Efficiency and technology gaps in achieving universal health coverage: a metafrontier data envelopment analysis of 189 countries

¹Independent freelancer, Abu Dhabi, United Arab Emirates, ²Department of Health, Abu Dhabi, United Arab Emirates, ³MatchingMind, London, UK

^&Corresponding author
Eyob Zere Asbu, Independent freelancer, Abu Dhabi, United Arab Emirates

Introduction    

Firms operating across different industries, regions, or countries face heterogeneous production opportunities because of differences in technology, resources, institutions, and market conditions [1]. In health systems, this heterogeneity may translate into distinct production frontiers across income groups (LICs, LMICs, UMICs, and HICs), reflecting differences in health system capacity, technology, financing, human resources, and institutional environments. These differences shape how efficiently countries transform health expenditure into UHC service coverage and the scope for performance improvement within each context.

Universal health coverage (UHC) is central to Sustainable Development Goal (SDG) 3, with progress monitored through SDG Indicator 3.8.1, the UHC Service Coverage Index (SCI), alongside financial protection under Indicator 3.8.2 [2,3] . The SCI ranges from 0 to 100 and summarizes coverage of essential health services across reproductive, maternal, newborn, and child health; infectious diseases; non-communicable diseases; and service capacity and access. Its use as an output measure provides a basis for assessing how effectively countries translate health-system resources into service coverage.

Inefficient allocation and use of scarce health resources remain widespread across health systems. The World Health Organization estimates that approximately 20%-40% of global health expenditure is lost to waste and inefficiency [4]. Recent evidence likewise indicates persistent inefficiencies and resource constraints in achieving UHC across income groups. For example, Eze et al. (2024), analyzing 191 countries using Data Envelopment Analysis and Tobit models, found that both managerial inefficiency and underinvestment constrain progress toward UHC [5]. These findings highlight the importance of distinguishing between inefficiency in the use of existing resources and constraints associated with the technology available to countries.

Recent cross-country studies, including Jordi et al. (2020) and Eze et al. (2024) [5,6], have used pooled-frontier DEA to assess the efficiency with which countries translate health-system resources into UHC-related outcomes. While a common frontier provides an overall measure of relative efficiency, it does not distinguish inefficiency relative to the technology available within a country’s peer group from gaps between that group-specific technology and global best practice. A metafrontier approach addresses this distinction by estimating group-specific frontiers alongside a common metafrontier, thereby decomposing overall efficiency into within-group technical efficiency and the technology gap ratio (TGR). Within the UHC framework, suboptimal service coverage relative to health expenditure may therefore reflect either technical inefficiency or technology gaps arising from differences in workforce capacity, infrastructure, institutional maturity, and service-delivery capabilities. Distinguishing between these sources is important for identifying whether performance improvements primarily require better use of existing resources or longer-term technological and systemic investment.

The metafrontier DEA framework provides a means of making this distinction[1,7]. It decomposes overall performance into group-specific technical efficiency, which captures how effectively countries operate relative to the best-performing countries within their income group, and the technology gap ratio (TGR), which measures the distance between the group frontier and the pooled metafrontier. This decomposition allows performance gaps to be interpreted in terms of managerial inefficiency versus technology-related constraints. This study assesses how efficiently countries across income groups convert health resources into UHC service coverage using a metafrontier DEA framework. Specifically, it estimates country-level efficiency relative to income-group frontiers, benchmarks performance against a pooled metafrontier, and measures technology gaps using the TGR. The central policy question is whether gaps in UHC coverage primarily reflect inefficient use of resources relative to the production possibilities available within countries’ peer environments, or whether those peer environments themselves are technologically disadvantaged relative to global best practice.

Methods     

Data and sources

Current health expenditure (CHE) per capita, expressed in purchasing power parity (PPP)-adjusted international dollars, was obtained from the WHO Global Health Expenditure Database [8]. Purchasing power parity adjustment facilitates cross-country comparison by accounting for differences in price levels and purchasing power. Data for SDG Indicator 3.8.1, the UHC Service Coverage Index (SCI), were obtained from the UHC2030 data portal [9]. The SCI is a composite index ranging from 0 to 100 that measures coverage of essential health services across four domains: reproductive, maternal, newborn, and child health; infectious diseases; non-communicable diseases; and service capacity and access.

Both datasets refer to 2023 and were extracted in January 2026 using the most recent available releases. Countries without observations for both variables were excluded from the analysis; no imputation was undertaken (including carry-forward, interpolation, or other methods for estimating missing observations). The resulting complete-case sample comprised 189 countries. Five WHO Member States were excluded because of insufficiently comparable data: Cuba, Democratic People’s Republic of Korea, Ukraine, Cook Islands, and Niue. Countries were assigned to World Bank income groups using the FY2025 classification applicable to the 2023 study period [10].

Analytical approach

Data envelopment analysis model specification

A non-parametric DEA framework was used to assess how efficiently countries convert health expenditure into service coverage. An output-oriented BCC model with variable returns to scale (VRS) was applied [11] recognizing that increases in spending do not necessarily translate into proportional gains in coverage across countries. The output orientation reflects the policy objective of maximizing service coverage, given largely fixed health budgets in the short to medium term [12]. For each country k, the output-oriented VRS DEA model is defined as:

Where: j = index of countries (1…,189), k =target country being evaluated, x_j = input for country j, y_j = output for country j, λ_j = intensity weight, φ represents the maximum proportional expansion in outputs for country k, holding inputs constant. Technical efficiency is defined as:

Where TE_k=1 indicates that country k lies on the efficient frontier, and values below 1 indicate scope to expand UHC service coverage without increasing expenditure.

Group-specific technology sets and frontiers

Group-specific technology sets and frontiers are used to capture heterogeneity in health systems across World Bank income groups. Each group is assumed to operate under a distinct production technology, reflecting differences in resources, institutions, and access to health technologies. We selected World Bank income status over WHO region as the metafrontier grouping variable based on tests of within-group homogeneity in both the input and output variables. The sample-weighted average CV for CHE per capita was 0.651 within World Bank income groups versus 1.270 within WHO regions, while the corresponding CVs for UHC SCI were 0.132 and 0.164, respectively. Thus, the income-group classification produces lower within-group dispersion for both the input and output, particularly for CHE. Moreover, every WHO region has a higher CHE CV than every World Bank income group. While this does not establish strict technological homogeneity, it provides empirical support for the relative coherence of the income-group partition in this application. For each income group g, the group-specific technology set is defined as:

This set represents the feasible combinations of health expenditure and service coverage within each income group. It is constructed from observed country performance and satisfies the standard DEA assumptions of convexity, free disposability, closedness, and boundedness. Group-specific frontiers are estimated using Equation (1), restricting the reference set to countries within the same income group. In doing so, the frontier represents best-practice performance under comparable technological conditions.

Metafrontier framework

Let T_j denote the technology set for income group g, as defined in equation (3). The metafrontier technology set is defined as:

where G is the number of income groups, denotes the union of the group-specific technology sets, and conv⁡ denotes their convex hull. Thus, the metafrontier encompasses the feasible technologies observed across all income groups and provides a common benchmark for comparing countries operating under different technological conditions. Metafrontier efficiency is estimated using the same output-oriented VRS DEA model specified in Equation (1), with all countries pooled in the reference set. This yields metafrontier technical efficiency, , while estimation using only countries in the same income group yields group-specific technical efficiency, .

Technology gap ratio

The technology gap ratio (TGR) measures the extent to which group-level technology falls short of the global best-practice technology, and is defined as:

Values close to one indicate that a country operates near the global technology frontier, while lower values reflect larger technology gaps. These gaps are primarily attributed to structural or technological constraints, rather than to managerial inefficiency alone.

Bootstrap correction

To account for statistical noise and the deterministic nature of DEA, we applied the Simar-Wilson smoothed bootstrap with 2,000 replications to the group-frontier and metafrontier estimates [13,14]. The bootstrap retained the output-oriented VRS specification used in the main analysis, with group frontiers estimated separately within World Bank income groups and the metafrontier estimated on the pooled sample. The procedure used a Silverman bandwidth and a fixed random-number seed of 123. Bootstrap draws were coordinated across the group and metafrontier estimations using a common resampling index and the same smoothed inverse-efficiency draw in each replication, thereby preserving comparability when calculating the technology-gap ratio. Bias correction was applied to the inverse efficiency measure, (φ=1/TE), as , followed by . Ninety-five percent confidence intervals were obtained from the 2.5^th and 97.5^th percentiles of the corresponding bias-adjusted bootstrap distributions. Full code and country-level data are provided in the supplementary materials.

Sensitivity analysis

We additionally applied Andersen-Petersen super-efficiency analysis to assess the influence of individual frontier-defining observations, using a threshold of 1.20 [15], and performed leave-one-out re-estimation of mean technical efficiency for each income group.

Inputs and outputs

A single-input, single-output specification is used. The input is current health expenditure per capita (PPP-adjusted), which represents the overall health system resource envelope. It serves as a proxy for the broader set of health system inputs, including human resources, infrastructure, medicines, and technologies, and is standard in cross-country efficiency analyses where disaggregated input data are limited [16,17]. It also avoids double-counting issues that can arise when components already embedded in the UHC Service Coverage Index (e.g., workforce and beds) are included separately. The variable is log-transformed to address skewness. The output is the UHC Service Coverage Index (SCI; SDG 3.8.1). Because the UHC SCI is bounded between 0 and 100, using it directly in an output-oriented DEA model could imply projections beyond its theoretical maximum. To avoid this, we defined the proportional shortfall from full coverage as:

And transformed the output as:

This strictly increasing transformation maps the bounded SCI onto a non-negative, unbounded scale while preserving country rankings. Its concave inverse implies that proportional expansion in y' corresponds to diminishing absolute gains in SCI as coverage approaches full saturation. Data envelopment analysis projections were subsequently back-transformed to the original SCI scale.

Data analysis

All analyses are conducted using R version 4.4.3 [18]. Data envelopment analysis and bootstrap inference are implemented using the Benchmarking package, applying an output-oriented VRS model [19]. Data management and descriptive analyses are carried out using dplyr [20], while visualization using ggplot2 [21]. Results were exported to Excel using the openxlsx package [22].

Results     

Descriptive analysis

Table 1 summarizes the descriptive statistics for the input and output variables by country income group. UHC SCI increases steadily with income, from a mean of 44.4 in LICs to 81.9 in HICs, while health expenditure per capita rises from Int$45 to Int$3,611. Across all 189 countries, health expenditure ranges from Int$17 in Madagascar to Int$13,473 in the United States, and UHC SCI ranges from 26 in Chad to 92 in Canada. Dispersion in health expenditure also widens across income groups, with the greatest variation observed among HICs. Figure 1 shows the association between CHE per capita (log scale) and UHC SCI. The two variables are strongly positively correlated (Pearson’s r = 0.849, p < 0.001), but the relationship is markedly nonlinear. The LOESS smoother (span = 0.75; 95% CI shaded) rises steeply at lower expenditure levels before showing a visible flattening around Int$165-200 per capita (log CHE ≈ 5.1-5.3). Considerable variation in UHC SCI is evident among countries with similar levels of health spending, particularly at lower expenditure levels. Among countries with log CHE below 4.5, for example, Syria, Lao PDR, Pakistan, and Rwanda are located above the fitted relationship, whereas Madagascar, Somalia, Chad, and Ethiopia are located below it. Similar dispersion is evident in the middle of the distribution, with Uzbekistan, Thailand, China, and Brazil above the fitted curve and Guatemala, Botswana, and Bosnia and Herzegovina below it. Among high-income countries, clustered along the flatter portion of the curve, UHC SCI ranges from approximately 76 to 92 despite several-fold differences in health spending.

Efficiency and technology gaps

Group-specific frontiers

The overall mean group-frontier technical efficiency score for the full sample was 0.714 (bias-corrected: 0.498). In the output-oriented framework, and after back-transforming the projected outputs from the log-shortfall scale back to the original UHC Service Coverage Index (SCI) scale, these estimates suggest that countries could achieve, on average, a potential 22.8% increase in UHC SCI (39.8% after bias correction) while maintaining their current levels of inputs. Importantly, these percentage improvements are expressed in the original UHC SCI metric and therefore represent potential improvements in UHC service coverage rather than the proportional output expansion measure (1/TE-1). The estimated gains represent ceteris paribus potential improvements toward the observed relative frontier, holding the observed level of inputs constant, rather than forecasts of realized gains. Accordingly, they indicate the scope for improvement implied by the efficiency model and should not be interpreted as causal or guaranteed increases in UHC service coverage.

This potential varies substantially across income groups, declining with income level. The estimated average potential increase in UHC SCI ranges from 41.0% (bias-corrected: 85.7%) in low-income countries to 7.3% (bias-corrected: 17.0%) in high-income countries, with lower-middle-income countries showing a potential gain of 26.1% (bias-corrected: 56.9%) and upper-middle-income countries 16.5% (bias-corrected: 28.5%). These results indicate substantially greater scope for improving UHC service coverage through increased efficiency in lower-income countries, while potential gains are more limited in higher-income countries. Andersen-Petersen super-efficiency identified Syria as the only frontier-defining country exceeding the prespecified 1.20 threshold (1.504). Leave-one-out re-estimation showed that excluding Syria increased the mean LIC group technical efficiency from 0.585 to 0.725 (Δ = 0.141).

Table 2 summarizes the group-frontier efficiency scores, metafrontier efficiency scores, and technology gap ratios (TGRs), reported as both original and bias-corrected estimates, by income classification. Panel A of the table shows a clear gradient in group frontier efficiency scores across income classifications, with technical efficiency increasing by income group. Based on the original estimates, mean technical efficiency rises from 0.585 in LICs to 0.783 in HICs. This is accompanied by a reduction in variability, as seen in the coefficient of variation (CV), which declines from 34.0% to 13.2%. A similar pattern is observed after bias correction, although mean technical efficiency levels are lower across all groups. Bias-corrected mean scores range from 0.401 in LICs to 0.552 in HICs, maintaining the same increasing pattern by income group. Substantial within-group variation remains, particularly in lower-income settings, as indicated by relatively high coefficients of variation. Country level efficiency scores, peer reference sets and λ weights, potential output gains, and the corresponding technology gap ratios are provided in Annex 1 A, Annex 1 B, Annex 1 C, Annex 1 D.

Metafrontier efficiency

Table 2, panel B reports metafrontier technical efficiency scores obtained by estimating a pooled DEA frontier over all 189 countries. In contrast to the group-specific frontiers, the metafrontier represents the global best-practice technology set against which all countries are evaluated. Original efficiency scores range from 0.233 to 1.000, with a mean of 0.674. Bias-corrected estimates are uniformly lower, averaging 0.452 across the full sample, consistent with adjustment for upward bias in DEA scores. Four countries-Canada (HIC), Madagascar (LIC), Syria (LIC), and Laos (LMIC)-define the metafrontier. All had Andersen-Petersen super-efficiency scores below the prespecified 1.20 threshold. The table further disaggregates these estimates by income group, allowing comparison of how closely each group approaches the global frontier. Metafrontier efficiency also exhibits a clear income gradient: mean uncorrected scores increase from 0.547 in LICs to 0.757 in HICs, with a similar pattern after bias correction, where mean scores range from 0.368 in LICs to 0.518 in HICs.Within the output‑oriented framework, the global mean bias‑corrected efficiency of 0.452 corresponds to a potential increase in UHC service coverage of approximately 43% relative to observed levels, given the existing level of inputs. Variation in efficiency declines with income, as reflected in the coefficient of variation, which decreases from 56.8% in LICs to 17.6% in HICs. To further examine the relationship between technical efficiency and the logarithm of current health expenditure per capita, Figure 2 plots bias-corrected technical efficiency scores against per capita health expenditure. Panel A presents estimates relative to group-specific frontiers, while panel B reports the corresponding estimates relative to the pooled metafrontier.

In panel A, technical efficiency is positively associated with the logarithm of current per capita health expenditure, though substantial heterogeneity remains among countries at comparable expenditure levels. Dispersion tends to be greater among lower-spending countries, particularly low- and lower-middle-income countries, whereas higher-spending countries generally exhibit higher efficiency scores and somewhat less dispersion. The LOESS curve rises across the lower and middle expenditure ranges before flattening at higher expenditure levels, indicating a weaker association between expenditure and efficiency among higher-spending countries. Efficiency scores also overlap substantially across income groups, particularly in the middle of the expenditure distribution.

Panel B shows a broadly similar expenditure-efficiency relationship when countries are evaluated against the common metafrontier. Metafrontier technical efficiency scores are generally lower than those estimated under the group-specific frontier, reflecting the more demanding benchmark set by the best-practice technology available across all countries. Dispersion is greater among lower-spending countries and generally narrows at higher expenditure levels, particularly among high-income countries, although substantial heterogeneity remains among countries with similar levels of health expenditure. The LOESS curve indicates a nonlinear relationship, with efficiency increasing across intermediate expenditure levels before flattening more modestly at higher expenditure levels than in panel A.

Taken together, the two panels show that the expenditure-efficiency relationship is broadly preserved when moving from group-specific frontiers to the common metafrontier. Although the metafrontier generally reduces measured efficiency, the resulting differences are modest relative to the substantial variation in efficiency across countries operating at similar expenditure levels. This pattern is consistent with the high technology-gap ratios reported in the main analysis and indicates that cross-country differences in the efficiency with which health expenditure is translated into UHC service coverage are more pronounced than differences in access to best-practice technology.

Technology gap ratio

Group-specific efficiency measures how efficiently countries convert health spending into UHC service coverage relative to peers at similar income levels, but do not distinguish between managerial inefficiency and differences in the technology available to countries. The TGR provides this distinction by comparing group-specific frontiers with the global metafrontier. Table 2, panel C summarises TGR values by income group. Original TGR values are high across all income groups, ranging from 0.921 in LMICs to 0.968 in HICs, with an overall mean of 0.942. Following bias correction, mean TGRs are lower, ranging from 0.885 in LMICs to 0.946 in HICs, with an overall mean of 0.912. Despite this reduction, group-specific frontiers remain relatively close to the global metafrontier, particularly in HICs. The lower TGRs among LICs and LMICs indicate somewhat larger technology gaps in these groups, although the differences remain modest overall.

Relative variability is low in the original estimates, with the coefficient of variation ranging from 2.7% in LMICs to 5.5% in LICs. After bias correction, dispersion increases, particularly in UMICs and HICs, where the coefficient of variation reaches 8.2%, compared with 6.4% in LICs and 3.7% in LMICs. Overall, the bias-corrected estimates suggest modest technology gaps, with greater between-country variation after bias correction.

The sensitivity analysis indicated that the LIC technology-gap estimate was influenced by Syria, which had an Andersen-Petersen super-efficiency score of 1.504. Excluding Syria reduced the mean LIC TGR from 0.927 to 0.762, while the overall mean TGR changed only slightly, from 0.942 to 0.948. TGRs in the other income groups remained close to unity (0.981 in LMICs, 0.967 in UMICs, and 0.976 in HICs), indicating that the sensitivity was largely confined to the LIC estimate.

Figure 3 plots bias-corrected group-frontier efficiency against the TGR, using the respective sample means to define four quadrants. The sample means serve as descriptive reference points for visual comparison and do not represent substantively defined or policy-relevant thresholds. TGR values are generally high and concentrated near the top of the plot, with a sample mean of 0.912, whereas group-frontier efficiency varies substantially across countries, with a mean of 0.498. The figure shows a broadly positive association between the two measures, but with considerable dispersion across countries. Countries are concentrated predominantly in the two upper quadrants: high-income countries are more prominent in the upper-right quadrant, while many low- and middle-income countries are located in the upper-left quadrant. A substantial number of countries with below-average group-frontier efficiency nonetheless have TGRs close to or above the sample mean, whereas some countries with relatively high group-frontier efficiency have TGRs below the sample mean. The lower-right quadrant contains relatively few observations compared with the upper quadrants.

Discussion     

This study applies a metafrontier technical efficiency framework to assess how countries convert health system resources into UHC service coverage. By benchmarking performance against both income-group and global frontiers, the analysis decomposes metafrontier efficiency into two distinct components: within-group inefficiency and technology-related gaps captured by the TGR. This distinction is critical, as it allows cross-country differences to be interpreted in terms of suboptimal resource use versus constraints in access to global best-practice technologies. Our findings extend previous cross-country DEA assessments of UHC efficiency by making this distinction explicit. Whereas a pooled-frontier DEA provides an overall measure of relative efficiency, the metafrontier decomposition separates inefficiency within the prevailing production environment from differences in the technology available to countries.

The results show a strong positive but nonlinear relationship between current health expenditure per capita and UHC service coverage, consistent with existing evidence [23]. Countries with higher income levels tend to achieve greater coverage, which may reflect greater fiscal capacity [24,25] as well as stronger institutional and service delivery systems. At the same time, the flattening of the expenditure–coverage relationship at higher spending levels is consistent with diminishing marginal returns observed in cross-country analyses [6], suggesting that progressively greater resources may be required to achieve additional improvements as coverage increases. The substantial variation in coverage among countries with similar levels of expenditure further indicates that spending alone does not fully explain differences in UHC performance.

This variation is also evident within income groups. Countries operating at comparable expenditure levels can achieve markedly different levels of UHC service coverage, particularly at lower expenditure levels. This finding reinforces the view that adequate spending, although important, is not sufficient to ensure high UHC service coverage [23]. The broader literature suggests that differences in coverage at comparable spending levels may also reflect variations in policy choices, including health financing arrangements (e.g., pooling and purchasing), and service delivery systems not captured by aggregate expenditure measures [26,27].

The group-specific frontier analysis indicates substantial scope to improve UHC service coverage by using existing resources more efficiently. The magnitude of this potential gain follows a clear income gradient, with larger gains in low-income countries and progressively smaller gains in higher-income countries, which operate closer to their respective frontiers. These findings suggest that improving the efficiency with which existing health-system resources translate into service coverage could yield particularly important gains in lower-income settings. They are broadly consistent with previous cross-country evidence, although differences in model specification and coverage limit direct comparability [28].

The substantial within-group variation in technical efficiency indicates that countries with broadly similar levels of health expenditure can nonetheless differ markedly in their ability to translate health expenditure into UHC service coverage. This variation is most pronounced in lower-income settings, where efficiency scores span a wide range, while higher-income countries exhibit a more concentrated distribution. Although the present analysis does not identify the specific institutional or organizational drivers of these differences, the pattern is consistent with evidence that governance quality, institutional capacity, financing arrangements, and service-delivery organization are associated with health-system efficiency [29-32]. These factors warrant further investigation as plausible explanations for the observed heterogeneity.

The metafrontier analysis provides an important complementary perspective. Although efficiency scores are generally lower when countries are evaluated against the common global frontier, the gap between group-specific and metafrontier performance is relatively modest compared with the substantial variation in technical efficiency within income groups. This pattern is consistent with the high TGRs across income groups and indicates that differences in how efficiently resources are converted into UHC service coverage are more pronounced than differences in the technology represented by the group-specific frontiers. The substantial overlap in efficiency across income groups further suggests that relatively strong performance is achievable across different income settings, rather than being confined to the highest-income countries. The TGR results reinforce this interpretation. TGRs were high across all income groups, indicating that the group-specific frontiers generally lie relatively close to the global metafrontier. Although technology gaps were somewhat larger among lower-income groups, the differences were modest overall. These findings suggest that, within the production technology captured by the model, limited distance from the global best-practice frontier is not the dominant explanation for cross-country differences in performance. Instead, greater variation is observed in the efficiency with which countries operate within their respective technology sets.

This interpretation is further supported by the joint distribution of technical efficiency and TGR in Figure 3. Many countries with below-average technical efficiency nonetheless have TGRs close to or above the sample mean, particularly among low- and middle-income countries. Conversely, relatively few countries combine high technical efficiency with substantially below-average TGRs. The latter group represents settings in which further improvements may depend more on narrowing the technology gap, whereas countries with high TGR but low technical efficiency have greater scope to improve performance through more effective use of existing resources. These patterns support a differentiated view of efficiency and technology constraints rather than a single explanation for UHC performance gaps.

The consistently high bias-corrected TGRs across income groups, with most estimates exceeding 0.8 and generally close to unity, indicate that, for UHC service coverage, technology gaps are generally modest relative to within-group inefficiency. Thus, the greater scope for improving performance appears to lie in how efficiently countries use the technology available to them, rather than in differences in the technology itself. This provides a different policy interpretation from a conventional pooled DEA ranking, in which these two sources of performance differences cannot be separately identified. The sensitivity analysis nevertheless indicates that the technology-gap estimate for LICs is influenced by the position of Syria on the LIC frontier. This qualification is important when interpreting the LIC-specific estimate, but does not change the broader conclusion that technology gaps are generally less pronounced than differences in technical efficiency.

Taken together, the efficiency and TGR results suggest that policy priorities will differ according to countries’ relative positions along the two dimensions. In settings where technical efficiency gaps are substantial but TGRs are relatively high, the greatest gains are likely to come from improving the use and organization of existing resources within the prevailing technology set. By contrast, in countries operating close to their efficiency frontier but facing larger technology gaps, expanding system capacity and upgrading available technologies may be more important. This distinction underscores the value of context specific strategies for improving UHC service coverage, rather than relying on uniform prescriptions focused solely on additional resources or technological upgrading.

Limitations of the study

This study has several limitations. First, it relies on cross-sectional data for a single year and therefore provides a static assessment of efficiency and technology gaps, without capturing changes over time. Second, although the data are drawn from international sources, differences in the quality, completeness, and accuracy of country reporting may affect the estimates. Third, the efficiency and technology-gap estimates remain sensitive to measurement error and model specification despite the use of bootstrap procedures for bias correction and statistical inference. Fourth, metafrontier estimates are conditional on the specification of inputs and outputs and on the grouping structure used to define the group-specific frontiers. We selected World Bank income groups as a theoretically motivated proxy for heterogeneous production environments and found lower within-group dispersion in health expenditure than under WHO regional groupings; nevertheless, alternative classifications could yield different estimates. Finally, the results should be interpreted as relative measures of performance conditional on the specified production technology and grouping structure, rather than as absolute measures of efficiency or technology gaps.

Conclusion     

This study demonstrates the value of distinguishing between technical inefficiency and technology gaps when assessing countries’ progress toward UHC service coverage. The metafrontier DEA framework shows that, although countries differ substantially in their efficiency relative to their income group frontiers, the technology gaps between group specific frontiers and the global metafrontier are generally modest. This indicates that, for most countries, greater gains in UHC service coverage are likely to come from improving the efficiency with which existing resources are used rather than from changes in the underlying technology represented by the frontier.

These findings support a differentiated approach to policy. In settings where technical inefficiency is substantial, efforts to strengthen health system organization, management, accountability, and service delivery may yield the largest improvements. By contrast, countries operating relatively close to their efficiency frontier but facing more pronounced technology gaps may require investments that expand service capacity and enhance production possibilities. The metafrontier decomposition therefore provides a useful diagnostic framework for distinguishing these sources of performance differences and for informing benchmarking and context specific policy responses, rather than relying on a single pooled efficiency ranking.

The findings should nevertheless be interpreted with appropriate caution. In particular, the sensitivity of the LIC technology-gap estimate to Syria's position on the LIC frontier highlights the influence that individual frontier observations can have on metafrontier analysis. Despite this qualification, the broader finding that technology gaps are generally less pronounced than differences in technical efficiency remains unchanged. Overall, the results suggest that accelerating progress toward UHC requires not only additional resources but also greater attention to how effectively existing resources translate into service coverage.

Competing interests     

The authors declare no competing interests.

Authors' contributions     

Eyob Zere Asbu compiled and analysed the data and led the drafting of the manuscript. Mohamed Abdulla Al Haj and Zainab Nader Khazaal critically reviewed and revised the manuscript. Bela Prasad contributed to the drafting, review, and revision of the manuscript. All authors have read and approved the final version of this manuscript.

Tables and figures     

Table 1: output and input descriptive statistics by country income group

Table 2: group-frontier efficiency, metafrontier efficiency, and technology gap ratio (TGR) by income classification

Figure 1: universal health coverage service coverage index versus log current health expenditure per capita (purchasing power parity)

Figure 2: technical efficiency and health expenditure: bias-corrected group frontier and metafrontier estimates across countries

Figure 3: group-frontier technical efficiency and technology gap ratio (bias-corrected estimates)

Annexes     

Annex 1: supplementary materials (PDF - 1252KB)

References