Date Published: February 22, 2019
Publisher: Public Library of Science
Author(s): Carrie B. Dolan, Ariel BenYishay, Karen A. Grépin, Jeffery C. Tanner, April D. Kimmel, David C. Wheeler, Gordon C. McCord, Luzia Helena Carvalho.
To test the impact of a nationwide Long-Lasting Insecticidal Nets [LLINs] distribution program in the Democratic Republic of Congo [DRC] on all-cause under-five child mortality exploiting subnational variation in malaria endemicity and the timing in the scale-up of the program across provinces.
Geospatial Impact Evaluation using a difference-in-differences approach.
Democratic Republic of the Congo.
52,656 children sampled in the 2007 and 2013/2014 DRC Demographic and Health Surveys.
The analysis provides plausibly causal estimates of both average treatment effects of the LLIN distribution campaign and geospatial heterogeneity in these effects based on malaria endemicity. It compares the under-five, all-cause mortality for children pre- and post-LLIN campaign relative to children in those areas that had not yet been exposed to the campaign using a difference-in-differences model and controlling for year- and province-fixed effects, and province-level trends in mortality.
We find that the campaign led to a 41% decline [3.7 percentage points, 95% CI 1.3 to 6.0] in under-5 mortality risk among children living in rural areas with malaria ecology above the sample median. Results were robust to controlling for household assets and the presence of other health aid programs. No effect was detected in children living in areas with malaria ecology below the median.
The findings of this paper make important contributions to the evidence base for the effectiveness of large scale-national LLIN campaigns against malaria. We found that the program was effective in areas of the DRC with the highest underlying risk of malaria. Targeting bednets to areas with greatest underlying risk for malaria may help to increase the efficiency of increasingly limited malaria resources but should be balanced against other malaria control concerns.
Malaria remains a major global health concern, and one where progress has recently stalled. In 2016, approximately 216 million cases of malaria occurred worldwide, an increase of 5 million cases from the previous year. Malaria exacts a disproportionately high burden in Sub-Saharan Africa [SSA], where two countries, Nigeria and the Democratic Republic of the Congo [DRC], account for more than 37% of the global total of estimated malaria cases. Of particular relevance is the DRC, where 40% of deaths among Congolese children are attributed to malaria. The DRC was once renowned in Africa for its clinics, quality of physicians, and primary healthcare systems. Since 1996, however, the DRC has experienced devastating and destabilizing conflict characterized by extreme violence, mass population displacements, and a collapse of public health services[3,4]. Gaining a better understanding of the impact of the existing efforts in the DRC is therefore an important priority for researchers and policymakers alike.
The empirical results presented in subsequent sections combine data from four sources: the DRC Demographic Health Survey [DDHS], PMI, the Malaria Ecology Index [MEI], and the AidData Aid Management Platform [AMP]. The authors complied with the conditions of use for each of the four datasets used in the analysis.
In line with other studies measuring the impact of a public health campaign,[24,38,49–51,53] we utilized a difference-in-difference approach augmented with a geospatial variable of the NMCP LLIN distribution campaign exploiting the different timing and locations of the campaign rollout during 2009–2013. By combining data on the distribution of LLINs and child mortality at precise geographic locations, we control for potential confounding factors correlated across children within a province. Since the program was rolled out differentially across provinces over time, we were able to control for factors that improved child mortality across the entire country over our study period. By integrating data on climatic and ecological drivers of mosquito breeding that vary over time and space, we identified the effects of LLIN distribution in locations where malaria transmission risk is higher. This difference-in-differences approach, augmented with a geospatial variable, compares pre- and post-intervention change in under-five, all-cause child mortality risk using a difference-in-differences model. We include year fixed effects, which flexibly de-trend the mortality data at the country level to prevent spurious correlation with the LLIN campaign rollout, as well as province fixed effects to absorb time-invariant omitted variables that could be correlated both to mortality risk and the timing of LLIN campaign. Finally, we include province-specific linear time trends, which account for differences in mortality trends before the LLIN campaign. Therefore, assuming there is no residual confounding, our estimate provides the causal effect of the LLIN campaign on child mortality risk.
Table 1 presents descriptive statistics on the sample characteristics by treatment status for the primary outcome and control variables for our sample. The child mortality sample from the DHS included 52,656 children, of which 40,167 [76%] were children living in rural areas.
This study evaluates the impact of a national LLIN campaign that took place in the DRC from 2009–2013 on all-cause mortality of children under five years of age using nationally representative survey data. We found that the campaign was associated with important declines in child mortality in areas with malaria ecology above the median. Specifically, children living in areas with a bednet campaign and with high measures of malaria transmission experienced a 41% lower rate in all-cause child mortality compared to children in similar areas that had not received the bednet campaign. The magnitude of the decrease is consistent with estimates of around 71% decline in malaria mortality in children under 5 attributed to scale up of malaria control programs.7 The results were robust to controls for household assets, the presence of other health aid programs, and limiting the sample to rural areas.