Targeted malaria elimination interventions reduce Plasmodium falciparum infections up to 3 kilometers away

Malaria elimination interventions in low-transmission settings aim to extinguish hot spots and prevent transmission to nearby areas. In malaria elimination settings, the World Health Organization recommends reactive, focal interventions targeted to the area near malaria cases shortly after they are detected. A key question is whether these interventions reduce transmission to nearby uninfected or asymptomatic individuals who did not receive interventions. Here, we measured direct effects (among intervention recipients) and spillover effects (among non-recipients) of reactive, focal interventions delivered within 500m of confirmed malaria index cases in a cluster-randomized trial in Namibia. The trial delivered malaria chemoprevention (artemether lumefantrine) and vector control (indoor residual spraying with Actellic) separately and in combination using a factorial design. We compared incidence, infection prevalence, and seroprevalence between study arms among intervention recipients (direct effects) and non-recipients (spillover effects) up to 3 km away from index cases. We calculated incremental cost-effectiveness ratios accounting for spillover effects. The combined chemoprevention and vector control intervention produced direct effects and spillover effects. In the primary analysis among non-recipients within 1 km from index cases, the combined intervention reduced malaria incidence by 43% (95% CI 20%, 59%). In secondary analyses among non-recipients 500m-3 km from interventions, the combined intervention reduced infection by 79% (6%, 95%) and seroprevalence 34% (20%, 45%). Accounting for spillover effects increased the cost-effectiveness of the combined intervention by 37%. Our findings provide the first evidence that targeting hot spots with combined chemoprevention and vector control interventions can indirectly benefit non-recipients up to 3 km away.


Figure S2. Spillover effect estimates on cumulative incidence within subgroups
Cumulative incidence ratios estimated with hierarchical TMLE; outcome models were fit with cohort-level data.Models were adjusted for covariates that were screened separately for each model using a likelihood ratio test.Models for rfMDA + RAVC vs. RACD were unadjusted due to data sparsity.Confidence intervals account for cohort overlap.For rfMDA and RACD arms, the analysis includes the period from 0-35 days following index case detection for direct effects and 21-56 days for spillover effects.For rfMDA+RAVC and RAVC only arms, the analysis includes the period from 0-6 months following index case detection for direct effects and 17 days to 6 months for spillover effects.Total effects analyses include the person-time for the direct effects and spillover effects analyses.Direct effect includes treated in target zone.Spillover effect includes intervention non-recipients up to 1km from an index case.Total effect includes all individuals (intervention recipients and non-recipients) up to 1km from index case.For the human intervention, confidence interval upper bounds were truncated at 16 for above median distance to the nearest health facility (observed value: 23).

Figure S3. Sensitivity analyses for spillover effects on cumulative incidence of malaria with different distance radii
For rfMDA and RACD arms, the primary analysis includes the period from 0-35 days following index case detection for direct effects and 21-56 days for spillover effects; the alternative observation period analysis includes the period from 0-21 days following index case detection for direct effects and 21 to 42 days for spillover effects.For rfMDA+RAVC and RAVC only arms, the primary analysis includes the period from 0-6 months following index case detection for direct effects and 17 days to 6 months for spillover effects; the alternative observation period analysis includes the period from 0-7 days following index case detection for direct effects and 17 to 90 days for spillover effects.Total effects analyses include the person-time for the direct effects and spillover effects analyses.Direct effect includes intervention recipients in target zone.Spillover effect includes intervention non-recipients up to 1km from an index case in the primary analysis and up to 2km or 3km in sensitivity analyses.Total effect includes all individuals (intervention recipients and non-recipients) up to 1km from index case in the primary analysis and up to 2km or 3km in sensitivity analyses.Includes cohort-level analyses for all estimates except spillover effects of the combined intervention.All incidence outcome models were fit with cohort-level data except for models of spillover effects of rfMDA vs. RACD and rfMDA + RAVC vs. RACD only.

Figure S4. Sensitivity analyses for effects on cumulative incidence of malaria
For rfMDA and RACD arms, the primary analysis includes the period from 0-35 days following index case detection for direct effects and 21-56 days for spillover effects; the alternative observation period analysis includes the period from 0-21 days following index case detection for direct effects and 21 to 42 days for spillover effects.For rfMDA+RAVC and RAVC only arms, the primary analysis includes the period from 0-6 months following index case detection for direct effects and 17 days to 6 months for spillover effects; the alternative observation period analysis includes the period from 0-7 days following index case detection for direct effects and 17 to 90 days for spillover effects.Total effects analyses include the person-time for the direct effects and spillover effects analyses.Direct effect includes intervention recipients in target zone.Spillover effect includes intervention non-recipients up to 1km from an index case.Total effect includes all individuals (intervention recipients and nonrecipients) up to 1km from index case.Sensitivity analyses for no overlap of spillover zones excluded any cohorts whose spillover zones overlapped spatially or temporally with other spillover zones.Sensitivity analyses for no overlap of target areas excluded any cohorts whose target areas overlapped spatially or temporally with other target areas.Some direct effects models could not be fit due to data sparsity.All incidence outcome models were fit with cohort-level data except for models of spillover effects of rfMDA vs. RACD and rfMDA + RAVC vs. RACD only.

Figure S5. Sensitivity analyses for direct effects including all intervention recipients
The observation period was 0-35 days for rfMDA and RACD arms and 0-6 months for rfMDA+RAVC and RAVC only arms.Black points indicate estimates from analyses including all intervention recipients, regardless of whether they resided within the target zone within 500m of index cases.Mauve points indicate estimates from analyses restricting to intervention recipients within 500m of index cases that triggered interventions.Analyses were performed at the cohort level.

Table S1. Two-by-two factorial study design of reactive focal interventions
Reactive case detection (RACD) involved administering rapid diagnostic tests for malaria to individuals living within a 500-m radius of an index case and treating individuals who tested positive with artemether-lumefantrine and single-dose primaquine.Reactive focal mass drug administration (rfMDA) involved presumptively treating individuals living within a 500-m radius of an index case with artemether-lumefantrine, without testing for malaria beforehand.Reactive focal vector control (RAVC) involved spraying the long-lasting insecticide, pirimiphos-methyl, to the interior walls of households located within a sevenhousehold radius of an index case.The effectiveness of three interventions were compared to three respective controls: (1)

Table S2. Baseline characteristics among intervention recipients
Includes data from intervention recipients in target areas located within 500m of an index case.

Table S3. Baseline characteristics among non-intervention recipients up to 1km away from index cases
Includes data from intervention non-recipients up to 1km from an index case that triggered interventions.S4.Direct effect, spillover effect, and total effect estimates on cumulative incidence of malaria infection For rfMDA and RACD arms, the analysis includes the period from 0-35 days following index case detection for direct effects and 21-56 days for spillover effects.For rfMDA+RAVC and RAVC only arms, the analysis includes the period from 0-6 months following index case detection for direct effects and 17 days to 6 months for spillover effects.Total effects analyses include the person-time for the direct effects and spillover effects analyses.Direct effect includes intervention recipients in the target zone.Spillover effect analyses includes intervention non-recipients up to 1km from an index case.Total effect includes all individuals (intervention recipients and non-recipients) up to 1km from index case.Models were fit with hierarchical targeted maximum likelihood.All outcome models were fit with cohort-level data except for models of spillover effects of rfMDA + RAVC vs. RACD only.Adjusted models were fit if there were fewer than 10 malaria cases per variable.Covariates were screened separately for each model using a likelihood ratio test.We separately fit individual-and cohort-level outcome models and report the model with the smaller cross-validated mean squared error.All models except spillover effects of the human and combined interventions were fit on cohortlevel data.S7.Direct effect, spillover effect, and total effect estimates on malaria prevalence measured by qPCR Prevalence was measured in a cross-sectional survey in a random sample of households at the end of the malaria season.Analyses were restricted to individuals located within 3 km of at least one intervention recipient.Direct effects include individuals with any intervention recipients within 500m, spillover effects include individuals with no intervention recipients < 500m and any intervention recipients 500m-3km, and total effects include individuals with any intervention recipients <3km during the study.Prevalence ratios were estimated using TMLE with individual-level data, and standard errors were adjusted for clustering at the enumeration area level.Adjusted analyses were not fit there were fewer than 30 observations within strata of the intervention and outcome.Adjusted models were not fit if the number of cases within treatment arm strata was <30.S8.Direct effect, spillover effect, and total effect estimates on household-level malaria prevalence of measured by qPCR Prevalence was measured in a cross-sectional survey in a random sample of households at the end of the malaria season.Analyses were run at the household level.Household-level malaria prevalence was the percentage of households with more than one malaria case detected in the prevalence survey by qPCR.Direct effects include households with any intervention recipients within 500m, spillover effects include households with no intervention recipients < 500m and any intervention recipients 500m-3km, and total effects include households with any intervention recipients <3km during the study.Prevalence ratios were estimated using TMLE with household-level data.Adjusted analyses were not fit there were fewer than 30 observations within strata of the intervention and outcome.Adjusted models were not fit if the number of cases within treatment arm strata was <30.

Table S10. Cost-effectiveness analysis
Prevalent cases averted were estimated using hierarchical TMLE models for prevalence measured by qPCR.The number of prevalent cases averted equaled the produce of the difference in prevalence between arms among intervention recipients and non-recipients by the estimated population size within target areas vs. spillover zones.The incremental cost effectiveness ratio is the ratio of the difference in cost between arms by the difference in prevalent cases averted in both target area and spillover zones within 3 km of index cases for rfMDA + RAVC vs. RACD.Original estimates were reported in Ntuku et al., 2022 10.1136/bmjopen-2021-049050.

Study population
This study analyzed data from a cluster-randomized trial of focal malaria interventions conducted in Zambezi region of Namibia from January 1 to December 31, 2017 (NCT02610400) (1,2).The region has seasonal malaria transmission that peaks between January and June.
Plasmodium falciparum is the dominant species, and annual Pf incidence was less than 15 per 1,000 from 2010-2015.In 2016, the incidence was 32.5 per 1,000 following an outbreak (3).In 2015, prevalence measured by loop-mediated isothermal amplification was 2.2% (4).In the study site, the Namibia Ministry of Health and Social Services routinely delivered case management and annual preseason household IRS with dichlorodiphenyltrichloroethane, with the exception of a small number of structures that were sprayed with deltamethrin.In addition, they offered reactive case detection (RACD) within 500 m of confirmed malaria cases, which included testing with rapid diagnostic tests and treatment with artemether-lumefrantrine and single-dose primaquine for those who tested positive.

Cluster-randomized trial design
The trial included 56 clusters defined based on census enumeration areas that were within the catchment area of study health care facilities.Enumeration areas were eligible for inclusion in the trial if they 1) were located in the catchment areas of 11 health facilities, 2) had complete incidence data from 2012-13, and 3) had at least one incident case during the trial.Using a twoby-two factorial design, the trial randomized 56 clusters to four arms: 1) RACD only, 2) reactive focal mass drug administration (rfMDA) only, 3) reactive vector control (RAVC) + RACD, 4) RAVC + rfMDA.rfMDA included presumptive treatment with artemether-lumefrantrine to individuals in target areas (Extended Data Table 1).The trial used restricted randomization with the following criteria: mean annual incidence in 2013 and 2014, population size, population density, and mean distance from the household to a health-care facility.It was not practical to blind study participants or field staff to intervention assignment, but laboratory analyses and primary statistical analyses were blinded.

Interventions
Field staff delivered interventions in response to passively detected malaria index cases that were confirmed by rapid diagnostic tests or microscopy if the case had resided in the study cluster at least one night in the prior 4 weeks.The trial delivered interventions in "target areas" within approximately 500 m of confirmed malaria cases detected through passive surveillance.
In the RACD arms, individuals were eligible to receive rapid diagnostic tests, and individuals who tested positive were eligible for treatment with artemether-lumefrantrine and single-dose primaquine (Coartem, Novartis Pharmaceuticals, Kempton Park, South Africa; or Komefan 140, Mylan Laboratories, Sinnar, India).In the rfMDA arms, individuals were eligible for presumptive treatment with artemether-lumefrantrine.In the RAVC arms, households were eligible for IRS with pirimiphosmethyl (Actellic 300CS, Syngenta, Basel, Switzerland).In all arms, study teams aimed to deliver interventions within 500 m of a clinical malaria case and within 7 days to 5 weeks of the case report.RACD and rfMDA interventions were delivered to at least 25 people within target areas and RAVC was delivered to at least seven households within target areas.
Over 80% of eligible confirmed malaria cases received interventions, and over 85% of eligible intervention recipients were covered by interventions (2).Since compliance was high, for intervention recipients, we analyzed treatment as randomly assigned.Field staff did not offer repeat interventions in response to subsequent index cases within 5 weeks for rfMDA and RACD and within the same malaria season for RAVC.Field staff recorded the household geocoordinates of the index case and intervention recipients.Additional details about the interventions were previously published (1,2).

Procedures
Prior to randomization, field staff conducted a geographic census and recorded the latitude and longitude of all households in the study area.During the trial, trial staff extracted data on confirmed incident malaria cases and travel history from the rapid reporting system.At the end of malaria season between May and August 2017, the study team collected an endline crosssectional survey to measure infection prevalence.Field staff collected dried blood spots on filter paper (Whatman 3 Corporation, Florham Park, NJ, USA) by finger prick from consenting individuals, and qPCR was performed targeting the acidic terminal sequence of the var gene.(5) Field staff also collected 250 ml of whole blood in BD Microtainer tubes with EDTA additive (Becton, Dickinson and Corporation, Franklin Lakes, NJ, USA) for serological analyses.Using human plasma, Luminex assays were performed to detect malaria antigens using previously described procedures (6,7).Field staff recorded the geocoordinates of all sampled households.

Informed consent
In the original trial, written informed consent was obtained from individual participants for rfMDA or RACD, and from heads of households (≥18 years of age) for RAVC.A parent or guardian was required to provide written informed consent for children younger than 18 years receiving rfMDA or RACD, and written assent for receiving these interventions was also obtained from children aged 12-17 years.

Construction of analytic cohorts for incidence analysis
To construct cohorts, we matched index cases and intervention recipients to individuals recorded in the baseline census using household geocoordinates, age, and sex.We required that geocoordinates be < 100m apart to allow for small deviations in the location of geocoordinate recordings.We excluded 32 cohorts from the analysis for which it was not possible to merge intervention recipient geocoordinates with index data geocoordinates.Because clusters were contiguous with no buffer zones between them, to capture potential dependencies across study clusters, we allowed cohorts to include individuals assigned to an adjacent cluster with a different treatment assignment from the triggering index case if it was within 1 km of an index case.

Follow-up periods for analytic cohorts
We pre-specified cohort follow-up length based on the period in which we expected each intervention to reduce malaria among intervention recipients (direct effects) and non-recipients (spillover effects).Day 0 for each cohort was the date of index case detection.For comparisons of rfMDA and RACD interventions, the direct effect follow-up period was 0 to 35 days, the length of intrinsic incubation period for Pf malaria (8).This is the period of time in which we would expect the intervention to interrupt the parasite life cycle in treated, infected individuals, and in turn, prevent symptoms and/or infectiousness.The spillover effect follow-up period was 21 to 56 days; the 3-week lag period allowed for gametocyte clearance in the treated individual, sporozoite development in mosquitos, and development of detectable merozoites in humans.
For RAVC interventions, the direct effects follow-up period was 6 months since IRS can remain effective for an entire transmission season (9).The spillover effects follow-up period was from day 17 to 6 months.A mosquito bite could hypothetically be prevented on the day of intervention, so the earliest secondary case could occur after sporozoite development in mosquitos (minimum 10 days), and development of detectable merozoites in humans (minimum 7 days).We conducted a sensitivity analysis with alternative follow-up lengths (rfMDA and RACD direct effects: day 0-21; spillover effects: day 21-42; RAVC direct effects day 0-7; spillover effects day 17-90).
Hierarchical TMLE We compared incidence between arms using hierarchical targeted maximum likelihood estimation (TMLE) (10).We fit propensity score models at the cohort-level since interventions were delivered to cohorts.Within study clusters and cohorts, we expected individuals' outcomes to be correlated due to interventions, social interactions, and local environmental factors.We fit two types of outcome models that accounted for statistical dependence in different ways (11).Cohort-level models allowed for statistical dependence between individuals in the same cohort without making any assumptions about the nature of the dependency.Individual-level models assumed that cluster-level and individual-level covariates removed any dependence between outcomes of individuals in nearby geographic areas (11).We separately fit individual-and cohort-level models and then chose the outcome model with the smaller cross-validated mean squared error.
We fit outcome and propensity score models using an ensemble machine learning algorithm (the Superlearner) (12).For propensity score models, learners included generalized linear models, least absolute shrinkage and selection operator (LASSO) (13), and elastic net regression (14).For outcome models, we used the same learners as well as extreme gradient boosting (15).We performed 10-fold cross-validation using a loss function at either the individual-or cohort-level (11).Validation samples were constructed from randomly sampled individuals or cohorts.Because comparisons of rfMDA + RAVC vs. RACD had rare outcomes and a smaller sample size, we used 30-fold cross-validation.
Adjusting standard errors for cohort overlap We adjusted standard errors to account for potential correlation due to overlap between some cohorts using a model of cohort-level influence curves analogous to variance-covariance models used in cross-random effects models (16,17).Specifically, we fit the model: where Di  Dj is the product of influence curves of cohorts i and j, d(i,j) is the distance between the location of the index case that triggered the intervention in each cohort, t(i,j) is the start date of the intervention in each cohort, and C is the cluster-level intervention assignment (18).Adjustment for intervention assignment accounted for correlation due to shared exposure to or receipt of the intervention.For cohorts with no overlap, we set Di  Dj to zero.The regression was implemented with a simplified SuperLearner library including the generalized linear models and LASSO (13).We calculated the variance accounting for outcome dependence as follows: where  ̂ is the estimator,  is the estimand, and N is the number of cohorts.
In both incidence and prevalence analyses, we excluded any categorical covariates with less than 5% prevalence to avoid positivity violations.To minimize empirical positivity violations (19), we only fit models if the number of outcome events per variable was 10 and only fit adjusted models if the number of observations per strata was 30 (20).

Table S6 . Percentage of cohorts overlapping with other cohorts
Overlap in target area was defined as index cases that triggered interventions located within <1km of each other and observation periods that temporally overlapped with another cohort's.Overlap in spillover zones was defined as index cases that triggered interventions located within 1-2km of each other and observation periods that temporally overlapped with another cohort's.The denominator was the total cohorts included in each analysis.

Table S9 . Direct effect, spillover effect, and total effect estimates on Etramp5.Ag1 seroprevalence Prevalence
was measured in a cross-sectional survey in a random sample of households at the end of the malaria season.Analyses were restricted to individuals located within 3 km of at least one intervention recipient.Direct effects include individuals with any intervention recipients within 500m, spillover effects include individuals with no intervention recipients < 500m and any intervention recipients 500m-3km, and total effects include individuals with any intervention recipients <3km during the study.Prevalence ratios were estimated using TMLE with individual-level data, and standard errors were adjusted for clustering at the enumeration area level.Adjusted analyses were not fit there were fewer than 30 observations within strata of the intervention and outcome.Adjusted models were not fit if the number of cases within treatment arm strata was <30.