Examining the spatial effect ranges of new classes of long-lasting insecticidal nets (LLINs) on malaria infection and vector abundance: an anisotropic semivariance analysis in the context of a phase-III cluster-randomised malaria vector control trial

Abstract Background At high levels of population coverage with long-lasting insecticidal nets (LLINs), reductions in malaria vector populations can create a community effect, protecting individuals in an area who do not sleep under a net themselves. In cluster-randomised trials evaluating new LLINs, spatially localised reductions in malaria transmission may impact nearby clusters unless carefully designed, leading to attenuated treatment effect estimates. Unsampled buffer areas are commonly used to mitigate contamination, but no consensus on optimal buffer size exists. Using data from a phase-III cluster-randomised trial in northern Tanzania assessing three dual active-ingredient LLINs against a pyrethroid-only LLIN, our study aims to determine the spatial form of LLIN intervention effects on malaria infection and vector abundance. Methods Using cross-sectional data (July 2020; 18 months post-intervention) from a cluster-randomised malaria vector control trial, we measured malaria infection and commonly associated clinical, household, environmental, and geographic factors in 5955 children from 3188 households across four geographically contiguous areas. Data on vector abundance and associated factors were additionally collected from a subset of 789 households between September and December 2020. Mixed-effects logistic and negative binomial regression models were constructed to predict malaria infection in children and vector abundance in households, respectively. Model residuals were used to compute robust semivariance estimates and variogram models, accounting for anisotropic trends present in the data. Results We identified malaria infection in 2738 children (46.0%) and captured an average of 4.13 malaria vectors per trap night—predominantly Anopheles funestus sensu lato (80.7%). Range estimates for malaria infection model residuals varied from 1452m (95%CI = 1360─1545; Py-Pyriproxyfen LLIN) to 7833m (4413─11250; Pyrethroid-only LLIN) and vector abundance model residuals varied from 1139m (1100─1178; Py-Pyriproxyfen LLIN) to 4085m (1216─6954; Py-Chlorfenapyr LLIN). The Pyrethroid-only LLIN and Py-Pyriproxyfen LLIN malaria infection models were largely non-spatial in nature (RNE = 88.4% (84.8─92.4) and 91.0% (88.9─93.0), respectively) and the Py-Chlorfenapyr LLIN vector abundance model had no significant spatial form (68.1%, 50.9–100.0). Conclusions The results of this study describe the spatial form of community effects induced by LLIN interventions in certain contexts—potentially driven by the relative efficacy of intervention effects on local vector populations. Modelled range estimates uniformly exceed common buffer distances, although the proportion of variation explained by spatial dependence was quite low for three models. In designing cluster buffers, future vector control trials should consider the spatial dispersion patterns of local vector populations and identify how they intend to control for any contamination that may occur.