality through better coordination.

Generally, it would be interesting to observe further indicators, especially measures of the outcome dimension of quality. In our own previous work, we have discussed a number of potential indicators and their feasibility in claims data (Hofmann and Muhlenweg, 2016). Also, the most recent German evaluation study of Gerlach and Szecsenyi (2016) uses a variety of quality indicators related to chronic conditions. However, quality indicators based on chronic conditions might be considered to be of limited validity: PCPs in the gatekeeping program receive higher lump-sum payments for patients they classify as suffering from a specific condition. Because of this clear monetary incentive, we suspect PCPs' diagnosing behavior to differ for patients participating in gatekeeping and for non-participants. As a stylized fact, in our data, we observe that less than 10% of patients in the treatment and control groups (5.6% and 4.1% respectively) are classified as suffering from chronic depressions when observed before introduction of the gatekeeping program in 2009. In contrast, in 2013, about 35% of the gatekeeping patients are classified as depressed (while the rate remains more stable in the control group, 6.5% in 2013). These facts hint to chronic conditions being more likely to be diagnosed in patients suffering from less severe chronic symptoms among participants as compared to non-participants. In this case, comparisons of clinical endpoints and further complications of participants and non-participants are not considered adequate means to evaluate the gatekeeping program.

Observed characteristics of participants and non-participants

Table 1 presents mean characteristics of future participants and non-participants in the gatekeeping program. The respective characteristics are observed prior to the gatekeeping choice (in 2009). All of the presented variables are used as control variables in our econometric analysis. Some of the respective variables also constitute outcome measures when observed in 2013. (19)

Prior to the actual decision to participate in the gatekeeping program, the average future participant already differs substantially from the average non-participant in terms of age, morbidity as well as utilization of healthcare services (Table 1). This is in line with existing evidence on self-selection of patients into the German gatekeeping program (Freund et al., 2010; Kurschner et al., 2011). Particularly, Table 1 reveals that future program participants are on average about five years older than non-participants; participants are significantly more often diagnosed with severe illnesses such as heart disease, COPD or diabetes. Accordingly, their number of PCP and specialist visits is higher (8.3 vs. 6.3 and 4.3 vs. 3.5), and annual ambulatory healthcare costs exceed those of future non-participants by about EUR 120. Also, future participants participate more often in disease management programs (DMPs). In general, future gatekeeping PCPs offer these programs more frequently per se. For future gatekeeping patients, about 86% of the respective PCPs also offer DMPs. For future non-participants, only 67% of PCPs participate in DMPs (see bottom row of Table 1). Future gatekeeping participants' PCPs also differ in that they tend to operate in larger offices (mean of 235 vs. 205 patients). (20)

4. Empirical Strategies

As illustrated in the previous section, participation in the gatekeeping program depends on individual characteristics that may also influence patients' health outcomes and healthcare costs. For example, one may expect that older or less healthy patients (i.e. those under gatekeeping) cause higher treatment costs, irrespective of the quality of the potential gatekeeper's services. Simply comparing health outcomes or costs of program participants and non-participants would lead to misleading results. In the following section, we draw on multiple identification strategies addressing different sources of biases in different ways. Particularly, the panel structure of our data that observes individuals at two points in time (and also within PCP offices) allows us to apply designs based on a difference-in-difference logic (1), an instrumental variable approach (2) as well as a PCP fixed effects strategy (3), where strategies (2) and (3) are combined with the difference-in-difference set-up. As the corresponding estimates relate to different treatment effects, robustness allows us to conclude on a high validity of our findings. Each of the three empirical approaches is addressed in the following.

First of all, the most straightforward way to correct for potential selection biases is to control for possible confounders as far as they are observable. This approach has been chosen in all of the previously conducted German gatekeeping studies taking selection into account. Observing the same individuals in at least two data years--before and after treatment choice--allows building identification on a difference-in-difference logic: In a first step, we regress our outcome measures on an indicator of program participation while controlling for a large set of observable characteristics and with individual (health) and PCP characteristics observed prior to the treatment decision, (21) i.e.:

[Y.sup.2013.sub.i] + [alpha] [beta][T.sup.2010-2013.sub.i] + [gamma] [X.sup.2009.sub.i] + [[epsilon].sup.2013.sub.i],

where [T.sup.2013.sub.i] represents individual i's respective health outcome in 2013, [T.sup.2010-2013.sub.i] is an indicator for participation in the gatekeeping program, [X.sup.2009.sub.i] covers our set of control variables (including individual characteristics prior to treatment, in 2009), [[epsilon].sup.2013.sub.i] is the error term and [alpha], [beta], [gamma] represent the estimated coefficients. Thus, [beta] is supposed to provide the average treatment effect of interest. Observable controls include the characteristics presented in Table 1 : We use age (third order polynomial), sex, region of residence (two-digit zip code area), a labor force indicator, patients' morbidity and anindicator on nursing care requirement as well as healthcare utilization, including ambulatory costs. All information refers to the year prior to the participation decision (in 2009). We assume that the controlled characteristics influence the participation decision as well as patients' future outcomes. The 17 morbidity indicators included are based on the morbidity categories commonly used for calculating the Charlson morbidity score (Charlson, 1987; Sundararajan et al., 2004). In order to account for a priori differences in PCP quality, we also control for characteristics of the PCP offices, i.e. an indicator variable for group practices, number of patients as well as offer of DMP programs (in 2009, i.e. before treatment).

One may question to what extent controlling for observed characteristics may reduce potential selection biases (e.g. Riens et al., 2010). Our estimates might still suffer from biases due to unobserved characteristics influencing both the outcome measures and program participation. Thus, we do not solely rely on the assumed comprehensive specification of (pre-treatment) individual and PCP characteristics causing the selection biases. As an additional strategy, we use an instrumental variable (IV) approach identifying a local average treatment effect (LATE) based on the spatial distribution of gatekeeping contracts (Imbens and Angrist, 1994), i.e.

[mathematical expression not reproducible],

where the instrument [R.sup.2010-2013.sub.i] is the program participation rate within detailed (5 digit) zip-code area and [[??].sup.2010-2013.sub.i] is the first stage participation estimate. The same set of control variables ([X.sup.2009.sub.i]) is included in the first and second stage regressions. [e.sup.2010-2013.sub.i] ([[??].sup.2009.sub.i]) denotes the first (second) stage error term, while [delta], [theta] and [mu] ([[alpha].sub.1], [[beta].sub.1], [[gamma].sub.1]) are the estimated first (second) stage coefficients.

As shown in Figure 1, program participation substantially varies between zip-code areas. The logic of our instrument [R.sup.2010-2013.sub.i] is that we expect patients' probability to participate ([T.sup.2010-2013.sub.i] ) to be higher in regions with high shares of other patients opting for the contract, i.e. where more information on the existence of these contracts is available. At the same time, we assume other patients' choices to be exogenous with respect to the individual's health status and healthcare costs. The instrument is justified based on the assumption that much of healthcare demand is in fact driven by (local) supply (cf. evidence reviewed in Scott, 2000). (22) Our estimated local average treatment effect (LATE) [[beta].sub.1] is supposed to be representative for patients opting for gatekeeping contracts based on "what their neighbors do" within the zip-code area. The first stage estimate 8 and weak instrument check are presented in Section 5, together with the LATE estimates of [[beta].sub.1] (Table 2). We also use an alternative version of the instrument drawing on the regional participation rate net of individuals ' own PCP.

We expect that our IV identification strategy will reduce the bias of the estimated treatment effects. However, the estimated effects will still be biased if the regional provision of contracts is driven by unobserved PCP characteristics also affecting other patients' participation. One might suspect that physicians offering gatekeeping differ from non-participating PCPs in characteristics also determining service quality. Therefore, we address the possibility of PCP quality biases more directly. To this end, we exploit the observation of PCPs treating participating as well as non-participating patients at the same time. We estimate a PCP fixed effects (FE) model which is combined with our difference-in-difference set-up as follows:

[mathematical expression not reproducible]

where pcp and i index PCPs and patients respectively, [bar.[Y.sup.2013.sub.pcp]], [bar.[Y.sup.2010-2013.sub.pcp]] [X.[Y.sup.2009.sub.pcp]] and [bar.[[omega].sup.2013.sub.pcp]] represent averages for PCP's over patients outcome measures, participation indicators, control variables and error terms respectively. Among the estimated coefficients, [[beta].sub.2] represents the treatment effect based on within (PCP) variation.

Thus, we control for within-practice quality in order to estimate a mere gatekeeping effect not confounded by PCP quality. One implication of this approach is that it allows discussing potential spill-over effects within PCP offices. For example, it is possible that PCPs offering the gatekeeping program adjust their behavior to treat all of their patients in the same, better coordinated, way. (23) We will discuss the insights of the fixed effects evidence in Sections 5 and 6.

5. Results

Table 2 summarizes the regression results based on our optional identification strategies. In the following paragraphs, we first discuss our findings from the basic regressions. Afterwards, we present additional insight from the IV and FE estimates.

The first set of outcome indicators refers to aspects of utilization and coordination of care (see Section 3). The corresponding results from the basic regressions in column (1) suggest that gatekeeping principally does what it is supposed to do. The number of PCPs visited within one year is significantly reduced (about 0.16) for patients participating in the program. Also, the number of specialist visits with referral is significantly higher (0.49) while the number of specialist visits without reported referral is lower (-0.43). Thus, the reported share of specialist visits with referral is higher for participants (43% vs. 31% for non-participants, not directly shown in Table 2). At the same time, ceteris paribus the total number of specialist treatment cases is not significantly different for participants and non-participants in the gate-keeping program.

In line with a better coordination of care, the probability to enroll in DMPs is significantly higher for participants in the gatekeeping program as compared to the control group. The impact amounts to a two percentage point increase for asthma, COPD and CHD. For diabetes type II, the impact amounts to 12 percentage points. These are substantial increases, given that the enrolled proportion of recently diagnosed patients is generally rather low (varying between 4% for asthma and 27% for type II diabetes in the control group).

The middle panel of Table 2 presents the estimates for the healthcare quality indicators. Our findings suggest that the use of preventive measures is significantly higher for gatekeeping participants than for non-participants. The impact amounts to seven percentage points for influenza vaccinations and to two percentage points for health check-ups. Again, these are substantial effects: 34% of the elderly patients receive the flu shot in the control group and about 27% get check-ups. Similarly, concerning the avoidance of unnecessary hospitalizations, the probability of ACSH is reduced by about 0.2% for gatekeeping participants. This corresponds to about 10%, referring to an average of 2% in the control group.

The coordination and quality effects seem to go along with significantly higher costs of ambulatory care. The respective estimate suggests that annual ambulatory healthcare costs are about 16% higher for gatekeeping patients as compared to the control group.

The IV regression results are presented in column (2) of Table 2. Information on the instrument is provided at the bottom of Table 2. The first stage coefficient of 1.16 implies that an increase in the regional participation rate of about 10% yields a 12% increase in the individual participation probability. The very high partial F-statistics suggests that the instrument does not suffer from a weak instrument problem (cf. Staiger and Stock, 1997; Stock, Wright, and Yogo, 2002).

In general, while being robust, the point estimates of the gatekeeping effects based on IV regressions tend to be higher as compared to the basic regression results in column (1). The estimates are consistently higher when looking at healthcare utilization. We assume that this is due to the specific local average treatment effect estimated by IV. Compilers in terms of our instrument ("opting for the contract when their neighbors do so") might be a group of patients also acting more compliant with respect to the gatekeeping contracts' conditions (e.g. contacting the gatekeeping PCP for referrals).

As an alternative to the considered instrument, we have used regional participation rates net of patients visiting the respective individual's PCP (see Appendix, Table Al). In this case, we still have a strong instrument (with a first stage coefficient of 0.5) and fairly robust second stage results.

Column (3) of Table 2 presents the respective estimates from the PCP fixed effects regressions. Again, the estimates are fairly robust as compared to the previous regression results. The coefficient on specialist visits with (without) referral is somewhat higher (lower) as compared to columns (1) and (2). This means that within PCP offices, gatekeeping participants and non-participants are treated in (more) differential ways regarding this aspect of the coordination of care. However, the point estimates of the quality indicators are strongly robust among regressions. Even within PCP offices (i.e. net of PCP quality), we observe significant and substantial quality impacts of patients' participation in the gatekeeping program. One interpretation of the robust finding is that there seem to be no positive quality spill-over effects from PCPs participation in the gatekeeping program (also see discussion in the next section).

6. Conclusions and Discussion

PCPs role as gatekeepers is discussed as a means to increase effectiveness in healthcare services in many countries (EXPH, 2014). In line with this, the primary objective of the German gatekeeping schemes is to increase overall quality of care while ensuring cost effectiveness (BT-Drs. 16/3100, 2006; BT-Drs. 16/7576, 2008).

The results of our evaluation study suggest that the German gatekeeping contract yields a somewhat higher coordination of care, improved quality (regarding prevention and avoidance of hospitalization) but also higher billed ambulatory costs. Our empirical results are largely robust among the identification designs. Also, the sizes of the quality effects are substantial: For influenza vaccinations among elderly patients, the average effect across specifications is about 0.07, corresponding to a 20% increase in the population mean. Similarly, gatekeeping patients more often obtain general health check-ups with a mean effect size of 0.02 (7% of the sample mean). Also, patients under gatekeeping suffer less often from avoidable hospitalizations (average over all estimation results of-0.003 or -17% of the sample mean).

The findings are in line with findings from the previous literature on gatekeeping in Germany. Particularly, the most recent and comprehensive study by Gerlach and Szecsenyi (2016) reports similar findings for avoidable hospitalizations (about 1%). For influenza vaccinations among the elderly, Gerlach and Szecsenyi (2016) find an effect of 4.7% (roughly comparable to our estimate of 6.9%). The study is based on claims data of another major insurance company in Germany. Due to the similarity of results, we infer that our gatekeeping effects may also apply to other health insurance companies' populations.

The overall positive quality effects also provide guidance toward the financial incentives potentially interacting with the gatekeeping effects. As detailed in Section 3, PCPs operate under two distinct financial schemes: While they are remunerated mainly on a fee-for-service scheme for the non-gatekeeping patients, they obtain higher capitation payments for the gate-keeping patients (see Section 3 for details). For health check-ups, PCPs are equally paid based on fee-for-service for both groups of patients; financial incentives may thus be considered of limited concern for identifying the respective gatekeeping effect. On the other hand, one may suspect a negative impact of the (gatekeeping) capitation system on overall healthcare quality: Profit-maximizing PCPs may be incentivized to limit treatment quality under the capitation scheme. However, we still observe a substantial beneficial gatekeeping impact on avoided hospitalizations. Thus, the capitation incentive of the gatekeeping contracts seems not to (entirely) counteract the potentially improved quality through better coordination. The overall positive quality estimates raise confidence in the gatekeeping services actually improving aspects of care quality.

A further insight is gained when considering the results from the PCP fixed effects specifications. Our findings of more pronounced effects for specialist visits and ambulatory costs in the fixed effects specification could be interpreted to point to resource allocation from patients in the control group to those in the treatment group (also within the same PCP office). However, estimates are rather robust over specification as we consider quality indicators.

Furthermore, one may assume that participating PCPs becoming gate-keepers for some of their patients generally adjust their treatment behavior. If these PCPs offered the same treatment to all of their patients, the fixed effects estimates would be biased towards zero. However, this also does not seem to be the case as some of the effects are more pronounced compared to the other specifications. Thus, we conclude that there are no positive spill-over effects within the PCP office. Particularly, with respect to the health check-up measure which is equally remunerated based on fee-for-service, the lack of positive spill-over effects may suggest that we observe a true quality effect of better coordination under gatekeeping.

Overall, our results suggest that the gatekeeping contract yields quality effects for the treatment group as envisioned by policy-makers supporting a strong primary care system. However, the limited set of quality indicators we observe only depicts selected elements of the overall care environment. Like-wise, the finding of higher ambulatory costs under the gatekeeping contract is only a part of the story. In order to come to a conclusion on the overall benefits of the system, a more extensive analysis of costs and benefits would be required. Similarly, the time horizon of our analysis is rather limited. Quality impacts may crystallize at a later point in time after introduction of the contracts (Gerlach and Szecsenyi, 2016). To this end, our current analysis of indicators relating to health coordination, process- and outcome-related quality and ambulatory costs for the years 2009-2013 provides first insights and may be viewed as being a first step towards a more comprehensive longer-run analysis.

Acknowledgements

This independent research was initiated from our previous project on behalf of the HAVG (German PCP association). We are also grateful to the IKK classic (the cooperating health insurance company) for data access and for excellent data support, particularly in coding the cost measures. Furthermore, we are grateful to Martin Karlsson and seminar participants and colleagues at the WifOR institute, the CINCH (University of Duisburg-Essen) and the Institute of Labor Economics at the Leibniz University of Hannover, to participants of the German Health Association Conferences (DGGO) in Bielefeld and Berlin as well as participants of the 11th European Health Economics Association Conference in Hamburg for valuable comments on earlier versions of this work. All remaining errors are our own. We are not aware of any conflict of interest.

NOTES

(1.) Public debates on potential advantages and disadvantages of the newly introduced contracts are rather inconclusive (Hofmann and Muhlenweg, 2016). Policymakers in power assume that physician-centered gatekeeping (German: Hausarztzentrierte Versorgung) will help strengthening the primary care system and improve healthcare quality (CDU/CSU/SPD, 2013; BT-Drs. 16/3100, 2006; BT-Drs. 16/7576, 2008).

(2.) Correlative empirical evidence suggests that information on the patients' treatment history and medical conditions may exert positive as well as negative impacts, e.g. due to shorter consultations or "wait and see" behavior (Vedsted and Olesen, 2011; Weiss and Blustein, 1996; Hjortdahl and Borchgrevink, 1991).

(3.) Gerlach and Szecsenyi (2016) do not clearly state the applied method but indicate that results are adjusted for control variables.

(4.) The most comprehensive evidence is provided in Gerlach and Szecsenyi (2013a/b, 2014, 2016). These findings relate to several data waves of a major health insurance company (the "AOK"). Included control variables in these studies are similar to the set of controls used in our study. Also, the data stems from the same Southern German state as the data we use in our study.

(5.) Further findings from the German gatekeeping literature relate to patients' satisfaction with services and measures of self-assessed health (e.g. Bocken, 2008, and Schnitzer et al., 2011). Results are inconclusive and potentially suffer from reporting biases (also see the discussion in Hofmann and Muhlenweg, 2016).

(6.) Self-employed persons and employees who earn above a certain threshold (about EUR 50,000 monthly gross earnings) are free to choose between statutory or private health insurance contracts. In 2011, about 13% of the working population was insured under private health insurance contracts (Destatis, 2013).

(7.) German gatekeeping contracts imply that patients may see all other medical specialists only based on referrals. However, patients may ignore this regulation and still receive treatment. To date, there are no penalties for not complying with the contract.

(8.) According to the National Association of Statutory Health Insurance Funds, about 70% of provided services are regulated in this budget (GKV-Spitzenverband, 2017). General prevention or vaccinations which we consider in our set of outcome measures are not affected by this cap.

(9.) After January 2011, the Act of Financing Statutory Health Companies (GKV-FinG) requires new contracts to guarantee cost neutrality (cf. [section] 73b Abs. 8 SGB V), i.e. higher costs are expected to be justified through higher effectiveness of the corresponding treatment.

(10.) The rate amounts to EUR 60 to EUR 300, depending on classification of chronical conditions (Hausarzteverband, 2016).

(11.) Information provided by the German PCP association (HAVG).

(12.) According to recent numbers, the IKK classic is the sixth largest statutory health insurance company in Germany covering about 3.5 million members (Krank-enkasseninfo, 2016).

(13.) Our samples only include persons visiting a PCP which is crucial in order to control for PCP characteristics. In a previous version of our study, we have estimated effects based on a larger sample of patients irrespective of actual PCP visits and without controlling for PCP characteristics. This specification yields robust results in a propensity score matching framework (results from our project report, available upon request). More generally, the propensity score matching results are robust to the regression results presented in this paper.

(14.) For treatment without referral also see footnote 6.

(15.) Concerning diabetes type I and II, we only observe new registration into the diabetes type II program in our data.

(16.) We use these measures because, to our knowledge, these are the only PCP-based prevention measures with consistent information for all patients in the claims data.

(17.) Sundmacher et al. (2015) develop an ACSH list for Germany. This list had not been published by the time we generated our indicator. Therefore, we rely on the NHS elaboration of relevant indications, following Sundmacher and Kopetsch (2015).

(18.) Directly comparing costs and benefits of the gatekeeping program reaches beyond the scope of the present paper, where we only observe a limited number of indicators on costs and quality. Our work can be seen as a first step towards a more comprehensive framework.

(19.) Including the 2009 values as control variables in our analysis follows the logic of a difference-in-difference identification strategy.

(20.) The number of patients within PCP office refers to patients insured with the health insurance company considered in this study and being treated in the respective PCP's office in 2009.

(21.) As an alternative to our linear regression analysis, previous studies use matching designs: In an earlier version of our study, we have conducted propensity score matching which provided robust results when compared to our regression results (own project report, results available upon request).

(22.) Patients are more likely to participate if they live in regions with high shares of participating physicians. In this respect, anecdotal evidence we are aware of (based on statements of health insurance representatives) suggests that the share of participating PCPs is highest in regions where representatives of the federal state PCP association are most influential within the regional PCP associations. Concerning our identification strategy, this would imply the assumption that the PCP association's influence exerts no direct impact on patients' outcomes (i.e. beside the effects driven by the organization of gatekeeping care).

(23.) In the extreme case, we will always obtain zero effects of the gatekeeping program and this might be considered as a source of bias of our identification strategy. However, this is not what we find in the empirical analysis.

REFERENCES

Abbasi, K. (2014). "The Demise of Gatekeeping in Primary Care," Journal of the Royal Society of Medicine 107(11): 427. DOI: 10.1177/0141076814557606.

Allard, M, I. Jelovac, and P. T. Leger (2011). "Treatment and Referral Decisions under Different Physician Payment Mechanisms," Journal of Health Economics 30(5), 880-893. DOI: 10.1016/j.jhealeco.2011.05.016.

AQUA (2009). QiSA Band El. Pravention. Qualitatsindikatoren fur die Vermeidung von Krankheiten. AQUA Institut fur Angewandte Qualitatsforderung und Forschung im Gesundheitswesen, AOK Bundesverband, Berlin.

AQUA (2013). Evaluation von funf Ersatzkassen-Hausarztvertragen auf Grundlage des [section]73b SGB V. AUQA Institut Gottingen, https://www.aqua-institut.de/aqua/upload/CONTENT/Projekte/2013-09-16_HzV-Evaluation_vdek.pdf, last accessed 11 April 2017.

Blomqvist, [Angstrom]., and P. T. Leger (2005). "Information Asymmetry, Health Insurance and the Decision to Hospitalize," Journal of Health Economics 24(4): 775-793.

Bocken, J. (2008). "Hausarztmodelle im Spannungsfeld zwischen ordnungs-politischem Anspruch und Versorgungsrealitat," in Bocken, I, B. Braun, and R. Amhof (eds.), Gesundheitsmonitor 2008. Verlag Bertelsmann Stiftung.

Brekke, K. R, R Nuscheier, and O. R Straume (2007). "Gatekeeping in Health Care," Journal of Health Economics 26(1): 149-170. DOI: 10.1016/j.jhealeco.2006.04.004.

Brosig-Koch, J., B. Hehenkamp, and J. Kokot (2016). "The Effects of Competition on Medical Service Provision," Ruhr Economic Papers 647.

BT-Drs. 16/3100 (2006). Drucksache des Deutschen Bundestages 16/3100 vom 24.10.2006. Entwurf eines Gesetzes zur Starkung des Wettbewerbs in der gesetzlichen Krankenversicherung (GKV-Wettbewerbsstarkungsgesetz - GKV-WSG).

BT-Drs. 16/7576 (2008). Drucksache des Deutschen Bundestages 16/7576 vom 02.01.2008. Antwort der Bundesregierung auf die Kleine Anfrage der Abgeordneten Frank Spieth, Klaus Ernst, Dr. Martina Bunge, und weiterer Abgeordneter.

CDU, CSU, SPD (2013). Deutschlands Zukunft gestalten--Koalitionsvertrag zwischen CDU, CSU und SPD, 18. Legislaturperiode, Verlag und Druck: Union Betriebs-GmbH, 12/2013, https://www.cdu.de/sites/default/files/media/dokumente/koalitionsvertrag.pdf, last accessed 11 April 2017.

Charlson, M. E., P. Pompei, K. L. Ales, and C. R. MacKenzie (1987). "A New Method of Classifying Prognostic Comorbidity in Longitudinal Studies: Development and Validation," Journal of Chronic Diseases 40(5): 373-383. DOI: 10.1016/0021-9681(87)90171-8.

Destatis (2013). Fachserie 13, Reihe 1.1. Sozialleistungen. Angaben zur Krankenversicherung (Ergebnisse des Mikrozensus). Statistisches Bundesamt.

Docteur, E., and H. Oxley (2003). Health-care Systems: Lessons from the Reform Experience. Paris: Organization for Economic Cooperation and Development.

Donabedian, A. (1988). "The Quality of Care--How Can It Be Assessed," Journal of the American Medical Association (JAMA) 260(12): 1743-1748. DOI: 10.1001/jama.260.12.1743.

EXPH (2014). Report on Definition of a frame of reference in relation to primary care with a special emphasis on financing systems and referral systems. Expert Panel on effective ways of investing in Health, European Union.

Forrest, C. B. (2003). "Primary Care Gatekeeping and Referrals: Effective Filter or Failed Experiment," British Medical Journal 326(7391): 692-694.

Freund, T., J. Szecsenyi, and D. Ose (2010). "Charakteristika von eingeschriebenen Versicherten eines flachendeckenden Vertrages zur hausarztzentrierten Versorgung," Medizinische Klinik 105(11): 808-811.

Garrido, M. V, A. Zentner, and R. Busse (2011). "The Effects of Gatekeeping: A Systematic Review of the Literature," Scandinavian Journal of Primary Health Care 29(1): 28-38. DOI: 10.3109/02813432.2010.537015.

Gaynor, M, and R. J. Town (2011). "Competition in Health Care Markets," Working Paper 17208, National Bureau of Economic Research (NBER).

Gerdtham, U., B. Jonssen, M. Macfarlan, and H. Oxley (1998). "The Determinants of Health Expenditure in the OECD Countries: A Pooled Data Analysis," in Zweifel, P. (ed.), Health, the Medical Profession, and Regulation. New York: Springer, 113-134.

Gerdtham, U.-G., and B. Jonsson (2000). "International Comparisons of Health Expenditure," in Culyer, A. J., and J. P. Newhouse (eds.), Handbook of Health Economics. Amsterdam: Elsevier, 11-53.

Gerlach, F., and J. Szecsenyi (2013a). Abschlussbericht zur Evaluation der Hausarztzentrierten Versorgung (HzV) nach [section] 73b SGB V in Baden-Wurttemberg (2010-2011), https://www.aok-gesundheitspartner.de/imperia/md/gpp/bw/arztundpraxis/hzv/bw_hzv_eval_2008_2010.pdf, last accessed 11 April 2017.

Gerlach, F., and J. Szecsenyi (2013b). "Hausarztzentrierte Versorgung in Baden-Wurttemberg --Konzept und Ergebnisse der kontrollierten Begleitevaluation," Zeitschrift fur Evidenz, Fortbildung und Qualitat im Gesundheitswesen 107: 365-371.

Gerlach, F., and J. Szecsenyi (2014). Evaluation der Hausarztzentrierten Versorgung (HzV) nach [section]73b SGB V in Baden-Wurttemberg (2013-2016). Ergebnisbericht, https://www.aok-gesundheitspartner.de/imperia/md/gpp/bw/arztundpraxis/hzv/bw_hzv_eval_2013-2014.pdf, last accessed 11 April 2017.

Gerlach, F., and J. Szecsenyi (2016). Evaluation der Hausarztzentrierten Versorgung (HZV) in Baden-Wurttemberg, Zusammenfassung der Ergebnisse--Ausgabe 2016. Frankfurt/Heidelberg (project report), http://neueversorgung.de/images/PDF/Evaluation 2016/HZV_Evaluation_Broschuere.pdf, last accessed 11 April 2017.

GKV-Spitzenverband (2017). Vergutung arztlicher Leistungen, https://www.gkv-spitzenverband.de/presse/themen/verguetung_aerztlicher_leistungen/s_thema_aerzteverguetung.jsp, last accessed 11 April 2017.

Godager, G., T. Iversen, and C. A. Ma (2015). "Competition, Gatekeeping, and Health Care Access," Journal of Health Economics 39(C): 159-170. DOI: 10.1016/j.jhealeco.2014.11.005.

Greenfield, G., K. Foley, and A. Majeed (2016). "Rethinking Primary Care's Gate-keeper Role," British Medical Journal 354: i4803. DOI: 10.1136/bmj.i4803.

Gross, P. A., A. W. Hermogenes, H. S. Sacks, J. Lau, and R. A. Levandowski (1995). "The Efficacy of Influenza Vaccine in Elderly Persons. A Meta-analysis and Review of the Literature," Annals of Internal Medicine 123(7): 518-527.

Hansen, J., P. P. Groenewegen, and D. S. Kringos (2015). "Strength of Primary Care and Health Outcomes for Chronically Ill People: Results from 27 EU Member States," Health Affairs 34(9): 1531-1537.

Harrison, M. J., M. Dusheiko, M. Sutton, H. Gravelle, T. Doran, and M. Roland (2014). "Effect of a National Primary Care Pay For Performance Scheme on Emergency Hospital Admissions for Ambulatory Care Sensitive Conditions: Controlled Longitudinal Study," British Medical Journal 349: g6423. DOI: 10.1136/bmj.g6423.

Hausarzteverband (2015). "Hausarztzentrierte Versorgung," Sonderbeilage in Der Hausarzt / Der Allgemeinarzt, Februar/Marz.

Hausarzteverband (2016). Gatekeeping contracts (in German), http://www.haus-aerzteverband.de/cms/Hausarztvertraege.988.0.html, last accessed 11 April 2017.

Hjortdahl, P., and C. F. Borchgrevink (1991). "Continuity of Care: Influence of General Practitioners' Knowledge about Their Patients on Use of Resources in Consultations," British Medical Journal 303: 1181-1184. DOI: 10.1136/bmj.303.6811.1181.

Hoebel, J., M. Richter, and T. Lampert (2013). "Social Status and Participation in Health Checks in Men and Women in Germany--Results from the German Health Update (GEDA), 2009 and 2010," Deutsches Arzteblatt International 110(41): 679-685.

Hofmann, S., and A. M. Muhlenweg (2015). Tur auf fur den Gatekeeper? Konzeption eines Indikatoren-Sets fur ein routinema[beta]iges Prozess-, Ergebnis- und Wirtschaft-lichkeits-Monitoring der hausarztzentrierten Versorgung. WifOR, mimeo.

Imbens, G., and J. Angrist (1994). "Identification and Estimation of Local Average Treatment Effects," Econometrica 62(2): 467-476. DOI: 10.2307/2951620.

Imbens, G. (2015). "Matching Methods in Practice. Three Examples," Journal of Human Resources 50(2): 373-419. DOI: 10.3368/jhr.50.2.373.

Krankenkasseninfo (2016). Mitglieder und Versicherte je Krankenkasse, http://www.krankenkasseninfo.de/zahlen-fakten/mitgliederzahlen/, last accessed 11 April 2017.

Kringos, D. S. et al. (2010). "The European Primary Care Monitor: Structure, Process and Outcome Indicators," BMC Family Practice 11: 81. DOI: 10.1186/1471-2296-11-81.

Kringos, D. S., W. Boerma, J. van der Zee, and P. Groenewegen (2013). "Europe's Strong Primary Care Systems Are Linked to Better Population Health but also to Higher Health Spending," Health Affairs 32(4): 686-694. DOI: 10.1377/hlthaff.2012.1242.

Kurschner, N., W. Weidmann, and S. Muters (2011). "Wer wahlt den Hausarzt zum 'Gatekeeper'? Eine Anwendung des Verhaltensmodells von Andersen zur Besch-reibung der Teilnahme an Hausarztmodellen in Deutschland," Bundesgesund-heitsblatt 54: 221-227.

Marinoso, B. G., and I. Jelovac (2003). "GP's Payment Contracts and Their Referral Practice," Journal of Health Economics 22: 617-635. DOI: 10.1016/S0167-6296(03)00008-0.

Mossialos, E., M. Wenzl, R. Osborn, and D. Sarnak (2016). 2015 International Profiles of Health Care Systems. The Commonwealth Fund, pub. 1857, January.

McGuire, T. (2000). "Physician Agency," in Culyer, A. J., and J. P. Newhouse (eds.), Handbook of Health Economics. Amsterdam: Elsevier, 461-536.

Nichol, K. L., J. Wuorenma, and T. von Sternberg (1998). "Benefits of Influenza Vaccination for Low-, Intermediate-, and High-risk Senior Citizens," Archives of Internal Medicine 158(16): 1769-1776. DOI: 10.1001/archinte.l58.16.1769.

Ose, D., B. Broge, B. Riens, and J. Szecsenyi (2008). "Mit Uberweisung vom Hausarzt zum Spezialisten - Haben Vertrage zur Hausarztzentrierten Versorgung (HzV) einen Einfluss?" Zeitschrift fur Allgemeinmedizin 84: 321-326.

Riens, B., B. Broge, P. Kaufmann-Kolle, B. Pohlmann, B. Grun, D. Ose, and J. Szecsenyi (2010). "Bildung einer Kontrollgruppe mithilfe von Matched-Pairs auf Basis von GKV-Routinedaten zur prospektiven Evaluation von Einschreibe-modellen," Das Gesundheitswesen 72: 363-370.

Scott, A. (2000). "Economics of General Practice," in Culyer, A. I, and J. P. New-house (eds.), Handbook of Health Economics. Amsterdam: Elsevier, 1175-1200.

Staiger, D., and J. H. Stock (1997). "Instrumental Variables Regression with Weak Instruments," Econometrica 65(3): 557-586. DOI: 10.2307/2171753.

Steiner, M., W. Riedel, and J. Maetzel (2013). Wissenschaftliche Begleitung des hausarztlichen Qualitats- und Kooperationsmodells Sudbaden, Abschlussbericht, Prognos AG, http://www.aokgesundheitspartner.de/imperia/md/gpp/bw/arztundpraxis/arztnetze/bw_arztnetze_abschlussbericht.pdf, last accessed 11 April 2017.

Stock, J. H., J. H. Wright, and M. Yogo (2002). "Survey of Weak Instruments and Weak Identification in Generalized Method of Moments," Journal of Business and Economic Statistics 20(4): 518-529. DOI: 10.2307/2171753.

Sundararajan, V., T. Henderson, C. Perry, A. Muggivan, H. Quan, W. A. Ghali, (2004). "New ICD-10 Version of the Charlson Comorbidity Index Predicted In-hospital Morbidity," Journal of Clinical Epidemiology 57: 1288-1294.

Sundmacher, L., and T. Kopetsch (2015). "The Impact of Office-based Care on Hospitalizations for Ambulatory Care Sensitive Conditions," European Journal of Health Economics 16: 365-375. DOI: 10.1007/sl0198-014-0578-4.

Sundmacher, L., D. Fischbach, W. Schuettig, C. Naumann, U. Augustin, and C. Faisst (2015). "Which Hospitalisations Are Ambulatory Care-sensitive, to What Degree, and How Could the Rates Be Reduced? Results of a Group Consensus Study in Germany," Health Policy 119(11): 1415-1423. DOI: 10.1016/j.healthpol.2015.08.007.

Vedsted, P., and F. Olesen (2011). "Are the Serious Problems in Cancer Survival Partly Rooted in Gatekeeper Principles? An Ecologic Study," British Journal of General Practice 61(589): e508-e512. DOI: 10.3399/bjgp11X588484.

Weiss, L. J., and J. Blustein (1996). "Faithful Patients: The Effect of Long-term Physician--Patient Relationships on the Costs and Use of Health Care by Older Americans," American Journal of Public Health 86: 1742-1747. DOI: 10.2105/AJPH.86.12.1742.

Wilde, J. A., J. A. McMillan, J. Serwint, J. Butta, M. A. O'Riordan, and M. C. Steinhoff (1999). "Effectiveness of Influenza Vaccine in Health Care Professionals: A Randomised Trial," Journal of the American Medical Association 281(10): 908-913. DOI: 10.1001/jama.281.10.908.

Appendix
Table A1 Robustness of estimated impacts of gatekeeping, alternative
IV results

                                          (1)          (2)
                                 Sample   Basic        Instrumental
                                 average  regressions  variable

Coordination/continuity of
health care
Number of PCPs                     1.55   -0.16 (***)      -0.30 (***)
                                          (0.00)           (0.01)
Specialist visits with referral    1.35    0.49 (***)       0.51 (***)
                                          (0.01)           (0.03)
Specialist visits without          2.70   -0.43 (***)      -0.55 (***)
referral
                                          (0.02)           (0.05)
Participation in DMPs of
newly diagnosed...
... DMP asthma                     0.04    0.02 (**)        0.04
                                          (0.01)           (0.03)
... DMP COPD                       0.05    0.02 (*)         0.04 (*)
                                          (0.01)           (0.03)
... DMP diabetes type II           0.29    0.12 (***)       0.10 (**)
                                          (0.02)           (0.04)
... DMP CHD                        0.08    0.02 (**)        0.05 (**)
                                          (0.01)           (0.02)
Health care quality
Influenza vaccination              0.35    0.07 (***)       0.05 (***)
                                          (0.00)           (0.13)
General health check-up            0.27    0.02 (***)       0.02 (***)
                                          (0.00)           (0.01)
ACSH                               0.02   -0.00 (**)       -0.01 (**)
                                          (0.00)           (0.00)
Cost indicator
Log ambulatory costs (A)         555.11    0.15 (***)       0.09 (***)
                                          (0.00)           (0.15)
First stage: regional                                       1.16 (***)
participation rate (instrument)                            (0.01)
First stage partial F-statistic                        17,895.6
Sample size                      281.635

                                    (3)
                                    Instrumental
                                    variable (II)

Coordination/continuity of
health care
Number of PCPs                      -0.55 (***)
                                    (0.03)
Specialist visits with referral      0.65 (***)
                                    (0.08)
Specialist visits without           -0.54 (***)
referral
                                    (0.12)
Participation in DMPs of
newly diagnosed...
... DMP asthma                       0.06
                                    (0.07)
... DMP COPD                         0.04
                                    (0.07)
... DMP diabetes type II             0.10
                                    (0.10)
... DMP CHD                          0.10
                                    (0.07)
Health care quality
Influenza vaccination               -0.00
                                    (0.03)
General health check-up              0.02
                                    (0.02)
ACSH                                -0.01 (***)
                                    (0.01)
Cost indicator
Log ambulatory costs (A)             0.07 (***)
                                    (0.04)
First stage: regional                0.50 (***)
participation rate (instrument)     (0.01)
First stage partial F-statistic  3,098.17
Sample size

Note: The alternative instruments are regional participation rates
(column 2, also shown in Table 2) and regional participation rates net
of own PCP office (instrumental variable II, column 3). The first
column refers to average numbers ((A) average, not in log, for cost
indicator). All other numbers correspond to estimated coefficients
(with standard errors in parentheses), unless otherwise stated. (*)
Significant at the 10% level, (**) 5% level, (***) 1% level.
Source: Own estimations based on data for a major health insurance
company in the German state of Baden-Wurttemberg (in 2013, control
variables relate to the year before introduction of the gatekeeping
program in 2009).


SARAH M. HOFMANN

WifOR Darmstadt; Universitat Duisburg-Essen

ANDREA M. MUHLENWEG

andrea.muehlenweg@wifor.de WifOR Darmstadt; Leibniz Universitat Hannover (corresponding author)

How to cite: Hofmann, Sarah M., and Andrea M. Muhlenweg (2017), "Primary Care Physicians as Gatekeepers in the German Healthcare System: Quasi-experimental Evidence on Coordination of Care, Quality Indicators, and Ambulatory Costs," American Journal of Medical Research 4(2): 47-72.

Received 21 February 2017 * Received in revised form 9 May 2017

Accepted 10 May 2017 * Available online 30 May 2017

Caption: Figure 1 Spatial distribution of gatekeeping contracts
Table 1 Observed characteristics of future participants and
non-participants (2009)

                                   Standard care        Gatekeeping
                                   Mean       (SD)     Mean        (SD)

Socio-demographic
characteristics
 Age                              47.11     (17.67)   52.37     (16.92)
 Female indicator                  0.48      (0.50)    0.48      (0.50)
 Labor force indicator (A)         0.54      (0.50)    0.48      (0.50)
Morbidity
Diagnosed with...
 Myocardial infarction             0.01      (0.09)    0.01      (0.11)
 Heart disease                     0.02      (0.14)    0.04      (0.19)
 Peripheral vascular disease       0.02      (0.14)    0.03      (0.17)
 Cerebrovascular disease           0.03      (0.16)    0.04      (0.21)
 Dementia                          0.00      (0.05)    0.00      (0.06)
 COPD                              0.08      (0.28)    0.12      (0.32)
 Connective tissue disease         0.01      (0.12)    0.02      (0.14)
 Ulcer                             0.01      (0.07)    0.01      (0.09)
 Liver disease                     0.00      (0.05)    0.00      (0.06)
 Diabetes mellitus type I or II    0.08      (0.28)    0.14      (0.35)
 Diabetes mellitus
 complications                     0.01      (0.12)    0.03      (0.16)
 Paraplegia                        0.01      (0.08)    0.01      (0.08)
 Renal disease                     0.02      (0.13)    0.02      (0.15)
 Cancer                            0.03      (0.18)    0.05      (0.22)
 Cancer (metastasizing)            0.00      (0.06)    0.01      (0.07)
 Liver disease (severe)            0.00      (0.02)    0.00      (0.01)
 HIV                               0.00      (0.03)    0.00      (0.02)
 In nursing care                   0.00      (0.07)    0.00      (0.04)
Healthcare utilization
 Ambulatory healthcare costs     429.61    (528.34)  550.45    (564.30)
 Medication costs                323.55  (1,295.36)  440.67  (1,427.43)
 Number of PCP visits              6.29      (6.01)    8.30      (6.31)
 Number of different PCPs          1.49      (0.97)    1.66      (0.98)
 Number of specialist visits       3.54      (3.78)    4.27      (4.13)
 Number of hospitalizations        0.16      (0.51)    0.20      (0.56)
 DMP asthma participation          0.00      (0.06)    0.01      (0.12)
 DMP COPD participation            0.00      (0.06)    0.01      (0.12)
 DMP diabetes type II
 participation                     0.04      (0.20)    0.10      (0.30)
 DMP diabetes type I
 participation                     0.00      (0.03)    0.00      (0.03)
 DMP CHD participation             0.01      (0.11)    0.04      (0.20)
PCP characteristics (B)
 Group practice                    0.39      (0.49)    0.41      (0.49)
 Number of patients treated      204.55    (136.08)  235.03    (142.66)
 Offer of DMP                      0.67      (0.47)    0.86      (0.34)
Number of observations                245,246               36,209

Note: (A) The labor force indicator refers to being an employee, as
only this information is documented in the claims data. (B) PCP
characteristics refer to the gatekeeping PCP for future gatekeeping
participants. For non-participants, PCP characteristics refer to the
patient's most frequently contacted PCP (> 50% of contacts).
Source: Own calculations based on data for a major health insurance
company in the German state of Baden-Wurttemberg (control variables
relate to the year before introduction of the gatekeeping program in
2009).

Table 2 Estimated impacts of gatekeeping based on optional empirical
strategies

                                      (1)          (2)
                             Sample   Basic        Instrumental
                            average   regressions  variable

Coordination/continuity of
healthcare
Number of PCPs                 1.55   -0.16 (***)      -0.30 (***)
                                      (0.00)           (0.01)
Specialist visits
with referral                  1.35    0.49 (***)       0.51 (***)
                                      (0.01)           (0.03)
Specialist visits
without referral               2.70   -0.43 (***)      -0.55 (***)
                                      (0.02)           (0.05)
Participation in DMPs of
newly diagnosed...
... DMP asthma                 0.04    0.02 (**)        0.04
                                      (0.01)           (0.03)
... DMP COPD                   0.05    0.02 (*)         0.04 (*)
                                      (0.01)           (0.03)
... DMP diabetes type II       0.29    0.12 (***)       0.10 (**)
                                      (0.02)           (0.04)
... DMP CHD                    0.08    0.02 (**)        0.05 (**)
                                      (0.01)           (0.02)
Healthcare quality
Influenza vaccination          0.35    0.07 (***)       0.05 (***)
                                      (0.00)           (0.13)
General health check-up        0.27    0.02 (***)       0.02 (***)
                                      (0.00)           (0.01)
ACSH                           0.02   -0.00 (**)       -0.01 (**)
                                      (0.00)           (0.00)
Cost indicator
Log ambulatory costs (A)     555.11    0.15 (***)       0.09 (***)
                                      (0.00)           (0.15)
First stage: regional
participation rate                                      1.16 (***)
(instrument)                                           (0.01)
First stage
partial F-statistic                                17,895.6
Sample size                  281.635

                            (3)
                            PCP
                            fixed
                            effects
Coordination/continuity of
healthcare
Number of PCPs              -0,15 (***)
                            (0,01)
Specialist visits
with referral                0,65 (***)
                            (0,02)
Specialist visits
without referral            -0,34 (***)
                            (0,02)
Participation in DMPs of
newly diagnosed...
... DMP asthma               0,01
                            (0,02)
... DMP COPD                 0,01
                            (0,02)
... DMP diabetes type II     0,09 (***)
                            (0,03)
... DMP CHD                  0,03 (**)
                            (0,02)
Healthcare quality
Influenza vaccination        0,08 (***)
                            (0,01)
General health check-up      0,02 (***)
                            (0,00)
ACSH                        -0,00 (***)
                            (0,00)
Cost indicator
Log ambulatory costs (A)     0,23 (***)
                            (0,01)
First stage: regional
participation rate
(instrument)
First stage
partial F-statistic
Sample size

Note: The first column refers to average numbers ((A) average, not in
log, for cost indicator). All other numbers correspond to estimated
coefficients (with standard errors in parentheses), unless otherwise
stated. (*) Significant at the 10 % level, (**) 5 % level, (***) 1 %
level.
Source: Own estimations based on data for a major health insurance
company in the German state of Baden-Wurttemberg (in 2013, control
variables relate to the year before introduction of the gatekeeping
program in 2009).


Please Note: Illustration(s) are not available due to copyright restrictions.
COPYRIGHT 2017 Addleton Academic Publishers
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2017 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Hofmann, Sarah M.; Muhlenweg, Andrea M.
Publication:American Journal of Medical Research
Date:Oct 1, 2017
Words:10958
Previous Article:WHY DIDN'T MORE ETHNIC MEXICANS SIGN UP FOR OBAMACARE?: EXAMINING PUBLIC HEALTH'S IMPACT OVER ONE HUNDRED FIFTY YEARS.
Next Article:AN EXAMINATION OF THE BARRIERS TO BREAST CANCER CARE IN CALIFORNIA.
Topics:

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters