Estimating the distribution of morbidity and mortality of childhood diarrhea, measles, and pneumonia by wealth group in low- and middle-income countries

With the ambitious goals of ending extreme poverty and fighting against inequity in low- and middle-income countries (LMICs) [1], international agencies, such as the World Bank, have advocated for improving access to essential services, such as healthcare and education, for the lowest 40% of income earners [2]. Policymakers in LMICs, where often vast inequities in health exist, are interested in enhancing equity for both ethical and political reasons [3, 4].

Vaccine programs have been recognized as one of the most successful interventions in improving population health worldwide. Efforts put forward by local governments and international agencies have contributed to raising childhood vaccine coverage in the last decade [5], though high child mortality is still observed in LMICs, with approximately 5.9 million under-five deaths in 2015 [6]. More recently, the Global Vaccine Action Plan and Gavi, the Vaccine Alliance, both listed equitable access to vaccination as their top priorities [7‐9]. However, it is unclear whether ensuring equitable access to vaccines would lead to health equity, which we define as equality of health outcomes across population subgroups. To answer this question, one needs to compare how disease burden is distributed across subgroups before and after the introduction of vaccines. However, such data on disease burden by socioeconomic strata before and after vaccines are not available empirically, making it difficult to design equitable policies for populations. Furthermore, while vaccines prevent diseases, access to treatment is needed in preventing deaths among people who already contracted the disease, which is often another source of inequity across population subgroups. The objective of this work is therefore to introduce an analytical approach to estimate the distribution of childhood disease cases and deaths and the benefits of vaccines and treatments by socioeconomic group.

We examined three major childhood vaccine-preventable diseases – measles, pneumonia, and diarrhea, selected as they represented 23% of deaths as of 2015 (1%, 13%, and 9%, respectively) occurring among under-five children in LMICs [6], and have well-established sets of risk and prognostic factors. We studied the three corresponding vaccines (and the treatment of the diseases they prevent), namely measles vaccine (routine first dose, MCV1), rotavirus vaccine (RV, against rotavirus diarrhea), and pneumococcal conjugate vaccine (PCV, against pneumococcal pneumonia).

The approach is based on the concept of population attributable fraction (PAF) [10]. Using PAF, we quantified the contribution of sets of risk and prognostic factors, defined as behaviors and characteristics of an individual that can be used to estimate the likelihood of contracting and dying from each disease. The proportion of cases and deaths that could be attributed to the exposure of selected sets of risk and prognostic factors are classified as ‘attributable’, and the remaining as ‘unattributable’ cases and deaths. Our model takes advantage of the differences in the prevalence of risk and prognostic factors across wealth strata to estimate the distribution of cases and deaths. A flow diagram outlining each step is presented in Fig. 1.

Estimates for all countries are available in Additional file 1: Section IV. To illustrate, we display our findings for each disease for three populous country examples – Nigeria, Pakistan, and Ethiopia.

Equity is increasingly gaining attention on the global development agenda [2, 4]. To understand whether certain health interventions can lead to changes in health equity, we developed an analytical approach to estimate the changes in the distribution of childhood disease-related incidence and mortality by wealth quintile with selected interventions. In particular, we applied this approach to examine how vaccine programs might affect health equity under different scenarios of intervention coverage and treatment utilization assumptions in LMICs.

Our study highlighted several factors that can substantially contribute to the unequal distribution of childhood disease. First, higher prevalence of risk factors, such as childhood malnutrition indicators, among the poor contribute to unequal distribution of childhood disease incidence and mortality, before any intervention (immunization or treatment) is even introduced. Second, large differences were observed in vaccine coverage across quintiles. In many countries, vaccine coverage among the top quintiles can be three-to-four times higher than among the bottom quintiles. Third, an unequal distribution of deaths was caused by the combination of unequal distribution of prognostic factors (thus an unequal mortality risk) and treatment coverage rates across quintiles. Our results suggest that the most appropriate strategy to remedy childhood disease inequities for each country would depend on country context, namely on how the baseline risks and current vaccine and treatment coverage are distributed across population subgroups.

The Global Vaccine Action Plan listed equity as a major principle in delivering universal access to immunization [7]; it emphasized “equitable access to immunization” as a core component of the right to health, and suggested that closing the coverage gap between the lowest and highest wealth quintiles would lead to greater equity. However, as shown in this study, merely ensuring equal access to vaccines will not reduce the health outcomes gap across wealth quintiles (see Scenario 3 in Figs. 2, 3, and 4). The poor face higher baseline risks, which are tied to social determinants of health such as wealth and education, in addition to lower treatment coverage rates. In our study, we find that Scenario 4 (distributing vaccines proportional to baseline risk) would often yield the most equitable distribution of disease burden, and that unequal distribution of vaccines that benefit the lowest wealth groups more would be required to achieve health equity.

This study makes an important contribution to the limited publications on the assessment of the distributional burden of childhood diseases and of the distributional impact of vaccines [20‐22]. It builds on previous work, extends the analysis to three childhood diseases and their corresponding vaccines, and is replicable to a large number of countries. Our methodology could, in fact, be extended in the future to examine the distribution of other diseases and the distributional impact of other health interventions, and across other population subgroups (e.g. region, sex).

One potential application of this study is to provide inputs to decision-makers on how to determine appropriate equity-enhancing strategies for countries. We presented estimates for several policy options, including providing equal vaccine coverage across quintiles, targeting the poor, and addressing the underlying risks before improving vaccine coverage. One key input required to determine strategies is costs. We do not have data on how much more it would cost to increase intervention coverage across different socioeconomic groups, nor do we know how much effort it would take to address the inequities in the baseline morbidity and mortality risks. Thus, while we can conclude from our study that certain scenarios would be more effective than others, data on costs would be required to determine which scenario would be most cost-effective and sustainable under countries’ budget constraints.

This study has several limitations. First, a key limitation lies in our simplification of disease progression. We assumed that whether one gets a disease solely depended on the prevalence and relative risk of risk factors and vaccine coverage. As described above, this paper does not involve dynamic transmission models, and therefore does not reflect the potential nonlinear effects of vaccines on disease transmissibility and herd immunity [11, 12]. We were only able to identify contact matrices between different age groups [23], but not between wealth quintiles. On a related note, we searched for risk factors related to crowding, household, or neighborhood density as an attempt to account for the size of susceptible populations. Households in the lower wealth quintiles are more likely to have more children, and therefore disease transmission rates in these quintiles may be higher than the higher wealth quintiles with fewer children. However, we were unable to find the relative risks of related risk factors and/or the prevalence of these risk factors by wealth quintile in DHS. Furthermore, non-specific effects of vaccines, such as their effect on overall mortality [24], as well as the timeliness of receiving the vaccines [25, 26], were not taken into account. We believe that including these factors would lead to a more skewed distribution of cases. Second, two other important modes of delivery for measles vaccination, measles second dose and supplementary immunization activities, were not included in the analysis due to lack of coverage data by quintile. Third, for some countries, the DHS did not have complete data on the prevalence of risk and prognostic factors by wealth quintile, so we assumed they were at the same levels as neighboring or similar countries. Fourth, when taking the PAF approach, it is important that the definition of exposure and the studied population sample match closely with the prevalence data. For simplicity, we selected input data sources for risk factors from highly cited systematic reviews [14, 27, 28], and they may not perfectly meet the definitions of the prevalence data from the DHS. Similarly, the results depend on the selected data sources such as inputs for vaccine efficacy and treatment effectiveness. Since, for simplicity, we assumed they were constant across quintiles, changing the data sources would not lead to large changes in the results. However, we acknowledge that there are other data sources that could be equally suitable for this study [29‐31]. We also reflected the uncertainties around these data sources by estimating uncertainty ranges for our results. However, we did not incorporate uncertainty around estimates of vaccine coverage, treatment coverage, and risk factor prevalence, due to the lack of immediate availability of uncertainty ranges from the original data sources. Finally, we were not able to validate the accuracy of the estimates since empirical data, especially for the null counterfactual scenario, were not available. One alternative for verification would be to collect disease-specific morbidity and mortality data at the subnational or national levels and examine the relationship between their respective wealth and socioeconomic levels. Our work points to these important data collection needs in the future.

Our findings contribute toward understanding how diseases and the benefits of health interventions might be distributed, specifically in relation to achieving Sustainable Development Goal 3 in ensuring essential health services are provided for all. Achieving equity in health outcomes will only occur in step-by-step processes, which is why this paper is important in illustrating the potential distributional results of different approaches. The outputs can provide decision-makers with information on the possible distributional impact of policies, and thereby help promote more equitable resource allocation, even when empirical data are unavailable. Furthermore, the pursuit of health equity requires more than ensuring equal access to one intervention, and rather a more systematic approach in addressing the health gaps between population subgroups.