Assessing the predictive causality of individual based models using Bayesian inference intervention analysis: an application in epidemiology

Authored by Aristides Moustakas

Date Published: 2018

DOI: 10.1007/s00477-018-1520-6

Sponsors: No sponsors listed

Platforms: No platforms listed

Model Documentation: Other Narrative

Model Code URLs: Model code not found

Abstract

Understanding dynamics in time and the predominant underlying factors that shape them is a central question in biological and medical sciences. Data are more ubiquitous and richer than ever before and population biology in the big data era need to integrate novel methods. Calibrated Individual Based Models (IBMs) are powerful tools for process based predictive modelling. Intervention analysis is the analysis in time series of the potential impact of an event such as an extreme event or an experimentally designed intervention on the time series, for example vaccinating a population. A method for big data analytics (causal impact) that implements a Bayesian intervention approach to estimating the causal effect of a designed intervention on a time series is used to quantify the deviance between data and IBM outputs. Having quantified the deviance between IBM outputs and data, IBM scenarios are used to predict the counterfactual. The counterfactual is how the IBM response metric would have evolved after the intervention if the intervention had never occurred. The method is exemplified to quantify the deviance between a calibrated IBM outputs and epidemiological data of Bovine Tuberculosis with changing the cattle TB testing frequency as the intervention covariate. The advantage of IBM data validation and uncertainty assessment as time series is also discussed.

Tags

Agent-based models Complexity Agent based models Epidemiology selection knowledge Big data time series model validation utility time Generality Causal inference Data availability Bovine-tuberculosis Counterfactual prediction