Post-deployment effectiveness of malaria control interventions on Plasmodium infections in Madagascar: A comprehensive phase IV assessment

Background: Because international funding for malaria control is plateauing, affected countries that receive foreign funding are expected to maintain a constant budget while continuing to reduce Plasmodium transmission. To investigate the appropriateness of a malaria control policy in Madagascar, the effectiveness of all currently deployed malaria control interventions (MCIs) was measured. Methods: A nationwide cross-sectional survey was conducted in 2012-2013 at 62 sites throughout Madagascar. A total of 15,746 individuals of all ages were tested for Plasmodium infection using rapid diagnostic tests and were interviewed about their use of long-lasting insecticidal nets (LLINs), indoor residual spraying (IRS), intermittent preventive treatment of pregnant women (IPTp), and exposure to information, education and communication (IEC) campaigns. The association between Plasmodium infection and MCI exposure was calculated using multivariate multilevel models, and the protective effectiveness (PE) of an intervention was defined as one minus the odds ratio of this association. Results: The individual PE of regular LLIN use was high and significant (41 %, 95 % confidence interval [CI] 23-54), whereas its community PE was not. The PE of IRS at the household level was significant in one transmission pattern only (44 %, 95 % CI 11-65), and the community PE with high IRS coverage (>75 %) was high and significant overall (78 %, 95 % CI 44-91). Using LLINs after IRS increased the PE, and the reciprocal was also true. The maternal PE of IPTp was high but non-significant (65 %, 95 % CI -32 to 91). The PE of IEC was low, non-significant and restricted to certain areas (24 %, 95 % CI -34 to 57). Conclusions: This snapshot of the effectiveness of MCIs confirms that integrated vector control is required in malaria control policies in Madagascar and suggests combining MCIs when one is questionable. Policymakers should consider the local effectiveness of all deployed MCIs through a similar phase IV assessment. A complete description of the methodology of MEDALI’s cross-sectional survey is published elsewhere [5]. Below are the key points. Study sites were selected from a pre-existing network of sentinel health centres (SHC) for the surveillance of fever-associated diseases [7]. One SHC in each locality where at least one SHC existed was selected, and two study sites were randomly selected near each of the 31 SHC, for a total of 62 study sites [5, 8] (Fig. 1). The two coastal regions (east and west) exhibit hyperendemic malaria patterns. In the central highlands and the south, transmission patterns are unstable, episodic or epidemic. In the fringe areas at intermediate altitudes, transmission is limited to the rainy season. Coastal areas were investigated during September–October 2012, and other areas were investigated between November 2012 and January 2013. Malaria transmission patterns in the districts of Madagascar and MEDALI study sites and their population densities A total sample size of 13,950 is sufficient to detect an odds ratio (OR) of 0.7 for malaria RDT positivity with a power of 80 % according to the following parameters: baseline proportion of positive RDTs [or parasite rate (PR)] of 5 %, intervention coverage of 50 %, cluster effect of 2, and alpha risk of 5 % [9]. Under the same assumptions, this sample size is sufficient to detect ORs of 0.75 and 0.8 with powers of 69 and 49 %, respectively. To achieve a total sample size of 13,950 individuals, at least 225 people from a minimum of 50 households in each of the 62 study sites were included. The inclusion criteria included the following elements: ≥6 months of age; signed informed consent; and ability to take per os treatment in the case of positive RDT. The head of household or a representative and all participants answered a questionnaire about socio-demographic features and exposure to MCIs. Bed net use was defined as “use every night during the last 3 months” [5]. Household socio-economic status (SES) quintiles were created using principal component analysis (PCA) as described previously [5, 10]. Similarly, quintiles of housing permeability to mosquitoes were created using PCA based on housing construction materials and structural holes. Categories of exposure to IEC malaria messages were calculated by PCA according to one of the following types of media and time since the previous exposure: radio, poster, mobile video unit (MVU), television, leaflet or written press article, or other media/presentation [e.g., hiragasy (traditional Malagasy theatre), puppets, theatre, etc.]. The complete definition of all variables is available in Additional file 1. Blood was drawn from all participants by finger or heel puncture for RDT (CareStart® Malaria, Access Bio Inc., Monmouth Junction, NJ, USA). The population density of each study site was determined from the WorldPop/AfriPop database [5, 11]. The surface area of the study sites was calculated by contouring clusters with a polygon extending through GPS coordinates of external households using QGIS version 2.2.0. Analyses were performed for the complete dataset (IEC) or limited to populations targeted by the interventions (LLIN, IRS, or IPTp). The outcome was the result of the RDT: negative versus HRP2 and/or pLDH positivity. To explore factors associated with RDT positivity, generalized estimating equation (GEE) models were fitted by considering an exchangeable within-site correlation structure using the gee function of R [12]. The explanatory variables were fit into backward stepwise logistic regression models, and two variables (transmission pattern and population density) were forced in all models. All of the multivariate model fits were evaluated using binned residual plots [13, 14]. Two variables were tested as potential effect modifiers for the effectiveness of the MCIs: (i) the malaria transmission pattern and (ii) individuals <5 years old. Whenever the p value of these interactions terms were <0.05, a separate analysis was conducted for the associated area or age group. The effectiveness of LLIN and IRS was tested at the individual and cluster level, i.e., high coverage (>75 % population) versus low coverage (≤75 %). The protective effectiveness (PE) of an intervention was defined as one minus the odds ratio of the exposure to this intervention as described previously [15]. The study followed ethical principles according to the Helsinki Declaration. Informed consent was obtained from the individuals or the parents/tutors of the children before inclusion. The protocol of the present study was approved by the National Ethics Committee of the Ministry of Public Health of Madagascar (approval #CNE 57/MSANP/CE, July 24th, 2012).