Efficacy of a Spatial Repellent for Control of Malaria in Indonesia: A Cluster-Randomized Controlled Trial

Abstract. A cluster-randomized, double-blinded, placebo-controlled trial was conducted to estimate the protective efficacy (PE) of a spatial repellent (SR) against malaria infection in Sumba, Indonesia. Following radical cure in 1,341 children aged ≥ 6 months to ≤ 5 years in 24 clusters, households were given transfluthrin or placebo passive emanators (devices designed to release vaporized chemical). Monthly blood screening and biweekly human-landing mosquito catches were performed during a 10-month baseline (June 2015–March 2016) and a 24-month intervention period (April 2016–April 2018). Screening detected 164 first-time infections and an accumulative total of 459 infections in 667 subjects in placebo-control households, and 134 first-time and 253 accumulative total infections among 665 subjects in active intervention households. The 24-cluster protective effect of 27.7% and 31.3%, for time to first-event and overall (total new) infections, respectively, was not statistically significant. Purportedly, this was due in part to zero to low incidence in some clusters, undermining the ability to detect a protective effect. Subgroup analysis of 19 clusters where at least one infection occurred during baseline showed 33.3% (P-value = 0.083) and 40.9% (P-value = 0.0236, statistically significant at the one-sided 5% significance level) protective effect to first infection and overall infections, respectively. Among 12 moderate- to high-risk clusters, a statistically significant decrease in infection by intervention was detected (60% PE). Primary entomological analysis of impact was inconclusive. Although this study suggests SRs prevent malaria, additional evidence is required to demonstrate the product class provides an operationally feasible and effective means of reducing malaria transmission.


INTRODUCTION 60
It has been nearly 75 years since the role of spatial repellency (deterrency or avoidance) 61 was first described as a potentially beneficial attribute in malaria control, showing chemicals 62 could effectively disrupt normal mosquito behavior and interrupt contact with humans, thus 63 preventing disease transmission without actually killing the insect. [1][2][3] Spatial repellency is used 64 here as a general term to refer to a range of insect behaviors induced by airborne chemicals that 65 ultimately result in a reduction in human-vector contact. This can include movement away from 66 a chemical stimulus, and interference with host detection (attraction-inhibition) and/or feeding 67 response. 4-7 Spatial repellency action can be measured and distinguished from other chemical 68 actions, primarily contact irritancy and toxicity, in the laboratory 8, 9 and in semi-field 69 evaluations. 10, 11 A repellent action drives mosquitoes away from a treated space. Toxics 70 chemical properties can ultimately kill mosquitoes while irritants rely on brief physical contact 71 between mosquito (tarsal segments) and a treated surface to promote excitation and avert 72 prolonged resting. Many compounds can exhibit two or more modes of action and can be 73 distinguished by the concentration or dose exposure required to achieve a specific outcome. 12 74 Currently, the majority of commercial spatial repellent products utilize either low concentrations 75 of short-duration USEPA registered synthetic pyrethroids (pyrethrin, metofluthrin or most most 76 recently transfluthrin 13 ) or botanical-based compounds. 14, 15 77 The World Health Organization (WHO) Vector Control Advisory Group (VCAG) is the 78 WHO advisory group that assesses evidence on the epidemiological effectiveness of new vector 79 control interventions and by doing so supports WHO's development of policy recommendations, 80 including the potential use of spatial repellents as a public health vector control strategy. 16 81 Reviewed by VCAG since 2014, the assessment of rigorous epidemiological evidence for 82 endorsing a policy recommendation of a spatial repellent intervention class remains limited and 83 deemed insufficient. A malaria prevalence study in China evaluating mosquito coils containing 84 0.03% transfluthrin demonstrated a 77% reduction in human Plasmodium falciparum cases 17 85 and the use of coils containing 0.00975% metofluthrin provided 52% protective efficacy (PE) 86 against new (incident) malaria infections in Indonesia. 18 Although primary outcomes from both 87 studies were encouraging, neither met the VCAG requirements of rigor due to study design 19 88 (non-RCT for China) and/or being underpowered. The importance of a WHO policy for 89 implementation of spatial repellents in malaria control programs could dramatically increase 90 investments and efforts by private industry to develop chemicals that operate through different 91 modes of actions other than acute kill. This would potentially introduce a new generation of 92 effective active ingredients and product formulations into the disease control/eradication 93 arsenal. 20,21 In combination with existing WHO-recommended malaria control interventions, 94 spatial repellents have the potential to add significant protective benefit, especially where 95 traditional long-lasting insecticidal nets (LLINs) or indoor residual spraying (IRS) may not be 96 protective, be unavailable, or are impractical and/or feasible. Control or elimination of malaria in 97 these circumstances will require innovative approaches, as example spatial repellents providing a 98 highly protective role against transmission and this may be where spatial repellents may be most 99 beneficial. 22, 23 100 The current claim of a spatial repellent intervention class for public health value is 101 'deployment of a spatial repellent will prevent human-vector contact to reduce pathogen 102 transmission'. With this epidemiological primary endpoint in mind, the current large-scale study 103 aimed to build upon previous epidemiological findings and provide rigorous evidence of a spatial 104 forested hillsides with perennial running streams and small rivers. At the time of the survey, 150 human-landing collections revealed An. subpictus and An. vagus the predominant species in the 151 upland interior locations and An. sundaicus the most common species along the coastal plain. 152 The majority of residents in study villages work in small-holder agriculturalist pursuits, lack 153 public electricity or articulated water supply, and reside predominately in traditional large, 154 thatched-roof homes raised ~1m above ground and averaging 80 m 3 in size with bamboo walls 155 and floor that offer little protection from mosquito entry (Figure 2). 156 157 Sample size. Previous malaria incidence rates collected in a portion of the study area in coastal 158 Kodi Subdistrict 18 were used to estimate the likely malaria attack rate in the current study 159 villages at 0.3 infections/person-year. Sample size determination was based on the hazard rate 160 comparison in the proportional hazards regression model. 28 The required number of first-time 161 infections was estimated at 417 to permit detection of a 30% protective effect by the spatial 162 repellent intervention compared to placebo with 80% power at the type-I error rate, assuming a 163 between-cluster coefficient of variance (CV) of 30% and a baseline hazard rate of 0.3 per person-164 year. With 12 clusters per treatment arm (active or placebo), with 2-month accrual and 22 165 months follow-up, and an estimated 20% loss-to-follow-up (LTFU) during intervention, a total 166 of 54 subjects per cluster was required (n=1,296). 167 168 Study design. The study was a cluster-randomized, double-blind, placebo-controlled, clinical 169 trial with a total of 24 clusters divided into 12 clusters per intervention arm. Clusters were based 170 on housing type, focusing on traditional houses (with or without thatch roofing) (Figure 2), 171 mitigation of potential chemical dispersion 'spillover' effect (i.e., distance between nearest homes 172 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity. is the (which was not peer-reviewed) The copyright holder for this preprint  (Figure 3). 181 Clusters were allocated to receive either active or placebo treatment using a random 182 number generator (https://www.random.org). The cluster allocation code was made available 183 from the intervention manufacturer to the Data Safety Monitoring Board (DSMB) for use in 184 safety assessments. The site database manager assigned a unique identification number to each 185 household (HIN) and the site intervention administrator coordinated distribution of blinded 186 active or placebo to enrolled households within each cluster corresponding to the pre-labeled 187 package code. Unblinded assignments were shared with a site administrator in a sealed envelope 188 placed in a secure location within the managing center of the research project (Jakarta) for 189 purposes of emergency unblinding related to adverse and/or serious adverse events. Thus, the 190 investigators, research team, study subjects, and residents were blinded as to which cluster 191 received active versus placebo devices until after completion of the study. 192 The primary endpoint for estimating the protective efficacy (PE) of the spatial repellent 193 intervention was malaria incidence (time to first infection) among the enrolled sentinel residents 194 study participants in each of the 24 clusters, i.e., 1,332 subjects in all. These sentinel subjects, 195 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint Intervention. Immediately after completion of radical cure, intervention was simultaneously 238 initiated in all subject households and non-subject households that consented to receive 239 intervention. The spatial repellent intervention is a transfluthrin-based passive emanator 240 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint produced by S.C. Johnson & Son, Inc (SCJ) designed to release a volatile chemical into the air 241 and prevent human-vector contact within the treated space. Transfluthrin is a registered 242 compound commonly found in commercially available mosquito coils globally based on WHO 243 specifications. 33 Emanators (active and placebo) of identical packaging and color were 244 distributed by study personnel every 2 weeks at the individual household application rate of 2 245 units / 9m 2 according to SCJ specifications. There was a median of 1 room (range 1,9)  Captured mosquitoes were immediately killed by organic compound vapor in the field and 282 initially identified to species (or species complex) using morphological characters. 34 All 283 specimens were transported to an on-site study base laboratory upon completion of the 12 h 284 collection and a random sample of representative anopheline species (up to 30 per cluster and 285 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity. is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint hour of collection) were dissected for parity and scored as either gravid/parous or nulliparous. 35 286 Partial (head-thorax for those dissected for parity) or whole anopheline specimens were placed 287 singly into individual Eppendorf® 1.5ml vials and stored over silica gel desiccant until further 288 processing at EIMB, Jakarta, for detection of malaria sporozoites and molecular-based species 289 identification, 36, 37 where applicable. Mosquito samples were evaluated for Plasmodium spp. period, all chemical and control specimens were stored individually over silica gel for analysis at 304 EIMB to confirm species identification and for detection of target-site mechanisms (e.g., kdr 305 mutations) of resistance. 306 307 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint Data management and verification. Data collection was designed around a tablet-based survey 308 platform linked to a custom-built database and web portal. CommCare (Dimagi Inc, MA) was 309 selected as the frontend form application, providing critical capabilities, such as: a parent-child 310 case structure, the ability to store forms when offline, update form versions after deployment, 311 build forms with complex logic in a web browser, and export form data to other tools. Data was 312 cleaned according to rules specified in the study protocol to ensure data integrity. Study data 313 related to participating subjects, participating households, intervention (placement/replacement), 314 collected mosquitos, and lab analyses was cross checked, identifying missing, incomplete, or 315 suspect data submissions. These were sent back to the site data manager to be resubmitted or 316 corrected. Once data correction was complete, data was verified and requested for analyses. 317 318 Statistical analyses (S1. Statistical Analyses Plan). Subjects were excluded from analyses when 319 a subject had no blood sample taken during the intervention period or when a household might 320 have contributed more than two subjects (rarely) when there was LTFU on the first recruited 321 subject and a 2 nd subject was recruited as a replacement. In the latter case, only the subject with a 322 longer follow-up period was used based on per-protocol allowing only one subject per household. 323 The primary hypothesis on PE against first-time malaria infection was tested by comparing 324 the hazard rates of the first-time malaria infection between active and placebo (blank) intervention. 325 The complementary log-log (cloglog) regression model 326 was. [42][43][44][45] is the discrete time hazard rate of subject in cluster I at time t, contains visit (as 327 a categorical predictor), the individual-level (age, gender), household-level (number of doors, open 328 eaves Y or N, wall type), and cluster-level (baseline incidence rate, cluster population size, 329 intervention group) covariates, and is the cluster-level random effect. PE was 330 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint estimated by ( 100% with a 90% confidence interval (CI) based on the Wald test, 331 where is the estimated regression coefficient associated with the intervention group, and 332 is the estimated hazard ratio (HR) between active and placebo. The null hypothesis of PE is 0 was 333 tested by the Wald's test , where is the estimated standard error of . It was concluded 334 that active intervention reduces the first-time malaria hazard rate compared to placebo if 335 ; otherwise, the study would not have enough evidence to suggest that active 336 intervention reduces the first-time malaria hazard rate compared to placebo at a one-sided 337 significance level of 5%. The Kaplan-Meier (KM) curves on first-time malaria infection per cluster 338 was provided for the active and placebo arms, respectively. 339 The statistical method for analyzing the secondary endpoint of the overall (total new) 340 malaria infections detected in study subjects was similar to that used for analyzing the primary 341 endpoint of the first-time malaria infection except that the above cloglog model has additional term 342 the random effect at the individual level to account for the dependence among multiple 343 malaria infections per individual. The same set of analyses as the above were performed in a 344 subgroup analysis in clusters with non-zero baseline incidence rates in clusters used for 345 entomology data collection. 346 The effects of active intervention on the secondary entomological endpoints were 347 estimated by applying the negative binomial regression model, if applicable. Specifically, the 348 anopheline landing rate (surrogate 'bite' based on HLC and indicator of human-vector contact) is 349 defined as the number of mosquitos caught during the 12-hr interval overnight. The covariates in 350 the model for analyzing the landing rate includes the fixed effects of intervention group, the 351 interaction between treatment and location of collection (inside or outside), visit (as categorical), 352 baseline incidence rate, baseline vector count, cluster population, and random effects for household 353 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Trial outcomes show baseline covariates regarding subject, house construction, population and 372 baseline malaria incidence between the active and placebo arms were balanced at the individual-, 373 household-, and cluster-level ( Table 2). The intervention coverage rate, defined as the proportion 374 of actually placed emanators over the total number required per household, ranged from 82.2% 375 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint to 98.6% by cluster over the entire intervention period, with a mean application rate of 93.2% 376 and 92.3% for the active and placebo arms, respectively. 377 There were 134 and 164 first-time infections in active and placebo treatment arms, 378 respectively, with a hazard ratio of 0.72 (90% CI: 0.43, 1.21) ( Table 3). A 27.7% decrease in the 379 first-time malaria hazard risk was estimated using active compared with placebo (90% CI: -380 21.3%, 56.9%). The 27.7% PE was not statistically significant at the 5% one-sided significance 381 level. The estimated PE of active intervention against overall (total) malaria infections (first and 382 subsequent) was 31.3% (90% CI: -10.8, 57.4%) with a one-sided p-value = 0.098 (Table 3).  (Table 3). 395 396 Subgroup analyses of protective effect against malaria infection among non-zero baseline 397 incidence and entomology clusters (not pre-planned). 398 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint Among the 24 clusters, there were five clusters with zero baseline incidence which had zero to 399 very low incidence rates during intervention. The first subgroup analysis used the 19 clusters 400 with non-zero baseline incidence rate as the remaining five clusters would have been excluded 401 from the intervention if baseline analysis had been performed before the randomization. 402 Excluding these five clusters, the estimated PE for overall (total new) infections in the remaining 403 subgroup of 19 clusters with non-zero baseline incidence rates was 40.9% (90% CI: 8.61, 61.8%) 404 resulting in a one-sided p-value of 0.0236 ( Table 3). The second subgroup analyses included 405 incidence from the 12 clusters having entomology collection data (i.e., mean anopheline landing 406 rate), and where the average baseline incidence was ~ 4 folds greater than the other clusters. The 407 PE using active intervention against time to first-event and overall (total new) malaria infections 408 in this subgroup was 55.3% and 66%, respectively ( Table 3) Table 4). The cumulative indoor and 420 outdoor anopheline biting rates are shown in Figure 6. There was a numerical reduction in 421 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint indoor and outdoor landing density from collections performed at sentinel households containing 422 active intervention as compared with those assigned to placebos. This difference resulted in 423 16.4% and 11.3% reduction in anopheline attack rate on collectors positioned indoors and 424 outdoors, respectively, at sentinel households with active intervention compared with placebo 425 houses, although this reduction was not statistically significant ( Table 5).  (Table 7). 442 443 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Adverse Events (AE) and Severe Adverse Events (SAE). 459
A total of 523 and 144 AEs were reported from non-participants (using government clinic health 460 records) and the intervention study cohort (collected by the study team), respectively. General 461 respiratory complaints were most common, followed by general fever. There were 6 total SAEs 462 reported in the study cohort during follow-up: 2 deaths of unknown cause; 1 case suspected brain underpowered. This study highlights three primary challenges for consideration in future spatial 488 repellent trials as well for new vector control intervention classes more broadly. Firstly, having 489 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint 'adaptive' study designs, especially for evaluation in low to moderate malaria transmission 490 settings and settings with large cluster-to-cluster variance on transmission intensity; second, 491 defining and identification of the 'key' entomological correlates of protection; and third, 492 ensuring reliability and feasibility in AE/SAE reporting for accurate safety assessment. 493 Regarding adaptive study designs, our assumptions for power and sample size 494 calculations were based on a previous proof-of-concept study that was spatially very near 495 (border) of the current study area 18 and study villages were selected with the assumption that all 496 clusters would experience malaria transmission during the intervention period, which, 497 unfortunately was not the case. Baseline data indicated zero-incidence for five of the 24 selected 498 study clusters with a 5-folds higher between-cluster variability than assumed in the original 499 sample size calculation. The 5 clusters with zero baseline incidence (two allocated to receive 500 active and three allocated to receive placebo) also had either zero or low incidence rates during 501 intervention and ranked in the top 8 among all 24 clusters in terms of lowness in intervention 502 incidence (data not shown). In retrospect, a pre-planned analysis of the baseline data would have 503 allowed adjustment of the study design by filtering out the zero-incidence clusters, or increasing 504 the study duration (time of follow-up), the number of subjects recruited per cluster, or the 505 number of clusters to better satisfy requirements of power and capture the necessary number of 506 outcome events (infections). However, given the updated baseline incidence rate of 0.131 per 507 person-year and 97.1% CV, to maintain 80% power with the originally assumed 30% PE, 100 508 clusters per arm with 144 households (HHs; i.e. subjects) would have been required to be 509 recruited to collect 5,550 first-time malaria events -an unmanageable scale that could not have 510 been supported due to geographical, logistical and funding constraints. Moreover, increasing 511 the follow-up time to collect more first-time malaria events would have neither counterbalanced 512 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint the large variability nor the longer duration required to collect the additional events due to the 513 seasonally-influenced low to moderate malaria transmission among clusters. For this reason, 514 when planning future trials in low or moderate endemic transmission settings, investigators 515 should consider a greater number of clusters per arm and/or building into the statistical analyses 516 plan an interim and final analyses of baseline incidence with pre-determined study design 517 adaptations, to include down-selection of clusters with predetermined incidence thresholds. Early 518 exercises of the impact of a lower than assumed incidence (or greater CV) during study planning 519 should be explored amongst investigators, industry and funding partners to ensure that the study 520 area context, intervention manufacturing, program period and/or funding can sufficiently meet 521 the demands of power requirements if adjustments are to be made once the study 522 begins. Stakeholders should also discuss supporting and adopting adaptive designs so to allow 523 decision-making after planned interim analysis of intervention data, to either stop the trial for 524 futility or to continue the trial with adjusted design parameters, such as sample size. 48,49 Perhaps 525 just as important in the context of vector control, although RCTs are rated as high-quality 526 evidence, 50 considerations to RCT alternatives are prudent in the trial planning phase as these 527 may offer assurances of adequate data rigor while balancing cost and time constraints of 528 traditional RCTs. 51, 52 Alternatives study designs include large observational studies for 529 detecting population-level effects using analytical cross-sectional studies, or operational 530 program-based evidence; the latter perhaps especially for interventions containing an existing 531 registered chemical active ingredient (i.e., meets human safety thresholds) and where the 532 intervention is implemented in pilot trials and impact monitored through case reports, as 533 compared to a contemporaneous control group. 52 534 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint The inclusion of entomological endpoints in the Sumba RCT was, in part, to help 535 understand and validate the intervention's mode of action for the VCAG claim of a health impact 536 through a reduction in human-vector contact. Spatial repellent chemicals may cause initial 537 knockdown and direct kill by exposure to toxic doses at close range to the active ingredient, or a 538 delayed kill or behavioral avoidance response, through exposure to sub-lethal doses at distances 539 further away from the stimulus source. Therefore, entomological measurements for detecting 540 reductions in human landing density with active intervention, and a possible change in 541 anopheline age structure (parity rate) indicating older populations not surviving as long as in 542 placebo areas, was built into this study. Additionally, a reduction in sporozoite infection rates 543 because of lower blood feeding success, and/or an inability to survive the required time interval 544 (parasite incubation period) to become infective to a susceptible host represent other endpoint 545 measures of impact on the vector. A causal relationship with one, several or all, of these 546 endpoints would allow future trials evaluating non-inferiority of a 'second-in-class' spatial 547 repellent to integrate entomological measures only to predict protective efficacy and provide 548 assurances of meeting minimum thresholds of acceptance for public health use. 549 Sporozoite positivity rates and EIR estimates from the current trial were low, with a total 550 of 42 of 11,650 (0.36%) anophelines tested during baseline (10mo) and 19 of 17,971 (0.11%) 551 found infected during intervention (24mo). These findings are not unusual or unexpected in low 552 to moderate malaria transmission settings. 25 As example, the previous Sumba study (using 553 metofluthrin coils) reported just 15 out of 1,825 (0.82%) HLC anopheline samples sporozoite 554 infected. 18 The challenges for assessing sporozoite infectivity and using EIR as a measure of 555 intervention effect in such settings should be factored into study pre-planning to carefully 556 balance cost of sample processing with potential information gained. 557 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint These study findings showed a 16.4% reduction in indoor landing mosquitoes at sentinel 558 collection houses with active treatment compared to placebo-control. This result differs from the 559 previous study on Sumba evaluating a metofluthrin-active coil where a significant 32% reduction 560 in An. sundaicus indoor landing rates was found compared to blank coil control. 18 In the previous 561 and smaller study, An. sundaicus was the overwhelmingly predominate human-feeding 562 anopheline; whereas in the current study this species was relatively uncommon during the 563 intervention trial. The greater complexity of anopheline diversity in the current study area may 564 have contributed to the discordant findings between the two studies. Specifically, the range of 565 ecologies (coastal plains to upland forested hills) varied greatly across the 12 clusters used for 566 entomology collections in this trial compared to that of the earlier Sumba study where collections 567 were confined to only 4 adjoining clusters along similar coastal habitats. This variability in 568 ecology witnessed in the current study resulted in a broad range of species diversity (19 species, 569 group species captured), each with unique habitat requirements, bionomic and behavioral 570 characteristics (e.g., biting habits, host preferences). Similar results may occur in future trials that 571 are conducted in settings with diverse vector populations and/or as cluster size increases thereby 572 increasing the probability of greater habitat diversity. Other potential reasons for differences in 573 the effect on anopheline landing rates between the two studies on Sumba include intervention 574 formulation (i.e., coil vs. passive emanator) as well as species-specific effects by active 575 ingredients (i.e., metofluthrin vs. transfluthrin). Perhaps just as important as measuring a 576 reduction in indoor landing rates was the ability to demonstrate an effect on outdoor landing 577 rates (11% reduction) -on the exposed verandah of houses. Measuring an apparent effect on 578 outdoor landing is encouraging as many Sumba residents often sleep on the verandah without the 579 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint protection of a bed net. Results from further data analyses on effects of this intervention on non-580 anopheline populations are forthcoming. 581 Pertaining to monitoring of insecticide resistance, this labor-intensive activity was 582 integrated into the study to characterize the wild-type anopheline populations and monitor 583 changes in susceptibility due to continual exposure of the transfluthrin-active intervention. 584 Although we are confident in the data presented, there were limitations in sampling from 585 multiple sites with multiple species during pre-, within, and post-intervention periods to obtain a 586 clear ("useful") profile of susceptibility background and shift in phenotypic (and molecular) 587 frequencies. Future trials requiring assessment of insecticide resistance should allow careful 588 planning and budgeting. 589 The third 'lesson-learned' from the current study relates to AE and SAE monitoring. 590 Reports of intervention safety in this study should take into account possible limitations of the 591 data collection. The DSMB was provided with clinic attendance data from January 2015 to 592 December 2018, for upper respiratory infection (URI), pneumonia, and malaria, as classified by 593 month, health center, and village within health center coverage. Although not gathered as part of 594 the study, the DSMB was keenly interested in these data because of the surprisingly low number 595 of illnesses/deaths reported from the study population during the entire 24-month intervention 596 period. For some periods, particularly for malaria infection, the case data were missing or the 597 health center totals were not available by village. Of the 13 villages represented in the dataset, 598 all but four comprised clusters (or parts of clusters) from both study intervention arms (i.e., some 599 allocated to active, and others to placebo); therefore, it was not possible to infer the allocation 600 status of the corresponding cases. For this reason, the DSMB was unable to make a reliable 601 assessment of any possible association between active intervention and clinic attendance for 602 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint those health conditions reported. The respiratory illnesses in the clinic data are more reflective 603 of the magnitude of numbers of cases one would expect from this population; however, the 604 DSMB was again unable to parse them into test and control clusters due to reasons stated above, 605 and therefore the data was of little use for monitoring purposes. The number of SAEs reported, a 606 total of only six, is well below the expected in a study population of this size. The WHO 607 estimates the crude death rate in Indonesia to be approximately 6.2 per thousand per year. 53 608 Based on this, for a population of 1,296 children enrolled (the sample size calculation in the 609 protocol), the probability of having zero deaths in a year is less than one in 1,000. Although open 610 commemoration of deaths is commonly practiced in the study area, it seems apparent that they 611 were not completely reported/recorded as SAEs. 612 Overall, the point to apply for future spatial repellent trials (and perhaps other trials of 613 new vector control interventions) is to improve the mechanism for capturing adverse events 614 before the initiation of the study such that all deaths and hospitalizations are reliably reported 615 and regular monitoring in place for unusual numbers of complaints. If these trials are regarded as 616 a clinical trial with a placebo that requires comparative AE an SAE data, then almost anything at 617 this scale could fail simply on the mass of data, however imperfectly collected. Any safety 618 signal detected would most likely be due to bias in reporting or collection and would have little 619 to do with the investigational intervention. Trials evaluating a spatial repellent intervention with 620 a currently registered active ingredient (i.e., a chemical meets regulatory approvals for 621 acceptable levels of human risk) such as transfluthrin, could focus on a small number of 622 complaints that might be expected due to inhalation (i.e., volatilization) and develop a 623 monitoring scheme that collects consistent data across the population. Clinic-based complaints 624 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint of respiratory illness such as pneumonia and asthma might be possible, but the variability in their 625 collection (clinic or survey) will likely be highly dependent on study site infrastructure. 626 In conclusion, while more evidence will be required to determine whether spatial 627 repellents can serve as a viable malaria control intervention, both the primary and secondary 628 results of this Sumba Island trial have generated valuable data and observations that can 629 contribute to the overall assessment and improvement of testing protocols of a spatial repellent 630 intervention class. The VCAG has recommended that data from at least one additional trial 631 evaluating the spatial repellent be generated , 54 and once available, the panel will be able to 632 assess the available evidence for judging public health value. If the crude estimate of PE shown 633 here, near 30%, is replicated in future statistically robust cluster-randomized trials, the 634 intervention prototype evaluated in this RCT would approximate the benefit associated with 635 LLINs. 55 Perhaps just as important for endemic countries, the difference in the overall (total 636 new) infection rate between transfluthrin-active and placebo-control -the number of cases 637 averted -indicates 361 expected cases per 1000 persons per year (0.361*1000*1) without the 638 spatial repellent and 152 less cases expected when using the active intervention ((0.361-639 0.209)*1000*1). The cost-savings of averting these 152 less cases per 1000 person-years is an 640 important consideration for health-systems strengthening. These results have encouraged further 641 and substantial investment to validate spatial repellent efficacy through larger RCTs, including 642 an investigation of possible diversionary effects on human health (i.e., greater than expected 643 malaria incidence in households near intervention not receiving product), and evaluation of the 644 optimal delivery systems for humanitarian assistance use-case scenarios. 56 645 646 . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint Acknowledgments: We extend sincere appreciation to DSMB members: Dennis Shanks, Chris 647 Drakeley, and Neal Alexander for assurances on data integrity and safety assessments. We also 648 express our gratitude to fhi360 partners Kathy Bowerman, Ahmed Goolam and Zen Hafy for 649 trial oversight and monitoring of protocol compliance. We thank the WHO VCAG for their 650 comments from recurrent project data assessments. We are grateful for the tireless efforts of CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint  . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint

A B
. CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint Figure 3. Location of 24 study clusters. Clusters were selected for enrolling the incidence cohort, each consisted of ca. 100 households with an average distance of 500m between clusters. A total of 48 sentinel houses from 12 clusters were selected for routine entomological collections.
. CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint  . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint

A B
. CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint  . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint Table 3. Summary first-time and overall (total) malaria incidence.

First-time infection
Overall (total new) infection . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint (0.02%) *Misidentification possible depending on physical condition of specimen, molecular methods required to separate out some members in species groups and complexes.
. CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/19003426 doi: medRxiv preprint