The relation between physician density and
health outcomes in different financing
systems
Master thesis Ankie van der Kolk
University of Groningen,
Faculty of Economics and Business
MSc Business Administration
Organizational and Management Control
June 2014
Name: Ankie van der Kolk
Student number: 1919105
Address: Vrydemalaan 222
Postal code: 9713 WT
City: Groningen
Mobile: 0634688288
E-mail: ankievdkolk@hotmail.com
Supervisor: Ms. van de Mortel
1 ABSTRACT
The escalating healthcare costs are putting pressure on the budgets of governments and healthcare organizations. Most studies attempting to contribute to a solution investigate the direct effect of (public/private) healthcare expenditure on health outcomes, leading to contradictory results. Recent analyses indicated that it is important to consider real rather than financial resources since these have a more direct effect on health outcomes. This study investigates the relation between physician density and health outcomes and the moderating impact of public financing in 34 OECD countries during 1960-2013. Regression results show that physician density is associated with beneficial health outcomes and that this relation is negatively moderated by the amount of public financing. Although more research is necessary the findings indicate that physicians are a valuable resource in improving health outcomes and governments could benefit of using more private financing in situations where physician density is high.
2 PREFACE
This thesis investigates the relation between physician density and health outcomes and the moderating impact of public financing. Writing this thesis was the final and most time consuming part of my master BA Organizational and Management Control. It attempts to shed some light on possibilities to optimize the functioning of the healthcare sector across countries in order to control costs while maintaining or improving quality. Besides controlling this topic reflects my interest in international economics and business, exact sciences, the public sector and especially in healthcare. The freedom granted to students when writing their thesis made it the most interesting, but also the most difficult part of my master. I would like to use this preface to thank several people that supported me during this process. First of all I would like to thank my supervisor Ms. van de Mortel for her supportive feedback. I would also like to thank Mr. van der Bij and Mr. Noseleit for the time they took each time that I dropped by their office with questions regarding the regression analysis. Finally I would thank my father for his financial support during my study and his emotional support during the writing of my thesis, especially in the last week. Despite the difficulties encountered writing this thesis was an instructive experience that allowed me dive deeper into the topics of my interest. I hope I have been able to transfer some of that interest to the people reading this thesis.
3 TABLE OF CONTENTS Abstract Preface 1. Introduction ………... 4 1.1 Motivation………. 4 1.2 Problem statement………. 5 1.3 Scientific relevance……… 8 1.4 Practical relevance………. 8
1.5 Structure of this thesis………... 9
2. Theory and hypotheses………...………... 10
2.1 The healthcare sector………. 10
2.2 Physician density and health outcomes………. 12
2.3 Public financing……….……… 15
2.4 Other mechanisms for cost containment………... 19
3. Methodology………... 21
3.1 Sample and data sources………... 21
3.2 Measurement of the variables……….... 22
3.3 Descriptive statistics………... 26
3.4 Model and validity………... 27
4. Results………. 30
5. Conclusion and discussion……….. 34
6. References………... 36
4 1. INTRODUCTION
1.1 Motivation
During the past years several developments have led to a rise in healthcare costs across Organization of Economic Cooperation and Development (OECD) countries. The number of elderly, who make most use of the healthcare system, is increasing. There is a high pace of technological innovation which enables more effective treatment of diseases. Consumer expectations are rising because of the increased possibilities for choice in other areas of life (Walshe & Smith, 2011). These developments led to an increase in life expectancy and lower mortality due to diseases such as cancer (OECD, 2014). However, they were accompanied by a sharp increase in costs that was combined with a decrease in Gross Domestic Product (GDP) during the economic downturn in 2008-2009. Because of this the ratio of health spending to GDP increased sharply, putting pressure on many government budgets (figure 1).
Figure 1: Average OECD public healthcare spending as a percentage of GDP1
Without effective policy action, costs are expected to increase further in the future. Several factors could even increase the rate at which costs are rising, such as an extended pre-death period of ill health as life expectancy increases, higher than expected technological costs if new demands can be met by innovations and an increased dependency due to obesity trends or dementia (De la Maisonneuve & Martins, 2013). Without an increase in cost-effectiveness, health care demand will undermine public finances (OECD, 2013a). Therefore controlling healthcare costs is a first-order policy issue for most governments and healthcare organizations (De la Maisonneuve & Martins, 2013).
1 Calculated as an average of countries for which data was available from the OECD database (OECD,
5 1.2 Problem statement
Many countries struggle to control their healthcare costs. This quest is further complicated by concerns that attempts to contain costs might reduce the quality of healthcare services (Litvak & Long, 2000). Quality is of central importance in the healthcare sector since people’s lives are at stake. Therefore quality should take a central role when considering healthcare reforms. Some studies suggest that there might still be possibilities to contain costs while maintaining or improving quality.
Countries that spend the most on healthcare are not necessarily those that perform the best in terms of health outcomes, which suggests there is scope to improve the cost-effectiveness of spending. It is estimated that in more than one third of OECD countries, exploiting efficiency gains in the healthcare sector would allow improving health outcomes as much as over the previous decade while keeping spending constant (Joumard, André & Nicq, 2010). Therefore researchers and practitioners are looking for possible policy adjustments that could enable governments to reach these outcomes. Because of the present focus among OECD countries on containing healthcare costs, this search is mainly guided by investigating the effectiveness of health expenditure (Anell and Willis, 2000). Several studies indicate that higher expenditure is associated with improved health outcomes (Berger & Messer, 2002; Crémieux, Ouellette & Pilon, 1999; Hitiris & Posnett, 1992; Joumard, André, Nicq & Chatal, 2008; Wolfe, 1986). Other studies find no relationship (Asiskovitch 2010; Judge, Mulligan & Benzeval, 1998; LeGrand, 1987; Self & Grabowski, 2003) or suggest effects are different across population subcategories (Nixon & Ulmann, 2006).
These studies do not reach consensus regarding the effectiveness of expenditure in improving health outcomes. It could be that this is caused by methodological issues, such as the different measures of health outcomes employed. It could also be that these studies have adopted a simple view by investigating the direct effect of levels of expenditure on health outcomes. These studies have adopted an input-output approach while expenditure can be transformed into different types of resources. Joumard, André, Nicq and Chatal (2010) find that OECD countries vary in their potential to transfer funds into health outcomes in terms of life expectancy, premature mortality and infant mortality. They suggest that this could be because how money is spent is at least as important as how much money is spent. Similarly, Anell & Willis (2000) pointed to the tendency of researchers to use healthcare expenditure to assess the effect of governments’ investment decisions in healthcare. However, expenditure can be transformed into different resources. The amount and mix of these resources will have a more direct impact on outcomes than amount of financing. They emphasize that they did not found correlations between healthcare expenditure and the stock of resources across several countries.
6 perspective. It is estimated that on average 40-50 percent of healthcare budgets exists of wages and salaries (Hernandez, Dräger & Evans, 2006). Especially physicians have large salaries, partially reflecting necessary training (OECD, 2011). Also, they have a central role in healthcare because of their high status position within the professional hierarchy (Nembhard & Edmonson, 2011). Therefore this thesis will investigate the relation between physician density and health outcomes.
Similarly as the tendency of researchers to investigate the direct impact of health expenditure, many studies focus on the direct effects of the financing mix. The financing mix is an important policy decision in the context of cost control. Healthcare expenditure can be separated into public and private components. While public financing exists of funds spent by governments or social security schemes, private spending includes voluntarily payments by the population (OECD, 2013b). Which combination (mix) is most appropriate to contain healthcare costs while at the same time maintaining or improving quality is a topic of debate.
7 Contradictory findings might be explained by incorporating real resources in the analysis (Anell & Willis, 2000; Joumard et al., 2010). Similarly as researchers focussing on total expenditure, studies investigating the financing mix rely on a simple perspective in which proportions of expenditure are linked directly to health outcomes. Therefore this study will investigate the moderating effect of the amount of public financing on the relation between physician density and health outcomes. It could be that physicians perform differently when different proportions of private and public expenditure are used.
Previous studies investigating these issues are scarce, especially recent ones. While most recent studies show a positive relation between physician density and health outcomes (Joumard et al., 2008; Kinfu, Dal Poz, & Evans; 2006; Nixon & Ullman, 2006; Speybroeck, Kinfu, Dal Poz & Evans; 2008) other studies contradict these results (Hertz, Herbert & Landon, 1994; Kim and Moody, 1992). Only two studies investigated the effect of the financing mix on this relation. Puig-Junoy (1998) investigated how the financing mix moderates the relation between a component of resources, including physicians, and health outcomes. Private financing appeared to strengthen the positive relation. In contrast, Or, Wang and Jaminson (2005) find that public financing does not significantly moderate the relation between physician density and several measures of health outcomes. The lack of studies investigating these issues and the conflicting results indicate that more research is needed.
This study will investigate the relation between physician density and health outcomes and the moderating impact of public financing. Although the financing mix is a central issue of debate among researchers and practitioners, other intervention mechanisms might offer possibilities for cost containment in the healthcare sector as well (Wranik, 2011). Therefore the moderating impact of several other intervention mechanisms will also be investigated
Research question
What is the relation between physician density and health outcomes in different financing systems?
To answer this, the following subquestions will be discussed: What is the healthcare sector?
What constitutes healthcare performance?
Why does the government interfere in the healthcare sector? What is healthcare financing?
Which other mechanisms are available for cost containment?
8 1.3 Scientific relevance
The investigation of the relation between physician density and health outcomes in different financing systems could have important scientific implications. Previous studies attempting to find solutions to the escalating healthcare costs mainly investigated the relation between (public/private) expenditure and health outcomes. Because expenditure can be transformed into different types of resources, this study attempts to use a more direct perspective by focusing on physicians. Furthermore, the potential for interaction effects between real and financial resources is investigated by incorporating the moderating impact of public financing in the analysis. It might be that different amounts of public financing influence how physicians perform. Results give an indication of the necessity of researchers to use more complex models when analysing health expenditure. 1.4 Practical relevance
9 1.5 Structure of this thesis
10 2. THEORY AND HYPOTHESES
This chapter will discuss previous theories about cost containment in the healthcare sector in order to formulate hypotheses that can subsequently be tested. The first section will explain the context of this discussion by elaborating on the healthcare sectors characteristics. In the next section, healthcare system performance and the relation between physicians and health outcomes will be discussed. The subsequent section will elaborate on the impact of public financing on this relation. Theories in favour of and against government intervention and applications to the healthcare sector will be discussed next. Subsequently, financing will be distinguished from other intervention mechanisms. Previous findings regarding public financing will be summarized. Besides adjustment of the financing mix there are several other government intervention mechanisms that may contribute to a solution to the problems within the healthcare sector. Therefore the final section will focus on alternative intervention mechanisms.
2.1 The healthcare sector
To investigate possibilities to improve the performance of the healthcare sector, it is important to define what constitutes healthcare. Healthcare is one of the largest sectors in most developed countries: it compromises 8 – 15 percent of those economies. On average one of ten workers in those economies is employed in the healthcare sector (Walshe & Smith, 2011). Therefore, healthcare plays a prominent role within countries. Healthcare can be defined as follows (The American Heritage Medical Dictionary, 2008, p5):
“Healthcare is the treatment and management of illness and the preservation of mental and physical well-being through the services offered by the medical and allied health
professions”
Healthcare includes a broad range of activities offered across a diverse set of institutions. Activities are often summarized as primary, secondary and tertiary care. Primary care involves the first contact of patients with a healthcare system. Patients consult a General Practitioner (GP), who may refer them to secondary care in which they are treated by specialists. Tertiary care involves treatment in a specialized center with highly trained specialists and advanced technology – usually on referral from other providers (The American Heritage Medical Dictionary, 2008). These activities are performed in a broad range of health institutions including specialized hospitals, rehabilitation centers, dental centers and other health institutions which have accommodation facilities. Other institutions provide health services without such facilities, for instance those attached to firms, schools, homes for the aged and labor organizations (United Nations, 2008). These organizations attempt to relieve health problems of varying severity across the population.
11 quality of care (Litvak & Long, 2000). Resistance is further strengthened by the large entrenchment of the healthcare sector into the populations across countries. Almost everyone uses healthcare or knows someone who is a significant healthcare user and many people are employed within the sector. This is sometimes demonstrated in large opposition to healthcare reforms (Walsche & Smit, 2011). Despite these concerns about the quality of care others emphasize that reforms are necessary because healthcare systems consume a large amount of resources and are inefficiently organized (Musgrove, 1996).
Because of these notions governments have been attempting to reform healthcare in the face of the rising expenditures. These reforms attempt to optimize the internal functioning of healthcare organizations in order to increase their efficiency in producing health outcomes. Evans (2005) notes that despite the unique set of conditions within countries, parallel developments of reform are broadly similar. When comparing developments across countries, two phases can be identified. The first phase existed of the establishment of a near universal and comprehensive system of collective payment for healthcare. This development was prompted by political events such as the Second World War. The second phase – from the mid 1970 onwards – was one of attempts to contain costs, which has been difficult because of opposition of the pharmaceutical industry, doctors and the population. The figure below (figure 2) illustrates these developments by showing the average percentage of health expenditure financed by the government across OECD countries during the past years. Public involvement sharply increased till mid-1970 after which it slightly decreased.
Figure 2: Average percentage of public financing in the OECD2
Although governments are in constant attempts to reform healthcare, Maynard (2005) notes that they are often supported by a weak evidence based which has resulted in a lack of clarity in defining public goals, establishing trade-offs and aligning incentives with those objectives. Therefore there is a need for studies to support policy decisions.
2
12 2.2 Physician density and health outcomes
To determine how policies could be adjusted to increase performance of the healthcare sector, this section will elaborate on previous findings regarding the effects of healthcare expenditure – in terms of real or financial resources – on health outcomes. Based on these findings a hypothesis will be formulated describing the expected relation between physician density and health outcomes.
To illustrate how this may contribute to increased performance of the healthcare system, it is important to define what constitutes performance. Performance can be broadly or narrowly defined depending on which outcomes are considered (Venkatraman & Ramanujam, 1986). The focus can be on operational and financial objectives but could also emphasis the multiple and conflicting nature of organizational goals and the influence of multiple stakeholders (Cameron & Whetten, 1983a; 1983b). Due to the semi-public nature of healthcare, governments usually focus on multiple objectives that take into account the interests of a range of stakeholders. In healthcare, governments usually strive for the objectives good health, cost control, equity in terms of healthcare access and consumer satisfaction (Musgrove, 1996). Different healthcare policy objectives can be conflicting, as illustrated by beneficial health outcomes that were combined with the increase in costs over the past decade. Therefore governments are forced to set priorities when reforming healthcare.
13 contemporary measures that provide alternatives to healthcare expenditure (Anell & Willis, 2000).
In an attempt to reach consensus on factors explaining health outcomes, authors emphasize that financial resources can be transformed in different combinations of real resources. Investigating the effect of these resources might clarify previous contradictory results (Anell & Willis 2000; Joumard, André, Nicq & Chatal, 2010). The figure below (figure 3) illustrates how countries may differ along a range of resources employed within their healthcare organizations (Anell and Willis, 2000).
Figure 3: Spider-web diagrams showing differences between the UK and the US along multiple categories of resources (Anell and Willis, 2000)3
% GDP = health exp. as % of GDP Beds/cap = number of hospital beds per capita Exp/cap = health exp. per capita Emp/cap = health care employment per capita Drugs/cap = drug expenditures per capita Phys/cap = number of physicians per capita MRIs = MRI units per capita Nurses/cap = number of nurses per capita CT Scanners = CT scanners per capita Emp = health care employment as % of
total employment
Studies investigating the effects of these resources on health outcomes are scarce. To my knowledge there was only one cross-country study investigating other resources than human and financial resources. This study by Or et al. (2005) concluded that CT scanners positively moderate the beneficial impact of human resources on health outcomes. Further studies are lacking, presumably because of restricted data availability (OECD, 2014) and the central importance of human resources within healthcare.
Those studies investigating real resources mainly focused on human resources. Human resources are often considered as the central resource of healthcare systems (Dussault & Dubois, 2003; Joint Learning Initiative on Human Resources for Health, 2004; Anand, & Bärnighausen; 2012). Every function in the health system is undertaken or mediated through the health worker (Anand, & Bärnighausen; 2012). Human resources include administrative workers and workers specialized in care such as physicians, midwives, nurses, dentists, pharmacists and physiotherapists (OECD, 2014). Most studies investigating real resources focus on physicians, although some studies incorporate other
3 Figures were normalized by the group maximum including other countries: Germany, Sweden, France
14 healthcare workers. Findings across type of healthcare worker are often broadly similar. Appendix 1 provides an overview of these studies, which will be discussed below. The beneficial effect of resources is underscored by several well-known theories such as the resource based view and resource dependence theory (Peteraf, 1993; Pfeffer & Salancik, 1978; Wernerfelt, 1984). Based on these theories and the central role that healthcare workers assume within the healthcare sector they could be expected to have a positive influence on health outcomes. In contrast with this view several studies found that physicians were not- or negatively related to health outcomes. The first study investigating the issue found a positive association between the density of physicians and mortality in younger age groups (Cochrane, Leger & Moore; 1978). The authors could not succeed to find an explanation for this so called ‘doctor anomaly’. After a few years authors reinvestigated the issue. In contrast to the first study, Kim & Moody (1992) concluded that physician density was insignificant in explaining infant mortality. This conclusion was confirmed by Hertz et al., (1994) who further found that there was also no relation with maternal mortality and general life expectancy at birth.
Subsequent findings were more promising. All recent studies used regression analyses which showed that physician density has beneficial effects on several measures of health including infant and under-five mortality, perinatal mortality, mortality by sex, life expectancy for males and females at birth and 65 and years of life lost by heart disease (Anand & Bärnighausen, 2004; Joumard et al., 2008; Or, 2000; Or et al. 2005; Nixon & Ullman, 2006; Robinson & Wharrad, 2000; Speybroeck et al., 2006). These studies are performed using different datasets containing OECD countries, EU countries and WHO countries. Cross-sectional studies and panel analysis were performed over different time periods using different control variables. The consistent findings using these different samples and measures offer proof for a positive relation between physician density and health outcomes.
Although earlier studies do not support this view, recent studies using a variety of methods agree that higher physician densities are associated with beneficial health outcomes. Therefore the first hypothesis is formulated as follows:
Hypothesis 1
Physician density is positively associated with health outcomes
15 2.3 Public financing
To determine whether the amount of public financing may influence the relation between physician density and health outcomes, this section will review previous findings concerning public financing. To understand how financing may influence the healthcare sector, general reasons for government intervention will be discussed first.
Governments attempt to determine the optimal degree of intervention in the market in order for them to achieve their objectives. When organizing the economies of their countries, governments pursue the objectives of macroeconomic stabilization, income redistribution and resource allocation (Musgrave, 1939; Musgrave & Musgrave, 1939). They have to decide whether such goals can be attained through free-market-principles or if government intervention is necessary.
Theories describe that this decision should be based on whether the market operates in the desirable state of Pareto-efficiency (Stilglitz, 1991). In a perfect market no government intervention is required to obtain this state. Such a market requires that several conditions are met, such as perfect market information and no participant power to set prices. The market forces consumers to express their preferences while producers are forced by competitors to organize their production process properly (De Kam, Koopmans & Wellink, 2008). This leads to an optimal allocation of resources. While this ideal situation seems promising, in reality the conditions for such a market rarely exist. Public interest theories of regulation describe characteristics of sectors which show market failures and therefore require government intervention to restore the state of Pareto-efficiency (de Kam et al., 2008).
A first situation that indicates the need for government intervention is the existence of collective goods. These goods are non-rivalrous, which means it is impossible to technically split those goods in units that can be sold on a market. They are also non-excludable: they are available to all consumers including those that do not choose to pay for it. In health the protection of food and water safety is an example of a (nearly) public good (Musgrove, 1996). Therefore government intervention in the healthcare sector could lead to beneficial outcomes.
16 patients are often not able to properly determine the necessity and quality of the healthcare they purchase. Therefore providers would have possibilities to demand unreasonable prices if governments would not interfere. The final reason for intervention in healthcare concerns problems with risk insurance. The healthcare risks are distributed unequally over population groups. Halve of the insured account for less than 5% of the costs. In contrast, only 10% of the insured accounts for 75% of the costs. Without government interventions insurers would therefore choose to ask different premiums for different risk groups or even refuse groups for insurance. In that situation high risk groups would not be able to purchase healthcare. Insurance also entails a moral hazard problem: insured could choose to live less healthy or purchase more healthcare because the additional benefits they derive exceed the marginal costs since they pay a fixed premium. Governments can counter such problems, for instance by regulation concerning deductibles. Finally, some risks would not be covered by insurers in a free market at all, such as the insurance of a child that is born with a severe chronic handicap (De Kam et al., 2008). These situations show that government intervention in the healthcare sector could be beneficial.
Although these theories sound promising, it is difficult to use them for policy adjustments. Because of the complexity of the healthcare sector the theories offers only general guidelines for the amount of government intervention. Furthermore, is impossible for governments to try to correct all market imperfections since they manifest themselves in a large amount of situations. Therefore correcting them is a matter of priorities (de Kam et al., 2008). Private interest theories emphasize that state involvement should not be justified based on these theories alone: regulators do not always have sufficient information and enforcement powers to enable them to correct market failures and their aim is not always to peruse the public interest (Den Hertog, 2010; Posner, 1974; Stigler, 1971; Maskin & Tirole, 2004). According to these theories too much reliance on the public sector should be avoided.
The situation is further complicated by the different range of government intervention mechanisms available. The following mechanisms can be distinguished, arranged from the least to the greatest intrusion in private decisions (Musgrove, 1996):
Inform: persuade but not obligate to do anything, such as publicizing the health risks of smoking
Regulate: determine how an activity should be undertaken, such as setting standards for doctors
Mandate: obligate to undertake activities, such as employer contribution to employees health insurance
Finance healthcare with public funds such as those obtained through taxes
Provide or deliver healthcare services using publicly owned facilities and civil service staff
17 illustrates this distinction by giving examples of combinations of financing and delivery (McIntosh, Forest & Marchildon, 2003).
________________________________________________________________________
Public financing Private financing
Public delivery National Health Service User fee for public services
Private delivery Public insurance Private insurance
________________________________________________________________________ Figure 4: Examples of the distinction between public and private methods of financing
and –delivery in healthcare (McIntosh, Forest & Marchildon, 2003)
Because public financing requires public resources, it can be considered the crucial choice about state action (Musgrove, 1996). Therefore financing is assumed to be the most relevant intervention mechanism in the context of the escalating healthcare costs. Consistent with most studies this thesis focusses on healthcare financing. This view focusses on the left and right part of the figure above, and ignores its upper and lower part.
The proportions of private- and public financing that are used differ substantially across countries, as illustrated by the figure below (figure 5).
Figure 5: Health expenditure as a percentage of GDP across OECD countries in 20104 This figure shows differences in total health expenditure and its public and private components as a percentage of GDP across OECD countries. The lowest and highest spending countries are respectively Estonia with a total healthcare expenditure of 6,3% of GDP and the United States with 17% of GDP. In terms of proportions of public expenditure Chile represents the minimum of 47,2% while the Netherlands represents the maximum of 86,1%.
4
18 Several studies have empirically investigating which proportions of public- and private financing were most beneficial. Similar to theories investigating total health expenditure they mainly focused on direct associations with health outcomes. A range of studies were performed, some associating public financing with beneficial outcomes (e.g. Tuohy et al., 2004) and others with disadvantageous outcomes (e.g. Hitiris & Posnett, 1992; Berger & Messer, 2002).
Although some authors have emphasized the potential to clarify results by incorporating real resources in the analysis, studies investigating these issues are scarce. There are only two studies that investigated moderating effects of the financing mix on the relation between physician density and health outcomes. Both studies used a Data Envelopment Analysis (DEA) approach in which differences in efficiency across countries are attributed to moderating factors. Although this approach offers insight into possible mechanisms to improve performance, it does not offer possibilities for model comparison (Johnes, 2006). Puig-Junoy (1998) shows that private financing increases efficiency of countries in transforming a component of real resources, including physicians, in male and female life expectancy at birth. In contrast, Or et al. (2005) conclude that public financing does not have a significant moderating impact on the relation between physicians and these and several other measures of health outcomes.
Consistent with theories describing government intervention, these empirical studies do not reach consensus regarding which financing mechanism is most optimal. There are indications that the amount of public financing could influence the relation between physicians and health outcomes, but there is no consensus regarding the type of effect. Therefore the second hypothesis is formulated as follows:
Hypothesis 2
The amount of public financing will moderate the relation between physician density and health outcomes
19 2.4 Other mechanisms for cost containment
Besides public financing governments have several other mechanisms at their disposal which could contribute to a solution to the escalating healthcare costs. An attempt was made to investigate some of these mechanisms. Within the healthcare literature there is especially debate regarding the effectiveness of fund collection mechanisms, gatekeeping, cost sharing mechanisms and provider reimbursement (Gusmano, Weisz & Rodwin, 2009; Or et al., 2005; Puig-Junoy, 1998; Wranik, 2011). These mechanisms are examples of how the government may intervene through regulating and mandating. Therefore these mechanisms may also moderate the relation between physicians and health outcomes:
Hypothesis 3
Other intervention mechanisms will moderate the relation between physician density and health outcomes
Unfortunately the variance of the data available on these mechanisms prohibited the proper analyses of their effects within this type of study and time frame. Therefore they were not included in the final model, as will be discussed in the result section. However, these mechanisms indicate the importance of adopting a broad perspective when attempting to find solutions for the problems within the healthcare sector. Furthermore, they illustrate data problems within healthcare and the need for alternative analyses. Therefore a brief description of them is retained within this thesis. Each mechanism will be discussed here and will be elaborated on in the remaining chapters.
Fund collection
The public part of healthcare financing within countries can be further subdivided into manners of fund collection. Countries employ either a Beverdige or a Bismarck system to raise the funds that governments spend on healthcare. A Beveridge system is a single-payer system funded through household taxes while a Bismarck system is funded through contributions by employers and employees. The Beverdige system could increase cost control because of the simplicity of its fund collection mechanism (Glied, 2009) but it could restrict consumer choice (Hussey & Anderson, 2003) and healthcare access (Gusmano et al, 2009).
Cost sharing
20 Gatekeeping
Some countries employ a gatekeeper role in their healthcare system, which entails that people need a referral by a primary physician to be entitled to specialist services (Bodenheimer & Casalino, 1999). Gatekeeping could increase coordination of the healthcare system and restrict unnecessary use of specialist services while it could also restrict customer access and choice (Gusmano et al., 2009). Previous empirical studies found no significant effect of gatekeeping on the relation between physicians and health outcomes (Puig-Junoy, 1998; Or et. al, 2005).
Payment of general practitioners and specialists
21 3. METHODOLOGY
This chapter will elaborate on the method that was used to test the hypotheses that were formulated in the previous section. The first section will discuss the sample that was used. The second section will discuss the measurement of the dependent, direct independent and moderator variables. Subsequently, descriptive statistics will be discussed. The final section will describe the model and discuss the steps taken to obtain valid results.
3.1 Sample and data sources
The sample that was used consists of 34 OECD countries over the period 1960-2013. Information about the alternative mechanisms for cost containment (discussed in paragraph 2.4) were gathered from Wranik (2011). The remaining data was obtained from the OECD database (OECD, 2014). This database contains information on a range of measures describing the health systems across developed countries. This data has some drawbacks such as differences in measurement across countries and limited data availability for some variables. Despite these disadvantages the data is robust compared to standards of statistical reporting (Reinhardt, Hussey & Anderson, 1999). Relatively old previous articles have investigated similar topics using OECD data over time periods that were not covered by the online database. Therefore attempts were made to obtain this data from authors and the OECD. Unfortunately this did not succeed, although the OECD quickly provided me with several other datasets. These datasets did not offer benefits compared to the online data for my topic of interest.5 Therefore only the online data was used.
The figure below (figure 6) shows life expectancy across all countries included in the dataset. The figure illustrates the diversity of the dataset since there are substantial differences in health outcomes across countries.
Figure 6: Life expectancy at birth across all countries in the dataset6
5
The data that de OECD provided was superior compared to the online database for coverage in early years. This data was lacking for recent years and because measurement differences it was not possible to combine the datasets.
6
22 3.2 Measurement of the variables
The dependent variable: health outcomes
Health outcomes are a complex concept because they include both quantitative (life expectancy) and qualitative (quality of life) aspects. The World Health Organization provides an ideal measure in this respect: Healthy Life Expectancy (HALE) is calculated by subtracting the numbers of ill health, weighted according to severity, from the expected life of individuals (World Health Organization, 2014). Because this measure is only available for 2002 and 2004 it does not offer the possibility to perform time series analysis. HALE seems to be highly correlated with other measures health (Joumard et al., 2008). Therefore other measures were used (OECD, 2014). Different measures were used in order to increase the robustness of the results. Male and female health outcomes were tested separately because previous research showed that females benefit more from healthcare systems (Elola, Daponte & Navarro, 1995). The following measures were used:
Male and female life expectancy at birth
This conventional measure of life expectancy was used as a general measure of health outcomes.
Male and female life expectancy at 65
Elderly make heaviest use of the healthcare system and countries are more divided across this measure compared to life expectancy at birth (Joumard et al., 2008). This increased variation could make this measure more appropriate to detect health outcome improvements.
Male and female mortality
Mortality describes the Years of Potential Life Lost (YPLL) per 1000 population. This measure captures the average length of time persons aged less than 70 would have lived if they did not die prematurely. This measure has the advantage of capturing only avoidable deaths. Some argue that in most industrialized countries life expectancy is already at or close to its maximum (Self and Grabowski, 2010). Mortality measures could be more appropriate to detect the consequences of system reforms because they have more variation.
The independent variables: physician density and the control variables
23 Besides physicians, several control variables were used. Previous studies show that health outcomes are determined by a combination of healthcare resources, lifestyle factors and socio-economic factors (Joumard et al., 2008). Based on a review of past evidence
(appendix 1) the control variables were selected that were presumed to be most important. These are GDP, alcohol, tobacco and pollution. GDP affects health by facilitating goods with a beneficial effect on health such as transportation, food and housing (Joumard et al., 2008). GDP was measured on a per capita basis at constant prices and PPPs to ensure comparability between countries. Alcohol and tobacco consumption are among the highest risk factors for health in developed countries (World Health Report, 2002). Alcohol was measured as the average consumption in liters per capita over the part of the population aged 15 and older. The OECD (2014) offers several potential measures of tobacco consumption. The average consumption in grams per capita over the part of the population aged 15 and older was chosen rather than other measures such as the
24 The moderating variables: public financing and other intervention mechanisms Public financing is the part of total health expenditure that is financed from public sources (OECD, 2013b). Consistent with the OECD definition (2013b), total health expenditure includes the sum of expenditure on activities that – through application of medical, paramedical and nursing knowledge and technology – has the goal of promoting health and preventing disease, curing illness and reducing premature mortality, caring for persons affected by chronic illness who require nursing care, caring for persons with health-related impairments, disability, and handicaps who require nursing care, assisting patients to die with dignity, providing and administering public health and providing and administering health programmes, health insurance and other funding arrangements. Public health expenditure (OECD, 2013b) is the proportion of expenditure on these activities that is incurred by public funds. Public funds are state, regional and local government bodies and social security schemes. Public capital formation on health includes publicly financed investment in health facilities plus capital transfers to the private sector for hospital construction and equipment.
The remaining intervention mechanisms were described in paragraph 2.4 and were measured as dummy variables (Wranik, 2011):
Fund collection
Through a Beveridge or a Bismarck system (0=Beveridge, 1=Bismarck) Gatekeeping
The physician performing the role of a gatekeeper (0=no gatekeeper function 1=gatekeeper function)
Cost sharing
Countries deploying cost sharing mechanisms such as co-payments (0=cost sharing not employed, 1=cost sharing employed)
General practitioner payment
Payment through fee-for-service or alternative mechanisms (0=alternative, 1=fee for service)
Specialist payment
25 A summary of the measures described in this section and the expected direct- or moderating impacts is shown in the table below (table 1).
________________________________________________________________________
Dependent variable Measurement
(health outcomes)
Life expectancy at birth (F) Average life expectancy at birth for females Life expectancy at birth (M) Average life expectancy at birth for males Life expectancy at 65 (F) Average life expectancy at 65 for females Life expectancy at 65 (M) Average life expectancy at 65 for males
Mortality (F) Average Potential Years of Life Lost for females Mortality (M) Average Potential Years of Life Lost for males
________________________________________________________________________
Direct independent variables Measurement Expected relation with
health outcomes
Physician density (PHYS) Average amount of physicians per 1000 + population
Control variables
Gross domestic product (GDP) Average GDP in constant prices and constant + PPPs
Alcohol consumption (ALC) Average alcohol consumption in liters per - capita in population aged 15+
Tobacco consumption (TOB) Average tobacco consumption in grams per - capita in population aged 15+
Pollution (POL) Emissions of nitrogen oxide in kg per capita - ________________________________________________________________________
Moderating variables Measurement Expected moderating
impact
Public financing (PUB) Percentage of health expenditure that is ? finance from public funds
Fund collection (FUND) Fund collection system (dummy variable ? 0=Beverdige 1=Bismarck)
Cost sharing (COST) Employment of cost sharing mechanisms ? (dummy variable 0= no cost sharing
1=cost sharing)
Gatekeeping (GATE) Physician performing gatekeeper role ? (0=no gatekeeping 1=gatekeeping)
General practitioner Payment of GP mechanism (0=alternative ? payment (GP) mechanisms 1=fee-for-service)
Specialist payment (SPEC) Specialist payment mechanism (0=alternative) ? 1=fee-for-service)
26 3.3 Descriptive statistics
The descriptive statistics of the variables are shown in the table below (table 2). The table shows the number of observations, the mean, the standard deviation and the minimum and maximum values for each variable. The minimum and maximum values may seem extreme, such as the 46,30 minimum life expectancy at birth. This is because these values are observations for a single country in a single year. The table reflects data restrictions for some variables, especially for pollution and for the dummies.
________________________________________________________________________
Dependent variables Obs Mean Std. dev. Min Max
Life expectancy at birth (F) 1596 77.42 4.77 50.3 86.4 Life expectancy at birth (M) 1593 71.17 4.96 46.3 80.7 Life expectancy at 65 (F) 1547 17.78 2.20 12.1 24 Life expectancy at 65 (m) 1545 14.47 1.95 10.2 19.3 Mortality (F) 1478 4556.61 2533.39 1615.7 25796.8 Mortality (M) 1478 8331.04 4047.43 2864.5.2 36385.1 _______________________________________________________________________
Direct independent variables Obs Mean Std. dev. Min Max
Physician density (PHYS) 1086 2.43 0.94 0.3 6.14
Control variables
Gross domestic product (GDP) 1334 22994.73 10094.14 2431.69 73912.59
Alcohol (ALC) 1554 9.79 4 0.8 20.8
Tobacco (TOB) 1028 2273.69 704.9 660 4463 Pollution (POL) 806 35.02 20.48 9.14 114.78 _____________________________________________________________________
Moderating variables Obs Mean Std. dev. Min Max
Public financing (PUB) 1204 71.77 14.75 0 98.29 Fund collection (FUND) 819 0.46 0.5 0 1 Cost sharing (COST) 819 0.74 0.44 0 1 Gatekeeping (GATE) 819 0.6 0.49 0 1 General practitioner payment (GP) 819 0.52 0.5 0 1 Specialist payment (SPEC) 819 0.33 0.33 0 1
27 3.4 Model and validity
The previous sections described measurement of the variables in this study. A regression analysis will be used to estimate the linear relation between the independent variables and the measures of health outcomes:
Y i t = αi + β1 · GDP i t + β2 · ALC i t + β3 · TOB i t + β4 · POL i t+ β5 · PHYS i t+ β6 · PUB it+
β7 · FUND i t+ β8 · COST i t+ β9 · GATE i t+ β10 · GP i t+ β11 · SPEC i t+ β12 · PHYS · PUB i t + β13 · PHYS · FUND i t+ β14 · PHYS · COST i t+ β15 · PHYS · GATE i t+ β16 · PHYS · GP i t
+ β17 · PHYS · SPEC i t + u i t
The equation shows the following variables, which were described in more detail in the previous sections:
Y i t= measure of health outcomes (male or female life expectancy at birth, life
expectancy at 65 or mortality)
GDP = Gross domestic product FUND = Fund collection ALC = Alcohol consumption COST = Cost sharing TOB = Tobacco consumption GATE = Gatekeeping
POL = Pollution GP = General practitioner payment PHYS = Physician density SPEC = Specialist payment
PUB = Public financing
The formula shows that the moderating factors are incorporated in the model both directly and as an interaction effect with physicians, consistent with standard procedures for moderating models. This allows to discriminate between direct and interaction effects (Wijnen, Janssens, Pelsmacker, Van Kenhove, 2002).
A regression analysis will be used to estimate a linear relation between the independent variables and each measure of health outcomes. The regression analysis estimates the coefficients (betas) for each of the variables in the equation. The beta represents the unit increase in health outcome as a consequence of a unit increase in the independent variable(s) corresponding to that beta (Brooks, 2008). The betas of independent variables should carry the opposite sign in the models using mortality rather than life expectancy since this corresponds to disadvantageous rather than beneficial health outcomes.
The regression equation was tested by expanding the model on a stepwise manner. Subsequently the following models were investigated:
Control variables only
Control variables and physicians
Control variables, physicians and public financing
28 The models were compared in order to determine the most appropriate model. The regression analysis was done using a fixed-effect Generalized Least Squares (GLS) model with correction for heteroskedasticity and autocorrelation. This is a variation of Ordinary Least Squares (OLS) in order to make sure the assumptions hold and the model fits the data (Brooks, 2008). The technique determines the linear equation by minimizing the sum of squared vertical distances between the observations in the dataset (adjusted to satisfy the assumptions) and those predicted by the approximation. General literature on statistics (Brooks, 2008) and articles investigating similar topics using similar datasets (Or, 2005; Nixon, Ulmann, 2006; Joumard et al., 2008) were reviewed in order to determine the steps to be taken in order to obtain valid results. Each of these steps will now be discussed.
Heteroskedasticity and autocorrelation
Assumptions of OLS are that the standard errors of the equation are uncorrelated and have constant variances. A likelihood ratio was performed to test if the variance of the panel standard errors was constant. This test indicated that the errors were instead heteroskedastic. A Durbin Watson test for panel data was performed to test if there was a correlation between standard errors over subsequent time levels. The Durbin Watson statistic was close to zero for all the equations tested, indicating positive autocorrelation. Therefore the data was estimated with a GLS model with correction for panel-level heteroskedasticity and first-order autocorrelation. This model entails a data transformation in order to make sure the assumptions for valid estimation hold.
Country fixed effects
Country fixed effects were applied because the member states of the OECD are a specific non-random set with cross-sectional heterogeneity. Fixed effects are country dummies that allow the testing of an equation with expected common betas and different intercepts across countries. Thereby it controls for time-invariant effects not captured by the model. This could be for example institutional differences of healthcare systems across countries. Normality
29 Mutlicollinearity
Multicollinearity is a situation in which there is a risk of biased regression coefficients due to associations between independent variables. Therefore correlations among the variables were examined (appendix 2). Correlations are low compared to standards (Wijnen et al., 2002; Vocht, 2011). Therefore there was no reason to assume that multicollenearity among independent variables was present. In some situations fixed effects may be collinear with moderating factors with little variation. To test for the possibility of multicollinearity between the country dummies and alternative mechanisms for cost containment alternative model specifications (omitting country dummies or using random effects) were tested. This did not materially alter conclusions. Therefore it was assumed that the country dummies were not collinear with the dummy variables representing alternative mechanisms for cost containment.
Alternative models
After the preferred model was determined using the stepwise regressions, alternative specifications were tested and documented to increase the robustness of the results. There are severe limitations in data availability for some of the control variables. Therefore there was a trade-off between the number of observations and countries that could be included in the analysis and the number of variables that could be controlled for. Controlling for more variables entails a risk of excluding observations/countries and therefore limiting generalizability of findings. Excluding control variables increases the risk of incorrect conclusions due to an omitted variable bias. Although the main regression includes all control variables, the final model was tested using several smaller sets of control variables to investigate the generalizability of the results. Furthermore, authors differ in opinion regarding the need to use a panel-specific rather than general first-order autoregressive correction in similar datasets (Or, 2005; Nixon, Ulmann, 2006; Joumard et al., 2008). Therefore an alternative specification of the preferred model shows if the results hold when using a panel-specific rather than general correction for autocorrelation. These models are shown in appendix 4 and will be discussed in the results chapter.
Goodness of fit
Goodness of fit of a regression equation is usually assessed using the R squared that indicates the variance that the model explains. When using GLS the variance cannot be meaningfully broken down into explained and unexplained parts. Therefore it does not provide an accurate measure of fit. Instead, an R squared was calculated that represents the squared correlation between the observed values of health outcomes and those predicted by the model (Nixon & Ulmann, 2006). Although this statistic cannot be interpreted as the explained variation, it does give some indication whether adding variables to a mode increases its accuracy. Therefore it was used for the stepwise expansion of the model.
30 4. RESULTS
This chapter will discuss which hypotheses are supported by the results of the regression analysis. The regression model estimates the betas in the equation that was discussed in paragraph 3.4 and will be repeated here for purposes of oversight:
Y i t = αi + β1 · GDP i t + β2 · ALC i t + β3 · TOB i t + β4 · POL i t+ β5 · PHYS i t+ β6 · PUB it+
β7 · FUND i t+ β8 · COST i t+ β9 · GATE i t+ β10 · GP i t+ β11 · SPEC i t+ β12 · PHYS · PUB i t + β13 · PHYS · FUND i t+ β14 · PHYS · COST i t+ β15 · PHYS · GATE i t+ β16 · PHYS · GP i t
+ β17 · PHYS · SPEC i t + u i t
Y i t= measure of health outcomes (male or female life expectancy at birth, life
expectancy at 65 or mortality)
GDP = Gross domestic product FUND = Fund collection ALC = Alcohol consumption COST = Cost sharing TOB = Tobacco consumption GATE = Gatekeeping
POL = Pollution GP = General practitioner payment PHYS = Physician density SPEC = Specialist payment
PUB = Public financing
The regression analyses will estimate the betas in this equation, as was explained in the previous chapter. To determine whether there is enough proof that the independent variables are indeed important predictors of health outcomes, the significance level will be used. This level indicates the chance that the independent variable will not be rejected as a predictor of health outcomes while in reality this should be the case. Most studies adopt a significance level of either 10, 5 or 1 percent. These levels are represented in the regression results by stars behind the estimated coefficients (betas). One star means that the result is significant at the 10% level; two stars means that it is significant at the 5% level and three stars means that it is significant at the 1% level.
31 were dropped from the regression model because they presumably did not have enough variation to be properly investigated using a regression analysis.
The third model, including the control variables, physicians and public financing is assumed to be most optimal since results are consistent with previous models and the R squared increased. The increase in the R squared is relatively small but this should not be interpreted as the increase in explained variation, as was discussed in the previous chapter. The results using this model across all measures of health outcomes are summarized below (table 3). The table shows that the regression results are significant and broadly similar across measures of health outcomes employed.
________________________________________________________________________________________________
Life expectancy at birth Life expectancy at 65 Mortality7
Female Male Female Male Female Male
________________________________________________________________________________________________ Constant 69.17163*** 63.61651*** 14.36036*** 11.86846*** 8.72805*** 9.10054*** PHYS 4.48380*** 4.48380*** 1.77216*** 1.39764*** -.38909*** -.00686*** PUB .12696*** .11721*** .04022*** .02505*** -.00809*** -.26519*** PHYS*PUB -.04893*** -.04445*** -.01733*** -.01096*** .00367*** .00259*** GDP .00018*** .00024*** .00014*** .00016*** -.00002*** -00002*** ALC -.12200*** -.18207*** -.05047 -.06876** .02729*** .02802*** TOB -.00055*** -.00075*** -.00039*** -.00041*** .00004*** .00007*** POL -.03123*** -.04412*** -.01772*** -.02557*** .00402*** .00491*** ________________________________________________________________________________________________ Country fixed yes yes yes yes yes yes
effects ________________________________________________________________________________________________ Observations 395 395 395 395 369 369 Countries 24 24 24 24 23 23 ________________________________________________________________________________________________ R2 observed/ .95368 .96446 .94652 .95936 .95432 .96451 predicted8 ________________________________________________________________________________________________
Table 3: Regression result for the preferred model using different measures of health outcomes9
Stars indicate significance: * at 10%; ** at 5% and *** at 1%
The control variables are all highly significant and carry the expected sign. The first hypothesis stated that physicians were positively associated with health outcomes. All coefficients for physicians (table 3) are positive and significant at the 1% level. This means that the results support the first hypothesis:
Result 1
Physician density is positively associated with health outcomes
7
Mortality variables are in log form.
8 The R squared was calculated by squaring the correlation between the predicted- and actual values of the
health outcomes measure (Nixon & Ulmann, 2006). Within this type of regression the R squared should not be interpreted as the percentage of variation explained by the model. However, it is provided to give some indication of goodness of fit and to compare the stepwise regression results in the appendix.
9 GLS country fixed effect model with correction for panel-level heteroskedasticity and first-order
32 Table 3 shows the direct and moderating effect of public financing. The coefficients for the direct effect are positive and significant across all measures and models, indicating a positive association between public financing and health outcomes. This direct effect is discriminated from the interaction effect with physicians. The second hypothesis stated that the amount of public financing was expected to moderate the relation between physician density and health outcomes, but did not specify the direction of this relation. The results show that the coefficients for the moderating effect are negative and
significant at the 1% level across all measures and models. This means that the results support the second hypothesis as follows:
Result 2
Public finance negatively moderates the relation between physician density and health outcomes
Combining the first two results shows that public financing negatively moderates the positive relation between physician density and health outcomes. This is illustrated by the figure below (figure 7).
Figure 7: Impact of public financing on life expectancy for high and low levels of physicians10
The figure shows the relation between physician density, public financing and life expectancy for high and low values of physician density and public financing. These values were calculated by taking the minimum and maximum cross-country averages across years. The figure shows that the relation between physician density and life expectancy is positive and that this relation becomes less strong for higher amounts of public financing.
Several alternative versions of the model including the control variables, physicians and public financing were used to test the robustness of the results. These models are shown in appendix 4. Authors differ in their opinion of the proper manner to correct for
10
34 5. CONCLUSION AND DISCUSSION
In the light of the escalating healthcare costs (Walshe & Smith, 2011) several authors investigated the effects of health expenditure on health outcomes, showing conflicting results (Asiskovitch 2010; Berger & Messer, 2002; Crémieux et al., 1999; Hitiris & Posnett, 1992; Joumard et al., 2008; Judge et al., 1998; LeGrand, 1987; Nixon & Ulmann, 2006; Self & Grabowski, 2003; Tuohy et al., 2004; Wolfe, 1986). Therefore authors emphasized the necessity to consider the effect of real rather than financial resources since these have a more direct effect on health outcomes (Anell & Willis, 2000; Joumard et al., 2010). This study investigated the relation between physician density and health outcomes and the moderating impact of public financing.
Consistent with well-known theories describing the importance of resources (Peteraf, 1993; Salancik, 1978; Wernerfelt, 1984) the density of physicians was positively associated with health outcomes. Although this contradicts findings of several empirical studies (Cochrane et al., 1978; Hertz et al., 1994; Kim and Moody, 1988) most work is consistent with this result (Anand & Bärnighausen, 2004; Joumard et al., 2008; Nixon & Ullman, 2006; Or, 2000; Robbinson & Wharrad, 2000; Speybroeck et al., 2006). Contrary to earlier research the findings do not indicate that females benefit more from a healthcare system than males (Elola et al., 1995).
The results showed that public financing has a negative moderating impact on this relation. While healthcare is often characterized by a shortage of medical specialists this problem is less urgent when physician density is high. It could be that this indicates a situation in which monopolies play a less prominent role and subsequently less government involvement is necessary (De Kam et al., 2008; Musgrave & Musgrave, 1989; Rosenthal, 2001). The results of this study are in line with findings by Puig-Junoy (1998) that indicated that private financing has a positive effect on a component of real resources including physicians. Results contradict recent findings by Or et al. (2005) that indicated that public financing is irrelevant in explaining physician performance. This study complements previous work by investigating this issue in a more extensive and recent database using a regression analysis rather than DEA approach. This enables the possibility for model comparison (Johnes, 2006).
35 improving health outcomes. Governments also strive for the objectives of cost control, equity and popular satisfaction. These objectives can be conflicting (Musgrove, 1996) such that one cannot conclude that optimisation of one objective will entail a benefit for the sector overall. Future research should take into account other aspects. As many studies (Götze & Schmid, 2012), this thesis focused on dichotomous concept discerning public from private funding sources. Several other categorizations of financing might be investigated, such as insurance versus out of pocket payments (Musgrove, 1996).
Furthermore, future research could focus on alternative manners of data collection and the harmonization of data between countries. Countries often report to organizations such as the OECD using their own standard procedures, which restricts comparability across countries. Data availability on many types of resources and alternative mechanisms for cost containment is often lacking. This study attempted to investigate the effects of fund collection, gatekeeping, cost sharing and physician reimbursement. Previous studies have investigated whether these mechanisms were important using DEA. Investigating them using a regression analysis appeared to be impossible due to the low variance of the data. Measuring these constructs on a continuous basis, for example the level of cost sharing rather than whether such mechanisms are employed or not, could solve this problem. This study made use of panel data in order to be able to take advantage of a large cross-national database. This approach enabled the discovery of broad trends that are similar across countries at the expense of analysing in-depth differences within countries. Differences in life expectancy across countries illustrate the diversity of countries within the data set. Because of unique circumstances there is no one-size-fits-all solution for healthcare systems. Policymakers should adopt practices to actual circumstances that exist within their country (OECD, 2010).
