Extension of efficacy range for targeted malaria-elimination interventions due to spillover effects

Global malaria eradication efforts have renewed focus, yet progress has stalled, with global cases increasing from 231,000 in 2015 to 249,000 in 2022 and deaths per 100,000 largely unchanged. In southern Africa, eight countries aim for elimination by 2030. Interventions ideal for elimination would protect recipients and reduce transmission to non-recipients via spillover (indirect/herd) effects, as seen with some vaccines. While spillover effects have been documented for interventions like mass drug administration for trachoma, school-based deworming, insecticide-treated nets, and indoor residual spraying, evidence is limited for targeted malaria interventions in low-transmission, near-elimination settings and often lacks randomized designs. Accounting for spillover can substantially improve cost-effectiveness, which is critical given the high costs of elimination. WHO recommends reactive (post-case detection) and focal (targeted to neighbors of index cases) approaches in low-transmission settings. A recent cluster-randomized trial in Namibia found reactive focal chemoprevention and vector control reduced incidence and prevalence. This study re-analyzes that trial to separately estimate direct effects among recipients and spillover effects among nearby non-recipients to determine whether these focal interventions reduce onward transmission beyond treated households.

Prior research has identified spillover effects for mass drug administration for trachoma, school-based deworming, insecticide-treated bed nets, and indoor residual spraying (IRS). Earlier bed-net trials in the 1990s showed community spillover, but contemporary effects may differ due to changes in net coverage, insecticide resistance, and mosquito behaviors. Recent studies on IRS spillover lacked randomization and are prone to residual confounding. No prior work has rigorously quantified spillover from reactive, focal malaria interventions in low-transmission, near-elimination contexts. Economic analyses indicate that considering spillover effects can markedly improve the cost-effectiveness of interventions, an important factor as elimination efforts will be more expensive than current control programs. The Namibia trial previously demonstrated overall benefits of reactive focal chemoprevention and vector control, motivating a re-analysis to isolate direct versus spillover effects.

Design and setting: Secondary analysis of a previously completed cluster-randomized, two-by-two factorial trial conducted in the Zambezi region of Namibia (1 Jan–31 Dec 2017; NCT02610400) in a low Plasmodium falciparum transmission setting. Fifty-six census enumeration-area clusters were randomized 1:1:1:1 to: (1) reactive case detection (RACD), (2) reactive focal mass drug administration (rfMDA), (3) reactive vector control (RAVC) plus RACD, or (4) RAVC plus rfMDA. Interventions were delivered within approximately 500 m of confirmed passively detected index cases. Interventions: RACD involved rapid diagnostic testing and treatment with artemether-lumefantrine plus single-dose primaquine for positives. rfMDA provided presumptive artemether-lumefantrine to eligible individuals in target areas. RAVC used indoor residual spraying with pirimiphos-methyl (Actellic 300CS). Interventions were generally delivered within a median of 16 days after index-case reporting; repeat interventions were not offered within 5 weeks (rfMDA/RACD) or within the same season (RAVC). Coverage among eligible individuals was high: 89% RACD, 87% rfMDA, 89% RAVC. Analytic approach for incidence: To mimic a ring-trial structure and capture spatial/temporal spread, person-time cohorts were constructed around each index case, including individuals residing within 1 km. Direct-effect cohorts included intervention recipients within 500 m of index cases; spillover-effect cohorts included non-recipients within up to 1 km. Follow-up windows reflected biological timelines: rfMDA vs RACD direct effects (days 0–35), spillover effects (days 21–56); RAVC direct effects (0–6 months), spillover effects (days 17–6 months). Analyses also estimated total effects among all individuals within 1 km. Data restructuring allowed cohorts to include individuals from adjacent randomized clusters when within 1 km of an index case, acknowledging potential inter-cluster dependence. Analytic approach for prevalence and serology: Cross-sectional endline survey (May–Aug 2017) measured infection prevalence by qPCR targeting var gene and seroprevalence (Etramp5.Ag1) by Luminex. Direct-effects analyses included individuals residing within 500 m of any intervention recipient; spillover effects included those with no recipients within 500 m but at least one recipient within 500 m–3 km; total effects included those with at least one recipient within 3 km. Statistical methods: Hierarchical targeted maximum likelihood estimation (TMLE) with ensemble machine learning (SuperLearner) was used for incidence, accommodating cluster-level exposures and potential within-cluster dependence. Models adjusted for pre-specified and screened covariates: cluster-level (baseline incidence, IRS coverage, rainfall, enhanced vegetative index, daytime land surface temperature, population size, elevation, distances to households/health facilities, time from detection to intervention) and individual-level (age, sex, calendar month, distance to index case, prior interventions received, counts of prior recipients within radii, local population). Propensity-score models and outcome regressions used GLMs, LASSO/elastic net, and XGBoost; V-fold cross-validation selected optimal learners. Standard errors were adjusted using influence-curve covariance models accounting for spatial-temporal cohort overlap. For prevalence/seroprevalence TMLE, individual-level data were used with standard errors clustered at enumeration areas; models were unadjusted when observations per stratum were fewer than 30. Sensitivity and subgroup analyses: Spillover radii of 2 km and 3 km were evaluated for incidence; alternative follow-up windows assessed periods of potentially stronger effects; analyses repeated including recipients living slightly beyond 500 m; and analyses excluding overlapping cohorts tested robustness to dependence. Prespecified effect-modification analyses examined baseline transmission, prior IRS coverage, environmental factors (temperature, rainfall, EVI, elevation), treatment coverage, gender, and health-facility proximity. Cost-effectiveness: Incremental cost-effectiveness ratios (ICERs) were calculated using intervention costs from 2017 USD and total-effect estimates on qPCR prevalence for both direct and spillover populations (within 500 m and 500 m–3 km), comparing arms to quantify cost per prevalent case averted. Confidence intervals derived by applying the ICER formula to bounds of the total-effect 95% CI. Contamination test: Log-linear models assessed whether cluster outcomes depended on outcomes in adjacent clusters; likelihood ratio tests evaluated evidence of between-cluster contamination.

• Direct effects: No statistically significant direct effects on incidence among recipients within 500 m for any intervention comparison; analyses likely underpowered due to small recipient sample size.
• Spillover on incidence (primary radius 1 km): Strongest evidence for the combined intervention (rfMDA + RAVC) with a 43% reduction in incidence among non-recipients (95% CI, approximately 21–58%); chemoprevention alone showed no clear spillover effect; vector control alone had an imprecise estimate including the null (incidence reduction 32%; 95% CI, 0–65%).
• Spillover on prevalence (endline survey): Chemoprevention reduced prevalence among non-recipients by 72% (95% CI, 31–88%); combined intervention reduced prevalence by 79% (95% CI, 6–95%). Effects attenuated with increasing distance from interventions, strongest within 500 m–1 km.
• Direct and spillover on seroprevalence (Etramp5.Ag1): Direct effects observed for chemoprevention (25% reduction; 95% CI, 14–34%) and combined intervention (34% reduction; 95% CI, 10–42%). Spillover effect for combined intervention reduced seroprevalence by 34% (95% CI, 20–45%).
• Heterogeneity: Spillover effects were more protective in areas with baseline incidence below the median (<14/1,000), e.g., combined intervention reduced incidence by 68% (95% CI, 35–84%) in lower-transmission strata, with little effect in higher-transmission strata. Stronger spillover with higher rainfall (e.g., median monthly rainfall >24 mm). Spillover from chemoprevention more evident among men than women.
• Spatial scale: For incidence, spillover effects were not evident at 2–3 km radii, potentially due to cohort overlap and lower infection intensity with distance. For prevalence, combined intervention showed benefits to 3 km, with decreasing effect size by distance.
• Cost-effectiveness: Incorporating spillover increased cost-effectiveness: ICERs were $144 (95% CI, $136–$153) for chemoprevention, $1,882 ($1,679–$2,111) for vector control, and $1,050 ($915–$1,231) for the combined intervention. Relative to original estimates without spillover, cost-effectiveness improved by 11%, 30%, and 42%, respectively.
• Contamination: No evidence that outcomes were correlated with adjacent clusters’ outcomes (incidence χ² = 0.540, P = 0.462; prevalence χ² = 0.0107, P = 0.9178).

This re-analysis demonstrates that reactive, focal interventions can yield substantial spillover benefits for nearby non-recipients, particularly when chemoprevention is combined with indoor residual spraying. The combined strategy likely acts synergistically: chemotherapy shortens infectiousness in human reservoirs while IRS suppresses vector populations, jointly reducing local transmission and limiting replenishment of parasite reservoirs. The strongest spillover effects occurred in lower-transmission areas and under environmental conditions conducive to vector breeding, aligning with expectations for focal transmission dynamics and vector ecology. Spillover effects were often similar to or stronger than direct effects, potentially because delays between case detection and intervention allowed transmission among close contacts before delivery, while interventions still prevented onward spread to neighbors further from the index case. The distance-gradient in prevalence effects and the 1 km range for incidence spillover clarify the spatial operating scale of reactive focal measures. Incorporating spillover into economic evaluations substantially improves the perceived value of these strategies, crucial for policy-making in elimination contexts with tight resource constraints. Gender differences suggest higher baseline risk in men, consistent with mobility and exposure patterns, which may influence both apparent effects and targeting considerations.

Targeted, reactive focal chemoprevention combined with indoor residual spraying can extend benefits beyond direct recipients, reducing incidence within 1 km and prevalence up to 3 km among non-recipients in low-transmission settings. Accounting for these spillover effects increases the cost-effectiveness of interventions, particularly for combined strategies, supporting their use for reactive control and outbreak response near elimination. Methodologically, a ring-trial-mimicking cohort approach with hierarchical TMLE enables separation of direct and spillover effects and can be applied to other interventions, including malaria vaccines. Future research should evaluate generalizability across different transmission intensities and ecologies, quantify thresholds of coverage needed for spillover, optimize timing to minimize delays from case detection to intervention, assess synergy with other tools (e.g., long-lasting insecticide-treated nets, larval source management, vaccines), and investigate targeted strategies tailored to high-risk subgroups such as mobile male populations.

• Precision was limited for some analyses due to rare outcomes; direct-effect incidence analyses among recipients were likely underpowered.
• Spatiotemporal overlap among analytic cohorts may have induced outcome dependence; although standard errors were adjusted, residual bias is possible; excluding overlap attenuated some combined-intervention estimates.
• High intervention coverage (>85%) may not be achievable programmatically; spillover magnitudes could be smaller at lower coverage, and effect modification by coverage could not be assessed.
• The study was conducted in a low-transmission area; findings may not generalize to higher-transmission settings or contexts where risk is driven by occupational/travel exposure rather than household-level transmission.
• For chemoprevention vs RACD, the comparator included single-dose primaquine for positives, potentially attenuating contrast with rfMDA; timing of effects (incidence windows soon after interventions vs end-season prevalence) may also explain differences across outcomes.
• Delays (median ~16 days) from index-case detection to intervention could reduce direct impacts among immediate contacts.
• Some models were unadjusted due to small strata (<30 observations), and several sensitivity analyses indicated reduced precision with larger spillover radii.