Optimizing community case management strategies to achieve equitable reduction of childhood pneumonia mortality: An application of Equitable Impact Sensitive Tool (EQUIST) in five low– and middle–income countries

Background The aim of this study was to populate the Equitable Impact Sensitive Tool (EQUIST) framework with all necessary data and conduct the first implementation of EQUIST in studying cost–effectiveness of community case management of childhood pneumonia in 5 low– and middle–income countries with relation to equity impact. Methods Wealth quintile–specific data were gathered or modelled for all contributory determinants of the EQUIST framework, namely: under–five mortality rate, cost of intervention, intervention effectiveness, current coverage of intervention and relative disease distribution. These were then combined statistically to calculate the final outcome of the EQUIST model for community case management of childhood pneumonia: US$ per life saved, in several different approaches to scaling–up. Results The current ‘mainstream’ approach to scaling–up of interventions is never the most cost–effective. Community–case management appears to strongly support an ‘equity–promoting’ approach to scaling–up, displaying the highest levels of cost–effectiveness in interventions targeted at the poorest quintile of each study country, although absolute cost differences vary by context. Conclusions The relationship between cost–effectiveness and equity impact is complex, with many determinants to consider. One important way to increase intervention cost–effectiveness in poorer quintiles is to improve the efficiency and quality of delivery. More data are needed in all areas to increase the accuracy of EQUIST–based estimates.

The case fatality rates (CFR) which were used to adjust quintile costs for number of episodes of pneumonia was modelled by calculating country-specific pneumonia case fatality rates from existing data 11 and then graphing this against the known U5MR for each country from SOWC 2009 and subsequently splitting the data by region, resulting in a graph with a trendline for each region. 37 A linear trendline was applied throughout. Although one of the regional graphs showed a slightly higher r 2 value with an exponential trendline, this was marginal and it was assumed that as the worldwide trendline was robustly linear, this higher r 2 finding was an artefact of poor data rather than reflecting an actual difference. The equation of the trendline was used to calculate the CFR for each quintile in each country from the quintile-specific U5MR.

Limitations
Cost modelling presents a potential issue for the robustness of this implementation of EQUIST overall as very little data were known and assumptions were made (see above).

B. Estimates of current coverage by the three interventions by wealth quintiles in five countries
For Nigeria Egypt, Cambodia and Peru the available DHS data were quintile specific and were therefore used directly. For Bangladesh however only overall data was reported.The distribution between the quintiles was performed according to relative number of pneumonia deaths in each quintile, as it was assumed that the CCM coverage percentages for each quintile would follow the quintile's pneumonia mortality level (i.e. if a quintile had lower pneumonia mortality it was then assumed to have higher CCM coverage level).

Limitations:
For all five countries estimates of U5MR and CCM coverage by quintile were obtained from DHS reports, which although extensive have been highlighted previously as potentially biased, with findings of significant under-reporting of indicators in poorer parts of the population. 36 If this is the case it might mean that this paper's estimates of cost-effectiveness for poorer quintiles are overly high. DHS data is also potentially flawed due to relatively small sample sizes. For example, the coverage data for CCM taken from DHS country reports was based on only 805 cases of suspected pneumonia in Egypt and only 690 in Nigeria. 15,16 It is suggested that future DHS should endeavour to extend surveys to a greater percentage of the population, thereby decreasing the likelihood of poor or skewed data.

C. Effectiveness Estimates
For CCM, linear, polynomial and exponential trendlines were tested on the graph described in the effectiveness step of the methods section. The trendline with the highest R 2 value (i.e., the one which statistically had the greatest predictive value) was applied. The equation of this line was used to calculate the effectiveness of CCM in each quintile for the exemplar countries depending on their U5MR. This was then revised upwards as described above.
Limitations: While overall effectiveness data for each intervention was known from Theodoratou et al's reviews 11, 12 , they were again completely unreported split by wealth quintile in any setting and so it was necessary to model this using expert opinion 32 . Although this is likely to give a valid estimate, significantly more data is needed in the area to confirm this.

D. Disease Proportion Estimates
This was the most complex modelling process undertaken in this study. Firstly, the distributions of under-5 mortality causes by country were established. These data were taken from the most recent CHERG report 11 . The U5MR for each country from the CHERG report was added using data from the UNICEF SOWC 2009 report 22 . Next for each country, the relative proportion of mortality from each cause of death in the CHERG report was calculated by dividing the number of deaths from each cause of death by the total under-5 deaths for each country.
Using the WHO regional groupings from the CHERG report 13 , regional graphs were made of relative disease proportion by U5MR for 9 causes of death: AIDS, diarrhoea, malaria, pneumonia, preterm birth complications, birth asphyxia, neonatal sepsis, congenital abnormalities and injury. This is a smaller number than the number of causes listed in the CHERG report but it was opted to move the categories 'Tetanus', 'Measles', 'Meningitis', 'Pertussis', 'Other Infections' and Other Non-Communicable Diseases' into a broad category of 'Other' for the sake of clearer analysis and model stability.
Trendlines were then fitted to each of these graphs. Exponential, linear and logarithmic trendlines were tested on each disease graph and the one with the greatest r 2 value was applied. For birth asphyxia it was found that a 2nd order polynomial trendline exhibited a significantly higher r 2 value than the other models in the regions of Eastern Mediterranean and Southest Asia and so it was decided to apply this to all regions apart from Africa which exhibited a higher r 2 value with a logarithmic trendline. This difference is thought to be potentially due to the absence of a secondary healthcare effect in Africa.
The next step was to use the equations of the trendlines to model relative disease proportions by quintile for this study's five exemplar countries. Individual disease proportions by wealth quintile were modelled by applying the trendline equations (see above) for each exemplar country's WHO region to the quintile-specific U5MR data described above. The relative proportion of deaths from causes categorised as 'other' for each quintile was calculated by taking the sum of the percentages calculated for the 9 specific causes and subtracting this from 100 (i.e. 100% of mortality). U5MR data were not available for Niue in the Western Pacific region and so this country was removed from the model. These country specific disease proportion estimates for each of the 5 different wealth quintiles were tabulated and graphed and then applied to the total number of under 5 deaths per year (which was calculated by applying the U5MR for each country to the number of live births for each country, both of which were taken from UNICEF SOWC 2009 23 ) to calculate the number of under 5 deaths from each disease in each country in this study's target year, split by wealth quintile.

E. Final Model
The number of deaths that would be saved through each of the approaches to scaling-up an intervention was calculated by multiplying the under 5 pneumonia deaths in 2008 for the particular quintile (as calculated above) by the percentage effectiveness of the intervention estimated in that quintile. This was then brought together with the estimated cost for scalingup by 10% in each approach to give a measure of cost-effectiveness (cost per life saved (in US$)): the final outcome of the EQUIST.
To model the "mainstream" approach, it was assumed that coverage scale-up would continue to follow current quintile-specific relative distribution and so used the model to split the 10% increase according to current relative distribution between quintiles. Which quintile the next wealthiest 10% scale-up would be in was established from coverage estimates. Calculating the cost of scaling-up by 10% for each approach was done by multiplying cost per quintile scaleup data for the necessary quintile by 0.5 to show how much it would cost to scale-up by 10% (1/2 a quintile). If scaling-up an intervention to the next uncovered 10%, the middle uncovered 10% or through the "mainstream approach" would involve scaling-up parts of more than one quintile then two individual cost calculations were completed for the fraction of the quintiles that would be covered and then the sum of these was calculated.