In each site study teams performed a baseline census. The study teams geo-localised all houses by GPS and assigned a unique identification number linked to the study number of each household member. The census recorded the de jure population in each village, which tallies people according to their regular or legal residence. During subsequent surveys, the study teams recorded the de facto population, i.e., the number of people sleeping in the village during the night before the survey.

At each site, formative research conducted in preparation for the study suggested that, in addition to the information content, the background of the person providing information was a critically important determinant of the effectiveness of the communications [38–43]. Establishing trust between community members and the study team was a key prerequisite for a functioning relationship with the study populations. Team members who resided in the village for extended periods were encouraged to participate in local life and community events, such as weddings, funerals, and festivals. Participation in daily community life allowed the team to develop an understanding of local hierarchies, the seasonality of village work, and the needs and preferences of community members [44,45].

To increase the uptake and thus impact of MDAs, the study teams conducted intensive community engagement at each study site, starting from the time of site selection. This included educational elements to ensure that every community member was informed about the purpose of the MDA and the need for the entire village to participate to maximise individual benefits, and education to allay fears about risks of the antimalarial drugs. Village leaders, village malaria workers, and volunteers formed committees that assisted the study teams in organising the survey and designing and implementing community engagement. The teams also organised community engagement meetings at different levels, from individual or household informal discussions to whole village assemblies. Children and young people were engaged through activities such as music festivals and theatre.

Study teams provided benefits and incentives for participation at all sites (S1 Table). They adapted incentives according to the wishes of the villagers, as discussed in community meetings, and the preferences of regulatory authorities. In some sites, the study teams provided individual cash or non-cash incentives combined with gifts and/or lottery tickets, whereas in other sites the study teams preferred community incentives, e.g., improved water supply for the entire village. The study teams provided basic primary healthcare in the study villages. This ancillary care was essential to establish trust between the study team and community members. Health education on topics unrelated to MDA, such as family planning, nutrition, and vaccinations, was provided to community members at their request. The research teams worked to ensure that all participating villages had uninterrupted access to early diagnosis, efficacious antimalarial treatment, and insecticide-treated bednets.

The primary target of the MDAs was falciparum malaria. Radical cure of latent P. vivax, which would consist of a 7- to 14-day course of the 8-aminoquinoline primaquine, was not provided. The study teams conducted DP MDAs at month 0 (M0), M1, and M2 in early (intervention) MDA villages, whereas deferred (control) MDAs were conducted at M12, M13, and M14 in all sites except for the study sites in Myanmar, where, for operational reasons—specifically the limited access to study sites during the anticipated heavy rains—the deferred MDAs were conducted at M9, M10, and M11 (S1 Fig). In each country we matched the village pairs by geographical proximity, population size, and parasite prevalence. In each pair we selected 1 village randomly to receive early MDA and the other village to receive deferred MDA [46]. The randomisation was based on computer-generated random numbers provided by the trial statistician. For each round of treatment, the study team set up a mobile clinic or specific malaria post. Participants received MDA at home if they were unable to come to the clinic. The drug regimen comprised 3 rounds of 3 daily doses of DP combined with a single low dose of primaquine (0.25 mg/kg) given at intervals of 1 month. For regulatory reasons, no primaquine was administered in Cambodia. One DP tablet contained 40 mg of dihydroartemisinin and 320 mg of piperaquine. A weight-based regimen aiming at a total dose of 7 mg/kg dihydroartemisinin and 55 mg/kg piperaquine phosphate was used (Eurartesim, Sigma Tau, Italy, or Guilin Pharmaceutical, Guilin, China). The study teams encouraged all residents in the study villages to take part in the drug administration except for children under the age of 6 months and pregnant women (first trimester pregnancies were excluded in Myanmar and Vietnam and all pregnant women were excluded in Cambodia and Lao PDR). All doses were administered under direct observation of the study teams. All drugs were provided with snacks to increase tolerability. For children unable to swallow the tablets, the staff crushed the tablets and administered them mixed with water. Newcomers to the village or returning residents arriving during the study period were contacted by village malaria workers and offered a single round of DP. No placebo was used in control villages.

The rationale for the treatment regimen has been described previously [47]. Briefly, a full 3-day treatment dose of DP was given in order to clear any P. falciparum infection. The rationale for 3 repeated monthly rounds (3× 3 doses) was to provide the 3 months of post-treatment prophylactic effect needed to interrupt malaria transmission. This period was considered necessary because infectious mosquitos can survive for 30 days, and some community members were absent during the drug administration (they were treated during subsequent rounds) [48–51]. A single low dose of primaquine is sufficient to sterilise and rapidly clear mature P. falciparum gametocytes, which are not susceptible to schizonticidal drugs [50].

The study teams observed participants for 1 hour after drug administration and in case of vomiting offered a full or half repeat dose depending on the interval since the primary administration. Study staff enquired about adverse events (AEs) using a structured questionnaire on days 2, 3, and 7, and again 1 month after drug administration. During the 3-month intervention, a mobile clinic staffed with a medical assistant was available in the village to provide free consultations to villagers and to assess participants presenting with AEs. All serious AEs (SAEs) including hospitalisations and deaths occurring within 3 months of the first drug administration were documented and investigated by the study team, and the association with treatment was assessed.

The surveillance period was 24 months in the study sites in Myanmar and Vietnam but stopped after 12 months in Cambodia and Lao PDR because of the delayed starts in Cambodia and Lao PDR (see also S1 Fig). Just before the MDA and then every 3 months after the MDA, the study teams invited all people residing in the study villages aged 6 months or older, including temporary residents and migrant workers arriving after the MDA, to participate in the cross-sectional prevalence surveys. The survey at M21 in Myanmar was omitted because access to the study sites became impossible for the study teams during heavy rains. At the quarterly surveys, the study teams recorded the presence or absence of each participant in the village during the preceding 3 months. Additional information was collected during home visits at 2-weekly intervals: a study staff member enquired about the presence of recorded household members. Study staff recorded all newcomers who intended to stay in the village for longer than 2 weeks. Demographic information was collected, and participants’ temperature (skin surface forehead or tympanic membrane), weight, and height were measured by the study team. The threshold for determining fever was 38°C for body surface temperature and 37.5°C for tympanic temperature. A rapid diagnostic test (RDT) was performed, and if the test was positive for malaria, patients were treated according to national guidelines. The study team collected venous blood (3 ml from all individuals aged ≥5 years, and 500 μl from children aged ≥6 months to 5 years) at each survey, and from any individuals who developed fever during the study period.

The study team stored blood samples in a cool box in the field and transported the samples within 12 hours to the local laboratory. The teams tested blood samples from all survey participants using standard microscopy and malaria RDTs (Myanmar, Lao PDR, and Vietnam: SD Bioline Malaria Ag P.f/Pan POCT, Standard Diagnostics, Yongin-si, Republic of Korea; Cambodia: Healgen Malaria P. falciparum/Pan 1-step RDT, Zhejiang Orient Biotech, China). Microscopists who had at least 5 years’ experience and/or were confirmed to be Level 2 or higher, as assessed by a standard WHO 55 slide set, performed the standard microscopy, counting the number of parasites per 500 white blood cells on Giemsa-stained peripheral blood thick films. After separation of plasma, buffy coat, and packed red blood cells, samples were frozen and stored at −80°C. The study teams transported frozen samples from Myanmar, Cambodia, and Lao PDR monthly on dry ice to the molecular laboratory in Bangkok, Thailand, and the samples from the Vietnam sites to Ho Chi Minh City, Vietnam, for DNA extraction and high-volume ultrasensitive quantitative PCR (uPCR).

We have previously reported a detailed description and evaluation of the uPCR methods [52]. In summary, we purified the DNA template for PCR detection and quantification of Plasmodium from the thawed packed red blood cell samples. The purified DNA was dried completely in a centrifugal vacuum concentrator and then suspended in a small volume of PCR grade water, resulting in a concentration factor defined by the original blood volume (100–2,000 μl) divided by the resuspended double distilled water volume (10–50 μl). We used 2 μl of resuspended DNA as template in the quantitative PCR reaction. We assessed the presence of malaria parasites and estimated the parasite load in each sample using an absolute quantitative real-time PCR method. The 18S rRNA–targeting primers and hydrolysis probes used in the assay have been validated and are highly specific for Plasmodium species [53]. The lower limit of accurate quantitation using this method is 22 parasites per millilitre of whole blood. We used a QuantiTect Multiplex PCR NoROX Kit (Qiagen, Hilden, Germany) in the Bangkok laboratory and an absolute quantitative real-time PCR (quantitative PCR) method (Roche, Basle, Switzerland) in the laboratory in Ho Chi Minh City. We determined the Plasmodium species in uPCR positive samples using nested PCR specific to P. falciparum (microsatellite marker Pk2), P. vivax (microsatellite marker 3.502), and P. malariae (18s rRNA) as described previously [53–55]. We reported positive samples for which there was insufficient DNA for species identification—or where no amplification was obtained in this step—as being of indeterminate species.

We assessed polymorphisms in the PfKelch13 gene by nested PCR amplification covering the full length of the gene (total 2,181 bp) and sequenced the gene by ABI Sequencer (Macrogen, Seoul, Republic of Korea) as described previously. We monitored cross-contamination by adding negative control samples in every run. Sequencing results were aligned against PfKelch13 of reference strain 3D7 (putative 9PF13_0238 NCBI Reference Sequence [3D7]: XM_001350122.1), using Bioedit software (Abbott, Santa Clara, CA, US). Two study technicians assessed polymorphic patterns blinded to the origin of the sample.

We quantified Pfplasmepsin2/3 gene copy number using relative quantitative real-time PCR based on Taqman probe on a Corbett Rotor-Gene Q (Corbett Research, Mortlake, NSW, Australia). Primers and probes have been described previously [56]. We performed amplification in triplicate on a total volume of 25μl as multiplex PCR using a QuantiTect Multiplex PCR NoROX Kit (Qiagen, Hilden, Germany). Every amplification run contained 9 replicates of calibrators and triplicates without template as negative controls. Plasmodium-specific beta-tubulin served as an internal standard for the amount of sample DNA added to the reactions.

We categorised each resident’s individual MDA exposure as (a) did not participate at all, (b) did not complete a single round (3 doses), (c) completed only 1 round, (d) completed only 2 rounds, or (e) completed all 3 rounds. For the estimation of MDA coverage, we defined the numerator as the number of participants during three MDA rounds and the denominator as the de facto population during the time of MDA rounds. We defined Plasmodium prevalence by the uPCR result, but in the absence of a uPCR result we considered a positive microscopy or RDT result as sufficient to classify an individual as infected.

A P. falciparum infection was defined by either a P. falciparum positive result or a mixed result of P. falciparum and P. vivax. The incidence was defined using the number of malaria infections as numerator and exposure time as denominator. The individual exposure time was defined as the number of days spent within the catchment area, i.e., the village and the surrounding farms. The exposure time was estimated in 3-month intervals. For example, if a resident was present during 2 sequential surveys, the exposure time was 90 days. If a resident was missing during a survey, we assumed he/she stayed in the village for 45 days after the last participation in a survey. Similarly, we assumed a new arrival had arrived 45 days before the first participation in a survey. We assumed both losses to follow-up and intermittent missing data were missing at random. Seasons were defined as wet or dry by country as described above.

The unit of randomisation and hence the unit of statistical inference was the village cluster. The primary approach to analysis was based on intention to treat (ITT). In order to assess the sensitivity of our assumptions in the ITT approach, we also performed a dose-related per protocol analysis. We compared changes in prevalence then in incidence of P. falciparum (including mixed P. vivax and P. falciparum) infections over 12 months between villages that received early MDA and those that received deferred MDA.

Before the data collection was completed we drafted a statistical analytic plan, which is included as S1 Text. We examined the impact of DP MDA on malaria using multilevel mixed-effects Poisson models to obtain incidence rate ratios (IRRs) of Plasmodium infections. For the multilevel models, level 1 was repeated measurements of villagers over the follow-up time, level 2 was participants from the same village, level 3 was the 16 randomised villages, level 4 was the 4 different countries.

First, we performed univariable analyses to obtain the unadjusted estimates of IRRs of the association between malaria infections and MDA status, followed by adjustment for variables prespecified in the statistical analytic plan that are predictive for outcome and potentially imbalanced, i.e., sex, age, fever, bednet use, season, and prevalence of P. falciparum infections in the village. In an alternative model we used MDA exposure, i.e., 0, 1, 2, or 3 completed rounds as the main independent categorical exposure variable. In the initial analysis, we included only the exposure time from the completion of the MDA at M3 to M12. In a secondary analysis, we included the period from M0 to M12, i.e., including the MDA implementation period. For the analyses reported here, focusing on P. falciparum, we excluded parasitaemias in which the species could not be identified (indeterminate species, Plasmodium spp.). As the lower limit of quantification is 22 parasites/ml, genome densities less than 22 parasites/ml (n = 95 participants at baseline) were not included in the analysis, but the infection status (infected or uninfected) was left unchanged. The control villages in Myanmar, which received MDA at M9 during the crossover period, were excluded from the analysis at M12. We provide the intra-cluster correlation coefficient (ICC) for the incidence of P. falciparum infections with village (cluster) as a unit of randomisation, accounting for the random effect of country, using the exact linearisation calculation approach [57].

The sample size, 4 village clusters per country, was chosen mainly for operational and practical reasons. A formal sample size calculation suggested that 16 villages would provide 80% power to detect a 95% fall in prevalence from a 10% initial prevalence, controlling for random changes in prevalence in the control groups, with a minimum of at least 152 individuals in each village recruited and followed up satisfactorily.

The study teams collected survey data on case record forms and entered the data on smartphones before exporting them into OpenClinica (OpenClinica, Waltham, MA, US). Graphical summaries have been presented to show prevalence and incidence patterns over time. Treatment and AE data were recorded on registers and then entered in Excel (Microsoft, Redmond, Washington, US). Analyses were performed in STATA 15.0 (StataCorp, College Station, Texas, US).

The studies were approved by the Cambodian National Ethics Committee for Health Research (0029 NECHR, dated 04 Mar 2013), the Institute of Malariology, Parasitology, and Entomology in Ho Chi Minh City (185/HDDD, dated 15 May 2013), the Institute of Malariology, Parasitology, and Entomology in Quy Nhon (dated 14 Oct 2013), the Lao National Ethics Committee for Health Research (Ref No 013-2015/NECHR), the Government of the Lao PDR, and the Oxford Tropical Research Ethics Committee (1015–13, dated 29 Apr 2013). Each participant, or parent/guardian in the case of minors, provided individual, signed, informed consent; illiterate participants provided a fingerprint countersigned by a literate witness (ClinicalTrials.gov Identifier: NCT01872702).

Of the 4,423 people residing during M0, M1, and M2 in the 8 villages randomised to early MDA, 3,790 (86%) completed at least 1 round (3 doses) of DP MDA: 635 (14%) completed 1 round of antimalarials, 635 (14%) completed 2 rounds, and 2,520 (57%) completed all 3 rounds (Fig 2). Thus, 633 residents (14%) did not complete a single round or took no antimalarials at all. In all, 2,707 residents lived in the 4 control villages in Myanmar and Vietnam where 24-month follow-up was conducted. Of these, 2,185 (81%) completed at least 1 round of MDA: 530 (20%) completed 1 round of antimalarials, 618 (23%) 2 rounds, and 1,037 (38%) all 3 rounds. Thus, 522 residents (19%) did not complete a single round or took no antimalarials at all (S2 Fig). In total, of the 8,749 study participants present during any time of the study period, 5,848 (67%) participated in at least 3 of the 5 possible surveys, and 2,815 (32%) in all 5 surveys (Table S2). The start and end date of each MDA is listed in S2 Table. The participation in follow-up surveys is shown in S3 Table.

Three months (M3) after the first round of drug administrations, the P. falciparum prevalence in the early MDA villages had fallen by 92%, from 5.1% (171/3,340) to 0.4% (12/2,828), while in control villages the P. falciparum prevalence had decreased by 29%, from 7.2% (246/3,405) to 5.1% (155/3,057; difference in differences −2.5%, 95% CI −3.9 to −1.1%, p < 0.001; Fig 3). The P. falciparum prevalence rose steadily over the following 9 months in the villages that received early MDA, to 3.3% (96/2,881) at M12, a 7-fold increase. The P. falciparum prevalence in the control villages, which received deferred MDA, rose from 5.1% (155/3,057) to 6.1% (128/2,101) during the same period, a 20% increase. The P. falciparum incidence after early MDA was 18 infections per 1,000 person-years and in control villages (deferred MDA) was 217 per 1,000 person-years (p < 0.001). Over the following 9 months, the P. falciparum incidence rose in early MDA villages to 142 per 1,000 person-years and in villages with deferred MDA to 262 per 1,000 person-years. There were 10 cases of acute falciparum malaria in intervention villages and 12 in control villages during the surveillance period.

Following the crossover, MDA surveillance continued for a further 12 months in 8 villages in Myanmar and Vietnam (S3 Fig). Following the deferred MDAs in Myanmar and Vietnam, the P. falciparum prevalence fell by 90% over 6 months, from 6.1% (128/2,101) to 0.6% (10/1,575), while in villages that had received early MDA during the previous year (but none in the current year), P. falciparum prevalence nearly doubled over 3 months, from 3.3% (96/2,881) at M12 to 6.1% (90/1,483) at M15. It fell to 3.3% (44/1,350) at M18 and then stabilised between 3.3% (44/1,350) and 3.5% (54/1,549) between M18 and M24. The incidence of P. falciparum infections also fell by 90% in the 6 months after the deferred MDA, from 262 to 27 per 1,000 person-years, while in villages that had received early MDA during the previous year, the P. falciparum incidence rose from 142 to 261 per 1,000 person-years, then dropped to 77 per 1,000 person-years by M24. The overall impact of MDA in reducing the incidence of P. falciparum infections was highly significant. The adjusted IRR was 0.41 (95% CI 0.20 to 0.84) over the 9 months following implementation (Table 2).

The impact of MDA on falciparum malaria varied by country. The greatest impact was in Lao PDR, followed by Cambodia and Myanmar, and there was little effect in Vietnam. This resulted in a country effect variance of 6.82 (p = 0.009; Fig 4). The impact was lower in villages with a baseline P. falciparum prevalence ≤ 5% (adjusted IRR 0.71, 95% CI 0.51 to 0.99) compared to villages with a baseline prevalence > 5% (adjusted IRR 0.13, 95% CI 0.02 to 0.79). The P. falciparum prevalence from 3 to 12 months after the MDA as assessed by uPCR was 0 or close to 0 (<1%) in 4 of 8 villages receiving early MDA and also in 4 of 8 control villages receiving deferred MDA (S4 Fig). The ICC for the incidence of P. falciparum infections for villages (as clusters), accounting for the random effect of country, was in the range from 0.06 to 0.32, estimated at baseline and every 3 months up to M12. The weighted-average ICC was 0.27 over 1 year of follow-up.

In regression models that used village allocation of MDA as the main covariate, male sex, age, and presence of fever were each independently and significantly associated with P. falciparum infections (Table 2, model A). In models that replaced village allocation of MDA with MDA coverage (Table 2, model B), there was a highly significant dose–response relationship between protection and number of completed rounds (IRR 0.63, 95% CI 0.56 to 0.72, p < 0.001). Protection against P. falciparum infection was lowest in people who had not participated in the MDA and was reduced in people who took 1 or 2 doses but did not complete a single 3-dose round. Protection against P. falciparum infection was highest in participants who completed all 3 rounds of the 3-dose regimen. Models in which the observation period included M0 to M12 showed similar exposure–response relationships (S4 Table). In a model including respondents of all ages, the protection for P. falciparum infection increased significantly with regular bednet use. In a model that included only children under 12 years of age, no protection attributable to bednets could be detected (S5 Table). As data for bednet use was missing for 35% (7,801/22,239) of the total observations of participants, this variable was not included in the multivariable analyses.

Before early and deferred MDAs, we identified 269 individuals with P. falciparum infections, of whom 258 (96%) participated in at least 1 round of DP MDA, i.e., received DP on 3 consecutive days (Fig 5). A follow-up blood specimen was obtained from 221 (86%) participants 1 month after MDA. There were 14 (6%) participants whose infections persisted after treatment (13 in Vietnam and 1 in Cambodia), while the remaining 207 (94%) participants had cleared their infection.

Ten of the 269 (4%) P. falciparum infections had the PfPailin genotype [8] (a long haplotype containing PfKelch13 C580Y, conferring artemisinin resistance, and multiple copies of the Pfplasmepsin2/3 genotype conferring piperaquine resistance): 5 of 113 (4%) were in Vietnam and 5 of 8 (63%) were in Cambodia (p < 0.001; Fig 5). All but 1 of the 10 participants with PfPailin cleared their parasitaemia after receiving at least 1 round of MDA (i.e., clearance rate 90%, 95% CI 56% to 100%). One subclinical PfPailin genotype infection in Cambodia persisted after 3 rounds of MDA but by M6 had cleared without further drug treatment.

Following the administration of DP (59,375 individual doses), 121 (0.2%) participants vomited. Of these, 104 vomited shortly after taking the drugs and were offered a repeat dose (Table 3). Data on dizziness and itching were not recorded in Vietnam, but in Cambodia, Lao PDR, and Myanmar, where a total of 26,898 doses of DP were administered, 586 (2.2%) events of dizziness and 12 (0.04%) events of itching were reported following drug administration on day 1. Within 1 month of the MDAs, 1,535 of 8,112 (19%) MDA participants recalled 2,577 AEs, of which 911 (35%) were considered related to the antimalarials; 592 (23%) of the 2,577 AEs were dizziness, 199 (8%) nausea, 96 (4%) vomiting, and 39 (2%) itching, and 1,653 (64%) participants reported a range of other minor complaints. There were no cases of severe haemolysis.

Within 3 months of the MDA, 6 SAEs leading to death or requiring hospitalisation were reported in villages receiving early MDA (6/4,135 participants; 0.15%), whereas in control villages with deferred MDA, 1 SAE (1/2,596 participants; 0.04%) was reported in the same time span (p = 0.187; Table 4). The investigators found none of the SAEs to be related to the administration of study drugs.