Menu
Menu
Menu
Menu
The move towards universal health coverage is premised on having well-functioning health systems, which can assure provision of the essential health and related services people need. Efforts to define ways to assess functionality of health systems have however varied, with many not translating into concrete policy action and influence on system development. We present an approach to provide countries with information on the functionality of their systems in a manner that will facilitate movement towards universal health coverage. We conceptualise functionality of a health system as being a construct of four capacities: access to, quality of, demand for essential services and its resilience to external shocks. We test and confirm the validity of these capacities as appropriate measures of system functionality. We thus provide results for functionality of the 47 countries of the WHO African Region based on this. The functionality of health systems ranges from 34.4 to 75.8 on a 0-100 scale. Access to essential services represents the lowest capacity in most countries of the region, specifically due to poor physical access to services. Funding levels from public and out-of-pocket sources represent the strongest predictors of system functionality, compared with other sources. By focusing on the assessment on the capacities that define system functionality, each country has concrete information on where it needs to focus, in order to improve the functionality of its health system to enable it respond to current needs including achieving universal health coverage, while responding to shocks from challenges such as the 2019 coronavirus disease. This systematic and replicable approach for assessing health system functionality can provide the guidance needed for investing in country health systems to attain universal health coverage goals.
Vital signs were constructed from indicators that relate to it. Multiple indicators were used for each vital sign to reduce the influence of a single indicator and ensure the score and resulting capacity index reflects the multiple actions that are needed to improve it. Given challenges in data availability across multiple indicators across 47 countries, proxy indicators were used, which were selected based on the following criteria: (1) it is thematically related to the vital sign; (2) there is country-level data for the indicator; and (3) the data are publicly available from a reputable source. The proxy indicators represent the closest data that exist, related to a specific vital sign and are shown in table 1. Proxy indicators by capacity and vital sign for monitoring overall health system functionality For the indicators identified, a preference was given to data sources that provide primary data, including household surveys and facility assessments. Beyond these, indicator data were obtained from publicly available sources, including the WHO Health Observatory, the United Nations Sustainable Development Goals database and the World Bank World Development Indicators. The only vital sign with no data from these sources was on non-specific resilience, which was derived from health facility resilience assessments conducted during routine disease surveillance activities among countries in the region.33 The data used in this study represent the latest data values available for each country between 2010 and 2018 and are shared in online supplemental appendix S2. In instances where data points were absent for some countries, we implemented Multivariate Imputation by Chained Equations (MICE) using R software to impute for the missing data; the methodology works under the assumption that given the variables we used in the imputation procedure, the missing data are missing at random. This implies that the probability that a value is missing depends only on observed values and not on unobserved values. bmjgh-2020-004618supp001.pdf We used a regression equation with the relationship that overall performance is explained by 34 proxy indicators. Based on this relationship, the missing variables were imputed using the predictive mean matching, a procedure implemented within the MICE package in R software. For each vital sign, a score was derived from the values of the proxy indicators in a stepwise manner. This involved: (1) normalisation of all the indicator values to a range of 0–100 to make them comparable; (2) centralisation of the indicator values for each vital sign into a score; (3) centralisation of the score values for each vital sign in a capacity to derive a capacity index; and (4) centralisation of the four capacity index values to derive the overall functionality index. Normalisation of the indicators was done using the formula: where xMinimum and xMaximum are the lowest and highest reported values, respectively. This standardisation made the indicators comparable, with each country’s value ranging from 0 to 100, representing the worst and best country values for the indicator. The score for each vital sign was derived as the arithmetic mean of the indicators constituting it. Similarly, the index for each capacity was derived as the arithmetic mean for each of the vital signs constituting it, with the overall system functionality index being the arithmetic mean of the four capacities constituting it. Finally, the regional values for each of the capacities and the overall index were calculated as the geometric mean of all country scores. We applied the same weighting at each stage to emphasise the importance of each indicator, vital sign and capacity, respectively. A differential weighting denotes an implicit ranking of importance, which is contrary to the assumption that they are all crucial for overall system functionality. We make several structural and conceptual assumptions, whose validity we tested to improve confidence in the approach. Specifically, we explored whether: (1) the four capacities are appropriate predictors of desired health outcomes on their own; (2) the applied proxy indicators are appropriate predictors for the vital signs for each capacity; and (3) the emerging results are associated with predictors of functionality, such as efficiency, equity and/or effectiveness. We calculated Spearman’s rank correlation coefficients (rho) between the computed index and certain key indicators to test face validity. Correlations were interpreted in a standard manner as defined in literature: <0.35 low, 0.36–0.67 moderate and 0.68–1.0 as high.35 We explore both the magnitude and the relative strengths of correlations in ascertaining face validity, as well as our conceptual approach and methods. See online supplemental appendix S3 for details. In ascertaining how appropriate the four capacities are in predicting desired health outcomes, we compared the emerging results with the UHC service coverage index, minus its service capacity and access component. This is constructed from multiple service coverage outcomes, providing an appropriate comparator for system functionality. We opted to focus our analysis of UHC solely on elements of service coverage, rather than financial risk protection, given the ongoing debate on the measurement challenges associated with current thresholds used in the calculation of the official SDG indicator on financial protection as a component of UHC.36–38 We removed the service capacity and access component as it is conceptually meant to measure the health system functionality component and thus may introduce predictable collinearity. We calculated values for spearman’s rho between the emerging functionality scores and the UHC index, postulating that a strong correlation would confer face validity of the emerging index as a good predictor of UHC attainment. We anticipate a reduced correlation between the emerging functionality index and health security/other health-related outcomes, as these require investments from other sectors whose magnitude varies in different countries and contexts. Looking at the correlation between the emerging results and the impact level, we postulate that there is a strong significant correlation between a functioning health system and better impact based on improved health outcomes and systems efficiency. We postulate that the emerging correlation, while strong, should be weaker than that with the UHC service coverage index (minus service capacity and access component). We calculated values of Spearman’s rho between the emerging functionality index and maternal mortality ratio (MMR), under-5 (U5) mortality rate and neonatal mortality rate. We also examined its relationship with independently derived efficiency levels of health sectors.39 To advance this efficiency analysis, we also examined the relationship between the functionality index and funding by source. This was to better understand which funding sources are most effective at improving health system functionality for attainment of universal health coverage. We calculated Spearman’s rank correlation coefficients between the system functionality index and public, private, out-of-pocket and external sources of funding. The emergent values are dependent on the decision of indicators, representativeness of data used, methodology for imputation of missing data points, index construction process (arithmetic vs geometric mean) and the methodology for standardisation of constituent indicators. Sensitivity analyses tests the robustness of the decisions relating to these variables. We recalculated the index using the arithmetic mean for regional values, as opposed to the geometric mean. We also assessed the effect of switching the data normalisation methodology to using a standardised minimum value of zero as compared with using the least value as the zero value based on the formula: In order to explore the appropriateness of the proxy indicators used to measure the respective vital signs and capacities, we drew on the work of Hogan et al40 to assess the sensitivity of the index to each indicator by assessing the impact on the value by dropping one indicator at a time. If a given indicator is not appropriate, its removal would lead to a significant change in the values of the vital sign and the overall index value for that capacity. Finally, to test the robustness of the index to alternate weighting schemes, we recomputed the index with an adjusted weighting scheme, applying weights generated through principal component analysis (PCA). The emergent index was then correlated with the original index that defaults to equal weighting across all indicators. In carrying out the PCA, we selected the first four components, as they cumulatively explained 50% of the variation. We then generated the indicator loadings (eigenvectors) for each of the selected four components. The weights for each indicator were calculated based on the square of the loading, multiplied by the normalised value of the indicator (see online supplemental appendix S3). Neither patients nor the public were directly involved in the design of the study as it was primarily analytical. The research question and measures were not informed by patients’ or public’s experiences, as this study was largely analytical and based on publicly available data.
N/A
