The study was performed within the Kenya Medical Research Institute (KEMRI) and Centers for Disease Control and Prevention (CDC) Health and Demographic Surveillance System (HDSS) in Siaya County, Kenya [8, 9]. Malaria transmission is high and perennial with peak prevalence during May–July and November–December, following the long and short rainy seasons, respectively. In July 2012, the population prevalence of malaria was 30.6% by microscopy, and 80.2% by 18S-nucleic acid sequence–based amplification [10]. In 2013, 55% of individuals with microscopically confirmed malaria infections reported being afebrile in the preceding 2 weeks, and increased with age to > 90% [11].

Following the 2014 Ministry of Health’s long-lasting insecticidal net (LLIN) distribution, 54.4% of households had access (1 LLIN for every 2 household inhabitants) to LLINs [12]; indoor residual spraying has never been conducted programmatically in this area. Artemether-lumefantrine was scaled up as the first-line antimalarial in 2006 [13], and while community case management of malaria was initially implemented in 2013 [14], in 2015, only 3.6% of febrile children aged < 5 years who sought care did so from a community health volunteer (CHV) [12].

A detailed description of the study procedures and methodology has been published [9]. In brief, 10 health facilities in the HDSS were purposively selected and adjacent villages within 3 km of each facility were grouped into 3 clusters that were randomly assigned to intervention, control, or future intervention. We decided not to implement a future intervention and the third cluster was merged with the control cluster for a total of 10 clusters per arm (Figure 1). To reduce the impact of parasite migration on the analyses, only individuals residing in compounds within a core area of each cluster, defined as ≥ 300 m from the cluster perimeter, were considered for sampling [15].

Six rounds of MTaT were performed in the intervention clusters between September 2013 and April 2015 (Supplementary Figure 1). The selection of the number and timing of rounds was informed by a mathematical model [9]. During MTaT rounds, CHVs visited every household in intervention clusters until they had tested each household member ≥ 1 month of age by RDT (Carestart Malaria HRP-2/pLDH [Pf/PAN] Combo Test RDT; Somerset, New Jersey), or made 3 attempts to do so. RDTs were used for MTaT rounds as point-of-care tests are needed for treatment decisions. This RDT was selected because it was the RDT procured and distributed by the Kenya Ministry of Health for use in public facilities and by CHVs at this time, received a positive recommendation by the World Health Organization (WHO), and had a sensitivity of 95%–99% and 99%–100% during WHO testing in samples with 200 parasites/μL and 2000 parasites/μL, respectively [16]. Those positive by RDT were treated with dihydroartemisinin-piperaquine (Eurartesim, Sigma-Tau, Pomezia, Italy; or Duo-Cotecxin, Holley-Cotec, China), selected due to its prolonged posttreatment prophylaxis window, or according to the study algorithm [9]. Dried blood spots were prepared on filter paper for future real time quantitative polymerase chain reaction (qPCR). This manuscript describes the impact of MTaT on malaria prevalence; the impact on malaria incidence is described elsewhere [17].

Population coverage of MTaT, adherence to treatment, and in- and out-migration by round have been published previously [17] and are presented in Supplementary Figure 2. In brief, MTaT coverage was defined as follows: Number of Individuals Tested in Round(Individuals previously living in intervention cluster)+(Individuals migrated in since previous round)−(died+migrated out)×100% Coverage ranged between 75.0% – 77.5% during year 1 rounds and increased to 81.9%–94.3% in year 2. Test positivity rate across the 6 rounds ranged from 35.6% to 48.6%, and self-reported adherence to treatment courses ranged from 91.5% to 95.4%. In- and out-migration were measured in each household at each round and ranged between 25.1% and 35.9%.

We estimated the number of infections missed by MTaT during rounds due to the limit of detection (LoD) of RDTs (compared to qPCR) and incomplete intervention coverage using previously published data [17]. The equations, assumptions, and results are presented in the Supplementary Methods and Supplementary Table.

The protocol was approved by the KEMRI institutional review board (IRB), the CDC IRB relied on KEMRI for approval, and the Kenya Pharmacy and Poisons Board approved the protocol and importation of Eurartesim. The trial was retrospectively registered at ClinicalTrials.gov (NCT02987270). Written informed consent was obtained from adult participants and parents/guardians of participating children. Additionally, written informed assent was sought for children 13–17 years of age.

The sample size was calculated using Bennett and Hayes’s [18] formula for RCTs assuming a malaria infection prevalence of 40% in the control arm, a type I error rate of 5%, and 80% power to detect a relative difference in malaria prevalence of 50% between arms in the final cross-sectional study. A coefficient of variation of 0.3 for between-cluster and compound variance was used. We performed a simple random sample of compounds, sampling all constituents. Assuming an average compound constituency of 4.5 individuals, we selected 20 compounds per cluster to attain our calculated sample size.

Cross-sectional studies were performed annually at peak malaria transmission seasons in July prior to the first round of MTaT in September 2013, and then 2 months after the completion of the last of 3 MTaT rounds in years 1 and 2. CHVs visited the randomly selected compounds from the core areas of each cluster, enrolled all residents ≥ 1 month of age, and administered a questionnaire to each participant or their caregiver. Excepting children ≤ 4 months of age, only data from individuals having lived in the study area for at least 4 months (usual residents) were included in the analyses. Global Positioning System geocoordinates were collected for each compound. CHVs collected a blood sample for RDT to prepare a thick and thin blood smear for microscopy, and RDT-positive individuals were treated according to an algorithm that incorporated age, pregnancy status, and history of drug reaction [9].

Preparation and examination of blood smears are described in detail elsewhere [9]. In brief, all blood smears were read by 2 microscopists who were blinded to study arm; discordant reads were evaluated by a third blinded microscopist. The procedures for qPCR × have been described elsewhere [19].

Prevalence estimates and 95% confidence intervals (CIs) accounting for clustering at the level of the health facility were calculated using Taylor series linearization [20]. Socioeconomic status (SES) wealth quintiles were assigned using multiple correspondence analysis models from data of household assets collected during cross-sectional surveys [21]. Values for enhanced vegetative index (eVI) were accessed for the 3 months preceding each survey [22] using compound geocoordinates; values associated with the best model fit were assigned to each observation.

The primary outcome was the all-ages effect size of MTaT on P. falciparum infection prevalence by microscopy from the cross-sectional studies between arms for each year with the baseline serving as the reference. The effect sizes were calculated as adjusted ratio of prevalence ratios (aRPRs) with 95% CIs of the exponentiated parameter estimates of the interaction between study arm and year by a log-binomial model using generalized estimating equations to account for clustering at the health facility level [23]. The aRPRs represent a ratio in the change in prevalence from baseline for each of the 2 study arms. A 3-way interaction term between reported net use, study arm, and year was assessed for effect modification. In the absence of evidence of this (P > .05), reported net use was included as a variable in the model along with age (categorized as < 5 years, 5–14 years inclusive, and ≥ 15 years), SES, and eVI [24].

As a planned secondary analysis, we performed age-stratified analyses of the primary outcome, and post hoc secondary analyses of clinical malaria prevalence, defined as microscopically confirmed malaria in the presence of axillary fever ≥ 37.5°C or history of fever in the previous 2 weeks, and clinical malaria as a proportion of all with malaria infection.

Demographic characteristics, reported history of fever and LLIN use, SES, and eVI were similar between arms at each round (Table 1). Sex and age structures of the sampled population were similar to those of the overall HDSS population in 2012 [25]. Approximately 55% of the population was female, 15% were aged < 5 years, 33% were aged 5–14 years, and 52% were aged ≥ 15 years. Reported LLIN use increased significantly in both arms in year 2 to 87.5% (95% CI, 82.4%–92.6%) vs 87.0% (95% CI, 82.1%–91.9%), after the Ministry of Health sponsored mass LLIN distribution. Reported LLIN use was lowest in those aged 5–14 years and highest in those aged ≥ 15 years.

The effect size of MTaT on the primary outcome of all-age malaria microscopy prevalence was nonsignificant after year 1 (aRPR, 0.93 [95% CI, .79–1.1]) and year 2 (aRPR, 0.92 [95% CI, .76–1.1]) (Figure 3). Though the study was not powered for age-stratified evaluations, there was a consistent, though not statistically significant, protective effect of MTaT in the age group 5–14 years (0.85 [95% CI, .68–1.07] and 0.80 [95% CI, .63–1.02] after year 1 and year 2, respectively).

The effect size of MTaT on the prevalence of clinical malaria was not significant. There was a significant reduction in the proportion of individuals with clinical malaria among those infected with malaria between year 1 and baseline (0.81 [95% CI, .66–.99]); however, there was no effect after 2 years (Figure 3).

The total number of individuals tested per round ranged between 23 226 and 26 342. We estimated that 12.6%–19.6% and 5.7%–25.0% of the infections were missed due to the LoD of RDTs as compared to qPCR among those tested and due to incomplete coverage by round, respectively (Supplementary Methods and Supplementary Table). Combining these, we estimate that 24.2%–36.9% of all of the infections were missed per round.

Missed infections are primarily due to the LoD of the diagnostic test used and incomplete coverage during rounds [29]. We estimate that we missed a total of 24.2%–36.9% of all circulating infections in each round; 12.6%–19.6% due to the LoD of RDTs, and 5.7%–25.0% of all infections due to incomplete intervention coverage (Table 1). A minimal cutoff of 80% coverage is suggested for effective rounds [7], which we did not achieve until year 2. However, despite coverage ranging from 81.9% to 94.3% in year 2, there was no evidence of increased efficacy; the aRPRs in year 1 and year 2 to baseline were 0.93 (95% CI, .79–1.08) and 0.92 (95% CI, .76–1.10), respectively. MTaT with ultrasensitive RDTs, which were not available at the time of this study, likely would have reduced the number of missed infections. However, it may be that in areas with high parasite reproductive rates, higher coverage levels with MDA, which in addition to treating all reached infections provides a chemoprophylactic effect on all treated, may be necessary to effectively reduce transmission [30].

We attempted to limit the impact of parasite migration on the analyses by selecting compounds from cluster core areas. Epidemiological [31] and entomological [32] data from our study area that indirectly demonstrated the mass effect of a community-based intervention (LLINs) extended to approximately 300-m informed our choice of distance; this distance may have been insufficient.

Additionally, modeling studies have concluded that parasite migration through human mobility is an important factor toward the success of MDA [33]. We found that an average of 31% of the population in our clusters migrated in or out between each round, a large proportion of which likely carried parasites into intervention arms. Additionally, individuals may have been exposed to infectious bites through daily commuting activities to a market, place of work, or school outside the cluster of residence as increased vector biting has been documented in this area in the early evening and late mornings when individuals are unlikely to be under bednets [34–36]. These exposures were unmeasured in our study. Our cluster size (3 villages in the intervention arms) and buffer area may not have been large enough to minimize the impact from these events, and may partially explain the differing results from our trial and one performed in a moderate- to high-transmission setting in Zambia, which found a significant impact of a single year of 3 rounds of MTaT on the prevalence of malaria in children < 5 years of age (adjusted odds ratio [aOR], 0.47 [95% CI, .24–.90]) [37]. There, cluster sizes were much larger (2–3 health facility catchment areas), and rather than a 300-m buffer zone, they had a 5-km buffer [37].

The effect size of MTaT on infection prevalence and clinical malaria did not change. These findings were corroborated with those from the incidence cohort (incidence rate ratio [IRR], 0.95 [95% CI, .87–1.04]) and from passive surveillance of clinical malaria at facilities (IRR, 0.79 [95% CI, .61–1.02]) [17]. However, clinical malaria was transiently reduced after the dry season (after rounds 2 and 5; IRR, 0.73 [95% CI, .54–.98] and 0.66 [95% CI, .49–.87], respectively) [17], supporting the importance of timing of rounds in relation to malaria seasonality. Additionally, after year 1 of MTaT, individuals with malaria infection were less likely to report febrile events within the previous 2 weeks (aRPR, 0.81 [95% CI, .66–.99]). While this may be interpreted as MTaT impacting the clinical presentation of malaria, it is difficult to make any conclusions as this finding did not persist after year 1.

We did not power our trial to assess age-stratified effects of MTaT; however, there was a consistent nonsignificant protective point estimate of MTaT (aRPR, 0.80 and 0.85) in the 5–14 age category after each year. This age category harbors the highest prevalence of infection and is the least likely to have clinical malaria and thus seek care when infected, and least likely to report LLIN use the previous night. While a trial of MTaT in Kenya among school-aged children showed a nonsignificant reduction on malaria parasitemia after 12 months of follow-up (aOR, 0.76 [95% CI, .46–1.11]), the authors suggest that this may have been the result of the intervention only being carried out in 2 classes within the school [38]. Our findings suggest that an active approach, rather than an intervention predicated on consistent and repeated behavioral patterns by the end user (such as LLIN use), may be effective in reducing malaria in this age category and could be trialed.

In summary, MTaT utilizing traditional RDTs performed 3 times per year for 2 years in an area of high transmission was not efficacious in reducing the prevalence of malaria infection. This is likely due to several factors including missed infections and the impact of human movement on parasite migration.