A fuzzy set qualitative comparative analysis of 131 countries: which configuration of the structural conditions can explain health better?

Today, there is a 34-year gap between countries with the highest and lowest life expectancies [38]. However, no biological explanation has not been found for such a difference [18, 41].

It is now generally accepted that inequalities root in the conditions under which people are born, grow up, live, work, and grow older. Such conditions of daily life, also referred to as social determinants of health (SDH), are linked to several systems and forces, including political systems, social policies, economic policies, and social norms [18, 29]. In 2005, the World Health Organization (WHO) established the Commission on Social Determinants of Health (CSDH) to attract the attention of the governments and civil society to the effects of social factors upon health. Under the broad category of health determinants, The CSDH categorizes health determinants into structural determinants (those leading to health inequalities), intermediary health determinants (including material, psychosocial, behavioral, and/or biological factors), and the health system itself as a social determinant [29]. The CSDH’s final report provided a large body of evidence on what can be done to reduce health inequalities within and between countries. The commission made it perfectly clear that policies for health equity involve a wide range of sectors with different core activities.

Figure 1 depicts the conceptual framework of the CSDH with all its elements. As seen, the interactions between various socioeconomic and political factors, such as gender, race/ethnicity, income, education, occupation, and other factors, form a set of socioeconomic positions which affect the intermediary determinants of health. These determinants, which reflect a person’s position in the social order, are influenced by one’s social status and experiences of exposure and vulnerability to health-compromising conditions. Overall, The CSDH has subcategorized health determinants as the sociopolitical context, structural determinants and socioeconomic position, and intermediary determinants [29].

It should be noted that this framework is distinguished from previous ones by its emphasis on the socioeconomic and political context and the structural determinants of health inequality. Therefore, the most important lesson learned by the CSDH during the dialogue between idea and evidences on social determinants of health was that in order to minimize health inequalities, governments should focus on not only intermediary determinants of health, but also the underlying structural determinants [17, 18, 29]. Since policies aiming at structural determinants of health inequalities require the involvement of main institutes distributing power and resources within the society, they should be carefully designed based on methodologies developed by social and political science and according to contextual specifications [18, 29].

Since the CSDH published its conceptual framework for tackling the social causes of health inequalities, extensive research has been conducted on the empirical relevance of the socioeconomic variables discussed in the CSDH report [1, 7, 10, 14, 42]. Such quantitative research assessed the relationship between socioeconomic disparities and health gradients and the degree to which social factors could explain health inequalities. The considerable evidence compiled by these studies identified the “net effect” of particular socioeconomic variables on health outcomes [20].

Nonetheless, the idea that “complexity defines health” is the core focus of the policies recommended by the CSDH [17, 29]. This notion implies that the production of positive health outcomes requires interaction and coordination between the health sector and other non-health-related sectors. Indeed, the CSDH conceptual framework presents a complex and nonlinear causal model of health equity. Accordingly, this study aimed to unravel these causal complexities by developing a fuzzy set Qualitative Comparative Analysis (fsQCA) [19, 20].

The outcome of interest for this internationally comparative study was life expectancy at birth. This focus was selected due to the status of life expectancy as the most general indicator of health outcomes and a main concern in health-related Sustainable Development Goals [29, 38]. Life expectancy at birth is an estimate of the average number of years that a newborn is expected to live if current mortality rates continue to apply. It indicates the overall mortality level of a population [36]. A large body of literature has shown the relationship between socioeconomic conditions and life expectancy, i.e. socioeconomic health disparities have been found to be reflected in life expectancy [2, 4, 7, 40, 42].

After an overview of all recommended policies discussed in the CSDH reports, some preliminary analysis¹ was conducted and the authors identified five conditions as the most theoretically and empirically important drivers of socioeconomic and political conditions within the macro-level context in the CSDH framework. These conditions included [1] economic development (country wealth), [2] education, [3] quality of governance, [4] income inequality, and [5] health systems as an intermediary factor [29]. Furthermore, data for these five indicators are available at the national level with reasonable amount of missing value.

We applied fsQCA in order to assess the configurational effects of these five macro-level socioeconomic conditions. Our reasons for selecting fsQCA were related to the CSDH’s approach which emphasizes the need for multiple, intersectoral policies and actions on health. The major advantage of utilizing fsQCA for our study was that it enabled us to deal with causal complexity [20]. Moreover, the use of fsQCA facilitated the evaluation of cases as configurations of conditions, rather than the net effect of each variable on outcomes. We were able to conduct context-specific assessments indicating the ways in which multiple causal recipes linked to a particular outcome. Another advantage of using fsQCA in this study was its compatibility with macro-level data analysis in small- or medium-N research [20]. In fact, using this technique, we could answer certain questions that could not be approached via conventional variable-oriented analysis [19, 20]. More precisely, following the CSDH’s recommendations and considering set-theoretic relations, we were concerned with two main questions:

First, the positive and negative forms of five conditions were included in the necessity analysis. Based on the existing recommendations, the minimum acceptable level of consistency for the test of necessity was set at 0.965 [31, 34]. Accordingly, no individual conditions were found necessary for high or low level of life expectancy (Table 3).

Since fsQCA is sensitive to researcher-specified thresholds and benchmarks related to 1) the calibration of raw data, 2) the minimum frequency of cases in each configuration, and 3) the consistency cutoff [12, 28], several robustness tests were conducted in this study. The crossover point varied ±10% for all conditions in both analyses [6, 30]. The lowest acceptable consistency for solutions was set at second-best choice that was 0.84. Finally, the minimum frequency threshold was set at one. The results reported above are robust to the use of these alternative specifications (see Additional file 5).

In this cross-national comparative analysis, countries were examined configurationally and different combinations of conditions linked to high/low life expectancy were identified. The first question in this research was “What combination of structural conditions is usually sufficient for high life expectancy?” The intermediate solutions resulted from fsQCA indicated that the combinations of high-quality governance, education, affluent health system, and high income were consistently sufficient to cause high life expectancy. It is interesting to note that this solution included a set of countries which all had high levels (≥ 0.9) of outcome of high life expectancy. It, hence, had high empirical relevance for the explanation of the outcome [20]. In response to the second question, we found that four configurations were consistently sufficient for low life expectancy. These results may provide an empirical support for the CSDH argument, which highlights the complexity of the interplay of contextual factors. The following sections will discuss the findings of the fsQCA in more details.

One of the most notable patterns that resulted from the fsQCA was related to the WHO European Region. This region contains 53 countries⁴ of which 46 were included in this study. The region faces issues of industrial and post-industrial society, as well as emerging democracies of central and Eastern Europe and the former Union of Soviet Socialist Republics (USSR) ([5, 16], Wilkinson, 2002). A close look at Table 4 shows that all European Union (EU) members, except Bulgaria, Romania, Latvia, and Lithuania, made up the configurations that were consistently sufficient for the outcome of high life expectancy (row 1 and 2). All of these countries, except Croatia, enjoyed high levels of education, governance, health system, wealth and low level of income inequality. Croatia, however, had a relatively high level of income inequality and was placed in row 2.

The former communist countries comprised another interesting group in this region. According to the fsQCA, the members of the former USSR in our sample, except Baltic states and Georgia (i.e. Armenia, Azerbaijan, Kazakhstan, Kyrgyz Republic, Moldova, Russian Federation, and Ukraine), along with other countries, made up the first configuration of intermediate solution for low life expectancy (see Table 8) This configuration indicated that weak governance in countries with a low level of income inequality was consistently sufficient for low life expectancy. Surprisingly, in this solution term, income inequality appeared in the unexpected theoretical direction. Nations that were officially ruled by a socialist ideology and were theoretically expected to have improved public health [3] had the lowest life expectancies in Europe [38]. Although the role of weak governance in the occurrence of low life expectancy was highlighted in this solution term, it must be noted that political conditions in countries are likely to rely on many structural and intermediary factors that might affect populations’ health. Nevertheless, such factors, e.g. weakness of the soviet health policy in dealing with social changes and crisis, environmental damage and occupational exposures (e.g. air pollution), and poor health behaviors including heavy alcohol drinking, smoking, physical inactivity, and unhealthy diet, were not incorporated in this study [3, 24]. Further in-depth research is thus needed to explain the mechanisms by which political conditions in such contexts lead to low life expectancy.

Baltic countries (i.e. Estonia, Lithuania, and Latvia) indicated the configuration of conditions linked to high life expectancy (see row 1 of Additional file 6). While the outcome was relatively high in Estonia, it was not as high as expected in Lithuania and Latvia and these two countries were more out of than in the set of countries with high life expectancy. They were both contradictory cases. The Baltic region witnessed dramatic socioeconomic and political changes following its independence during World War I and II. The USSR later reoccupied the region and the region switched to market economy in the early 1990s. It entered the EU in 2004 and found the chance to continue more favorable trends in life expectancy before Russia [35]. These major historical changes exerted strong positive/negative effects on the process of health transition in the region. Further in-depth analysis is, hence, required to explain the differences between the level of life expectancy in these three states where the outcome was not explained by the configuration detected here.

In contrast to the former USSR members, the former Yugoslavian Countries in our sample (i.e. Croatia, Bosnia and Herzegovina, Montenegro, Serbia, and Slovenia) and Albania had relatively high life expectancy. As can be seen in row 1 and 2 of Table 4, the combination of conditions that is consistent with being usually sufficient caused high life expectancy in Croatia, Montenegro, and Slovenia. In comparison, this study could not explain the relatively high level of life expectancy in Bosnia and Herzegovina, Serbia, and Albania by the condition used.

Overall, the relatively low life expectancy in Latvia, Lithuania (also Romania and Bulgaria), where the configuration was linked to high life expectancy, is a matter of question. It is also interesting how countries such as Bosnia and Herzegovina and Albania achieved a relatively high life expectancy, while they made up a configuration linked to the low level of outcome. These contradictory configurations indicated the presence of other causal pathways to the outcome of life expectancy. More detailed research would, thus, be required to identify the potentially causal conditions that could be added to the truth table analysis [20, 26]. The final report of the CSDH argued that health equity and well-being is influenced by a set of combinatory factors, including biological and psychosocial factors, the social environment, material circumstances, and behaviors. These factors are, in turn, affected by social position which is itself under the influence of gender, ethnicity, race, education, occupation, and income. All of such effects would definitely depend on the governing sociopolitical and sociocultural context [16, 29].

Another interesting place to explore is the American Region of the WHO. This region contains 35 countries⁴ of which 26 were included in our sample. As shown in row 1 and 2 of Table 4, Barbados, Canada, Uruguay, and the United States displayed configurations linked to the outcome of high life expectancy. However, a closer look at the data (see Additional file 1) indicated that some countries in this region had higher level of outcome but were not included in our identified configurations linked to high life expectancy. For instance, life expectancy is higher in Chile and Costa Rica than in the United States. However, the United States obviously outperformed both of them in wealth, education, quality of governance, and affluent health system. It would be interesting to study what other causally relevant conditions might contribute to the high level of life expectancy in Chile and Costa Rica. For instance, according to previous studies, in case of Costa Rica, one case-specific explanation might rely on lower socioeconomic gradients in the main causes of death (e.g. lung cancer and heart disease) compared to the United states ([25], Rosero-Bixby and Dow, 2016). Moreover, much steeper socioeconomic disparities were seen in the United States, compared to Costa Rica, for behavioral risk factors such as smoking, obesity, health insurance coverage, and uncontrolled dysglycemia and hypertension (Rosero-Bixby and Dow, 2016).

There were also some contradictory configurations in this region. The relatively high level of life expectancy in Panama, Mexico, Ecuador, Jamaica, and Peru and the relatively low level of life expectancy in Brazil were not explained by configurations detected in this study. Therefore, more in-depth research is warranted to explain their level of outcome.

The Western Pacific Region of WHO includes 27 countries⁴ of which 10 were recruited in our sample. Australia, Japan, and South Korea were cases included in the configuration that were consistently sufficient for high life expectancy. It should be noted that New Zealand and Singapore were cases of missing data. Two cases, i.e. China and Vietnam, had contradictory outcomes and their relatively high level of life expectancy could not be explained by this study. According to the existing literature, the great concern about socioeconomic determinants of health in both of these countries has led to the emergence of policy priorities for governance, especially regarding poverty reduction and access to health care for the poor [39]. Thus, a sole case study with more details on factors such as health gap reduction policies (e.g. targeted subsidies, investments in infrastructure, training at the grassroots level, disease prevention and health promotion in maternal and child health, and infectious diseases) may help explain their level of life expectancy [36, 39].

Malaysia is another country that may require a case study to explain its relatively low level of life expectancy. Malaysia made up the configuration of relatively high-level education, governance, and economic development combined with relatively poor health system and high income inequality. This configuration was not consistent for being sufficient to either high or low life expectancy (see row 5 of Table 4).

The fsQCA indicated that among the six WHO regions, none of the Eastern Mediterranean, South-East Asian, and African countries in the sample were included in configurations that were consistently sufficient for high life expectancy. However, our study could not help to explain the level of life expectancy in Iran and Tunisia. They indicated configurations linked to low life expectancy (see row 7 and 8 of Additional file 6), while there were more in than out in the set of countries with high life expectancy. Due to such contradictions, we excluded these countries from the sample [26].

The primary goal of this configurational comparative study was to assess the connection between combinations of socioeconomic conditions and life expectancy at birth and to compare the results with the CSDH conceptual framework. The results of the fsQCA indicated that high or low life expectancy could be affected by combination of socioeconomic and political conditions. These results supported the CSDH’s argument that policies and actions on health, as a complex phenomenon, should be intersectoral in nature. Critics of QCA would argue that fsQCA is made to find configurational effects even where none may exist in reality [13]. Yet, Ragin stated that QCA privileges complexity, while conventional quantitative methods privilege simplicity. When substantive knowledge tells us that causation is complex or combinatorial, it is very unlikely to identify it using conventional quantitative methods. Indeed, what exists in reality is best described by substantive knowledge and is not the property of an analytical method [23].

According to the assessment of the combinations of conditions leading to high or low life expectancy, the configurations linked to high life expectancy were not the opposite of those associated with low life expectancy (see Table 5 and Table 7). This is in accordance with the concept of asymmetric causality [20].

Overall, this study evaluated the configurational effects of structural conditions, instead of estimating the “net effect” of each factor on life expectancy. As Ragin noted, “policy discourse often focuses on categories and kinds of people (or cases), not on variables and their net effects across heterogeneous populations” [20]. While it is useful to assess the impact of socioeconomic factors on population health, it could be more helpful for policymakers to know under what conditions such factors play a role. As discussed, a context-specific assessment is needed to answer such questions and this is possible when cases are studied as configurations (as done in this study) [20].

Finally, based on our findings, further in-depth research and detailed examinations are required in particular countries to explain their levels of life expectancy.

This research had some limitations regarding problems of causal ambiguity, operationalization, and generalizability.

The most notable limitation was the problem of causal ambiguity. Even though we used the fsQCA as an appropriate methodological tool for analyzing causal complexity, we could not prove causality. fsQCA identifies configurations that are consistent with usually being sufficient to cause outcomes. It does not eliminate the chance of spurious conjunction of configurations and outcomes [30]. Moreover, while health is mainly affected by the socioeconomic position of individuals [15, 18, 29], it, in some cases, causes social and economic consequences, i.e. it actually determines individuals’ socioeconomic position [29].

Since such conditions may lead to health inequality, there is still a chance of reverse causation (see Fig. 1).

Table 4 shows that a combination of all five conditions was consistently sufficient for high life expectancy (row 2). This could be the result of a mathematical quirk of the fsQCA, rather than an indicator of real causation. As mentioned earlier, according to the way that logical AND accomplished, creating a combination of unions of all five conditions meant that each case entered the analysis with its lowest possible configurational set score. The analysis of sufficiency in fsQCA then tested whether the cases’ scores for given configuration were equal to or lower than their scores for the outcome condition. Thus, larger number of conditions included in the solution would increase the risk of the sufficiency test yielding a spurious result. It is worth noting again that substantive knowledge is of the greatest importance. If the five-condition recipe made good sense and could be connected to cases, these facts undercut the argument that the recipe is spurious. However, as recommended by Ragin, we computed the degree of membership in each case’s second lowest membership score (of the five conditions). By comparing consistency and coverage scores, we found that having four out of five conditions was good enough to explain the outcome of high life expectancy.

The next challenge in this analysis was related to the operationalization and measurement. We were not certain if it was theoretically justifiable to give equal weight to all the sub-components of the scales used to measure the components (e.g. quality of governance and health system). Ideally, each sub-component had to be assigned a weight indicating the level of importance it had in affecting life expectancy.

Another limitation of the data used in this research was the selection biases that decreased the generalizability of the results. As mentioned earlier, cases with missing data were removed from the study sample. For instance, all members of the Gulf Cooperation Council (i.e. Bahrain, Kuwait, Oman, Qatar, Saudi Arabia, and the United Arab Emirates) were excluded. Some contradictory cases were also removed from the sample. Excluding these cases, which might have been systematically different from those included in the sample, decreased the generalizability of the results [30].

Please find the link below for the data that support the findings of this study.