Assessment of the level of adjustment of three epidemiological models in the analysis of epidemics with incidences less than 100% such as the lethal wilt of oil palm (Elaeis guineensis Jacq.)

Abstract

The production of oil palm is a major agricultural activity in Colombia. Lethal wilt (LW) of the oil palm is one of the most devastating diseases in the Eastern zone of the country. Several epidemiological models used in epidemic analyses assume that there is a constant area where the host will become diseased at the end of the epidemic (maximum incidence of disease = 100%). Based on the analysis of three different epidemics, we demonstrated the error in the application of the model that best fits the observed data when the maximum incidence of the disease (Kmax) is below the assumed. We assessed the fit of the monomolecular, logistic, and Gompertz models at different final incidence values of the disease including the maximum observed (y1 + 0.1). We analyzed the data with linear regression and residuals variance and distribution. We measured the relative quality level of fit of the model for each Kmax by determining coefficients (R2) and the Akaike and Bayesian information criteria (AIC & BIC). The monomolecular model showed a tendency to increase the level of adjustment when Kmax assumed values were close to 1 while the logistic and Gompertz models remained stable regardless of the evaluated Kmax values. The consequences of assuming a Kmax with values equal to 1 reflected not only the erroneous estimation of parameters such as y0 and r but also gave rise to a misinterpretation of the temporal behavior of the epidemic.