Study site
Households were enrolled into the RCD study in the catchment area of Macha Hospital in Choma District, Southern Province, Zambia between January 12, 2015 and July 26, 2017 [18, 25, 27, 28]. The region has a tropical savannah climate with the rainy season occurring from December to April, followed by a cool dry season from May to August, and a hot dry season from September to November as previously described [18, 22, 15, 26]. Malaria transmission is propagated by the primary vector Anopheles arabiensis, which peaks during the rainy season. Infections are almost exclusively due to P. falciparum [18, 25, 29]. The major malaria control interventions are case management with artemisinin-based combination therapy (ACT) introduced in 2004, long-lasting insecticide-treated nets (LLINs) that were introduced in 2007 and redistributed approximately every three years with the most recent being in November 2017, and targeted mass drug administration (MDA) and indoor residual spraying (IRS) largely outside the study area [18, 26].
Reactive case detection
RCD eligibility and enrollment started at thirteen health centres and 23 health posts within the study catchment area where symptomatic individuals positive for malaria by PfHRP2-based RDT (index cases) triggered follow-up visits by a CHW and study team from Macha Research Trust [15, 18]. The study field team received notifications of an index case through SMS text messages from the health centre staff, after which they visited the household of the index case as well as secondary households located within a 250 metres radius of the index case within one week of notification [15]. The RCD radius was expanded from the government recommended 140 to 250 metres for the study. If the index case travelled outside their home district in the previous month and stayed overnight, they were not eligible for RCD screening through the government program. The field team was trained to administer consent, perform RDT testing, provide ACT for uncomplicated malaria, collect finger prick blood on filter paper, administer surveys, collect data using electronic tablets, and educate participants on malaria transmission and prevention [18].
Study population
The study population consisted of residents in an index case household and secondary households within 250 metres of an index case. Households were single or multiple houses belonging to the main and extended family [30]. When index and secondary households were screened, all members of a household were eligible for enrollment. After written informed consent, including parental permission and assent for older children, a questionnaire was administered to obtain demographic information, knowledge of malaria transmission, malaria symptoms, travel history, and malaria prevention methods [18]. Parents or guardians completed surveys on behalf of participants younger than sixteen years. A blood sample was collected for a PfHRP2-based RDT (SD Bioline, Standard Diagnostics Inc, Gyeonggi-do, Republic of Korea) and as dried blood spots (DBS) on Whatman 903™ Protein Saver cards (GE Healthcare Bio-Sciences Corporation, Piscataway, NJ) for quantitative PCR (qPCR) [18, 19]. Household residents found to be RDT positive were offered artemether/lumefantrine (Coartem®), the standard treatment for uncomplicated malaria in Zambia. Global positioning system (GPS) coordinates were obtained at each household using hand-held GPSMAP® 62 devices (Garmin Ltd, Olathe, Kansas) and mapped using ArcGIS version 10.5 (Environmental Systems Research Institute, Redlands, California) on a high resolution Quickbird™ satellite image of the catchment area [25, 28].
Parasite prevalence
Parasite prevalence was determined using the PfHRP2-based RDT results and detection of P. falciparum mitochondrial cytochrome b gene (cytb) by qPCR. PfHRP2 RDT readings were performed according to the manufacturer’s instructions [31]. DBS samples for qPCR were stored in plastic bags with desiccants and transported to the laboratory at Macha Research Trust for further drying. Samples were re-packed and stored at -20˚C until parasite DNA extraction was performed using the Chelex® method [28, 32]. qPCR was performed with the Applied Biosystems StepOnePlus™ Real-Time PCR System (Thermo Scientific, Waltham, MA, USA). Primers specific to P. falciparum cytb were used to amplify, detect and quantify P. falciparum DNA using SYBR® Green fluorescence [30, 33, 34]. Filter paper spotted with laboratory-cultured parasites (NF54) and dilutions of 3D7 genomic DNA were used as standards [32]. The limit of detection was established as one parasite/uL [18]. The qPCR reaction consisted of 5 µL DNA template, 5 µL SYBR® Green PCR Master Mix (ThermoFisher), 200 nM forward primer (5' CCT GAT AAT GCT ATC GTA 3'), and 200 nM reverse primer (5' TAA TAC AAT TAC TAA ACC AGC 3’) [18]. Amplification with correct melting temperature was considered positive and the amplicon was further confirmed on a 4% agarose gel [18].
Environmental risk factors
A Quickbird™ satellite image of the 1,200 km2 catchment area provided by the GeoEye-1 satellite (DigitalGlobe Services, Inc., Denver, Colorado) in 2017 and comprised of four-band 1.64-metre spatial resolution and 0.41-metre resolution panchromatic single-band imagery. Remote sensing data was imported into ArcGIS version 10.5 to geocode index and secondary households [25, 28]. All data layers were projected onto the Universal Transverse Mercator (UTM), Southern Hemisphere, Zone 35, WGS1984. A digital elevation model (DEM) with 90-metre resolution was obtained from Shuttle Radar Topography Mission (SRTM) version 3, processed in ERDAS Imagine 2011 software (Hexagon Geospatial, Madison, Al) and imported into ArcGIS [18, 26]. The ArcHydro Tools module of ArcGIS was used to build a stream network according to the Strahler stream classification that assigned order values of 1, 2, 3, etc. based on a hierarchy of tributaries, such that two small first-order streams join to form a second-order stream [19, 24, 35]. A shapefile for roads was created by digitizing roads in ArcGIS based on a 1:50,000 topographic map of Zambia and the satellite image. Households with one or more malaria positive individuals by RDT or qPCR (excluding the index case) were classified as positive households [19, 25]. Index and secondary households were compared based on the following baseline characteristics: median age of residents per household, number of individuals per household, number of individuals five years and younger per household, number of parasitaemic individuals, insecticide-treated bed net ownership, floor material, and cooking energy source [21].
Secondary households with and without parasitaemic individuals were also compared using the same variables. In addition to household descriptive variables, environmental risk factors characterizing the surroundings of secondary households and previously shown to be associated with malaria risk were evaluated on the following levels: (1) household-level, defined as environmental risk factors within 100-metre radius of a household; (2) cluster-level, defined as environmental risk factors within 250-metres radius of a household; and (3) neighbourhood-level, defined as environmental risk factors outside the 250-metre screening radius.
Household-level risk factors included the number of animal pens within a 100-metre radius of the main house structure and distance to nearest animal pens [36, 37]. Cluster-level risk factors included distance to index households and elevation difference with index households [20, 23, 24]. Neighbourhood-level risk factors included distance to the main road and distance to streams [19, 23]. If the distance between the index and secondary households was more than 300 metres, coordinates were manually cross-referenced and re-mapped by the field team when necessary. Elevation differences were generated by taking the difference in elevation as recorded by GPS devices from each secondary household and its corresponding index household. Missing elevation coordinates were extracted from the DEM. Animal pens were manually digitized in ArcGIS and were defined as enclosed, dark- or light-brown, oblong, circular, or rectangular roofless structures of any size within a 100-metre radius of the main house structure. Animal pens that were visually problematic to identify in ArcGIS were cross-referenced with Google Earth Images captured in 2017. From the individual stream order distances, the closest stream to the secondary household was identified. All risk factors generated using ArcGIS were imported into STATA 14.2 for statistical analyses. Fig. 1 illustrates secondary households within the 140- and 250-metre screening radius of the positive index household and the proximity of animal pens and streams.
Statistical analysis
The chi-square test for proportions and Wilcoxon rank-sum for means were used to compare household descriptive variables between index and secondary households, as well as secondary households with and without parasitaemic individuals. The analysis was restricted to all participants in secondary households who provided consent, completed the survey, and had RDT or PCR results. Generalized linear models with inference based on generalized estimating equations (GEE) were used to estimate the cross-sectional population average effect for the difference in odds of a positive vs. negative secondary households for each environmental risk factor. The GEE model was chosen for its ability to account for the clustering of secondary households around the index household and to more accurately estimate standard errors. The outcome was a binary variable, distinguishing secondary households with parasitaemic individuals and those without. The household-, cluster-, and neighbourhood-level environmental risk factors were assessed for collinearity using variance inflation factor values. Variables included in the model were: distance to index household (per 50 metres), distance to main road (per 50 metres), elevation difference with index household (per 10 metres), number of animal pens within 100 metres, presence of animal pen (yes vs. no), season, and a categorical variable identifying nearest streams order 1 through 5. Model fit was evaluated using the Hosmer-Lemeshow goodness of fit test and significance was evaluated at a p-value of 0.05.