Welfare and healthcare typologies

Several strategies to measure and classify healthcare systems have been proposed since the 1970s, often based on healthcare expenditure, healthcare financing, service provision, and access regulation and resulting in versions of three healthcare system ideal types closely connected to Esping-Andersens welfare state regimes: voluntary insurance, social health insurance, and national health service [7]. Reibling, Ariaans, and Wendt [28] used 13 country-level variables to construct a typology of healthcare systems across 29 high-income countries. Health expenditure per capita and the number of GPs per population indicated healthcare supply, the financial and human resources spent on health. The role of the state and the public/private mix in healthcare was indicated by the public share of health expenditure, the share out-of-pocket payments, and the remuneration of specialists as a measure of cost sharing. Access regulation was measured by indicators of healthcare coverage and choice restrictions. Expenditure on outpatient-care and their GP-to-specialist ratio indicated primary care orientation. Finally, healthcare performance was measured by indicators of tobacco and alcohol consumption and a quality sum index based on avoidable hospital admissions. Here, tobacco and alcohol consumption were used as proxies for the effectiveness of a healthcare system’s preventive efforts, as adequate data on regulatory and monitoring activities was not available. Factor analyses of these indicators resulted in a five-fold typology of healthcare systems (countries included in our data in bold):

1. Type 1 –Supply- and choice-oriented public systems (Australia, Austria, Belgium, Czech Republic, France, Germany, Iceland, Ireland, Luxembourg, Slovenia): Primarily public funded social insurance systems. Characterized by medium to high levels of financial and human resources, free choice, and access regulation only by cost sharing. Performance scores are mediocre with regards to both prevention and healthcare quality.
2. Type 2 –Performance- and primary-care-oriented public systems (Finland, Japan, New Zealand, Norway, Portugal, South Korea, Sweden): Public funded high-performing healthcare systems. The state has a strong role in regulating access and in the payment of medical specialists. Primary care has high priority.
3. Type 3 –Regulation-oriented public systems (Canada, Denmark, Italy, Netherlands, Spain, United Kingdom): Primarily public funded healthcare systems. Medium level of resources, low levels of out-of-pocket payments, and high level of access regulation and limitation of choice. Lower priority of primary care and lower performance than Type 2.
4. Type 4 –Low-supply and low performance mixed systems (Estonia, Hungary, Poland, Slovakia): Mostly public funded healthcare systems with low levels of financial and human resources, high levels of out-of-pocket spending, strong access regulations, and low performance on prevention and quality of care.
5. Type 5 –Supply- and performance-oriented private systems (Switzerland, United States): Healthcare systems with a strong role of private financing and out-of-pocket payments. Public resources are in the majority, with high supply and expenditures. Access is regulated by sharing regulations such as deductibles. This type shows high quality-of-care performance.

Since we wanted to utilize the full range of our data, and to avoid calculating with single-country clusters, we grouped Lithuania (which is not included in the data of Reibling et al. [28]) in Type 4, and Switzerland (which is the only Type 5 country in our data) in Type 1. This is done based on an assessment of key indicators used in the initial factor analysis. Subsequently, only four of the five healthcare systems types were included in our analysis. As results from research using welfare state regimes to compare health inequalities have been largely inconclusive, our contribution with this article is to use a validated and more specific health outcome–amenable mortality rather than self-reported health or limiting longstanding illness–and a typology more directly related to health–Reibling and colleagues’ [28] healthcare system types.

Expectations

Our study design is not suited for predicting inequality effects of specific health policies. However, we expect inequality rates to vary across countries and healthcare system types, and results from previous research allow us to formulate some modest expectations with regards to this variation. First, low education can be associated with poor health by being an indicator of material disadvantage. Financial strain due to e.g. unemployment or low income may matter more in a context with scarce healthcare resources and high out-of-pocket payments. Blom, Huijts, and Kraaykamp’s [29] analyses of repeated cross-sectional survey data revealed that high total and state provision of healthcare, measured as total and governmental healthcare expenditure, was associated with smaller educational inequalities in self-rated health, while specific inequality-reducing health policies had a less substantial effect. This leads us to expect that low public funding, as found in the low supply and low performance mixed systems (Type 4), is associated with higher levels of inequalities.

Second, the impact of strong access regulation and choice restriction, as found in the performance- and primary-care-oriented public systems (Type 1) and the regulation-oriented public systems (type 3), appears less clear. On the one hand, regulations may enhance health equality, ensuring equal access and preventing overconsumption of services. On the other hand, to maneuver a bureaucracy-governed healthcare system may (unintentionally) reward immaterial resources typically associated with high socioeconomic position, such as health literacy, social networks and the ability to “work the system” [30].

Third, people of low socioeconomic position have tended to be more intensive users of general practitioners, mainly due to a higher disease prevalence [31,32]. High priority of primary care, as found in the performance- and primary-care-oriented public systems (Type 2), could therefore also be associated with lower inequalities.

Data

The EURO-GBD-SE project collected and harmonized mortality data from the 21 European countries for which comparable data was available. This article utilizes all available data, covering time periods between 1998 and 2006, depending on country (see S1 Table). This data is to our knowledge the latest individual-level mortality dataset encompassing a majority of European countries. The datasets included four Nordic countries (Finland, Sweden, Norway, and Denmark), six Western European populations (England & Wales, Scotland, Belgium, France, Switzerland, and Austria), four Southern European populations (Barcelona, Basque Country and Madrid (Spain) and Turin (Italy)), four Central/Eastern European countries (Slovenia, Hungary, Czech Republic, and Poland) and two Baltic countries (Estonia and Lithuania). The data covered the entire national, regional (Madrid, the Basque Country) or urban (Barcelona and Turin) populations. The data from Spain and Italy only covers parts of the population, which prevents us from generalizing to the whole countries. These populations are therefore excluded when we estimated relative and absolute inequalities for the different healthcare system types but are displayed in tables and figures as a reference point.

Mortality data for Hungary, the Czech Republic, Poland and Estonia came from cross-sectional (CS) unlinked mortality studies. Data for Barcelona and Madrid was derived from a cross-sectional census linked studies. Data for other European countries has a longitudinal design. In the cross-sectional unlinked mortality studies, information on socioeconomic position was derived separately from death certificates and census records. In the longitudinal studies, mortality was linked to socioeconomic position determined during a census. An overview of the mortality data sources is displayed in S1 Table.

The Finnish dataset included only 80% of the Finns. The Swiss dataset excluded Non-Swiss nationals, the French dataset excluded those born outside mainland and the Dutch dataset excluded people from institutions. The 100% linkage between the population and death registries was achieved in most of the included populations. In countries where the default in linkage was lower than 5% no corrections were applied. In countries and areas such as Austria, Barcelona, the Basque Country, and Madrid, where a higher percentage of deaths that could not be matched with the mortality registry, we introduced a correction factor. In Austria, the correction factor was broken down by sex and 5-year age group. In Barcelona, the Basque Country and Madrid, there were no variations by age and sex for excluded deaths. The correction factor was therefore equal to 1.06 (1/0.946) for Barcelona and the Basque Country and 1.25 (1/0.8) for Madrid.

The causes of death amenable to healthcare were selected on basis of the publications by Stirbu et al. (2010) and the AMIEHS (2011) report from the European Union's Public Health Programme. In public health research, the terms “avoidable”, “amenable”, and “preventable” have been associated with some ambiguity, and often been used interchangeably [33]. Piers, Carson, Brown, and Ansari [34] have argued that avoidable mortality includes amenable and preventable conditions, where deaths can be averted from the former, while the latter can be prevented from occurring altogether. Others have attempted to classify mortality according to the relevant level of healthcare intervention: primary, secondary, and tertiary avoidable mortality [35], and health policy and medical care indicators of avoidable mortality [36]. For example, Perez and colleagues’ [37] analysis of avoidable mortality in Spain showed that figures on avoidable mortality could be affected by different processes such as healthcare interventions, prevention and promotion strategies, or by intersectoral policies. The authors argued that the concepts (and sub-concepts) of amenable and avoidable mortality have tended to blur the image of the prevalence and trends of specific causes of death. Nolte and McKee [33] have further questioned the underlying assumption of these classifications: that health outcomes can be attributed to specific elements of healthcare. For several conditions, there are discrepancies in the literature regarding the effect of public health and medical interventions, and thus also the nature of their preventability. Additionally, the classification of amenable mortality may to a certain extent suffer from systematic cross-national variation in diagnosis, death certification, and cause of death classification [27]. When assessing amenable mortality in the different healthcare system types, we will also contrast these estimates with inequalities in all-cause mortality.

Our classification leans on the precedence set by previous cross-national comparisons of amenable mortality (cf. [38–40]). One contested measure has been to classify ischemic heart disease and heart failure as non-amenable. It has been argued that the impact of medical treatment on these causes of death is unclear, while the association with lifestyle factors such as smoking, alcohol consumption and obesity is strong. Causes of death classified as amenable to healthcare are reported in Table 1. Other scholars have used different versions of the same data with similar classifications. Stirbu et al. [41] found educational inequalities in mortality amenable to medical care across all European countries, particularly pronounced in Central-/Eastern-, and Baltic European countries; Plug et al. [42] found that these inequalities were not associated with inequalities in healthcare use; Mackenbach et al. [15] compared mortality amenable to behavior change, amenable to medical intervention, amenable to injury prevention, and non-preventable mortality, finding the smallest inequalities in the latter category, and the steepest gradient in the former; Mackenbach et al. [43] found that mortality declined faster among the higher than among the lower educated and that educational inequalities in mortality decline were similar between causes of death amenable to behaviour change and medical care.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Causes of death amenable to medical care according with ICD10 codes.

https://doi.org/10.1371/journal.pone.0234135.t001

We used educational attainment as a measure of socioeconomic position. This was categorized according to the International Standard Classification of Education as low (no or primary education and lower secondary education, ISCED 0–2), middle (upper secondary education, ISCED 3–4) and high (tertiary education, ISCED 5–6) education. In order to create comparability across countries, we needed the same educational grouping in all countries. These three groups were what national educational classifications allowed us to create, and this division is also utilized in the studies cited above. Table 2 displays the amenable mortality rates by educational level.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Mortality rates by educational level standardized to the European Standard Population.

https://doi.org/10.1371/journal.pone.0234135.t002

Analyses

All analyses were conducted separately for women and men aged 35–79 years (age interval depending on country) and age-standardized with the European Standard Population as reference [44]. Individuals whose educational attainment was unknown were omitted from the analyses. The magnitude of relative educational inequalities in mortality amenable to healthcare across European countries and across healthcare systems was calculated by relative indices of inequality (RII) by means of Poisson regression. The RII is a regression-based measure that accounts for the distribution of the population by educational groups using rank of educational attainment as a dependent variable [45]. The educational rank was calculated over all three educational groups defined above. The resulted RII represents the risk of death at the lowest educational level as compared to the highest educational level in the population. Values larger than 1 indicate a disadvantage for the low educated, values smaller than 1 a disadvantage for the high educated. The magnitude of absolute educational inequalities was calculated by Slope Index of Inequality (SII), a regression-based measure that takes into consideration the entire distribution of education; its values indicates differences in predicted values between low and high educated. Positive values indicate a disadvantage for the low educated, negative values a disadvantage for the high educated.

To further test the applicability of the different typologies, meta-analyses and analysis of variance (ANOVA) was performed on RII and SII estimates. Meta-analyses are common in systematic reviews and aim to synthesize data from multiple studies [46]. In this article, pooled estimates were calculated for each healthcare system type through meta-analysis techniques; each country estimate was weighed with its inversed variance to calculate effect summary with standard errors and confidence intervals. Since the inequality rates were estimated from different populations, we calculated random effects models when heterogeneity was not too low. When performing ANOVA analyses, we used F-tests to compare the RII and SII means of the healthcare systems, and to determine whether between-group variance was larger than within-group variance. Meta- and ANOVA analyses utilize tests of statistical significance, but with a small country-level sample size, estimates are bound to be surrounded by uncertainty [47]. We therefore avoid using these analyses as tests of whether differences between healthcare system types are significant or non-significant. Fig 1 displays statistical uncertainty as 95% confidence intervals, while S3–S5 Tables includes the p-values from the ANOVA analyses.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. RII and SII estimates (95% CIs).

Healthcare system types in parentheses.

https://doi.org/10.1371/journal.pone.0234135.g001

Limitations

The approach of classifying countries into typologies or regimes has been subject to debate. As Wendt [6] has demonstrated, several typologies with different healthcare system types and varying country classifications have been proposed during the last few decades (e.g. [7,59–61]). Although typologies inherently capture a broad range of interrelated dimensions, they also always depend on the extent to which dimensions are emphasized or de-emphasized in the operationalization. Apparently similar programs and policies may be differently organized, and indicators upon which a typology is based, for instance choice restrictions and funding, may be confounded. However, the healthcare system typologies first developed by Wendt [6] and later followed up by Reibling et al. [28] is to our knowledge the most comprehensive typology to our knowledge, aiming to intercept all important aspects of a healthcare system.

To adapt the Reibling et al. [28] typology to our available data material, we classified Lithuania and Switzerland as respectively Low-supply and low performance mixed systems and Supply- and choice-oriented public systems. Classification was done by key indicators from the initial factor analyses of Reibling et al. [28]. Additional meta-analyses and ANOVA tests showed that including these countries in their respective clusters affected meta-analysis estimates, but the overall differences between the estimates remained similar, while results from ANOVA tests excluding Switzerland and Lithuania indicated that the within-type variation was not lower than the between-type variation, similar to the analyses of amenable mortality.

Some compatibility issues occurred between the country-level healthcare system typology and the individual-level cause-specific mortality data. The Reibling et al. [28] typology is based on data from 2011 to 2014, while the mortality data covers the period 1998 to 2006 (depending on country, see S1 Table). Though the 2019 healthcare system types have similarities with earlier typologies (cf. [6,62]), this partial incompatibility weakens the link between our two data levels. Most all analyses combining data from the individual and country level face similar constraints; the influence of country-level variables on mortality is hard to narrow down in general, as numerous policies affect one’s health over the life course. In our discussion, we have met this limitation by using the typologies to describe variations rather than assigning direct effects to specific policies.

The 20% of Finns excluded from the data was a random sample and results should not be affected. Related is the exclusion of non-Swiss nationals from the Swiss data. The impact of this potential bias is unclear; our analyses may over- or underestimate the magnitude of inequalities in mortality in Switzerland as a whole, depending on inequalities in mortality in the excluded population compared to Swiss nationals. As aforementioned, meta-analyses and ANOVA with and without Switzerland returned similar results, but this exclusion nevertheless limits our conclusions. Non-linkage represents another limitation; applying the correction factor provides a more accurate result but will not remove a systematic non-linkage bias–we do not know the composition of the non-linked populations. Lastly, the “No education” and “Missing education data” categories may be heterogenous; Flanagan and McCartney [63] have demonstrated how differentiation across categories and missing data on educational attainment has varied between censuses in England and Wales from 1971 to 2001. The ISCED categories provides comparability across countries, but national differences in questioning, coding, and organization of the education system are still unaccounted for.

The applied definition of amenable mortality and the indicators used to construct a typology may also be conflicting. An apparent example is that consumption data on alcohol and tobacco are used to measure for healthcare prevention performance, while mortality directly related to lifestyle traits was excluded from the analyses. Variation in countries’ performance in preventing smoking and alcohol use may thus not be reflected in the mortality numbers. On the other hand, Reibling et al. [28] included these indicators as proxies; they are meant to indicate general preventive care performance. Further, only mortality directly attributed to smoking and alcohol use was excluded; we included causes of death indirectly associated with lifestyle, which again could be related to the performance of a country’s preventive services.

The concepts of amenable mortality and healthcare system types offers both the advantages and disadvantages associated with combining several dimensions in one encompassing classification. Originally, amenable mortality was intended to be useful in terms of policy intervention, with an aim to distinguish those forms of mortality that a more effective organization of the healthcare system could deal with. However, such classifications may also hide variation between the different causes of death–within and across countries. Though amenable mortality was originally proposed as an indicator of healthcare quality, Nolte and McKee [33] have suggested–on the basis of the ambiguous operationalisations and evidence–that it rather should be treated as a starting point for further research and an indicator of concern. Although our analysis may suffer from crude divisions of mortality, we argue that these were necessary steps for the cause of overview and comparison, and as a point of departure for discussing how healthcare systems may produce health inequalities. We urge future research to derive more specific policy recommendations based on empirical analyses focusing on specific aspects of healthcare systems and detailed forms of amenable mortality. This will require the availability of rich data at the individual level as well as the national level for a large number of countries to improve statistical power.