# A conceptual and computational framework for modeling the complex, adaptive dynamics of epidemics: The case of the SARS-CoV-2 pandemic in Mexico

## TL;DR

This paper introduces a hybrid intelligence model combining human and AI to better understand and predict pandemic dynamics, using Mexico's SARS-CoV-2 pandemic as a case study.

## Contribution

The novel hybrid intelligence framework integrates diverse data sources and human interpretation to model complex pandemic dynamics.

## Key findings

- EPI-PUMA integrates public data sources to predict pandemic outcomes with high accuracy.
- The model identifies differential predictive value of various factors influencing pandemic spread.
- Classifiers in the system achieve ROC areas between 0.8 and 0.9, indicating strong predictive performance.

## Abstract

In the quest to ensure adequate preparedness for health emergencies caused by infectious disease pandemics, there is a need for tools that can address the myriad relevant questions related to the spread and trajectory of pandemics. A hybrid intelligence model that combines human and artificial intelligence may provide a viable solution, as it can process data from models that comprehensively integrate contextual and direct factors, effectively mimicking the social processes surrounding transmission while incorporating human interpretation to enhance our understanding of pandemics. Using data from the COVID-19 pandemic, we demonstrate the implementation of this approach with the publically available EpI-PUMA (Epidemiological Intelligence Platform for the Universidad Nacional Autónoma de México (“PUMA”)) project and platforms, where a user may create their own hybrid intelligence Bayesian classifier models for a range of epidemiological indicators of interest. EPI-Puma integrates data from various public sources (including the national registry of SARS-CoV-2 cases, census data, poverty indicators, climate data and data related to atmospheric contaminants), enabling the deployment of models that predict a range of relevant outcomes. The main criteria for the data included was its coverage (at least at the municipality level) and availability (public data). EPI-Puma was able to identify both the differential predictive value of the different sets of factors related to the epidemic path and well as anticipate with a high probability the path of the pandemic (typical areas under the ROC curve for the associated classifiers being 0.8–0.9).