Geospatial joint modeling of vector and parasite serology to microstratify malaria transmission

Abstract

The World Health Organization identifies a strong surveillance system for malaria and its mosquito vector as an essential pillar of the malaria elimination agenda. Anopheles salivary antibodies are emerging biomarkers of exposure to mosquito bites that potentially overcome sensitivity and logistical constraints of traditional entomological surveys. Using samples collected by a village health volunteer network in 104 villages in Southeast Myanmar during routine surveillance, the present study employs a Bayesian geostatistical modeling framework, incorporating climatic and environmental variables together with Anopheles salivary antigen serology, to generate spatially continuous predictive maps of Anopheles biting exposure. Our maps quantify fine-scale spatial and temporal heterogeneity in Anopheles salivary antibody seroprevalence (ranging from 9 to 99%) that serves as a proxy of exposure to Anopheles bites and advances current static maps of only Anopheles occurrence. We also developed an innovative framework to perform surveillance of malaria transmission. By incorporating antibodies against the vector and the transmissible form of malaria (sporozoite) in a joint Bayesian geostatistical model, we predict several foci of ongoing transmission. In our study, we demonstrate that antibodies specific for Anopheles salivary and sporozoite antigens are a logistically feasible metric with which to quantify and characterize heterogeneity in exposure to vector bites and malaria transmission. These approaches could readily be scaled up into existing village health volunteer surveillance networks to identify foci of residual malaria transmission, which could be targeted with supplementary interventions to accelerate progress toward elimination.

Keywords: Anopheles salivary antibodies; disease mapping; geospatial; malaria.

Fig. 1.

Overall anti-SG6 IgG seroprevalence, with (A) P. falciparum and (B) P. vivax prevalence and anti-CSP IgG over time. Fig. 1A shows the seroprevalence and 95% CI of IgG to P. falciparum transmission stage (PfCSP) and the vector salivary (SG6) antigens (Left y axis), as well as the prevalence (95% CI) of P. falciparum infection (Right y axis), over the 15-mo study period. Fig. 1B shows the seroprevalence (95% CI) of IgG to P. vivax transmission stage (PvCSP) and the vector salivary (SG6) antigens (Left y axis), as well as the prevalence (95% CI) of P. vivax infection (Right y axis), over the 15-mo study period. Vertical dotted lines indicate typical season.

Fig. 2.

Predicted anti-SG6 IgG seroprevalence and model uncertainty for Bago (East), Kayah and Kayin, with model validation. Geospatial maps showing (A) the predicted posterior mean probability and (B) SD of anti-SG6 IgG seropositivity. Estimated using a geospatial model that adjusts for rainfall, distance to water, potential evapotranspiration, tree coverage, and diurnal temperature difference. Model validation (C): The model is trained on data from a random sample of 90% of villages (20 repeats), with internal validity assessed as the correlation between the observed vs. predicted SG6 seroprevalence (with 95%Crl) for each of the omitted 10% of villages (pink crosses) and for omitted villages grouped in a series of bins (deciles) by predicted seroprevalence (black dots). Pearson correlation used to estimate r = 0.731

Fig. 3.

Predicted seroprevalence of anti-SG6 and CSP IgG antibodies and predicted prevalence of PCR-detectable Plasmodium spp. infections after joint modeling of these outcomes. Estimated using a geospatial model that adjusts for distance to water, topographical wetness index, slope, tree coverage fraction, inaccessibility to cities, and night-time lights (models were fitted to data from participants in all villages who had observations for all outcomes, n = 11,988).

Fig. 4.

Directed acyclic graph of the joint model. Dotted circles represent covariates, dashed circles represent latent outcome variables, and solid circles represent measured outcome variables.