Synthetic Population Generation with Public Health Characteristics for Spatial Agent-Based Models

Emma Von Hoene,Amira Roess,Hamdi Kavak,Taylor Anderson

DOI: https://doi.org/10.1101/2024.09.18.24312662

2024-09-19

Abstract:Agent-based models (ABMs) simulate the behaviors, interactions, and disease transmission between individual agents within their environment, enabling the investigation of the underlying processes driving disease dynamics and how these processes may be influenced by policy interventions. Despite the critical role that characteristics such as health attitudes and vaccination status play in disease outcomes, the initialization of agent populations with these variables is often oversimplified, overlooking statistical relationships between attitudes and other characteristics or lacking spatial heterogeneity. Leveraging population synthesis methods to create populations with realistic health attitudes and protective behaviors for spatial ABMs has yet to be fully explored. Therefore, this study introduces a novel application for generating synthetic populations with protective behaviors and associated attitudes using public health surveys instead of traditional individual-level survey datasets from the census. We test our approach using two different public health surveys (one national and the other representative of the study area, Virginia, U.S.) to create two synthetic populations representing individuals aged 18 and over in Virginia, U.S., and their COVID-19 vaccine attitudes and uptake as of December 2021. Results show that integrating public health surveys into synthetic population generation processes preserves the statistical relationships between vaccine uptake and attitudes in different demographic groups while capturing spatial heterogeneity at fine scales. This approach can support disease simulations that aim to explore how real populations might respond to interventions and how these responses may lead to demographic or geographic health disparities. Our study also demonstrates the potential for initializing agents with variables relevant to public health domains that extend beyond infectious diseases, ultimately advancing data-driven ABMs for geographically targeted decision-making.

What problem does this paper attempt to address?

The problem this paper attempts to address is: In Spatial Agent-Based Models (ABMs), how to generate synthetic populations with realistic health attitudes and protective behaviors to better simulate disease spread and assess the effectiveness of interventions.

Specifically, existing ABMs often overlook the statistical relationships between key variables such as health attitudes and vaccination status, as well as the spatial heterogeneity of these variables when initializing agents. This leads to overly simplified and homogeneous agent behaviors in the models, which cannot accurately reflect the complex situations in the real world. Therefore, this study proposes a new method to generate synthetic populations with realistic health attitudes and protective behaviors using public health survey data, thereby improving the accuracy and practicality of the models.

The study achieves this goal through the following steps:
1. **Data Sources**: Using two different public health survey datasets, one is national (Household Pulse Survey, HPS), and the other is local survey data representing Virginia.
2. **Population Synthesis Method**: Employing the Iterative Proportional Fitting (IPF) algorithm to combine individual-level survey data with spatially aggregated demographic data, generating synthetic populations with detailed health attitudes and protective behaviors.
3. **Validation of Results**: Comparing the generated synthetic population with actual vaccination rates to validate the effectiveness of the method.

Through this method, the study aims to generate more realistic synthetic populations to more accurately reflect the behavioral characteristics and spatial distribution of different groups in disease spread simulations, thereby supporting more effective policy-making and intervention assessment.