Robustness of Eco-Epidemiological Capture-Recapture Parameter Estimates to Variation in Infection State Uncertainty

Abstract

Estimating eco-epidemiological parameters in free-ranging populations can be challenging. As known individuals may be undetected during a field session, or their health status uncertain, the collected data are typically “imperfect”. Multi-event capture-mark-recapture (MECMR) models constitute a substantial methodological advance by accounting for such imperfect data. In these models, animals can be “undetected” or “detected” at each time step. Detected animals can be assigned an infection state, such as “susceptible” (S), “infected” (I), or “recovered” (R), or an “unknown” (U) state, when for instance no biological sample could be collected. There may be heterogeneity in the assignment of infection states, depending on the manifestation of the disease in the host or the diagnostic method. For example, if obtaining the samples needed to prove viral infection in a detected animal is difficult, this can result in a low chance of assigning the I state. Currently, it is unknown how much uncertainty MECMR models can tolerate to provide reliable estimates of eco-epidemiological parameters and whether these parameters are sensitive to heterogeneity in the assignment of infection states. We used simulations to assess how estimates of the survival probability of individuals in different infection states and the probabilities of infection and recovery responded to (1) increasing infection state uncertainty (i.e., the proportion of U) from 20 to 90%, and (2) heterogeneity in the probability of assigning infection states. We simulated data, mimicking a highly virulent disease, and used SIR-MECMR models to quantify bias and precision. For most parameter estimates, bias increased and precision decreased gradually with state uncertainty. The probabilities of survival of I and R individuals and of detection of R individuals were very robust to increasing state uncertainty. In contrast, the probabilities of survival and detection of S individuals, and the infection and recovery probabilities showed high biases and low precisions when state uncertainty was >50%, particularly when the assignment of the S state was reduced. Considering this specific disease scenario, SIR-MECMR models are globally robust to state uncertainty and heterogeneity in state assignment, but the previously mentioned parameter estimates should be carefully interpreted if the proportion of U is high.

Uncovering ecological state dynamics with hidden Markov models

“…Individual‐level data often contain additional information about developmental states such as those related to size (Nichols et al ., 1992), reproduction (Nichols et al ., 1994), social groups (Marescot et al ., 2018) or disease (Benhaiem et al ., 2018). However, assigning individuals to states can be difficult when traits such as breeding (Kendall et al ., 2012), infection (Chambert et al ., 2012), sex (Pradel et al ., 2008) or even species (Runge et al ., 2007) are ascertained through observations in the field.…”

Modelling pest dynamics under uncertainty in pest detection: the case of the red palm weevil

“…Multievent capture recapture models have been recently used to study of demographic parameters of animal populations (Frederiksen et al 2014), including the epidemiological dynamics of animals in which the sanitary status of individuals cannot be assessed with total certitude (Conn and Cooch 2009;Santoro et al 2014;Benhaiem et al 2018;Simon et al 2018). Here, we show that multievent models can also be used to study the plant-insect interaction, analyzing the infestation data of the plant.…”

“…missing data) and assessment of the infectious state of the host (i.e. insect detection) (Pradel 2005;Conn and Cooch 2009;Santoro et al 2014;Benhaiem et al 2018;Simon et al 2018). In addition, the flexibility of multievent models allows assessing the influence of external variables or individual covariates as predictors of model parameters.…”

Newcastle disease virus transmission dynamics in wild peridomestic birds in the United Arab Emirates

“…It can also be linked with the re-capture rates as the birds were not sampled if the time interval between two consecutive samples was smaller than a month, francolins and pigeons being the most frequently re-captured birds. Similarly, this probability was dependent on the year of sampling, being larger in 2017 than in 2016, linked with improvement of techniques, which is often encountered in similar studies 47 .…”