Estimates of Child Deaths Prevented From Malaria Prevention
Estimates of Child Deaths Prevented From Malaria Prevention
This analysis, using the LiST model, focused on estimating the impact of ITNs for preventing post-neonatal (one-59 months of age) child malaria deaths from 2001–2010, as compared to a baseline of 2000. A baseline of 2000 was chosen as this is the year prior to any scale-up of ITNs or malaria prevention in pregnancy interventions in most African countries. Forty-three countries in sub-Saharan Africa are included in this analysis; 43 for the ITN analysis and 32 for the malaria prevention in pregnancy analysis (Figure 1).
(Enlarge Image)
Figure 1.
Malaria-endemic countries included in estimates of malaria intervention scale-up and resultant malaria-caused child deaths prevented 2001–2010.
This analysis of malaria prevention during pregnancy (with either IPTp or ITNs) for preventing all-cause < 5 child deaths through prevention of low birth-weight (LBW) was limited to 32 countries in Africa with stable malaria transmission. These countries accounted for 88% of the population in SSA at risk of malaria and 90% of the malaria-caused child deaths in 2000 in SSA. A list of excluded countries from the analyses, as well as the rationale for exclusion, can be found in Additional file 1.
The LiST model is based on the earlier work on effectiveness of interventions developed by WHO and UNICEF's Child Health Epidemiology Reference Group (CHERG) and RBM's Monitoring and Evaluation Reference Group (MERG). The model starts with estimates and assumptions within each country on the profiles for population and population growth, under-five cause of death structure, disease-specific mortality rates, and current coverage estimates of key child survival interventions. The model used in this analysis and accompanying documentation can be downloaded.
The model estimates child deaths prevented (within specific cause of death categories) due to intervention scale-up within a specified country as a function of three primary parameters, as outlined in details below: 1) the number of child deaths by cause projected to occur in each year (including population growth parameters over time); 2) the protective effect (PE) on cause-specific mortality (PE = 1-relative risk*100) for each intervention being scaled-up; and 3) increases in population coverage of each intervention. The model computes the number of deaths prevented by cause each year, accounting for population growth, as the difference between the estimated deaths that occur with intervention scale-up and the estimated deaths that would have occurred without intervention scale-up beyond the coverage at a baseline year. Further details of the LiST model estimation methods are presented in the Additional file 1.
Within the LiST model, the total number of < 5 child deaths for the baseline year of 2000, by age, is based on estimates of < 5 mortality produced by the UN Inter-agency Group for Child Mortality Estimation (IGME). The number of post-neonatal child malaria deaths in our baseline year of 2000 was estimated as the proportion of child deaths one-59 months attributable to malaria, multiplied by the total number of all-cause < 5 child deaths one-59 months in the year 2000. Based on this method, the LiST model started with 870,928 malaria-caused deaths one-59 months in 43 African countries and 1,034,828 all-cause neonatal deaths in 32 African countries in the baseline year of 2000 (Additional file 2).
As described in detail elsewhere, the PE of ITNs for preventing post-neonatal child malaria deaths has been estimated to be 55% (range 49–60%) based on a systematic review and meta-analysis of three trials.
Malaria in pregnancy increases the risk of LBW, resulting primarily from intrauterine growth retardation (IUGR) in areas of stable P. falciparum malaria transmission, while LBW has been demonstrated to be a significant risk factors for neonatal and infant mortality. The PE of malaria prevention during pregnancy for preventing LBW has been estimated to be 35% (95% confidence interval [CI] 23–45%) during the first two pregnancies in malaria endemic areas based on a systematic review of 2 ITN trials and 3 IPTp trials conducted in 1998 and 2002 in Kenya and 2004 in Mozambique, after the onset of sulphadoxine-pyrimethamine (SP) resistance. The effect of malaria prevention interventions during pregnancy on LBW in the LiST model acts solely through intrauterine growth retardation (IUGR). IUGR in the LiST model acts mostly through neonatal mortality, increasing the risk of dying during this period due to diarrhoea [RR = 2.0], sepsis/pneumonia (RR = 2.0), and asphyxia (RR = 2.3). During the post-neonatal period, IUGR slightly increases the risk of dying due to measles, malaria, diarrhoea and pneumonia, via links between IUGR and stunting. In this analysis, the effect of IPTp and ITNs acted only on child deaths from the first two pregnancies of women in each country.
Estimates of household ITN coverage used in this analysis were based upon previous publications that modeled the proportion of household with ≥ 1 ITN each year in each country in Africa based on survey data and ITN procurement and distribution data, with uncertainty. These estimates are adjusted for each country's population at-risk for malaria such that coverage reflects the assumption that all nets are received and owned by households in malaria endemic areas of the country.
The LiST model uses ITN household possession instead of ITN use by children because the trials from which 55% PE was derived all used intention-to-treat analyses, meaning the estimated effects (relative risk) were based on whether or not a child lived in a village that had high ITNs household possession. Results from the analysis of the association between ITNs and all-cause post-neonatal mortality across 29 cross-sectional datasets in Africa showed very little difference in the effect regardless of whether ITN household possession or ITN use by children the previous night was used, further supporting the use of household ITN possession instead of child use.
The impact of malaria prevention during pregnancy on LBW and subsequent < 5 child survival was estimated using the higher of the following two coverage indicators: 1) proportion of pregnant women using an ITN the previous night or; 2) the proportion of women who had a live birth in the past two years who received ≥ 2 doses of SP during an ante-natal care (ANC) visit. The authors are unaware of published yearly national estimates of malaria protection during pregnancy. Estimates of coverage of malaria prevention during pregnancy used in this analysis were derived from nationally representative household surveys, including the Demographic and Health Survey (DHS), the Multiple Indicator Cluster Survey (MICS), the Malaria Indicator Survey (MIS) and the AIDS Indicator Survey (AIS), as summarized below and described in detail elsewhere (Additional file 1). Because the majority of malaria deaths and the burden of malaria in pregnancy is concentrated in rural areas, and as intervention coverage in rural areas has often lagged behind urban areas in many countries in Africa, the level of malaria prevention during pregnancy in rural areas was used for estimating the malaria deaths prevented from malaria prevention during pregnancy scale-up at the national level.
If there was no survey data point for IPTp/ITN coverage in 2000, coverage was assumed to be 0%. Linear interpolation was used from the first year of measured IPTp/ITN coverage, or from 0% coverage in 2000, to the next available survey point estimate. This slope was then used to inform the increase for years beyond the most recent household survey through 2010. In the case that there was only one household survey in the country, coverage was assumed to be 0% in the year IPTp was implemented as policy, and a linear slope was calculated from that year to the household survey data point. For countries projected to reach ≥ 80% coverage, maximum coverage estimates were limited to the proportion of women attending ANC at least once in last nationally representative household survey (see Additional file 1 for details on methods, as well as Additional files 3 and 4 for results).
The LiST model analysis conducted here has several important assumptions. First, the LiST model assumes that the PE of interventions is constant across each country in this analysis. While the PE of ITNs interventions varies across level of intervention coverage in the population, child age and malaria transmission level, the PE of ITNs used in this analysis is assumed to represent a valid estimate of the mean effect across these effect modifiers. Results from an analysis of 29 cross-sectional country datasets in Africa since 2000 with varying levels of ITN coverage and across varying levels of malaria transmission showed the average effectiveness of ITN under programme conditions to be very similar to the trials for reducing all-cause one-59 month child mortality, which helps support the use of a fixed 55% PE for ITNs used here. A previous validation study of the LiST model against measured reductions in all-cause child mortality following the scale-up of ITNs, across a range of transmission settings, further validates the use of the fixed PE of 55% for ITNs used here. Second, while there is evidence that malaria acts as a significant risk factor for dying from conditions other than malaria during childhood, the LiST model constrains deaths prevented by malaria control within the envelope of malaria-attributable < 5 deaths. Deaths associated indirectly with malaria as a contributory cause were not counted, although the ITN trials had shown that this indirect contribution of malaria may be an equally important additional contribution of malaria's overall effect on < 5 mortality as the directly malaria-attributed deaths.. Third, intervention coverage for a given year is assumed to be a valid estimate for the mean coverage over the 12-month period. Fourth, due to a lack of data to quantify effect modification, the model does not account for any possible synergistic or saturation effect between ITNs and IPTp for preventing child deaths. And lastly, coverage by a malaria prevention intervention is assumed independent from coverage by another malaria prevention intervention, which is likely not true.
Percent reduction in malaria deaths were estimated by dividing the number of malaria deaths estimated to occur with intervention coverage scale-up in a given year (or over a period of years) by the number of malaria deaths estimated to have occurred in that year (or over that period) had no intervention scale-up occurred from 2000.
Uncertainty bounds about total estimated post-neonatal child malaria deaths prevented from ITNs 2001–2010 were based on a non-probabilistic sensitivity analysis of the uncertainty of the three primary model parameters. This resulted in a worst-case (i.e. low-impact) and best-case (i.e. high-impact) scenario, with the worst-case using the lower bound of estimated post-neonatal malaria deaths in 2000 in each country, the lower bound of the PE for ITNs (49%), and the smallest net increase (slope) in ITNs scale-up 2001–2010 for each country based on the uncertainty of the coverage estimates by Flaxman and colleagues. The best-case scenario resulted from using the upper bound of estimated post-neonatal malaria deaths in 2000 in each country, the upper bound of the PE for ITNs (60%), and the largest net increase (slope) in ITNs scale-up 2001–2010 for each country based on uncertainty about the coverage estimates (Additional file 1 for details on methods).
Uncertainty bounds about total estimated < 5 deaths prevented from malaria prevention during pregnancy 2001–2010 were obtained based on a similar non-probabilistic sensitivity analysis of the uncertainty of two primary model parameters: lower and upper bound of the estimated PE of malaria prevention in pregnancy on preventing LBW, and the smallest and largest net increases in intervention coverage changes within each country from 2001–2010 based on sampling errors (see Additional file 1 for details on methods, as well as Additional file 4 for resultant uncertainty of malaria prevention in pregnancy scale-up for each country 2000–2010). This resulted in a worst-case and best-case scenario as described above.
A review of existing literature on the costs and cost effectiveness of ITN delivery [ITN or long-lasting ITN (LLIN)] in Africa since 2005 was conducted in PubMed and the grey literature; 13 studies were identified and included. Costs of ITN programmes were separated into delivery and commodity components, and used to estimate a median cost of ITN delivery, and of ITNs procured for programmes. Total costs were then calculated over the period 2006–2009 following previous methodological guidelines. Costs were combined with estimates of lives saved and Disability Adjusted Life Years (DALYs) over this time period to produce cost effectiveness results. A one-way sensitivity analysis was also conducted (see Additional file 1 and 5 for additional details).
Methods
This analysis, using the LiST model, focused on estimating the impact of ITNs for preventing post-neonatal (one-59 months of age) child malaria deaths from 2001–2010, as compared to a baseline of 2000. A baseline of 2000 was chosen as this is the year prior to any scale-up of ITNs or malaria prevention in pregnancy interventions in most African countries. Forty-three countries in sub-Saharan Africa are included in this analysis; 43 for the ITN analysis and 32 for the malaria prevention in pregnancy analysis (Figure 1).
(Enlarge Image)
Figure 1.
Malaria-endemic countries included in estimates of malaria intervention scale-up and resultant malaria-caused child deaths prevented 2001–2010.
This analysis of malaria prevention during pregnancy (with either IPTp or ITNs) for preventing all-cause < 5 child deaths through prevention of low birth-weight (LBW) was limited to 32 countries in Africa with stable malaria transmission. These countries accounted for 88% of the population in SSA at risk of malaria and 90% of the malaria-caused child deaths in 2000 in SSA. A list of excluded countries from the analyses, as well as the rationale for exclusion, can be found in Additional file 1.
The LiST model is based on the earlier work on effectiveness of interventions developed by WHO and UNICEF's Child Health Epidemiology Reference Group (CHERG) and RBM's Monitoring and Evaluation Reference Group (MERG). The model starts with estimates and assumptions within each country on the profiles for population and population growth, under-five cause of death structure, disease-specific mortality rates, and current coverage estimates of key child survival interventions. The model used in this analysis and accompanying documentation can be downloaded.
The model estimates child deaths prevented (within specific cause of death categories) due to intervention scale-up within a specified country as a function of three primary parameters, as outlined in details below: 1) the number of child deaths by cause projected to occur in each year (including population growth parameters over time); 2) the protective effect (PE) on cause-specific mortality (PE = 1-relative risk*100) for each intervention being scaled-up; and 3) increases in population coverage of each intervention. The model computes the number of deaths prevented by cause each year, accounting for population growth, as the difference between the estimated deaths that occur with intervention scale-up and the estimated deaths that would have occurred without intervention scale-up beyond the coverage at a baseline year. Further details of the LiST model estimation methods are presented in the Additional file 1.
Parameter 1- Within Country Estimates of Cause-specific Child Deaths
Within the LiST model, the total number of < 5 child deaths for the baseline year of 2000, by age, is based on estimates of < 5 mortality produced by the UN Inter-agency Group for Child Mortality Estimation (IGME). The number of post-neonatal child malaria deaths in our baseline year of 2000 was estimated as the proportion of child deaths one-59 months attributable to malaria, multiplied by the total number of all-cause < 5 child deaths one-59 months in the year 2000. Based on this method, the LiST model started with 870,928 malaria-caused deaths one-59 months in 43 African countries and 1,034,828 all-cause neonatal deaths in 32 African countries in the baseline year of 2000 (Additional file 2).
Parameter 2- Estimates of Intervention Effectiveness
As described in detail elsewhere, the PE of ITNs for preventing post-neonatal child malaria deaths has been estimated to be 55% (range 49–60%) based on a systematic review and meta-analysis of three trials.
Malaria in pregnancy increases the risk of LBW, resulting primarily from intrauterine growth retardation (IUGR) in areas of stable P. falciparum malaria transmission, while LBW has been demonstrated to be a significant risk factors for neonatal and infant mortality. The PE of malaria prevention during pregnancy for preventing LBW has been estimated to be 35% (95% confidence interval [CI] 23–45%) during the first two pregnancies in malaria endemic areas based on a systematic review of 2 ITN trials and 3 IPTp trials conducted in 1998 and 2002 in Kenya and 2004 in Mozambique, after the onset of sulphadoxine-pyrimethamine (SP) resistance. The effect of malaria prevention interventions during pregnancy on LBW in the LiST model acts solely through intrauterine growth retardation (IUGR). IUGR in the LiST model acts mostly through neonatal mortality, increasing the risk of dying during this period due to diarrhoea [RR = 2.0], sepsis/pneumonia (RR = 2.0), and asphyxia (RR = 2.3). During the post-neonatal period, IUGR slightly increases the risk of dying due to measles, malaria, diarrhoea and pneumonia, via links between IUGR and stunting. In this analysis, the effect of IPTp and ITNs acted only on child deaths from the first two pregnancies of women in each country.
Parameter 3- Changes in Coverage of Malaria Prevention Interventions
Estimates of household ITN coverage used in this analysis were based upon previous publications that modeled the proportion of household with ≥ 1 ITN each year in each country in Africa based on survey data and ITN procurement and distribution data, with uncertainty. These estimates are adjusted for each country's population at-risk for malaria such that coverage reflects the assumption that all nets are received and owned by households in malaria endemic areas of the country.
The LiST model uses ITN household possession instead of ITN use by children because the trials from which 55% PE was derived all used intention-to-treat analyses, meaning the estimated effects (relative risk) were based on whether or not a child lived in a village that had high ITNs household possession. Results from the analysis of the association between ITNs and all-cause post-neonatal mortality across 29 cross-sectional datasets in Africa showed very little difference in the effect regardless of whether ITN household possession or ITN use by children the previous night was used, further supporting the use of household ITN possession instead of child use.
The impact of malaria prevention during pregnancy on LBW and subsequent < 5 child survival was estimated using the higher of the following two coverage indicators: 1) proportion of pregnant women using an ITN the previous night or; 2) the proportion of women who had a live birth in the past two years who received ≥ 2 doses of SP during an ante-natal care (ANC) visit. The authors are unaware of published yearly national estimates of malaria protection during pregnancy. Estimates of coverage of malaria prevention during pregnancy used in this analysis were derived from nationally representative household surveys, including the Demographic and Health Survey (DHS), the Multiple Indicator Cluster Survey (MICS), the Malaria Indicator Survey (MIS) and the AIDS Indicator Survey (AIS), as summarized below and described in detail elsewhere (Additional file 1). Because the majority of malaria deaths and the burden of malaria in pregnancy is concentrated in rural areas, and as intervention coverage in rural areas has often lagged behind urban areas in many countries in Africa, the level of malaria prevention during pregnancy in rural areas was used for estimating the malaria deaths prevented from malaria prevention during pregnancy scale-up at the national level.
If there was no survey data point for IPTp/ITN coverage in 2000, coverage was assumed to be 0%. Linear interpolation was used from the first year of measured IPTp/ITN coverage, or from 0% coverage in 2000, to the next available survey point estimate. This slope was then used to inform the increase for years beyond the most recent household survey through 2010. In the case that there was only one household survey in the country, coverage was assumed to be 0% in the year IPTp was implemented as policy, and a linear slope was calculated from that year to the household survey data point. For countries projected to reach ≥ 80% coverage, maximum coverage estimates were limited to the proportion of women attending ANC at least once in last nationally representative household survey (see Additional file 1 for details on methods, as well as Additional files 3 and 4 for results).
Key Assumptions of this LiST Analysis
The LiST model analysis conducted here has several important assumptions. First, the LiST model assumes that the PE of interventions is constant across each country in this analysis. While the PE of ITNs interventions varies across level of intervention coverage in the population, child age and malaria transmission level, the PE of ITNs used in this analysis is assumed to represent a valid estimate of the mean effect across these effect modifiers. Results from an analysis of 29 cross-sectional country datasets in Africa since 2000 with varying levels of ITN coverage and across varying levels of malaria transmission showed the average effectiveness of ITN under programme conditions to be very similar to the trials for reducing all-cause one-59 month child mortality, which helps support the use of a fixed 55% PE for ITNs used here. A previous validation study of the LiST model against measured reductions in all-cause child mortality following the scale-up of ITNs, across a range of transmission settings, further validates the use of the fixed PE of 55% for ITNs used here. Second, while there is evidence that malaria acts as a significant risk factor for dying from conditions other than malaria during childhood, the LiST model constrains deaths prevented by malaria control within the envelope of malaria-attributable < 5 deaths. Deaths associated indirectly with malaria as a contributory cause were not counted, although the ITN trials had shown that this indirect contribution of malaria may be an equally important additional contribution of malaria's overall effect on < 5 mortality as the directly malaria-attributed deaths.. Third, intervention coverage for a given year is assumed to be a valid estimate for the mean coverage over the 12-month period. Fourth, due to a lack of data to quantify effect modification, the model does not account for any possible synergistic or saturation effect between ITNs and IPTp for preventing child deaths. And lastly, coverage by a malaria prevention intervention is assumed independent from coverage by another malaria prevention intervention, which is likely not true.
Calculation of Percent Reduction in Malaria Deaths
Percent reduction in malaria deaths were estimated by dividing the number of malaria deaths estimated to occur with intervention coverage scale-up in a given year (or over a period of years) by the number of malaria deaths estimated to have occurred in that year (or over that period) had no intervention scale-up occurred from 2000.
Uncertainty about LiST Estimates
Uncertainty bounds about total estimated post-neonatal child malaria deaths prevented from ITNs 2001–2010 were based on a non-probabilistic sensitivity analysis of the uncertainty of the three primary model parameters. This resulted in a worst-case (i.e. low-impact) and best-case (i.e. high-impact) scenario, with the worst-case using the lower bound of estimated post-neonatal malaria deaths in 2000 in each country, the lower bound of the PE for ITNs (49%), and the smallest net increase (slope) in ITNs scale-up 2001–2010 for each country based on the uncertainty of the coverage estimates by Flaxman and colleagues. The best-case scenario resulted from using the upper bound of estimated post-neonatal malaria deaths in 2000 in each country, the upper bound of the PE for ITNs (60%), and the largest net increase (slope) in ITNs scale-up 2001–2010 for each country based on uncertainty about the coverage estimates (Additional file 1 for details on methods).
Uncertainty bounds about total estimated < 5 deaths prevented from malaria prevention during pregnancy 2001–2010 were obtained based on a similar non-probabilistic sensitivity analysis of the uncertainty of two primary model parameters: lower and upper bound of the estimated PE of malaria prevention in pregnancy on preventing LBW, and the smallest and largest net increases in intervention coverage changes within each country from 2001–2010 based on sampling errors (see Additional file 1 for details on methods, as well as Additional file 4 for resultant uncertainty of malaria prevention in pregnancy scale-up for each country 2000–2010). This resulted in a worst-case and best-case scenario as described above.
Cost Effectiveness
A review of existing literature on the costs and cost effectiveness of ITN delivery [ITN or long-lasting ITN (LLIN)] in Africa since 2005 was conducted in PubMed and the grey literature; 13 studies were identified and included. Costs of ITN programmes were separated into delivery and commodity components, and used to estimate a median cost of ITN delivery, and of ITNs procured for programmes. Total costs were then calculated over the period 2006–2009 following previous methodological guidelines. Costs were combined with estimates of lives saved and Disability Adjusted Life Years (DALYs) over this time period to produce cost effectiveness results. A one-way sensitivity analysis was also conducted (see Additional file 1 and 5 for additional details).
Source...