Mapping the Spatial Distribution of a Disease-Transmitting Insect in the Presence of Surveillance Error and Missing Data

Abstract

Maps of the distribution of epidemiological data often ignore surveillance error or possible correlations between missing information and outcomes. We analyse presence-absence data at the household level (12050 points) of a disease-carrying insect in Mariano Melgar, Peru, collected as part of the Arequipan Ministry of Health's efforts to control Chagas disease. We construct a Bayesian hierarchical model to locate regions that are vulnerable to under-reporting due to surveillance error, accounting for variability in participation due to infestation status. The spatial correlation in the data allows us to identify relative inspector sensitivity and to elucidate the relationship between participation and infestation. We show that naive estimates of prevalence would be biased by surveillance error and missingness at random assumptions. We validate our results through simulations and observe how randomized inspector assignments may improve prevalence estimates. Our results suggests that bias due to imperfect observations and missingness at random can be assessed and corrected in prevalence estimates of spatially auto-correlated binary variables.

A real-time search strategy for finding urban disease vector infestations

“…In addition, we assume perfect inspections and that insecticide application is one hundred percent effective. Although the inspectors are highly skilled, there is always the possibility of heterogeneity in the detection accuracy of active vector surveillance, as found in other studies [21]. It is especially difficult to detect early stage nymphs, which are small and difficult to see.…”

“…Despite these preliminary limitations, our approach has valuable aspects that could improve vector surveillance and control, especially in cases where detailed data are not available. For example, we commonly observe the infection status of a subset of individuals at some time point after the infection occurred [1, 21, 22], and the actual timepoints of each infection are often unknown. Methods have been developed to handle incomplete data from infectious disease outbreaks, including unobserved infection times and incomplete epidemics [7, 9].…”

Two‐scale dispersal estimation for biological invasions via synthetic likelihood

“…In the first survey, 96 households (18%) could not be entered and in the second survey, 131 households (24%). We have shown previously that households that do not participate in vector surveys are less likely to be infested than those that participate (Hong et al ). Based on this finding, we assume that T. infestans are absent from households we did not enter.…”

Residual Infestation and Recolonization during UrbanTriatoma infestansBug Control Campaign, Peru1

“…In practice, no household was observed to be infested during both treatments of the initial treatment phase and surveillance inspections, which suggested effectiveness of the second treatment. In addition, because there is a strong correlation between infestation and participation ( 15 , 21 ), households that never participated might have initially a lower prevalence of infestation than those that participated once, in contrast to the equal prevalence we assumed in our analysis. Assuming an overly conservative 5-fold lower prevalence among never-treated houses relative to once-treated households, the never-treated households would still represent >85% of the residual infestation after the treatment phase (http://www.spatcontrol.net/articles/Barbu2014/suppMet.pdf).…”

“…Unless otherwise noted, we used a random effect term to control for potential similarity of households in a locality ( 20 , 21 ). …”