Back

Epidemics of chikungunya, Zika, and COVID-19 reveal bias in case-based mapping

Bustos Carrillo, F. A.; Lopez Mercado, B.; Carey Monterrey, J.; Collado, D.; Saborio, S.; Miranda, T.; Barilla, C.; Ojeda, S.; Sanchez, N.; Plazaola, M.; Suazo Laguna, H.; Elizondo, D.; Arguello, S.; Gajewski, A. M.; Maier, H. E.; Latta, K.; Carlson, B.; Coloma, J.; Katzelnick, L.; Sturrock, H.; Balmaseda, A.; Kuan, G.; Gordon, A.; Harris, E.

2021-07-26 epidemiology
10.1101/2021.07.23.21261038 medRxiv
Show abstract

Accurate tracing of epidemic spread over space enables effective control measures. We examined three metrics of infection and disease in a pediatric cohort (N {approx} 3,000) over two chikungunya and one Zika epidemic, and in a household cohort (N=1,793) over one COVID-19 epidemic in Managua, Nicaragua. We compared spatial incidence rates (cases/total population), infection risks (infections/total population), and disease risks (cases/infected population). We used generalized additive and mixed-effects models, Kulldorfs spatial scan statistic, and intracluster correlation coefficients. Across different analyses and all epidemics, incidence rates considerably underestimated infection and disease risks, producing large and spatially non-uniform biases distinct from biases due to incomplete case ascertainment. Infection and disease risks exhibited distinct spatial patterns, and incidence clusters inconsistently identified areas of either risk. While incidence rates are commonly used to infer infection and disease risk in a population, we find that this can induce substantial biases and adversely impact policies to control epidemics. Article summary lineInferring measures of spatial risk from case-only data can substantially bias estimates, thereby weakening and potentially misdirecting measures needed to control an epidemic.

Matching journals

The top 11 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.