How Do Health System Features Influence Health System Performance?

3. Do financial incentives to providers improve performance?

Abstract

This chapter examines the relationship between healthcare system design – with a particular focus on financial incentives and payment methods – and treatable mortality rates across OECD countries. Using cluster analysis, countries were grouped into three clusters based on their approaches to provider financial incentives and payment systems. The analysis found that health systems using strong financial incentives for quality to providers showed lower treatable mortality rates compared to systems with weak/limited quality incentives. This difference was particularly significant in systems that combined weak quality incentives with fee‑for-service payment methods. However, these relationships should not be interpreted as causal due to the lack of an appropriate counterfactual and potential unmeasured system features.

This chapter looks at whether differences in treatable mortality rates across countries can be explained by clustering health systems based on actionable policy levers such as financial incentives to providers to improve healthcare quality and volume incentives embedded in physicians’ payment methods.

Avoidable deaths are categorised as those that are either preventable or treatable. A death is considered preventable if it can be avoided through effective public health and primary prevention interventions. On the other hand, a treatable death is a premature death which could be avoided through timely and effective healthcare interventions, including secondary prevention (Nolte and McKee, 2008[1]) (Tobias and Yeh, 2009[2]).

While preventable deaths indicate the state of public health, treatable deaths reflect the availability, accessibility, and quality of healthcare interventions.

In this set of analyses, treatable mortality1 was used as outcome variable, assuming that a lower rate of deaths amenable to healthcare can indicate an improvement in quality of care. Treatable mortality was identified by OECD and WHO as a key performance indicator that works as tracer for access to and quality of care (Figueras et al., 2023[3]).

Financial incentives to providers have been associated with measures of better access and higher quality of healthcare. An early review by the Cochrane Collaboration based on 4 systematic reviews of 32 studies concluded that financial incentives are effective at changing healthcare professional practice, but, given the paucity of studies, it is still unclear whether they contribute to improved patient outcomes (Flodgren et al., 2011[4]). Along similar lines, a more recent systematic review of studies conducted by (Heider and Mang, 2020[5]) indicated the positive effects on the quality of care of certain types of financial incentives provided to healthcare staff, although it is noteworthy that other studies are less clear with respect to the direction of the effect (for the United Kingdom see (Mandavia et al., 2017[6])). With respect to the use of mixed payment schemes, a growing body of literature has previously demonstrated that combining capitation with fee‑for-service (FFS) incentives mitigates significantly the under provision of medical services observed with capitation and the overprovision of services observed with FFS (Brosig-Koch et al., 2013[7]). Bundled payment schemes consistently report increases in efficiency and corresponding cost savings of health systems (Feldhaus and Mathauer, 2018[8]). Regarding patient outcomes, studies have shown that blended capitation payment is associated with improvements in some aspects of diabetes care (Bamimore et al., 2021[9]). Showing similar results but taking the debate a step further, Li et al. conclude that when patients are characterised by lower disease severity and resource consumption is relatively small, FFS may be a more suitable payment method (Li et al., 2022[10]).

Following these findings, we visually analysed the relationship between indicators constructed using country responses to the HSC survey (see Annex A) and the output variable (treatable mortality). This analysis helped identify five indicators to which the output variable is particularly sensitive: volume incentives embedded in physicians’ payment schemes, volume incentives embedded in hospital payment schemes, financial incentives for healthcare quality, recruitment and remuneration of hospital staff and the degree of regulation of prices/fees for providers. However, recruitment and remuneration of hospital staff was not used to group health systems due to low face validity (Table 3.1).

Table 3.1. Description of the indicators used for a cluster analysis on access to high-quality services

Indicator	Description
Volume incentives embedded in physician payment schemes	Evaluates the predominant mode of payment for physicians in primary care, community, and hospitals (i.e. fee‑for-service, capitation, salary), as a proxy for incentives to generate volumes of services. A higher value of the indicators means a larger use of fee‑for-service.
Volume incentives embedded in hospital payment schemes	Indicates the likely impact of hospitals’ predominant payment method (i.e. line‑item budget, per diem, global budget, DRGs, per procedure) on volume of care. A higher value of the indicator means a larger use of activity-based payments
Financial incentives for healthcare quality	Shows whether financial bonuses for primary care physicians, specialists and hospitals are provided, related to preventive care, management of chronic diseases, patient satisfaction, uptake of IT services. A higher value of the indicator means a larger use of pay-for-performance
Degree of price/fee regulation	Indicates the degree of price regulation of payments to providers by purchasers. A higher value means that prices are set unilaterally by purchasers at central level

A data driven cluster analysis based on these indicators (see Annex 3.A) indicated that OECD health systems2 can be grouped into three clusters mainly based on financial incentives to providers to improve quality of care and volume incentives embedded in physicians’ payment methods. Expert judgement confirmed the face validity of groups that display the following key features (Figure 3.1):

• Cluster 1: A first group of countries – Chile, Czechia (2023), France, Korea, Portugal, Spain and the United Kingdom – use large financial incentives to providers to improve quality of care.

• Cluster 2: A second group of countries – Austria, Colombia, Costa Rica, Estonia, Finland, Greece, Hungary, Iceland, Israel, Italy, Lithuania, Mexico, Norway, Poland, Slovenia and Sweden are characterised by weak/limited incentives for quality and use blended payment arrangements for physicians (i.e. FFS and capitation).

• Cluster 3: A third group consisting of Australia, Belgium, Canada, Czechia (2016), Denmark, Germany, Latvia, Luxembourg, the Netherlands and Switzerland use weak/limited incentives for quality; however, in contrast to the previous group, their health system is characterised by the use of fee‑for service as the predominant payment method for physicians.

Figure 3.1. Groups of health systems with similar financial incentives to providers to improve quality of care and physicians’ payment methods

Statistical methods

A pooled panel regression model was used to understand whether clusters of health systems sharing similar financial incentives and payment methods to providers help explain differences in treatable mortality – the output variable.

The model controlled for lifestyle, environmental and socio‑economic factors (Table 3.2). These variables are consistent with those included in previous empirical analyses and reflect the key determinants of health outcomes as identified in the relevant literature. Furthermore, it was assumed that if characteristics for a given country did not change between 2016 and 2023, then the health system was assigned to the same cluster for the whole period in study. The model also controlled for two features that were identified as non-actionable from a policy perspective in the absence of large‑scale structural reform: the overall type of coverage (residence‑based/single payer versus multiple insurers) (Paris et al., 2016[11]) and the degree of decentralisation of spending autonomy in health (Dougherty and Phillips, 2019[12]). Finally, a dummy variable was used to capture the impact of COVID‑19.

Table 3.2. Control variables used in the panel regression model

Variable	Indicator	Reference
Wealth	GDP per capita (PPP constant)*	(Hajat et al., 2010[13])
Economic activity	Unemployment rate	(Clemens, Popham and Boyle, 2014[14])
Inequality	Gini index of household income distribution	(Kondo et al., 2009[15])
Education	% of the population 25‑65 with tertiary education	(Balaj et al., 2024[16])
Risk factors	Percentage of obese population Percentage of population 15+ daily smokers	(Wang et al., 2014[17]) (Glei, Lee and Weinstein, 2022[18])
Environmental hazard	Pollution (PM2.5 kg per km2)	(Fuller et al., 2022[19])

Note: GNI – instead of GDP – was used for Luxembourg.

Key findings

In line with expectations, the panel regression analysis indicated that lower treatable mortality rates are reported for health systems with higher GDP per capita, higher level of educational attainment and lower obesity rates. Treatable mortality rates were lower in the “multiple insurers and high decentralisation” group of health systems as compared to the “residence‑based/single payer and high decentralisation” group of health systems (Table 3.3).

Compared to cluster 1, where health systems use strong incentives for quality, health systems in cluster 2 and cluster 3, which are characterised by weak/limited incentives for quality, reported higher treatable mortality rates. However, the finding for cluster 2 was not statistically significant (Table 3.3).

Table 3.3. Variables with a statistically significant coefficient in the regression model

Variable	Estimate	Standard Error	P-value
GDP per capita	‑9.25 E‑06	2.16 E‑06	0.000***
Education	‑0.01	0.002	0.000***
Obesity	0.029	0.007	0.001**
NonModChar3 (vs. NonModChar1)	‑0.18	0.088	0.06’
Cluster 2 (versus cluster 1)	0.092	0.07	0.19
Cluster 3 (versus cluster 1)	0.33	0.09	0.000***

Note: Significant result at ‘0.1, *0.01, **0.001, ***0.000 level. The model uses the Arellano method for heteroskedasticity-consistent standard errors (White) clustered at country level. The outcome of the model is log transformed. The model is also controlled for level of unemployment, Gini, PM2.5, tobacco, COVID‑19 and non-modifiable characteristics. The full model, together with the functional form, estimation method and assumption testing can be found in Annex 3.A.

Health systems in cluster 3, which combine weak/limited incentives for quality with a predominant use of fee‑for-service to pay physicians, reported a statistically significant potential for outcome increase (Figure 3.2).

Figure 3.2. Average difference in treatable mortality rates by cluster

Note: 95% confidence intervals are displayed.

Sensitivity analyses

Sensitivity analyses were conducted to assess the robustness of results to adding covariates such as hospital beds per 1 000 population and workforce per 1 000 population, time fixed effect and year as a numeric variable (see Annex 3.A). The direction and significance of coefficients remained robust across these specifications. However, hospital beds and workforce – metrics that capture capacity – influence the effect of the non-modifiable and modifiable characteristics (clusters) variables.

Financial incentives to providers to increase quality of care are associated with better performance

Because of the lack of an appropriate counterfactual, the relationship between the key variables in this analysis should not be interpreted as causal. Moreover, it is possible that the observed effect of clusters on the outcome variable may reflect underlying health system features that were not captured or controlled for in the analysis. Despite these limitations, the results suggest an important association: higher use of financial incentives for providers to enhance quality is linked to a reduction in deaths that should be preventable with high-quality, timely and effective healthcare.

Annex 3.A. Clustering and sensitivity analyses

Output variable

The cross-country comparison of treatable mortality rates per 100 000 population – the output variable of this set of analysis – is shown in Annex Figure 3.A.1.

Annex Figure 3.A.1. Treatable mortality rates by country, 2021

Source: OECD Health Statistics, January 2024.

Clustering

Ward’s method was employed to group countries into clusters based on four indicators that could be influenced by actionable policy levers: volume incentives embedded in physician payment schemes; volume incentives embedded in hospital payment schemes; financial incentives to increase healthcare quality; intensity of price/fee regulation (Annex Table 3.A.1). A hierarchical clustering algorithm was used to create a dendrogram (Annex Figure 3.A.2).

Annex Table 3.A.1. Score of indicators by cluster

Indicator		Mean
	Cluster 1	Cluster 2	Cluster 3
Volume incentives embedded in physician payment schemes	0.30	0.19	0.65
Volume incentives embedded in hospital payment schemes	0.65	0.54	0.60
Financial incentives to increase healthcare quality	0.90	0.07	0.26
Intensity of price/fee regulation	0.79	0.76	0.72

Annex Figure 3.A.2. Dendogram

Three clusters were then identified as containing elements that were similar among themselves and dissimilar to elements belonging to other groups (Annex Figure 3.A.3).

Annex Figure 3.A.3. Health systems by cluster. Data driven approach

Statistical approach

Pooled ordinary least squares regression

A pooled ordinary least squares regression was used to estimate the relationship between various socio‑economic, environmental, and health-related factors and treatable mortality, defined as the mortality amenable to quality healthcare. The dependent variables, treatable mortality rate is operationalised as the logarithm to improve model fit. The explanatory variables include Gross Domestic Product (GDP) per capita, adjusted for purchasing power parity, which serves as a control for wealth of the country. The unemployment rate was also included, in order to capture the economic activity. The Gini index was used to measure income inequality. Educational attainment was represented by the percentage of the population aged 25‑65 with tertiary education, which reflects the overall education level of the country. Health-related behaviours were captured through variables such as the percentage of the obese population and the percentage of daily smokers aged 15 and above. Environmental exposure was represented by the mean annual concentration of particulate matter (PM2.5 in mg per m^3) per squared kilometre.

In addition to the model variables outlined above, the model included care capacity indicators of hospital beds per 1 000 population (Beds) and Primary care practitioners (General practitioners, paediatricians and gynaecologist) per 1 000 population (PCP). In time, capacity indicators (rate of workforce and hospital beds per 1 000 population) were included only as sensitive analysis.

Additionally, Nonmodifiable health system characteristics (NonModChar) were included to account for aspects of the health system that are not subject to change in the short term (Annex Table 3.A.2).

Annex Table 3.A.2. Non-modifiable health system characteristics by country

	Spending autonomy in health
Coverage	High decentralisation	Low decentralisation
Residence‑based – single payer	Australia, Canada, Denmark, Finland, Italy, New Zealand, Spain, Sweden Türkiye, United Kingdom	Costa Rica, Estonia, Greece, Hungary, Iceland, Ireland, Korea, Latvia, Lithuania, Luxembourg Norway, Poland, Portugal, Slovenia
Multiple insurers	Chile, Colombia, France, Germany, Israel, Japan, Netherlands, Slovak Republic, the United States	Austria, Belgium, Czechia, Mexico, Switzerland

Source: OECD Health Systems Characteristics Survey; Dougherty, S. and L. Phillips (2019[12]), “The spending power of sub-national decision makers across five policy sectors”, https://doi.org/10.1787/8955021f-en.

Finally, the model used dummy variables (Clusters) to represent the three clusters of countries (Equation 3). The analysis is conducted using the “plm” package in R.

Equation 3

${\mathrm{l}\mathrm{o}\mathrm{g}(TreatMort}_{it})=$

$\alpha +{\beta }_0{GDP}_{it}+{{\beta }_1 Education}_{it}+{\beta }_2{Unemployment}_{it}+{\beta }_3{Gini}_{it}+{\beta }_{4}Pollution+{\beta }_5{Obesity}_{it}+{\beta }_6{tobacco}_{it}$$+{{\beta }_{7}COVID}_{it}+{\beta }_{8}NonModChar+{\beta }_9\mathit{C}\mathit{l}\mathit{u}\mathit{s}\mathit{t}\mathit{e}\mathit{r}\mathit{s}{+\epsilon }_{it}$

With $i$ representing OECD countries and $t$ the year in the 2016‑22 period.

Assumption testing

Several tests were conducted to assess the underlying assumptions of our panel models. We first performed a Chow test for poolability using the pooltest function (R package plm), which examines whether the panel data model benefits from individual effects beyond the pooled model. For both models, this test suggested potential individual effects across entities, significant at 5% error. Due to the static nature of our interest variables, we could not be adequately controlled by a fixed effects model at country level without losing these variables from the analysis, thus becoming one of the limitations of our approach. Nevertheless, we clustered our standard errors at the country level to account for potential country level fixed effects and not overestimate the significance of the results. Similarly, we tested the significance of and included time fixed effects when appropriate by calculating F Test for Individual and/or Time Effects. In the final models, time fixed effects were not significant when adding the COVID‑19 variable.

Additionally, Pesaran’s test of cross-sectional dependence was applied to investigate the independence of errors across panel units. For heteroskedasticity, a Breusch-Pagan test indicated the presence of heteroskedastic error terms in both models, which was addressed by computing robust standard errors using the Arellano method. The Breusch-Godfrey test was conducted for autocorrelation, which checks for serial correlation in the residuals, and assessed multicollinearity among independent variables using the Variance Inflation Factor. The results of the main model are summarised in detail in Annex Table 3.A.3. The variables in the model were removed one by one, with the overall model conclusions remaining strong to the different specifications. The final model corresponds to the one showing the highest R squared.

Annex Table 3.A.3. Panel regression results

Variable	Estimate	SE	P-values
(Intercept)	3.88E+00	2.71E‑01	2.14E‑14***
GDP	‑9.25E‑06	2.18E‑06	2.21E‑04***
Education	‑1.31E‑02	3.01E‑03	1.58E‑04***
Unemployment	‑1.81E‑02	8.58E‑03	4.34E‑02*
Gini	1.29E+00	6.45E‑01	5.48E‑02
PM2.5_km	‑1.52E‑04	1.91E‑04	4.32E‑01
Obese	2.92E‑02	7.85E‑03	8.86E‑04***
Tobacco	1.61E‑02	8.34E‑03	6.40E‑02
Post_Covid years	4.22E‑02	4.02E‑02	3.02E‑01
NonModChar2 (vs. NonModChar1)	9.70E‑02	9.59E‑02	3.20E‑01
NonModChar3 (vs. NonModChar1)	‑1.76E‑01	8.83E‑02	5.59E‑02’
NonModChar4 (vs. NonModChar1)	‑1.06E‑01	9.64E‑02	2.83E‑01
Cluster 2 (versus cluster 1)	9.92E‑02	7.36E‑02	1.89E‑01
Cluster 3 (versus cluster 1)	3.26E‑01	9.98E‑02	2.88E‑03**
R2	0.89
R2adj	0.84
F-statistic	17.7085 on 13 and 28 DF, p-value 3.72E‑10

Note: Significant result at ‘0.1, *0.05, **0.01, ***0.001 level. The model uses the Arellano method for heteroskedasticity-consistent standard errors (White) clustered in time (year) and group (Country). The outcome of the model is log transformed.

NonModChar1: Residence‑based or single payer + Relatively high spending power in health of subnational decision makers.

NonModChar2: Residence‑based or single payer + Relatively low spending power in health of subnational decision makers.

NonModChar3: Multiple insurers + Relatively high spending power in health of subnational decision makers.

NonModChar4: Multiple insurers + Relatively low spending power in health of subnational decision makers.

Sensitivity analyses

The model was tested with various specifications, incorporating variables such as hospital beds per 1 000 population and workforce per 1 000 population, time fixed effects and year as a numeric variable.

The direction and significance of the coefficients remained robust across these specifications. However, healthcare capacity variables (workforce and beds) influence the effect of the nonmodifiable and modifiable characteristics (clusters) variables (Annex Table 3.A.4). These findings are consistent with the expected mediation effect of health system inputs on the relationship between characteristics and output.

Annex Table 3.A.4. Sensitivity analysis: Panel regression results

Variable	Estimate	SE	P-values
(Intercept)	2.86E+00	3.34E‑01	4.63E‑09***
GDP	‑1.46E‑05	2.72E‑06	1.23E‑05***
Education	‑1.41E‑02	2.30E‑03	1.79E‑06***
Unemployment	‑7.78E‑03	6.15E‑03	2.17E‑01
Workforce	4.66E+00	8.35E‑01	7.38E‑06***
Beds	‑3.00E‑04	1.38E‑04	3.91E‑02*
Gini	2.63E‑02	7.15E‑03	1.06E‑03**
PM2.5_km	‑7.02E‑03	9.80E‑03	4.80E‑01
Obese	2.86E+00	3.34E‑01	4.63E‑09***
Tobacco	‑1.46E‑05	2.72E‑06	1.23E‑05***
Workforce	1.45E‑01	2.94E‑02	7.75E‑02
Beds	2.88E‑02	3.79E‑02	4.23E‑05***
Post_Covid years	7.93E‑02	7.49E‑02	4.54E‑01
NonModChar2 (vs. NonModChar1)	‑1.45E‑01	7.24E‑02	2.99E‑01
NonModChar3 (vs. NonModChar1)	‑2.66E‑01	7.70E‑02	5.58E‑02
NonModChar4 (vs. NonModChar1)	1.00E‑01	7.25E‑02	1.88E‑03**
Cluster 2 (versus cluster 1)	3.67E‑01	9.50E‑02	1.79E‑01
Cluster 3 (versus cluster 1)	1.45E‑01	2.94E‑02	6.68E‑04***
R2	0.93
R2adj	0.90
F-statistic	25.3665 on 15 and 26 DF, p-value 9.35E‑12

Note: Significant result at ‘0.1, *0.05, **0.01, ***0.001 level. The model uses the Arellano method for heteroskedasticity-consistent standard errors (White) clustered in time (year) and group (Country). The outcome of the model is log transformed.

NonModChar1: Residence‑based or single payer + Relatively high spending power in health of subnational decision makers.

NonModChar2: Residence‑based or single payer + Relatively low spending power in health of subnational decision makers.

NonModChar3: Multiple insurers + Relatively high spending power in health of subnational decision makers.

NonModChar4: Multiple insurers + Relatively low spending power in health of subnational decision makers.

References

[16] Balaj, M. et al. (2024), “Effects of education on adult mortality: a global systematic review and meta-analysis”, The Lancet Public Health, Vol. 9/3, pp. e155-e165, https://doi.org/10.1016/s2468-2667(23)00306-7.

[9] Bamimore, M. et al. (2021), “Quality of Diabetes Care in Blended Fee-for-Service and Blended Capitation Payment Systems”, Canadian Journal of Diabetes, Vol. 45/3, pp. 261-268.e11, https://doi.org/10.1016/j.jcjd.2020.09.002.

[7] Brosig-Koch, J. et al. (2013), “How to Improve Patient Care? - An Analysis of Capitation, Fee-for-Service, and Mixed Payment Schemes for Physicians”, SSRN Electronic Journal, https://doi.org/10.2139/ssrn.2278841.

[14] Clemens, T., F. Popham and P. Boyle (2014), “What is the effect of unemployment on all-cause mortality? A cohort study using propensity score matching”, European Journal of Public Health, Vol. 25/1, pp. 115-121, https://doi.org/10.1093/eurpub/cku136.

[12] Dougherty, S. and L. Phillips (2019), “The spending power of sub-national decision makers across five policy sectors”, OECD Working Papers on Fiscal Federalism, No. 25, OECD Publishing, Paris, https://doi.org/10.1787/8955021f-en.

[8] Feldhaus, I. and I. Mathauer (2018), “Effects of mixed provider payment systems and aligned cost sharing practices on expenditure growth management, efficiency, and equity: a structured review of the literature”, BMC Health Services Research, Vol. 18/1, https://doi.org/10.1186/s12913-018-3779-1.

[3] Figueras, J. et al. (eds.) (2023), Assessing Health System Performance: Proof of Concept for a HSPA Dashboard of Key Indicators, OECD Publishing, Paris/World Health Organization, Geneva, https://doi.org/10.1787/4e6b28c0-en.

[4] Flodgren, G. et al. (2011), “An overview of reviews evaluating the effectiveness of financial incentives in changing healthcare professional behaviours and patient outcomes”, Cochrane Database of Systematic Reviews, https://doi.org/10.1002/14651858.cd009255.

[19] Fuller, R. et al. (2022), “Pollution and health: a progress update”, The Lancet Planetary Health, Vol. 6/6, pp. e535-e547, https://doi.org/10.1016/s2542-5196(22)00090-0.

[18] Glei, D., C. Lee and M. Weinstein (2022), “Assessment of Mortality Disparities by Wealth Relative to Other Measures of Socioeconomic Status Among US Adults”, JAMA Network Open, Vol. 5/4, p. e226547, https://doi.org/10.1001/jamanetworkopen.2022.6547.

[13] Hajat, A. et al. (2010), “Long-Term Effects of Wealth on Mortality and Self-rated Health Status”, American Journal of Epidemiology, Vol. 173/2, pp. 192-200, https://doi.org/10.1093/aje/kwq348.

[5] Heider, A. and H. Mang (2020), “Effects of Monetary Incentives in Physician Groups: A Systematic Review of Reviews”, Applied Health Economics and Health Policy, Vol. 18/5, pp. 655-667, https://doi.org/10.1007/s40258-020-00572-x.

[15] Kondo, N. et al. (2009), “Income inequality, mortality, and self rated health: meta-analysis of multilevel studies”, BMJ, Vol. 339/nov10 2, pp. b4471-b4471, https://doi.org/10.1136/bmj.b4471.

[10] Li, X. et al. (2022), “Effects of fee-for-service, diagnosis-related-group, and mixed payment systems on physicians’ medical service behavior: experimental evidence”, BMC Health Services Research, Vol. 22/1, https://doi.org/10.1186/s12913-022-08218-5.

[6] Mandavia, R. et al. (2017), “Effectiveness of UK provider financial incentives on quality of care: a systematic review”, British Journal of General Practice, Vol. 67/664, pp. e800-e815, https://doi.org/10.3399/bjgp17x693149.

[1] Nolte, E. and M. McKee (2008), “Measuring the health of nations: updating an earlier analysis”, Health Affairs, Vol. 27/1, pp. 58-71, https://doi.org/10.1377/hlthaff.27.1.58.

[11] Paris, V. et al. (2016), “Health care coverage in OECD countries in 2012”, OECD Health Working Papers, No. 88, OECD Publishing, Paris, https://doi.org/10.1787/5jlz3kbf7pzv-en.

[2] Tobias, M. and L. Yeh (2009), “How much does health care contribute to health gain and to health inequality? Trends in amenable mortality in New Zealand 1981–2004”, Aust N Z Public Health, Vol. 33/1, pp. 70-78, https://doi.org/10.1111/j.1753-6405.2009.00342.x.

[17] Wang, X. et al. (2014), “Fruit and vegetable consumption and mortality from all causes, cardiovascular disease, and cancer: systematic review and dose-response meta-analysis of prospective cohort studies”, BMJ, Vol. 349/jul29 3, pp. g4490-g4490, https://doi.org/10.1136/bmj.g4490.

Notes

← 1.  COVID‑19 deaths are not included in the list of causes of treatable mortality.

← 2.  Ireland, Japan, New Zealand, the Slovak Republic and Türkiye are not included in this analysis as there are missing values for the indicators used to cluster health systems.