Social Science & Medicine 124 (2015) 18e28
Contents lists available at ScienceDirect
Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed
Paying for primary care: A cross-sectional analysis of cost and morbidity distributions across primary care payment models in Ontario Canada David Rudoler a, b, e, *, Audrey Laporte a, b, e, Janet Barnsley a, b, Richard H. Glazier a, b, c, d, Raisa B. Deber a, b, e a
Institute of Health Policy, Management and Evaluation, University of Toronto, Canada Institute for Clinical Evaluative Sciences, Ontario, Canada Centre for Research on Inner City Health, and St. Michael's Hospital, Ontario, Canada d Family and Community Medicine, University of Toronto and St. Michael's Hospital, Canada e Canadian Centre for Health Economics, Canada b c
a r t i c l e i n f o
a b s t r a c t
Article history: Received 29 November 2013 Received in revised form 22 October 2014 Accepted 4 November 2014 Available online 5 November 2014
Policy-makers desire an optimal balance of financial incentives to improve productivity and encourage improved quality in primary care, while also avoiding issues of risk-selection inherent to capitationbased payment. In this paper we analyze risk-selection in capitation-based payment by using administrative data for patients (n ¼ 11,600,911) who were rostered (i.e., signed an enrollment form, or received a majority of care) with a primary care physician (n ¼ 8621) in Ontario, Canada in 2010/11. We analyze this data using a relative distribution approach and compare distributions of patient costs and morbidity across primary care payment models. Our results suggest a relationship between being in a capitation-based payment scheme and having low cost patients (and presumably healthy patients) compared to fee-for-service physicians. However, we do not have evidence that physicians in capitationbased models are reducing the care they provide to sick and high cost patients. These findings suggest there is a relationship between payment type and risk-selection, particularly for low-cost and healthy patients. © 2014 Elsevier Ltd. All rights reserved.
Keywords: Canada Ontario Primary care Payment incentives Case mix Relative distribution Physician behavior
1. Introduction Payment incentives are one intervention that policy-makers use to achieve reform objectives. With respect to primary care, there have been a number of recent examples: the Quality for Outcomes Framework (QOF) in the UK linked physician payment to measures of performance (Roland, 2004; Doran et al., 2006); and the Shared Savings Program for US Medicare has provided financial incentives for providers to work together to reduce costs and improve quality of care (Fisher et al., 2009). Despite the focus on incentives as a policy tool, there are gaps in our understanding about the success and failure of such policies, and about the optimal way to pay doctors. Robinson (2001) provides a comprehensive overview of physician payment models; he suggests the worst forms of * Corresponding author. Institute of Health Policy, Management and Evaluation, University of Toronto, 155 College Street, Toronto, Ontario, Canada M5T 3M6. E-mail address: [email protected] (D. Rudoler). http://dx.doi.org/10.1016/j.socscimed.2014.11.001 0277-9536/© 2014 Elsevier Ltd. All rights reserved.
payment are ‘pure’ payment schemes (including fee-for-service (FFS), capitation (CAP), and salary), as opposed to blended models. It is well accepted, for instance, that FFS encourages increased service volume, but raises issues with respect to overtreatment, and generates budgetary uncertainty for funders ger, 2008). On the other hand, such prospective (Evans, 1974; Le payment models as CAP and salary discourage over-treatment, but may result in under-treatment and risk selection based on patient ger, 2008). complexity (Bloor and Maynard, 2006; Le Risk selection has been an important issue in payment reform for primary care physicians; for instance, Gravelle et al. (2010) found evidence of general practitioner gaming based on patient characteristics after the introduction of the QOF in the UK. To some extent, this can be minimized by risk-adjusting payments, but the state of risk-adjustment is still under development; most rely on age and sex, which has been shown to be inadequate to adjust for patient morbidity (Sibley and Glazier, 2012). However, it is important to note that the relationship between payment and physician
D. Rudoler et al. / Social Science & Medicine 124 (2015) 18e28
behavior predicted by economic models is often constrained by ethical and professional responsibilities (Ellis and McGuire, 1986). Nonetheless, payment models must find an optimal balance of incentives in order to avoid issues of overuse, underuse and misuse (Institute of Medicine, 2001; Deber et al., 2008). Some have proposed that the optimal payment model includes a mix of FFS and prospective payment (e.g., capitation, salary) (Ellis and McGuire, 1986; Ma, 1994). Payment reform for primary care physicians (PCPs) in Ontario, Canada (Canada's most populous province), presents an interesting context to study the relationships between payment and physician behavior. The province of Ontario has undertaken a number of reforms to the way PCPs are compensated, adding mixed payment schemes to traditional FFS (Hutchison and Glazier, 2013). These payment schemes rely on formal patient rostering, which involves having patients sign an enrollment form that designates a particular PCP as their usual source of primary care. Formal rostering was introduced as a way of making the relationship between the PCP and their patients explicit and formalizing the obligations of both parties. It also allowed for the transition from FFS to alternative forms of payment, such as mixed capitation that is largely based on age and sex adjusted payments and on PCPs having a definable patient population (CHSRF, 2010). However, PCPs could still be underpaid for high complexity patients and overpaid for low complexity patients, further exacerbating incentives to risk-select. In addition, healthcare utilization and costs are highly skewed even within age-sex groups (Deber and Lam, 2009), with a small number of patients accounting for a disproportionate amount of healthcare costs. PCPs need only avoid a small number of patients to significantly reduce the demand on their resources. There is considerable empirical literature on the relationship between payment and healthcare provider behavior. A number of studies consider the relationship between payment and health system outcomes. For instance, Hutchison et al. (1996) conducted a retrospective cohort study to determine the impact of capitation payment on hospital utilization; they found physician payment did not have an effect on utilization rates. In addition, there is considerable literature on physician productivity under different forms of remuneration (e.g., Evans, 1974; Brown and Lapan, 1979; Ellis and McGuire, 1986; Thornton and Eakin, 1997; Conrad et al., 1998, 2002; Fortin et al., 2008; Dumont et al., 2008). There has also been a significant amount of research done, particularly in the US, on risk selection and adverse selection amongst hospitals and insurance plans (e.g., Ellis, 1998; Frank et al., 2000; Luft and Miller, 1988). To our knowledge, there are comparatively fewer studies on the relationship between payment and physician selection of patients. Sorbero et al. (2003) examined patient selection of PCPs using case mix, payment type, and healthcare utilization as explanatory variables in three independent practice associations in the US. The authors found that high users of healthcare services were more likely to change physicians if their current physician received capitation than if they received FFS payment. The authors conjectured that this was either due to PCP risk selection of their “high cost” patients, or that high users were more sophisticated consumers of care and were more easily dissatisfied. Glazier et al. (2009) compared patient and practice characteristics across compensation schemes in Ontario. The authors used cross-sectional administrative data to find that PCPs in CAP models serve patient populations with higher income, and lower morbidity and co-morbidity levels. Devlin and Sarma (2008) and Sarma et al. (2010) conducted studies in the Canadian context using data from a cross-sectional national survey of physicians to determine the impact of remuneration on the quantity of visits provided. They compared FFS
19
remuneration with alternative remuneration schemes and found that PCPs compensated via FFS have significantly higher levels of output, even after controlling for the self-selection of physicians into different payment schemes. As mentioned earlier, Gravelle et al. (2010) found a relationship between the introduction of pay for performance incentives through the QOF in the UK, and gaming based on patient characteristics. Specifically, they found GPs would report a higher number of patients that were eligible for treatments targeted by financial incentives. Kralj and Kantarevic (2013) compared the quantity and quality of primary care services in mixed CAP and mixed FFS schemes in Ontario. The authors used a longitudinal administrative dataset to follow a cohort of physicians between 2006 and 2009. They found PCPs who receive CAP payment provide approximately 6% fewer services and visits per day, and were 8%e15% more likely to obtain bonuses for preventative care (e.g., payment for cancer screening), which the authors associate with high quality care. The authors also considered differences in the complexity of formal patient rosters (measured by age and sex adjusted multipliers) and found no significant difference across payment models. However, age and sex does not fully capture complexity (Sibley and Glazier, 2012; Deber and Lam, 2009). In this paper we examine the extent to which selection is occurring and study its implications for physician costs. The existing evidence, particularly in the Canadian context, seems inconclusive with respect to the existence of patient selection by providers. We also conduct this analysis to determine if there is sufficient evidence to further model and empirically analyze physician behavior under different forms of remuneration. While local context certainly plays a role, we believe the results of this study will have important implications for payment reforms for healthcare providers, particularly in jurisdictions contemplating the introduction of multiple voluntary payment schemes. This includes the UK and the US, where payment reforms are frequent and ongoing, as well as such jurisdictions as Norway, where physicians also receive combinations of FFS and capitation payment (Lindahl and Ringard, 2013). 2. Ontario's payment models In Canada, healthcare falls under provincial jurisdiction, and there can be considerable variation in how care is delivered and paid for within and between provinces/territories (Marchildon, 2013). To receive full federal transfer payments, all provincial/territorial insurance plans must fully cover all “insured services” provided to all “insured persons” (defined as Canadian residents). For historical reasons, the definition of insured services includes all medically necessary care delivered in hospitals or by physicians; provinces/territories can insure beyond these requirements, but are not required to. Canada uses what the OECD calls a public contracting model, whereby private providers (including physicians) receive public payment for insured services (Docteur and Oxley, 2003). Ontario thus uses a single payer insurance model, where all legal residents of the province are enrolled into the Ontario Health Insurance Plan (OHIP); patients must provide their OHIP number to receive insured services. Traditionally, most physicians in Canada were paid on a FFS basis, using a fee schedule jointly negotiated by the provincial ministry of health and the provincial medical association. The fee schedule did not incorporate financial incentives to maintain ongoing relationships with patients, practice in groups or hire multidisciplinary providers, although physicians were able to do so should they so desire. Although non FFS models had long existed in Ontario, they involved only a small proportion of PCPs. Starting in the early
20
D. Rudoler et al. / Social Science & Medicine 124 (2015) 18e28
2000s, the Ontario government began to introduce a series of alternatives to FFS in an attempt to encourage uptake of primary care by medical students, control costs, and provide incentives to improve access and quality of care. These new payment models were introduced as voluntary models and were not imposed upon PCPs. One key element of many of these models was formal rostering. As mentioned earlier, formal rostering involves having patients sign an enrollment form that designates a particular PCP as their usual source of primary care, and capitation payments are only made for patients that are formally rostered. However, it is important to note that PCPs can still provide care to non-rostered patients and receive FFS payment. The key models examined in this study are briefly described below. In 2002, the Ontario government introduced the Family Health Network (FHN) e a blended capitation (CAP) payment model e that established requirements for PCPs to practice in groups of three or more physicians. The capitation component of the model comprises an age and sex adjusted monthly payment for each patient formally rostered to the individual PCP (rather than to the practice). Monthly payments for each patient formally rostered beyond a cap of 2400 patients are reduced by 50 percent. All PCPs who moved into the FHN model were obligated to offer formal rostering to their existing patients in order to maintain continuity of care, but patients could decline to be formally rostered. The capitation payment in a FHN covers a ‘basket’ of 56 services considered part of delivering comprehensive primary care. The FFS component of the FHN model is for all services outside of this ‘basket,’ and for all services provided to non-rostered patients; this component was capped at $56,000 per year. Therefore, PCPs could continue to provide care to a limited number of patients who remained off the roster, as long as the total amount billed did not exceed this cap. In 2003, the Ontario government introduced an enhanced feefor-service (EFF) payment model that was seen as a compromise between CAP and FFS. The model, called the Family Health Group (FHG), also requires PCPs to work in groups of three or more, and to provide extended hours. PCPs are also strongly encouraged to formally roster patients. In 2005, another EFF model, the Comprehensive Care Model (CCM) was introduced; this is a solo-practice model which is similar to the FHG but does not require group practice. In 2006, the Ontario government introduced a variation on the CAP model, called the Family Health Organization (FHO). This model is very similar to the FHN, but includes a larger ‘basket’ of 119 services, and a higher monthly capitation payment rate. In 2002, 94% of all Ontario PCPs were receiving traditional FFS. By 2012, only 24% were receiving traditional FFS; 39% of all Ontario PCPs were receiving CAP payment, 29% were receiving EFF payment; the remainder received salary, blended salary and other specialized payments (Hutchison and Glazier, 2013). In this study we focus on FFS, EFF and CAP payment given these models include over 90% of all PCPs in Ontario. We also consider PCPs who work under the Family Health Team (FHT) model. The FHT was introduced in 2004 and is a practice model in which PCPs receive additional funding from the Ontario government to hire interdisciplinary providers (e.g., nurses, pharmacists, dietitians, and others). However, the FHT is not a payment model; rather, the vast majority of FHT PCPs receive CAP payment (approximately 93%) (Glazier et al., 2012). A breakdown of our included models is provided in Table 1. We have excluded several other Ontario primary care models from our study, including the Community Health Centre (CHC), Group Health Centre (GHC), and Rural Northern Physician Group Agreement both due to data availability (in the case of the CHC),
and because they include a very small proportion of both PCPs and the Ontario population (~1%) (Glazier et al., 2012).
3. Analytical approach We analyzed patient selection using a non-parametric approach e the relative distribution approach developed by Handcock and Morris (1999) e that allowed us to analyze the relationship between physician payment, risk selection and healthcare costs. Contoyannis and Wildman (2007) used this approach to study changes in obesity in Canada and England, and we closely follow their presentation of these methods in our paper. The relative distribution approach is a non-parametric statistical tool that allows the researcher to compare the distributions of two groups with respect to a discrete or continuous variable (e.g., morbidity and costs). This approach allows the researcher to represent the values of one group (called the “comparison group”) within the distribution of another group (called the “reference group”). Handcock and Morris (1999) explain the approach as follows. Let the value of a random variable for the reference group and the comparison group be represented as Y0 and Y respectively. The relative distribution of Y and Y0 is defined in terms of the random variable R:
R ¼ FO ðYÞ
(1)
R is a value within the distribution of the comparison group (Y) transformed by the cumulative distribution function (CDF) for the reference group (F0). Since R is the reference group quantile where a value Y from the comparison group lies, it is a random variable which has a probability density function (PDF) (the relative density) and a CDF (the relative distribution). The CDF of R is as follows:
G r ¼ F F01 ðrÞ ;
0r1
(2)
This is also the quantile function of F0, and r is any rank on the CDF of Y0. The PDF of R is the derivative of G(r) with respect to r:
f F01 ðrÞ ; gðrÞ ¼ f0 F01 ðrÞ
0r1
(3)
Equation (3) can be interpreted as the ratio of the densities of the comparison group and the reference group evaluated at the quantiles of the reference group, but is also a PDF in that it integrates to 1. For interpretation of the results below, it is important to note that values of g(r) will equal 1 when there are no distributional differences between the comparison and reference groups. By displaying the PDF of the relative distribution, we can clearly observe where there are differences in the two distributions in one graphical figure since values >1 mean that observations at that point in the distribution are more likely to be observed in the comparison group, while values
Contents lists available at ScienceDirect
Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed
Paying for primary care: A cross-sectional analysis of cost and morbidity distributions across primary care payment models in Ontario Canada David Rudoler a, b, e, *, Audrey Laporte a, b, e, Janet Barnsley a, b, Richard H. Glazier a, b, c, d, Raisa B. Deber a, b, e a
Institute of Health Policy, Management and Evaluation, University of Toronto, Canada Institute for Clinical Evaluative Sciences, Ontario, Canada Centre for Research on Inner City Health, and St. Michael's Hospital, Ontario, Canada d Family and Community Medicine, University of Toronto and St. Michael's Hospital, Canada e Canadian Centre for Health Economics, Canada b c
a r t i c l e i n f o
a b s t r a c t
Article history: Received 29 November 2013 Received in revised form 22 October 2014 Accepted 4 November 2014 Available online 5 November 2014
Policy-makers desire an optimal balance of financial incentives to improve productivity and encourage improved quality in primary care, while also avoiding issues of risk-selection inherent to capitationbased payment. In this paper we analyze risk-selection in capitation-based payment by using administrative data for patients (n ¼ 11,600,911) who were rostered (i.e., signed an enrollment form, or received a majority of care) with a primary care physician (n ¼ 8621) in Ontario, Canada in 2010/11. We analyze this data using a relative distribution approach and compare distributions of patient costs and morbidity across primary care payment models. Our results suggest a relationship between being in a capitation-based payment scheme and having low cost patients (and presumably healthy patients) compared to fee-for-service physicians. However, we do not have evidence that physicians in capitationbased models are reducing the care they provide to sick and high cost patients. These findings suggest there is a relationship between payment type and risk-selection, particularly for low-cost and healthy patients. © 2014 Elsevier Ltd. All rights reserved.
Keywords: Canada Ontario Primary care Payment incentives Case mix Relative distribution Physician behavior
1. Introduction Payment incentives are one intervention that policy-makers use to achieve reform objectives. With respect to primary care, there have been a number of recent examples: the Quality for Outcomes Framework (QOF) in the UK linked physician payment to measures of performance (Roland, 2004; Doran et al., 2006); and the Shared Savings Program for US Medicare has provided financial incentives for providers to work together to reduce costs and improve quality of care (Fisher et al., 2009). Despite the focus on incentives as a policy tool, there are gaps in our understanding about the success and failure of such policies, and about the optimal way to pay doctors. Robinson (2001) provides a comprehensive overview of physician payment models; he suggests the worst forms of * Corresponding author. Institute of Health Policy, Management and Evaluation, University of Toronto, 155 College Street, Toronto, Ontario, Canada M5T 3M6. E-mail address: [email protected] (D. Rudoler). http://dx.doi.org/10.1016/j.socscimed.2014.11.001 0277-9536/© 2014 Elsevier Ltd. All rights reserved.
payment are ‘pure’ payment schemes (including fee-for-service (FFS), capitation (CAP), and salary), as opposed to blended models. It is well accepted, for instance, that FFS encourages increased service volume, but raises issues with respect to overtreatment, and generates budgetary uncertainty for funders ger, 2008). On the other hand, such prospective (Evans, 1974; Le payment models as CAP and salary discourage over-treatment, but may result in under-treatment and risk selection based on patient ger, 2008). complexity (Bloor and Maynard, 2006; Le Risk selection has been an important issue in payment reform for primary care physicians; for instance, Gravelle et al. (2010) found evidence of general practitioner gaming based on patient characteristics after the introduction of the QOF in the UK. To some extent, this can be minimized by risk-adjusting payments, but the state of risk-adjustment is still under development; most rely on age and sex, which has been shown to be inadequate to adjust for patient morbidity (Sibley and Glazier, 2012). However, it is important to note that the relationship between payment and physician
D. Rudoler et al. / Social Science & Medicine 124 (2015) 18e28
behavior predicted by economic models is often constrained by ethical and professional responsibilities (Ellis and McGuire, 1986). Nonetheless, payment models must find an optimal balance of incentives in order to avoid issues of overuse, underuse and misuse (Institute of Medicine, 2001; Deber et al., 2008). Some have proposed that the optimal payment model includes a mix of FFS and prospective payment (e.g., capitation, salary) (Ellis and McGuire, 1986; Ma, 1994). Payment reform for primary care physicians (PCPs) in Ontario, Canada (Canada's most populous province), presents an interesting context to study the relationships between payment and physician behavior. The province of Ontario has undertaken a number of reforms to the way PCPs are compensated, adding mixed payment schemes to traditional FFS (Hutchison and Glazier, 2013). These payment schemes rely on formal patient rostering, which involves having patients sign an enrollment form that designates a particular PCP as their usual source of primary care. Formal rostering was introduced as a way of making the relationship between the PCP and their patients explicit and formalizing the obligations of both parties. It also allowed for the transition from FFS to alternative forms of payment, such as mixed capitation that is largely based on age and sex adjusted payments and on PCPs having a definable patient population (CHSRF, 2010). However, PCPs could still be underpaid for high complexity patients and overpaid for low complexity patients, further exacerbating incentives to risk-select. In addition, healthcare utilization and costs are highly skewed even within age-sex groups (Deber and Lam, 2009), with a small number of patients accounting for a disproportionate amount of healthcare costs. PCPs need only avoid a small number of patients to significantly reduce the demand on their resources. There is considerable empirical literature on the relationship between payment and healthcare provider behavior. A number of studies consider the relationship between payment and health system outcomes. For instance, Hutchison et al. (1996) conducted a retrospective cohort study to determine the impact of capitation payment on hospital utilization; they found physician payment did not have an effect on utilization rates. In addition, there is considerable literature on physician productivity under different forms of remuneration (e.g., Evans, 1974; Brown and Lapan, 1979; Ellis and McGuire, 1986; Thornton and Eakin, 1997; Conrad et al., 1998, 2002; Fortin et al., 2008; Dumont et al., 2008). There has also been a significant amount of research done, particularly in the US, on risk selection and adverse selection amongst hospitals and insurance plans (e.g., Ellis, 1998; Frank et al., 2000; Luft and Miller, 1988). To our knowledge, there are comparatively fewer studies on the relationship between payment and physician selection of patients. Sorbero et al. (2003) examined patient selection of PCPs using case mix, payment type, and healthcare utilization as explanatory variables in three independent practice associations in the US. The authors found that high users of healthcare services were more likely to change physicians if their current physician received capitation than if they received FFS payment. The authors conjectured that this was either due to PCP risk selection of their “high cost” patients, or that high users were more sophisticated consumers of care and were more easily dissatisfied. Glazier et al. (2009) compared patient and practice characteristics across compensation schemes in Ontario. The authors used cross-sectional administrative data to find that PCPs in CAP models serve patient populations with higher income, and lower morbidity and co-morbidity levels. Devlin and Sarma (2008) and Sarma et al. (2010) conducted studies in the Canadian context using data from a cross-sectional national survey of physicians to determine the impact of remuneration on the quantity of visits provided. They compared FFS
19
remuneration with alternative remuneration schemes and found that PCPs compensated via FFS have significantly higher levels of output, even after controlling for the self-selection of physicians into different payment schemes. As mentioned earlier, Gravelle et al. (2010) found a relationship between the introduction of pay for performance incentives through the QOF in the UK, and gaming based on patient characteristics. Specifically, they found GPs would report a higher number of patients that were eligible for treatments targeted by financial incentives. Kralj and Kantarevic (2013) compared the quantity and quality of primary care services in mixed CAP and mixed FFS schemes in Ontario. The authors used a longitudinal administrative dataset to follow a cohort of physicians between 2006 and 2009. They found PCPs who receive CAP payment provide approximately 6% fewer services and visits per day, and were 8%e15% more likely to obtain bonuses for preventative care (e.g., payment for cancer screening), which the authors associate with high quality care. The authors also considered differences in the complexity of formal patient rosters (measured by age and sex adjusted multipliers) and found no significant difference across payment models. However, age and sex does not fully capture complexity (Sibley and Glazier, 2012; Deber and Lam, 2009). In this paper we examine the extent to which selection is occurring and study its implications for physician costs. The existing evidence, particularly in the Canadian context, seems inconclusive with respect to the existence of patient selection by providers. We also conduct this analysis to determine if there is sufficient evidence to further model and empirically analyze physician behavior under different forms of remuneration. While local context certainly plays a role, we believe the results of this study will have important implications for payment reforms for healthcare providers, particularly in jurisdictions contemplating the introduction of multiple voluntary payment schemes. This includes the UK and the US, where payment reforms are frequent and ongoing, as well as such jurisdictions as Norway, where physicians also receive combinations of FFS and capitation payment (Lindahl and Ringard, 2013). 2. Ontario's payment models In Canada, healthcare falls under provincial jurisdiction, and there can be considerable variation in how care is delivered and paid for within and between provinces/territories (Marchildon, 2013). To receive full federal transfer payments, all provincial/territorial insurance plans must fully cover all “insured services” provided to all “insured persons” (defined as Canadian residents). For historical reasons, the definition of insured services includes all medically necessary care delivered in hospitals or by physicians; provinces/territories can insure beyond these requirements, but are not required to. Canada uses what the OECD calls a public contracting model, whereby private providers (including physicians) receive public payment for insured services (Docteur and Oxley, 2003). Ontario thus uses a single payer insurance model, where all legal residents of the province are enrolled into the Ontario Health Insurance Plan (OHIP); patients must provide their OHIP number to receive insured services. Traditionally, most physicians in Canada were paid on a FFS basis, using a fee schedule jointly negotiated by the provincial ministry of health and the provincial medical association. The fee schedule did not incorporate financial incentives to maintain ongoing relationships with patients, practice in groups or hire multidisciplinary providers, although physicians were able to do so should they so desire. Although non FFS models had long existed in Ontario, they involved only a small proportion of PCPs. Starting in the early
20
D. Rudoler et al. / Social Science & Medicine 124 (2015) 18e28
2000s, the Ontario government began to introduce a series of alternatives to FFS in an attempt to encourage uptake of primary care by medical students, control costs, and provide incentives to improve access and quality of care. These new payment models were introduced as voluntary models and were not imposed upon PCPs. One key element of many of these models was formal rostering. As mentioned earlier, formal rostering involves having patients sign an enrollment form that designates a particular PCP as their usual source of primary care, and capitation payments are only made for patients that are formally rostered. However, it is important to note that PCPs can still provide care to non-rostered patients and receive FFS payment. The key models examined in this study are briefly described below. In 2002, the Ontario government introduced the Family Health Network (FHN) e a blended capitation (CAP) payment model e that established requirements for PCPs to practice in groups of three or more physicians. The capitation component of the model comprises an age and sex adjusted monthly payment for each patient formally rostered to the individual PCP (rather than to the practice). Monthly payments for each patient formally rostered beyond a cap of 2400 patients are reduced by 50 percent. All PCPs who moved into the FHN model were obligated to offer formal rostering to their existing patients in order to maintain continuity of care, but patients could decline to be formally rostered. The capitation payment in a FHN covers a ‘basket’ of 56 services considered part of delivering comprehensive primary care. The FFS component of the FHN model is for all services outside of this ‘basket,’ and for all services provided to non-rostered patients; this component was capped at $56,000 per year. Therefore, PCPs could continue to provide care to a limited number of patients who remained off the roster, as long as the total amount billed did not exceed this cap. In 2003, the Ontario government introduced an enhanced feefor-service (EFF) payment model that was seen as a compromise between CAP and FFS. The model, called the Family Health Group (FHG), also requires PCPs to work in groups of three or more, and to provide extended hours. PCPs are also strongly encouraged to formally roster patients. In 2005, another EFF model, the Comprehensive Care Model (CCM) was introduced; this is a solo-practice model which is similar to the FHG but does not require group practice. In 2006, the Ontario government introduced a variation on the CAP model, called the Family Health Organization (FHO). This model is very similar to the FHN, but includes a larger ‘basket’ of 119 services, and a higher monthly capitation payment rate. In 2002, 94% of all Ontario PCPs were receiving traditional FFS. By 2012, only 24% were receiving traditional FFS; 39% of all Ontario PCPs were receiving CAP payment, 29% were receiving EFF payment; the remainder received salary, blended salary and other specialized payments (Hutchison and Glazier, 2013). In this study we focus on FFS, EFF and CAP payment given these models include over 90% of all PCPs in Ontario. We also consider PCPs who work under the Family Health Team (FHT) model. The FHT was introduced in 2004 and is a practice model in which PCPs receive additional funding from the Ontario government to hire interdisciplinary providers (e.g., nurses, pharmacists, dietitians, and others). However, the FHT is not a payment model; rather, the vast majority of FHT PCPs receive CAP payment (approximately 93%) (Glazier et al., 2012). A breakdown of our included models is provided in Table 1. We have excluded several other Ontario primary care models from our study, including the Community Health Centre (CHC), Group Health Centre (GHC), and Rural Northern Physician Group Agreement both due to data availability (in the case of the CHC),
and because they include a very small proportion of both PCPs and the Ontario population (~1%) (Glazier et al., 2012).
3. Analytical approach We analyzed patient selection using a non-parametric approach e the relative distribution approach developed by Handcock and Morris (1999) e that allowed us to analyze the relationship between physician payment, risk selection and healthcare costs. Contoyannis and Wildman (2007) used this approach to study changes in obesity in Canada and England, and we closely follow their presentation of these methods in our paper. The relative distribution approach is a non-parametric statistical tool that allows the researcher to compare the distributions of two groups with respect to a discrete or continuous variable (e.g., morbidity and costs). This approach allows the researcher to represent the values of one group (called the “comparison group”) within the distribution of another group (called the “reference group”). Handcock and Morris (1999) explain the approach as follows. Let the value of a random variable for the reference group and the comparison group be represented as Y0 and Y respectively. The relative distribution of Y and Y0 is defined in terms of the random variable R:
R ¼ FO ðYÞ
(1)
R is a value within the distribution of the comparison group (Y) transformed by the cumulative distribution function (CDF) for the reference group (F0). Since R is the reference group quantile where a value Y from the comparison group lies, it is a random variable which has a probability density function (PDF) (the relative density) and a CDF (the relative distribution). The CDF of R is as follows:
G r ¼ F F01 ðrÞ ;
0r1
(2)
This is also the quantile function of F0, and r is any rank on the CDF of Y0. The PDF of R is the derivative of G(r) with respect to r:
f F01 ðrÞ ; gðrÞ ¼ f0 F01 ðrÞ
0r1
(3)
Equation (3) can be interpreted as the ratio of the densities of the comparison group and the reference group evaluated at the quantiles of the reference group, but is also a PDF in that it integrates to 1. For interpretation of the results below, it is important to note that values of g(r) will equal 1 when there are no distributional differences between the comparison and reference groups. By displaying the PDF of the relative distribution, we can clearly observe where there are differences in the two distributions in one graphical figure since values >1 mean that observations at that point in the distribution are more likely to be observed in the comparison group, while values
