Measurement of the subnational effect of vector control interventions on malaria infection

When the Roll Back Malaria Partnership was launched in 1998, insecticide-treated nets (ITNs) were seen as the key malaria prevention tool to halt the raging malaria epidemic in Africa.¹ A systematic review of several studies in endemic areas had shown that the use of ITNs reduced malaria mortality by 17% and clinical episodes by half.² Since 2000, almost 2 billion US dollars have been invested in malaria control globally.³ Between 2004 and 2012, almost 600 million ITNs have been distributed in Africa. Over this period, studies have shown significant reductions in malaria infection prevalence in the continent⁴ and ITN coverage is likely to be an important driver of this decline.⁵

Despite the general consensus on the role of ITNs in reducing malaria infection rates in Africa, few studies have analysed their effect subnationally. Such an analysis is especially important given the spatially and temporally heterogeneous nature of malaria transmission and the dependence of the effectiveness of ITNs on the transmission intensity of an area.⁶ A major limitation has been the lack of nationally representative spatially and temporally comparable data that concurrently measure parasitaemia and access or use of ITNs among the population. Such data have become increasingly available since 2005, mainly through malaria indicator surveys and some demographic and health surveys.⁷

In this issue of The Lancet Global Health, Federica Giardina and colleagues⁸ report their analysis of national household survey data from six countries (Angola, Liberia, Mozambique, Rwanda, Senegal, and Tanzania) that have had at least two surveys with data on both the use or access of ITNs, indoor residual spraying (IRS), and malaria infection rates among children younger than 5 years over the period 2006–12. Giardina and colleagues developed a Bayesian geostatistical approach to estimate the spatial effects of ITNs and/or IRS on malaria parasitaemia, after adjusting for time of survey, climatic factors, urbanisation, and socioeconomic status. They first used Bayesian geostatistical interpolation to predict malaria risk at 1×1 km spatial resolution at the two time periods for each country. From these they estimate the probability of parasitaemia risk reduction and the difference in the total number of children infected. They use a Bayesian variable selection approach to determine the most appropriate intervention indicator to the changing infection rates in a country. The analysis shows variable national and subnational levels of reductions of infection rates and effects of the vector control interventions. At the country level, the estimated decline between survey periods in the number of infections among children younger than 5 years was about 52% in Angola, 15% in Liberia, 42% in Rwanda, 40% in Senegal, and 30% in Tanzania; no change was seen in Mozambique. Changing ITN coverage seemed to have a significant effect on infection rates in Angola and Senegal but not in the other countries. Interestingly, however, in each country, subnational analysis showed reduction in parasitaemia and significant associations with ITN and/or IRS in some areas.

This study provides an innovative model to harness national household survey data to quantify the effects of the various malaria control interventions. The model accounts for the spatial dependence in parasitaemia as a consequence of the heterogeneous distribution of malaria transmission and its drivers and also allows for the estimation of varying subnational effects of the interventions. Giardina and colleagues argue that these variations are likely to be as a result of levels of ITN/IRS coverage and the intensity of transmission where, for a unit change in intervention coverage, a greater effect is seen in moderate transmission areas than in those of high transmission. This suggestion is supported by both the theoretical and empirical literature.⁶ However, there are other factors that have not been included in this study that could have influenced the estimated effect of the vector control interventions on changing levels of malaria infection. Many of these are acknowledged by Giardina and colleagues and include the timing of ITN scale-up campaigns and that of surveys where a short window might not be enough to observe effect on infection rates; the age and condition of nets, which affects their effectiveness;⁹ the changing vector distribution and bionomics as a consequence of exposure to various vector control interventions; and insecticide resistance.¹⁰

Finally, Giardina and colleagues rightly emphasise the potential policy use of the results of their study. However, careful interpretation of these results is required. For example, where ITNs are not associated with reductions in parasitaemia, should countries stop issuing them? If the intrinsic transmission is extremely low such that, epidemiologically, ITNs have minimal impact, this might be a sensible course of action. However, where transmission is currently extremely low due to the scale-up of vector control interventions over the past few years, beginning even before the study period, or remains moderate or high, the scale up of ITNs must continue. For these decisions, a careful comparison of transmission intensity currently and in the preintervention period (circa 2000) is required.⁴