Mathematical Biology Seminar

Join Virtual Seminar Here

Binod Pant, Northeastern University
Human behavior and uncertainty quantification in the context of epidemiological models
Human behavior has been attributed as one of the major reasons why models perform poorly when forecasting. The first half of this talk will focus on modeling the interplay between human behavior and disease outbreaks. In a retrospective study, we show that models incorporating human behavior change capture disease trajectories better than equivalent models without behavior change. Further, I will present a study characterizing population-level hu- man behavior change, as inferred through survey-collected behavior data from all 50 US states during the first two years of the COVID-19 pandemic. Identifiability issues, a common problem in mathematical biology, have also been attributed to why models fail to forecast properly and struggle to correctly characterize disease transmission even in retrospective studies. Using model-generated synthetic data where ground truth is known, we investigate the inference of epidemiological quantities of interests when only fitting to detected incidence data. Case detected ratios are often far below one for various diseases, yet models are routinely fitted without accounting for undetected infections. We show that ignoring case underdetection entirely yields parameter errors exceeding 1000%, even when model fits appear visually reasonable. Moreoever, even when models explicitly account for underdetection by including case detection ratios as unknown parameters, an infinite spectrum of epidemiologically distinct scenarios can fit the data equally well due to structural unidentifiability. Strikingly, incorporating even a single seroprevalence measurement alongside detected cases considerably shrinks parameter uncertainty, even under reasonable noise conditions.