Abstract Reducing health inequality is a major policy concern for low- and middle-income countries (LMICs) on the path to universal health coverage. However, health inequality impacts are rarely quantified in cost-effectiveness analyses of health programmes. Distributional cost-effectiveness analysis (DCEA) is a method developed to analyse the expected social distributions of costs and health benefits, and the potential trade-offs that may exist between maximising total health and reducing health inequality. This is the first paper to show how DCEA can be applied in LMICs. Using the introduction of rotavirus vaccination in Ethiopia as an illustrative example, we analyse a hypothetical re-designed vaccination programme, which invests additional resources into vaccine delivery in rural areas, and compare this with the standard programme currently implemented in Ethiopia. We show that the re-designed programme has an incremental cost-effectiveness ratio of US$69 per health-adjusted life year (HALY) compared with the standard programme. This is potentially cost-ineffective when compared with current estimates of health opportunity cost in Ethiopia. However, rural populations are typically less wealthy than urban populations and experience poorer lifetime health. Prioritising such populations can thus be seen as being equitable. We analyse the trade-off between cost-effectiveness and equity using the Atkinson inequality aversion parameter, ε, representing the decision maker’s strength of concern for reducing health inequality. We find that the more equitable programme would be considered worthwhile by a decision maker whose inequality concern is greater than ε = 5.66, which at current levels of health inequality in Ethiopia implies that health gains are weighted at least 3.86 times more highly in the poorest compared with the richest wealth quintile group. We explore the sensitivity of this conclusion to a range of assumptions and cost-per-HALY threshold values, to illustrate how DCEA can inform the thinking of decision makers and stakeholders about health equity trade-offs. Cost-effectiveness analysis, policy implementation, equity, health inequalities Key Messages Policy makers in low- and middle-income countries on the path to universal health coverage face hard choices about which services to prioritise and how to scale up delivery. These choices can involve equity trade-offs between improving total health and reducing social inequalities in health. We show how such equity trade-offs can be quantified, and how policy makers can use this information to make transparent decisions, using the example of rotavirus vaccination in Ethiopia. We do this using a new method called distributional cost-effectiveness analysis, which we adapt for use in low- and middle-income country settings. Introduction Tackling inequalities is a key feature of global health policy agendas and underpins the sustainable development goals and associated universal health coverage movement (Marmot et al. 2012; Ottersen et al. 2014; United Nations 2015). However, methods of cost-effectiveness analysis (CEA) used in mainstream healthcare decision making focus on the objective of maximising population health rather than reducing unfair health inequalities (Sassi et al. 2001; Weatherly et al. 2009; Johri and Norheim 2012). Recent methodological advances have enhanced CEA methods to enable them to go beyond the mythical ‘average’ citizen and consider the social distribution of costs and benefits (Cookson et al. 2017). Notable amongst these methodological developments are ‘extended’ cost-effectiveness analysis (ECEA), which provides breakdowns of costs and benefits by social groups (e.g. income, area of residence and gender), and ‘distributional’ cost-effectiveness analysis (DCEA) which, in addition, provides summary measures of health inequality impact and analyses potential trade-offs between increasing total health and reducing health inequality (Asaria et al., 2015; Verguet et al. 2015a). ECEA expands traditional CEA to examine the effects of an intervention on financial risk protection (safeguarding against financial hardship associated with paying for health services) as well as on health outcomes (Verguet et al. 2015a; Verguet et al. 2016). It breaks down the costs, health benefits and financial risk protection benefits by social groups (usually income groups). However, ECEA does not account for health opportunity costs of displaced expenditure within the health sector budget neither does it provide any guidance on how to resolve equity-efficiency trade-offs (Levin et al. 2015; Pecenka et al. 2015; Verguet et al. 2015b). DCEA provides an explicit framework for analysing the social distribution of health benefits and opportunity costs and the equity trade-offs that may arise between improving total health and reducing health inequality (Asaria et al. 2015). It examines the distribution of outcomes similarly to ECEA but goes further by enabling transparent analysis of health opportunity costs, summary measures of health inequality impact and equity trade-offs, with the aim of providing decision makers with a clearer understanding of the health inequality impacts and trade-offs and the implications of alternative social value judgements about equity (Asaria et al. 2016). Social factors, such as income, education, socioeconomic status can impact on the health of individuals over their lifetime and cause inequalities in lifetime health between social groups. DCEA facilitates the inclusion of impacts on overall inequality in lifetime health, rather than on inequality in health gains directly from the intervention, by accounting for the effects of an intervention on the distribution of lifetime health (Asaria et al. 2014). However, DCEA does not account for non-health benefits such as financial risk protection. To date, DCEA has primarily been applied to research in England (Asaria et al. 2015), while ECEA has been used extensively to explore the distribution of outcomes of health policies in low- and middle-income countries (LMICs) (Verguet and Jamison forthcoming); though some previous research in LMIC settings has used some components of DCEA (Johansson and Norheim 2011; Ngalesoni et al. 2016). In this article, we apply DCEA to an illustrative example of scaling up rotavirus vaccination in Ethiopia and show how the different components of DCEA can be applied in LMIC settings. In doing so we aim to demonstrate how to overcome the various challenges of applying these data intensive methods in relatively data sparse contexts and showcase some of the insights that such an analysis can provide to health policy makers in such contexts. Methods Publicly funded rotavirus vaccination Rotavirus is responsible for around a third of global diarrhoea-related deaths, the majority of which occur in LMICs (Liu et al. 2012; Tate et al. 2012). Before vaccination was introduced in 2013, Ethiopia had the fifth highest number of rotavirus related deaths (28 218 per year) worldwide. The vaccine has been found to be effective in many countries around the world, reducing severe rotavirus disease by > 60% in the first year of life (Armah et al. 2010; Madhi et al. 2010). The most recent estimate of vaccine coverage from Ethiopia Demographic and Health Survey (DHS) 2016 is 56%, however, coverage varies across groups (Central Statistical Agency (CSA) [Ethiopia] and ICF 2016). Since its introduction, coverage has been lowest amongst the poor in Ethiopia (Central Statistcal Agency [Ethiopia] 2014). As more of the population in poorer wealth quintile groups live in rural areas, by targeting these areas it may be possible to increase coverage amongst the poor (Central Statistcal Agency [Ethiopia] and Living Standards Measurement Study (LSMS) 2017). However, there are higher delivery costs in rural areas due to a range of logistical challenges. Comparing DCEA and ECEA To illustrate the value of DCEA, we built on a previous ECEA conducted before the introduction of the rotavirus vaccine in Ethiopia (Verguet et al. 2013). We began by evaluating the existing rotavirus vaccination programme from 2013 to 2016 (the ‘standard’ vaccination programme), compared to the prior situation of no vaccination in 2012, using the same assumptions as the original ECEA model, but updating the coverage assumptions to the levels actually achieved in different groups by 2016. We then evaluated a hypothetical alternative re-designed vaccination programme from 2013 to 2016 (the ‘pro-poor’ vaccination programme) that would have made proportionally more effort to deliver vaccination in rural areas. We incorporated health opportunity costs into our analyses to compute ‘net’ health gains by social group. We placed these health gains into the broader context of differences in lifetime health, and finally examined the trade-offs between cost-effectiveness and health equity. We focused on equity in terms of the distribution of health between population wealth quintiles; however, we note this is only one possible dimension by which equity may be considered. It is possible to conduct similar analysis by other dimensions of interest, such as, gender, age, ethnicity, geography, if the relevant data is available. Parameters Parameters used to evaluate each vaccination programme are presented in Table 1. Rotavirus death rate, relative risk of rotavirus mortality and vaccine effectiveness were taken from the original ECEA study (Verguet et al. 2013). All other parameters were updated to use more recent estimates or to represent the pro-poor hypothetical programme. Differential health risks reflected the fact that poorer children were more likely to contract rotavirus and more likely to die if they did. The vaccine was assumed to have equal effectiveness in all groups, instantaneous scale-up was assumed, with the primary health outcome being rotavirus diarrhoea deaths averted (Verguet et al. 2013). Table 1. Base case parameters used to model each scenario Parameters that are the same across vaccination programmes Source Population (annual cohort of live births) 2 800 000 Based on Central Statistical Agency [Ethiopia] (2014) Urban 467 622 Rural 2 332 378 Rotavirus death rate per 1000 live births 5.4 Based on Liu et al. (2012) and Tate et al. (2012) Relative risk ratio of rotavirus mortality Based on Rheingans et al. (2012) Ratio of poorest to richest quintile group 2.9 Risk index, poorest to richest quintile group 1.34, 1.23, 1.06, 0.91, 0.46 Vaccine effectiveness (%) (per 2-dose course) 49 Madhi et al. (2010) Vaccine price (per 2-dose course) $5.00 (Global Alliance for Vaccines and Immunization 2016) Vaccine price with GAVI subsidy (per 2-dose course) $0.40 Cost-effectiveness threshold (base case) $50 Based on Woods et al. (2016) Cost-effectiveness threshold range (sensitivity analyses) $10–$1857 Based on Woods et al. (2016) and Evans et al. (2005) Distribution of opportunity cost (proportion of cost borne by the poorest to richest quintile group) 0.185, 0.21, 0.21, 0.21, 0.185 Assumed Parameters that differ across vaccination programmes No vaccination Standard vaccination Pro-poor vaccination Source Incremental vaccination delivery cost (per 2-dose course) Based on Atherly et al. (2012), Wolfson (2008), Le Gargasson et al. (2015), Schütte et al. (2015) Urban $0 $0.50 $0.50 Rural $0 $0.50 $1.00 Vaccination coverage (%) (average proportion receiving the 2-dose course) 0 56 62 Central Statistical Agency (CSA) [Ethiopia] and ICF (2016)1; Hypothetical2 Within group vaccination coverage (%) (poorest to richest quintile group) 0 46, 49, 52, 63, 78 56, 56, 56, 63, 78 (Central Statistical Agency (CSA) [Ethiopia] and ICF (2016)1; Hypothetical2 Parameters that are the same across vaccination programmes Source Population (annual cohort of live births) 2 800 000 Based on Central Statistical Agency [Ethiopia] (2014) Urban 467 622 Rural 2 332 378 Rotavirus death rate per 1000 live births 5.4 Based on Liu et al. (2012) and Tate et al. (2012) Relative risk ratio of rotavirus mortality Based on Rheingans et al. (2012) Ratio of poorest to richest quintile group 2.9 Risk index, poorest to richest quintile group 1.34, 1.23, 1.06, 0.91, 0.46 Vaccine effectiveness (%) (per 2-dose course) 49 Madhi et al. (2010) Vaccine price (per 2-dose course) $5.00 (Global Alliance for Vaccines and Immunization 2016) Vaccine price with GAVI subsidy (per 2-dose course) $0.40 Cost-effectiveness threshold (base case) $50 Based on Woods et al. (2016) Cost-effectiveness threshold range (sensitivity analyses) $10–$1857 Based on Woods et al. (2016) and Evans et al. (2005) Distribution of opportunity cost (proportion of cost borne by the poorest to richest quintile group) 0.185, 0.21, 0.21, 0.21, 0.185 Assumed Parameters that differ across vaccination programmes No vaccination Standard vaccination Pro-poor vaccination Source Incremental vaccination delivery cost (per 2-dose course) Based on Atherly et al. (2012), Wolfson (2008), Le Gargasson et al. (2015), Schütte et al. (2015) Urban $0 $0.50 $0.50 Rural $0 $0.50 $1.00 Vaccination coverage (%) (average proportion receiving the 2-dose course) 0 56 62 Central Statistical Agency (CSA) [Ethiopia] and ICF (2016)1; Hypothetical2 Within group vaccination coverage (%) (poorest to richest quintile group) 0 46, 49, 52, 63, 78 56, 56, 56, 63, 78 (Central Statistical Agency (CSA) [Ethiopia] and ICF (2016)1; Hypothetical2 Parameter uncertainty ranges are not included as this is an illustrative example. 1 These references provide source information for the standard vaccination parameters. 2 These references provide source information for the pro-poor vaccination parameters. Costs were calculated as the vaccine cost plus the cost of delivering the vaccine and are presented in US dollars. These were applied to the population receiving the vaccine from an approximate annual birth cohort of 2 800 000, taking account of differential fertility rates by wealth quintile group (Central Statistical Agency [Ethiopia] 2014). The benefits of the vaccine were calculated as deaths averted in children under 5 years old, which were converted to health-adjusted life years (HALYs) (essentially the same as quality-adjusted life years, see Supplementary Appendix 1 for some of the methodological nuances). Consideration was given to how health opportunity costs, i.e. health forgone elsewhere in the system as a result of introducing the vaccine programme, were distributed among the population. The assumed distribution of the opportunity costs of reduced public healthcare expenditure is shown in Table 1. This inverted ‘U’ shape was based on the assumption that the poorest group had the lowest opportunity costs, as government health expenditure does not often reach them, while the richest group had similarly low opportunity costs as they were most likely to opt out of government funded healthcare preferring instead to purchase healthcare from private providers. Alternative assumptions were considered and are presented in Supplementary Appendix 3 along with their effect on results. The cost-effectiveness threshold (CET) was used to assess whether an intervention improved total health, after allowing for other forgone investments. There are a range of possible values that could be used to represent the CET for Ethiopia. For the base case analysis we made the assumption that the CET for Ethiopia was $50 per HALY. This CET was chosen for two reasons: first, it falls within the range recommended ($10–$255) in the analysis by Woods et al. (2016) and therefore represents a plausible empirical estimate of health opportunity cost in Ethiopia; second, it was chosen as it enabled us to clearly illustrate the trade-off between equity and efficiency in this example. It is notable, however, that the WHO recommended threshold range of 1–3 time GDP per capita is substantially higher ($619–$1875). Consequently, we assess the impact of alternative CETs within the range $10–$1857 in sensitivity analyses. The baseline distribution of lifetime health The baseline distribution of lifetime health describes how the overall burden of mortality and morbidity is distributed among different social groups within the general population. This allows consideration of the effects of an intervention on inequality in lifetime health, by comparing the distribution of lifetime health at baseline to that which would result following a new health intervention. Mortality and morbidity may vary by gender, age, ethnicity, wealth and many other factors. The factors that are used to estimate the distribution of lifetime health in a population will depend on the data available and the dimensions deemed relevant for the distributional evaluation undertaken. The baseline health distribution must be estimated for the general population, not just for those receiving the intervention. This is because policy concern for inequality encompasses inequality within the entire general population, and the opportunity costs of displaced resources will likely fall on members of the wider population including those who do not directly benefit from the intervention being evaluated (Asaria et al. 2016). In the absence of reliable census and vital statistics data in Ethiopia (as is also the case in many other low-income countries), we used indirect estimates of mortality differences by social strata drawing on WHO data on healthy life expectancy (HALE) and DHS data on morbidity and mortality by household asset wealth group in order to estimate the distribution of lifetime health. Population average HALE values, measured in HALYs, were taken from the WHO Life Expectancy database (Salomon et al. 2012; World Health Organisation 2015). These average HALE values were weighted by mortality based on previously modelled life expectancy by socioeconomic group, drawing on DHS data on child mortality by household asset group (Tranvag et al. 2013). Finally, the values were weighted by morbidity based on available prevalence data by socioeconomic group (GBD 2010; IHME 2015). The groups were then ordered from least to most healthy and population weighted to create the baseline distribution of HALE at birth. Detailed descriptions of these calculations are provided in Supplementary Appendix 1. Equity impact analysis Using the same modelling assumptions as Verguet et al. (2013), costs (see Table 1), calculated as the vaccine cost plus the delivery cost and health effects, calculated as deaths averted due to the vaccine, were estimated for each wealth quintile group (see Supplementary Appendix 2). Costs were assigned based on the proportion of each quintile group living in urban/rural areas (Central Statistical Agency [Ethiopia] 2014). Opportunity costs and health benefits were calculated in terms of HALYs. Deaths averted were multiplied by the HALE values for the corresponding quintile group from the baseline health distribution to estimate total HALYs gained in each group, from which per capita HALY gain was calculated. Assuming a constant population over time, the additional HALYs gained were added to the baseline HALE to give an estimation of the HALE distribution after the introduction of the vaccination programme. Using the net benefit approach, total costs of the vaccine programme were divided by an assumed base-case CET value of $50 per HALY to convert costs into health opportunity costs (alternative thresholds were assessed in sensitivity analyses). Net health benefits ( NHB) were thus calculated for each wealth quintile group (J) as: NHBJ=ΔHALYsJ-(Δ cos tsJ/CET) (1) Equity trade-off analysis To evaluate any trade-offs between improving total health and reducing inequality in health, we used an Atkinson social welfare function with relative inequality aversion parameter, ε, to represent the decision maker’s strength of concern for reducing health inequality. This parameter describes the decision maker’s willingness to make trade-offs between improving total health and reducing health inequality (Asaria et al. 2015). We focused on this single index of inequality aversion in order to keep the paper simple and concise. However, different social welfare functions can be used in DCEA to represent different kinds of inequality concern, for example, absolute health inequality, as illustrated by Asaria et al. (2015). The nature of the inequality concern should drive the selection of inequality indices and it may be useful to employ several such measures in sensitivity analyses. We used DCEA to compare the two alternative vaccination strategies in equity-cost-effectiveness space using the ‘health equity impact plane’ in which the effect on total health was quantified as in standard CEA, and the health inequality impact is quantified by the reduction in the Atkinson index offered by the ‘pro-poor’ vaccination programme compared with the ‘standard’ programme. To use the Atkinson index one must choose a value for the inequality aversion parameter. A value for the inequality aversion parameter (ε) equal to 0 indicates there is no aversion to inequality and a Utilitarian perspective is represented. As the value of ε increases, higher priority is given to transfers lower in the distribution (the worse off). An empirical study in England has estimated an Atkinson relative health inequality aversion parameter as 10.95 (Robson et al. 2016). This was used as a tentative reference point for the analysis. Extensive sensitivity analysis was conducted to explore the impact of using alternative values of the inequality aversion parameter. The equity-efficiency trade-off can be depicted by calculating the equally distributed equivalent (EDE) level of health that, if obtained by every individual, would enable a society to reach the same level of overall health as the current modelled distribution of health (Asaria et al. 2015). This allows alternative interventions or strategies to be compared for any given level of inequality aversion. The difference between a population’s mean level of health and the EDE for health indicates the average amount of health per person that society (or the decision maker) is willing to sacrifice to achieve an equal distribution of health for a given level of inequality aversion, conditional on the current inequality in the population health distribution (Asaria et al. 2015). We therefore plotted the EDE for each strategy at a range of levels of inequality aversion, ε, to show which option would be preferred taking account of the trade-off between overall health and health equity. We conducted equity trade-off analysis for the base case at a CET value of $50. In addition a range of alternative CET values were also used to explore the equity-efficiency trade-off in sensitivity analyses. Results The baseline distribution of lifetime health Figure 1 shows the estimated baseline distribution of HALE at birth in each wealth quintile group of the Ethiopian population compared with the life expectancy in each group. The HALE values are lower than the life expectancy values because of the adjustment for morbidity. This adjustment is greater in the lower wealth quintile groups due to higher levels of morbidity in these groups. Figure 1. View largeDownload slide Population distribution of health in Ethiopia: HALE at birth compared with life expectancy (life expectancy from Tranvag et al. 2013). Figure 1. View largeDownload slide Population distribution of health in Ethiopia: HALE at birth compared with life expectancy (life expectancy from Tranvag et al. 2013). Equity impact analysis Step by step calculation of the distribution of NHB resulting from the equity impact analysis is given in Table 2 and the resulting distribution is shown in Figure 2, for each vaccination programme compared to no vaccination. This shows that, compared with the standard programme, the pro-poor programme provides greater gains to the lowest wealth quintile groups at the expense of the higher wealth quintile groups. Table 2. Equity impact analysis Wealth quintile group Population births1 Population after rotavirus deaths (no vaccination) Cost to vaccinate group2 (US$) Deaths averted Standard Pro-poor Difference4 Standard Pro-poor Difference4 Q1 (poorest) 677 419 673 367 1 881 870 2 191 311 309 441 919 1112 193 Q2 643 548 639 829 1 899 755 2 210 809 311 054 897 1021 124 Q3 598 387 595 182 1 863 377 2 159 122 295 745 815 880 65 Q4 564 516 561 764 2 137 258 2 451 673 314 415 851 851 0 Q5 (richest) 316 129 314 738 1 481 381 1 570 513 89 132 532 532 0 Total 2 800 000 2 784 880 9 263 642 10 583 428 1 319 786 4014 4395 381 Wealth quintile group Population HALYs gained Health opportunity costs (HALYs)3 Net health effect (HALYs) Standard Pro-poor Difference4 Standard Pro-poor Difference4 Standard Pro-poor Difference4 Q1 (poorest) 43 642 52 786 9144 34 275 39 159 4884 9367 13 627 4260 Q2 45 647 51 956 6309 38 907 44 450 5543 6740 7506 766 Q3 44 920 48 469 3549 38 907 44 450 5543 6013 4019 −1994 Q4 46 115 46 115 0 38 907 44 450 5543 7207 1664 −5543 Q5 (richest) 32 065 32 065 0 34 275 39 159 4884 −2211 −7094 −4883 Total 212 389 231 390 19 001 185 273 211 669 26 396 27 116 19 722 −7394 Wealth quintile group Population births1 Population after rotavirus deaths (no vaccination) Cost to vaccinate group2 (US$) Deaths averted Standard Pro-poor Difference4 Standard Pro-poor Difference4 Q1 (poorest) 677 419 673 367 1 881 870 2 191 311 309 441 919 1112 193 Q2 643 548 639 829 1 899 755 2 210 809 311 054 897 1021 124 Q3 598 387 595 182 1 863 377 2 159 122 295 745 815 880 65 Q4 564 516 561 764 2 137 258 2 451 673 314 415 851 851 0 Q5 (richest) 316 129 314 738 1 481 381 1 570 513 89 132 532 532 0 Total 2 800 000 2 784 880 9 263 642 10 583 428 1 319 786 4014 4395 381 Wealth quintile group Population HALYs gained Health opportunity costs (HALYs)3 Net health effect (HALYs) Standard Pro-poor Difference4 Standard Pro-poor Difference4 Standard Pro-poor Difference4 Q1 (poorest) 43 642 52 786 9144 34 275 39 159 4884 9367 13 627 4260 Q2 45 647 51 956 6309 38 907 44 450 5543 6740 7506 766 Q3 44 920 48 469 3549 38 907 44 450 5543 6013 4019 −1994 Q4 46 115 46 115 0 38 907 44 450 5543 7207 1664 −5543 Q5 (richest) 32 065 32 065 0 34 275 39 159 4884 −2211 −7094 −4883 Total 212 389 231 390 19 001 185 273 211 669 26 396 27 116 19 722 −7394 The rich gain the least even with lower opportunity cost because they are the least at risk. 1 Based on fertility rate by quintile (DHS Ethiopia 2011). 2 Based on within group coverage, proportion in urban/rural areas and associated cost. 3 Total opportunity cost is calculated as the total cost (from column 3 of the top half of the table) and a cost per HALY of $50 (Woods et al. 2016). Opportunity costs for each group are calculated as the total opportunity cost multiplied by the fraction of opportunity cost that falls to each quintile, assuming an inverted ‘U’ shaped distribution of opportunity cost (the distribution of opportunity cost is reported in Table 1). 4 Difference = pro-poor − standard. Figure 2. View largeDownload slide Net health effect of each programme compared with no vaccination (λ = $50). Figure 2. View largeDownload slide Net health effect of each programme compared with no vaccination (λ = $50). Equity trade-off analysis The cost-effectiveness plane, Panel a of Figure 3, shows the pro-poor vaccine compared to the standard vaccine as would be presented in standard CEA. The pro-poor vaccine is not cost-effective at a threshold of $50 per HALY as compared to the standard vaccination programme. However, using the ‘health equity impact plane’ shown in Panel b of Figure 3 we compare the two strategies in equity-cost-effectiveness space. The pro-poor vaccine falls in the south-east ‘lose-win’ quadrant (CET = $50), demonstrating that relative to the standard vaccination programme it has a positive impact on health equity despite its negative impact on total health. Thus, a trade-off occurs between improving total health and reducing socioeconomic inequality in health. Figure 3 also presents the equity trade-off analysis at the four CET values selected for the sensitivity analyses to show how the choice of threshold impacts on the results. Figure 3. View largeDownload slide Incremental analysis of pro-poor vaccine compared with standard vaccine: cost-effectiveness plane vs. health equity impact plane. Panel (a) plots the gross health effect; Panel (b) plots the net health effect (taking account of opportunity cost). The Atkinson index of inequality is normally scaled from 0 to 1, where 1 represents full inequality, but we have reversed the scale so that a positive impact represents improved equity. Figure 3. View largeDownload slide Incremental analysis of pro-poor vaccine compared with standard vaccine: cost-effectiveness plane vs. health equity impact plane. Panel (a) plots the gross health effect; Panel (b) plots the net health effect (taking account of opportunity cost). The Atkinson index of inequality is normally scaled from 0 to 1, where 1 represents full inequality, but we have reversed the scale so that a positive impact represents improved equity. The difference in the EDE health of the two vaccination strategies at a range of levels of relative inequality aversion, ε, are depicted in Figure 4. If inequality aversion is zero then the EDE is equal to the mean health and, at a CET of $50, the standard programme would provide 7395 more population HALYs than the pro-poor programme. For inequality aversion parameters greater than ε = 5.66, the point at which the line crosses the x-axis, the pro-poor programme is preferred, implying that the decision maker is willing to sacrifice the additional 7395 HALYs in pursuit of lower health inequality. At this level of inequality aversion this would require that health gains in the poorest wealth quintile group were weighted at least 3.86 times more highly than health gains in the richest. No research exists on the level of inequality aversion in Ethiopia. The appropriate level of this parameter is a matter for decision makers in Ethiopia, and we acknowledge both that English views are not directly relevant to Ethiopia and that there is uncertainty and potential for bias in the findings for England. However, if there is similar inequality aversion in Ethiopia as has been estimated in England, the concern for inequality is large enough to choose the pro-poor vaccination programme at the CET of $50. Figure 4. View largeDownload slide Equity trade-off analysis. (1) The health inequality aversion (x-axis) represents the strength of concern for reducing health inequality rather than improving total health. (2) At the point the line crosses the x-axis: to the left it is better to implement the standard programme, to the right it is better to implement the pro-poor programme. Figure 4. View largeDownload slide Equity trade-off analysis. (1) The health inequality aversion (x-axis) represents the strength of concern for reducing health inequality rather than improving total health. (2) At the point the line crosses the x-axis: to the left it is better to implement the standard programme, to the right it is better to implement the pro-poor programme. The equity trade-off analysis at the four CET values selected for the sensitivity analyses are also depicted in Figure 4. At a CET of $10, the line lies above the x-axis and never crosses it meaning that no matter how great the inequality aversion is, the pro-poor programme would never be preferred because at this threshold the net health benefits are large enough that they are valued more highly than the gains in equity. For CETs of $255, $619 or $1857, the opposite is true. These lines lie below the x-axis and never cross it meaning that no matter how small the inequality aversion the pro-poor programme will always be preferred because at these thresholds the net health benefits are small enough in relation to the equity benefits that the gains in equity are valued more highly than the net health benefits. For the comparison between the standard- and the pro-poor-vaccine programme, a trade-off between equity and efficiency exists for thresholds between $27.95 and $69.45. Discussion In this work, we illustrated the use of DCEA in LMIC settings, using the example of alternative designs of the rotavirus vaccination programme in Ethiopia and focusing on the added value and insights that may be obtained from the methods of DCEA and ECEA in designing and evaluating such programmes. The results of the DCEA study confirmed those of the previously conducted ECEA study (Verguet et al. 2013), that the standard vaccination programme introduced in Ethiopia was cost-effective as compared with the prior situation of no vaccination coverage. The analysis also confirmed that the vaccination programme delivered greater health benefits to poorer groups despite introducing inequality in coverage levels favouring wealthier social groups and urban areas. We built on this analysis to examine an alternative design to the vaccination programme to address the social inequality in coverage. DCEA analysis of this re-designed programme found that such a programme would not be cost-effective relative to the standard programme but would reduce both inequality in coverage, and in health. By analysing the trade-off between cost-effectiveness and equity we found that the more equitable, pro-poor, programme would be preferred by a policy maker whose degree of inequality concern was greater than ε = 5.66. At current levels of health inequality in Ethiopia, this implied that additional health gains in the poorest wealth quintile group would need to be weighted approximately 4 times as highly as health gains in the richest for the re-designed programme to be preferred. Unlike ECEA, the DCEA methodology focused solely on health impacts and did not capture financial risk protection benefits or other non-health benefits related to household wellbeing. Our study has a number of limitations. We made simplifying assumptions around issues such as: cost savings from future treatments, morbidity benefits, herd effects and transmission effects. These all will result in the gains form the vaccine being underestimated. However, age at death was not incorporated into the analysis, instead, overall HALE was used to account for death averted which will result in the gains from the vaccine being overestimated. The lack of a vital registration system for adult mortality in Ethiopia meant that we had to instead rely on the use of child health data for modelling health. Whilst rotavirus diarrhoea mainly affects children under 5 years old, the lack of adult data may be a more significant limitation for evaluations of interventions for diseases that affect older age groups. Additionally, to account for morbidity in the baseline health distribution, we used prevalence data for only three diseases, though these were chosen based on their contribution to the burden of disease in Ethiopia (GBD 2010; IHME 2015). Another issue requiring value judgement is the choice of health outcome metric, and how far this may indirectly discriminate against disadvantaged groups. There is a large ethics and economics literature on how the HALY and other outcome metrics commonly used in health economic evaluation may implicitly discriminate against preventing mortality among relatively unhealthy population groups such as the poor, the elderly and the disabled (Edlin et al. 2013). This indirect discrimination occurs because relatively few years of healthy life are gained from averting the death of a relatively unhealthy individual. It is important for producers and users of health economic evaluation to consider this ethical issue, and how far it may or may not be relevant to the case in hand. Where the issue is considered relevant by stakeholders, analysts can address it using sensitivity analysis based on simple binary outcome metrics, such as mortality or cases of disease averted, which do not indirectly discriminate in this way although they provide an incomplete and potentially misleading picture of the overall health gains in other ways, as documented in the standard health economic evaluation literature (Drummond et al. 2015). The main strengths of DCEA are that it provides a summary measure of health inequality impact and allows explicit analysis of the trade-offs that can occur between maximising total health and reducing unfair inequality in health. There may be cases where such trade-offs do not occur, for example, if policies fall in the ‘win-win’ quadrant of the equity-impact plane, improving both total health and reducing health inequality. However, it will often be the case that by paying closer attention to additional delivery costs to reach the most disadvantaged groups and accounting for these implementation costs at the analysis stage will lead to more equitable policies being pursued. DCEA serves a valuable purpose, particularly for LMIC contexts, as it facilitates the consideration of fairer options so that the consequences of alternative uses of very scarce resources can be explicitly quantified in terms of both equity and cost-effectiveness. Through the analysis of trade-offs, DCEA can help with the assessment of policy alternatives so that informed decisions can be made when considering policy implementation. Additionally, use of equity parameters can help with the development of equity ‘benchmarks’ for policy makers to compare across different decisions and policy choices. This can be particularly useful as a way to understand the impact of policy decisions and facilitates transparency and consistency of decision making. A number of future research directions would be of value in this area. More research on the distribution of opportunity costs and also on levels of health inequality aversion would be of value. This is true for both high-income countries and LMICs. While there has been research in this area conducted for England (Robson et al. 2016), research of this kind repeated in other settings would inform further research and analyses using DCEA in other global contexts. Finally, as has been noted elsewhere (Grimm et al. 2010), data availability is often a fundamental limitation to such analyses both in terms of vital statistics collection but also in terms of variation in vital statistics and delivery costs by equity-relevant parameters. Conclusion Policy makers in LMICs face difficult choices about which services to cover and how to scale up on the path to universal health coverage. This article has outlined how DCEA can be used to extend current methods of economic evaluation of healthcare in LMICs so that trade-offs between improving total health and reducing health inequalities can be quantified and analysed in order to inform the decision making process. Acknowledgements For helpful comments to ensure this article is accessible to a wide audience the authors would like to thank Claire Hulme. The authors would also like to thank Matthew Robson for his assistance with the equity weight calculations. Richard Cookson is supported by the National Institute for Health Research (Senior Research Fellowship, Dr Richard Cookson, SRF-2013-06-015). The views expressed in this publication are those of the authors and not necessarily those of the NHS, the National Institute for Health Research or the Department of Health. Supplementary Data Supplementary data are available at Health Policy and Planning online. Conflict of interest statement. None declared. References Armah GE, Sow SO, Breiman RF et al. 2010. Efficacy of pentavalent rotavirus vaccine against severe rotavirus gastroenteritis in infants in developing countries in sub-Saharan Africa: a randomised, double-blind, placebo-controlled trial. Lancet (London, England) 376: 606– 14. Google Scholar CrossRef Search ADS PubMed Asaria M, Cookson R, Griffin S. 2014. Incorporating health inequality impacts into cost-effectiveness analysis. In: Culyer AJ (ed). Encyclopaedia of Health Economics . San Diego: Elsevier, 22–6. Asaria M, Griffin S, Cookson R. 2016. Distributional cost-effectiveness analysis: a tutorial. Medical Decision Making: An International Journal of the Society for Medical Decision Making 36: 8– 19. Google Scholar CrossRef Search ADS PubMed Asaria M, Griffin S, Cookson R, Whyte S, Tappenden P. 2015. Distributional cost-effectiveness analysis of health care programmes—a methodological case study of the UK Bowel Cancer Screening Programme. Health Economics 24: 742– 54. Google Scholar CrossRef Search ADS PubMed Atherly DE et al. 2012. Projected health and economic impact of rotavirus vaccination in GAVI-eligible countries: 2011–2030. Vaccine 30(Suppl. 1): A7– 14. Central Statistical Agency [ Ethiopia] 2014. Ethiopia Mini Demographic and Health Survey 2014. Addis Abba, Ethiopia. Central Statistcal Agency [Ethiopia] and Living Standards Measurement Study (LSMS), W. B. 2017. LSMS-Integrated Surveys on Agriculture, Ethiopia Socioeconomic Survey (ESS) 2015/2016. Central Statistical Agency (CSA) [ Ethiopia] and ICF 2016. Demographic and Helath Survey 2016: Key Indicators Report. Addis Ababa, Ethiopia, and Rockville, Maryland, USA: World Bank. Cookson R, Mirelman AJ, Griffin S et al. 2017. Using cost-effectiveness analysis to address health equity concerns. Value in Health: The Journal of the International Society for Pharmacoeconomics and Outcomes Research 20: 206– 12. Google Scholar CrossRef Search ADS PubMed Drummond MF, Sculpher MJ, Claxton K, Stoddart GL, Torrance GW. 2015. Methods for the Economic Evaluation of Health Care Programmes . Oxford: Oxford University Press. Edlin R, Mccabe C, Round J et al. 2013. Understanding Harris’ understanding of CEA: is cost effective resource allocation undone? Journal of Health Services Research & Policy 18: 34– 9. Google Scholar CrossRef Search ADS PubMed Evans DB, Lim SS, Adam T, Edejer TT-T. 2005. Evaluation of current strategies and future priorities for improving health in developing countries. BMJ 331: 1457– 61. Google Scholar CrossRef Search ADS PubMed Global Alliance for Vaccines and Immunization. 2016. Available from: www.gavialliance.org. Global Burden of Disease (GBD). 2010. GBD Profile: Ethiopia [Online]. http://www.healthdata.org/sites/default/files/files/country_profiles/GBD/ihme_gbd_country_report_ethiopia.pdf, (accessed 22 July 2015). Grimm M, Harttgen K, Klasen S et al. 2010. Inequality in human development: an empirical assessment of 32 countries. Social Indicators Research 97: 191– 211. Google Scholar CrossRef Search ADS PubMed Institute for Health Metrics and Evaluation (IHME). 2015. Ethiopia [Online]. http://www.healthdata.org/ethiopia, (accessed 22 July 2015). Johansson KA, Norheim OF. 2011. Problems with prioritization: exploring ethical solutions to inequalities in HIV care. The American journal of bioethics: AJOB 11: 32– 40. Google Scholar CrossRef Search ADS PubMed Johri M, Norheim O. 2012. Can cost-effectiveness analysis integrate concerns for equity? Systematic review. International Journal of Technology Assessment in Health Care 28: 125– 32. Google Scholar CrossRef Search ADS PubMed Le Gargasson JB, Nyonator FK, Adibo M et al. 2015. Costs of routine immunization and the introduction of new and underutilized vaccines in Ghana. Vaccine 33: A40– 6. Google Scholar CrossRef Search ADS PubMed Levin CE, Sharma M, Olson Z et al. 2015. An extended cost-effectiveness analysis of publicly financed HPV vaccination to prevent cervical cancer in China. Vaccine 33: 2830– 41. Google Scholar CrossRef Search ADS PubMed Liu L, Johnson HL, Cousens S et al. Child Health Epidemiology Reference Group of WHO and UNICEF. 2012. Global, regional, and national causes of child mortality: an updated systematic analysis for 2010 with time trends since 2000. Lancet (London, England) 379: 2151– 61. Google Scholar CrossRef Search ADS PubMed Madhi SA, Cunliffe NA, Steele D et al. 2010. Effect of human rotavirus vaccine on severe diarrhea in African infants. The New England Journal of Medicine 362: 289– 98. Google Scholar CrossRef Search ADS PubMed Marmot M, Allen J, Bell R, Bloomer E, Goldblatt P. 2012. WHO European review of social determinants of health and the health divide. The Lancet 380: 1011– 29. Google Scholar CrossRef Search ADS Ngalesoni FN, Ruhago GM, Mori AT, Robberstad B, Norheim OF. 2016. Equity impact analysis of medical approaches to cardiovascular diseases prevention in Tanzania. Social Science & Medicine (1982) 170: 208– 17. Google Scholar CrossRef Search ADS PubMed Ottersen T, Norheim OF; World Health Organization Consultative Group on Equity and Universal Health Coverage. 2014. Making fair choices on the path to universal health coverage. Bulletin of the World Health Organization 92: 389. Google Scholar CrossRef Search ADS PubMed Pecenka CJ, Johansson KA, Memirie ST, Jamison DT, Verguet S. 2015. Health gains and financial risk protection: an extended cost-effectiveness analysis of treatment and prevention of diarrhoea in Ethiopia. BMJ Open 5: e006402. Google Scholar CrossRef Search ADS PubMed Rheingans R, Atherly, D, Anderson, J. 2012. Distributional impact of rotavirus vaccination in 25 GAVI countries: estimating disparities in benefits and cost-effectiveness. Vaccines 30(Suppl. 1): A15– 23. Robson M, Asaria M, Tsuchiya A, Ali S, Cookson RA. 2016. ‘ Eliciting the Level of Health Inequality Aversion in England’ CHE Research Paper, no. 125 . Centre for Health Economics, University of York: York, UK, 1–19. Salomon JA, Wang H, Freeman MK et al. 2012. Healthy life expectancy for 187 countries, 1990-2010: a systematic analysis for the Global Burden Disease Study 2010. Lancet (London, England) 380: 2144– 62. Google Scholar CrossRef Search ADS PubMed Sassi F, Archard L, Le Grand J. 2001. Equity and the economic evaluation of healthcare. Health Technology Assessment 5: 1– 138. Google Scholar CrossRef Search ADS PubMed Schütte C et al. 2015. Cost analysis of routine immunisation in Zambia. Vaccine 20 33(Suppl. 1): A47– A52. Tate JE, Burton AH, Boschi-Pinto C et al. 2012. 2008 estimate of worldwide rotavirus-associated mortality in children younger than 5 years before the introduction of universal rotavirus vaccination programmes: a systematic review and meta-analysis. The Lancet. Infectious Diseases 12: 136– 41. Google Scholar CrossRef Search ADS PubMed Tranvag EJ, Ali M, Norheim OF. 2013. Health inequalities in Ethiopia: modeling inequalities in length of life within and between population groups. International Journal for Equity in Health 12: 52. Google Scholar CrossRef Search ADS PubMed United Nations 2015. The Millennium Development Goals Report 2015 . New York: United Nations. Verguet S, Jamison DT. forthcoming. Applications of extended cost-effectiveness analysis (ECEA) methodology in DCP3. In: Jamison DT, Nugent R, Gelband H, Horton S, Jha P, Laxminarayan R (eds). Disease Control Priorities . Washington, DC: World Bank. PubMed PubMed Verguet S, Kim JJ, Jamison DT. 2016. Extended cost-effectiveness analysis for health policy assessment: a tutorial. Pharmacoeconomics 34: 913– 23. Google Scholar CrossRef Search ADS PubMed Verguet S, Laxminarayan R, Jamison DT. 2015a. Universal public finance of tuberculosis treatment in India: an extended cost-effectiveness analysis. Health Economics 24: 318– 32. Google Scholar CrossRef Search ADS Verguet S, Murphy S, Anderson B et al. 2013. Public finance of rotavirus vaccination in India and Ethiopia: an extended cost-effectiveness analysis. Vaccine 31: 4902– 10. Google Scholar CrossRef Search ADS PubMed Verguet S, Olson ZD, Babigumira JB et al. 2015b. Health gains and financial risk protection afforded by public financing of selected interventions in Ethiopia: an extended cost-effectiveness analysis. The Lancet. Global Health 3: e288– 96. Google Scholar CrossRef Search ADS Weatherly H, Drummond M, Claxton K et al. 2009. Methods for assessing the cost-effectiveness of public health interventions: key challenges and recommendations. Health Policy 93: 85– 92. Google Scholar CrossRef Search ADS PubMed Wolfson L. 2008. WHO Immunization Coverage Estimates and Trajectories (WHO ICE-T) . Geneva: World Health Organization, Department of Immunization, Vaccines, and Biologicals. Woods B, Revill P, Sculpher M, Claxton K. 2016. Country-level cost-effectiveness thresholds: initial estimates and the need for further research. Value in Health 19: 929– 35. Google Scholar CrossRef Search ADS PubMed World Health Organisation 2015. Life Expectancy Data by Country, Global Health Observatory data repository. http://www.who.int/gho/mortality_burden_disease/life_tables/en/. © The Author(s) 2018. Published by Oxford University Press in association with The London School of Hygiene and Tropical Medicine. All rights reserved. For permissions, please e-mail: email@example.com
Health Policy and Planning – Oxford University Press
Published: Apr 1, 2018
It’s your single place to instantly
discover and read the research
that matters to you.
Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.
All for just $49/month
Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly
Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.
Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.
Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.
All the latest content is available, no embargo periods.
“Hi guys, I cannot tell you how much I love this resource. Incredible. I really believe you've hit the nail on the head with this site in regards to solving the research-purchase issue.”Daniel C.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud
“I must say, @deepdyve is a fabulous solution to the independent researcher's problem of #access to #information.”@deepthiw
“My last article couldn't be possible without the platform @deepdyve that makes journal papers cheaper.”@JoseServera