The tMDA campaign
With the assistance of the Malaria Zero Consortium, the National Malaria Control Program in Haiti has implemented a tMDA campaign using sulfadoxine-pyrimethamine (SP) and single low-dose (SLD) primaquine in a single round. The target dose for SP was 25/1.25 mg/kg, the approved therapeutic dose in Haiti for second-line treatment. The target dose for SLD primaquine was 0.25 mg/kg, lower than the recommended therapeutic dose and used in the tMDA to clear late-stage gametocytes. The campaign took place just before the peak of the annual rainy season, October 10–November 6, 2018, with the intention of accelerating progress towards malaria elimination [11, 12]. It was implemented in conjunction with an indoor residual spraying (IRS) campaign that used a pyrethroid insecticide (Actellic CS) [13].
The study was conducted in a pilot area covering five communes of the Grande-Anse Department (first-level administrative division), which is the Department (among 10) with the highest malaria incidence rate in the country (18.1 per 1000 in 2017) [14]. Within the pilot area, the tMDA campaign was delineated to 12 operational units, defined as the contiguous polygonal areas of approximately 2000 residents with the highest predicted reproductive numbers. Models that integrated population density, surveillance data, population mobility scores, and ecological factors were used to predict risk of transmission and rank operational units. The targeted areas that comprised these 12 operational units covered ~ 100 km2 with an estimated population of 46,372. All individuals in the targeted areas aged ≥ 6 months were offered directly observed, age-appropriate treatment of SP + SLD primaquine. Women in their first trimester of pregnancy and participants with signs of severe illness, known allergies to the treatment, specific medical conditions (e.g.: being in the first trimester of pregnancy), or using contraindicated medications (e.g.: sulfanomide or any other antimalarial medication) were excluded. More information about the tMDA/IRS campaign and the pilot area are available elsewhere [13, 15].
Study design and recruitment
This study was conducted in 25 schools that serve as easy access groups; in comparison to population-based household surveys, logistical issues and costs are considerably reduced in easy access groups surveys, while they have proven to be effective proxies for measuring the effectiveness of malaria interventions [16]. The schools were selected by stratified random sampling after a census of all schools with at least 100 pupils in the pilot area (which includes but is not limited to the areas targeted for MDA). Equal distribution of the schools across communes and by remoteness was ensured. Surveys were conducted in 2017 (November 6–December 7, i.e., pretest) and 2018 (November 12–December 13, i.e., posttest). The only eligibility criterion was that the children had to be regularly enrolled in the school. All pupils present at school at the time of the survey were therefore eligible to participate, irrespective of their age or symptoms. The target sample size was 2500 participants each year, or ~ 100 children per school. When the total number of children present at a school exceeded 150, a simple random sample of 150 children were selected from the entire pupil population at that school. Only pupils present at the time of the survey were sampled; for logistical reasons, it was not feasible to return on another day to the site to survey children who were temporarily absent.
Since the same set of schools was surveyed both years, it was possible to identify a nested cohort of children. Matching between the 2017 and 2018 lists of participants was performed independently by two team members who compared the school, the children’s full name, and their date of birth. The present study is restricted to the matching subset of children who were surveyed twice. Among these, children who self-reported in the 2018 survey that they had taken the drugs distributed during the tMDA campaign constituted the exposure group; children who self-reported that they did not constituted the control group. This was a natural experiment conducted independently from the intervention; the study was not initially designed to assess the effects of the tMDA campaign.
More information about the recruitment procedures and the selection of the schools is available elsewhere [17]. Rate of refusal to participate in the survey was < 1% both years. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines were followed.
Survey procedures
A sociodemographic questionnaire was administered to all participating children, with standardized questions about their age, history of fever, use of bed nets, travel, treatment-seeking practices, and exposure to tMDA. Siblings from the same household were gathered together, and the oldest sibling was asked to help the younger ones with their responses. Data were automatically entered on a mobile data collection platform installed on tablets and uploaded to a secure cloud-based server.
A capillary blood sample from a finger prick was taken to perform a conventional histidine-rich protein 2 (HRP2)-based RDT (Standard Diagnostic Bioline Ag. Pf, South Korea) to test for P. falciparum parasite infection. Finger prick blood was also spotted on Whatman 903 cards (GE Healthcare), dried overnight at ambient temperature, and packed the next day with silica gel. Dried blood spots were stored at + 4 °C and transported to the national laboratory (Laboratoire National de Santé Publique) in Port-au-Prince weekly.
Laboratory procedures
A set of 17 P. falciparum antigens were included in the panel, with antigens covalently coupled to unique bead regions [18]. Complete information regarding the creation and production of this panel of P. falciparum antigens has been published previously [4, 10, 18]. Immunoglobulin G data were collected for all participants. Antibody levels were measured using a bead-based assay in which sample and secondary antibody were incubated simultaneously overnight. Median fluorescence intensity (MFI) was recorded and corrected for background reactivity to give a final value of MFI minus background. Alongside participant samples, a six-point serial dilution of a hyperimmune positive control pool was added to each plate to assess plate-to-plate variation in data collection. More information about the laboratory procedures, including quality control, is available elsewhere [10, 18, 19].
Serological outcomes
Antibody concentrations from the 2017 and 2018 surveys were aligned using a procedure detailed elsewhere [10]. Seropositivity was defined separately for each antigen using a mixture model approach. Two distributions were identified in the log-transformed median fluorescence intensity (MFI) data, and the threshold for positivity was set at the mean of the lower distribution plus five standard deviations [19]. Based on an antigen selection process described elsewhere [10], five antigens were selected for further analysis: MSP1-19, AMA-1, Etramp5 antigen 1 (ETR51), HSP40, and GLURP-R0. Among these, three (HSP40, GLURP-R0, and ETR51) were previously shown to be short-term markers, defined as informative of recent exposure (i.e., exposure that occurred within the previous three months) [5, 20]. ETR51 showed the highest accuracy in predicting current or recent infection to P. falciparum in the Haitian context [4, 5, 10]. The remaining two antigens (MSP1-19 and AMA-1) are usually considered as long-term markers in the adult population, but results are less conclusive in children and they were therefore added to the analysis [21,22,23].
Analyses
Exposure to tMDA was assessed by each pupil’s self-report. Descriptive analysis to assess pre-post differences in socio-demographic characteristics between the intervention and comparison groups was performed using Chi-square tests or Fishers’ tests, for frequencies, and t-tests, for means. For each of the five antigens, a multilevel logistic regression model was fitted with seropositivity of the participants as the dependent variable. A set of time-varying potential confounding variables were tested in the models: socioeconomic characteristics of the households (size, occupation of the head of household, owning of livestock and bed nets), household’s exposure to indoor residual spraying in the previous three months, individual use of a bed net the night before, travel history, and total rainfall at the recruitment site during the previous two months. To facilitate comparison, the same set of variables was kept in the five final models, with the best-fitting model selected according to the Akaike information criterion values. Random intercepts at the individual, household, and commune levels were also included, and robust variance estimators were used across all analyses [24, 25].
Treatment effects were assessed following a difference-in-differences approach, which allows controlling for observed and unobserved time-invariant confounders [26]. Pre-post changes in seropositivity were therefore compared between the exposure and control groups by adding an interaction term between time period and exposure. Because impact estimates sometimes vary depending on the type of measure, three different indicators were used in the analysis as recommended: risk difference, odds ratio, and relative risks [27]. Relative risks were derived from logistic regression models by computing marginal standardized probabilities and using the margins and nlcom Stata commands [28]. All analyses were performed using Stata version 14.0 software (StataCorp LLC, College Station, Texas).
Ethical considerations
The study was approved by the National Bioethics Committee in Haiti (1516–30), the London School of Hygiene and Tropical Medicine Ethics Committee (103939), and the Tulane University Institutional Review Board (795709). Participation in the study was not remunerated. Activity did not constitute engagement in human subjects research as determined by the human subjects office of the US Centers for Disease Control and Prevention Center for Global Health (2016-135a).
Written assent was sought for children > 6 years of age. Participants could choose to give thumbprint consent/assent if they could not sign. An opt-out method was used to obtain consent from the children’s parents, as described elsewhere [17]. Written informed consent was also obtained from each school director after consultations and consent from the Department of Education and local leaders.