Source-linked AI summary
ABM-SIRTEM: A Hybrid Agent-Based and Epidemiological Model for Pandemic Response
Sheryl Paul, Samuel Williams, Preetom K. Biswas, Giulia Pedrielli, Jyotirmoy V. Deshmukh
TL;DR
Pandemic-response models must account for disease spread alongside heterogeneous behavior and economic effects, especially for populations and jobs disproportionately affected by restrictions. ABM-SIRTEM combines an epidemiological model with an agent-based representation of occupations, welfare, and evolving compliance, then calibrates transmission using historical test counts from four U.S. states. The framework supports studying feedback between compliance, socioeconomic behavior, and disease dynamics, while its reported comparisons assess calibration fit rather than held-out forecasting performance.
Problem
Pandemic responses require analysis of interactions among disease spread, individual behavior, and economic activity, but population models simplify heterogeneity and detailed ABMs can be computationally demanding.
Method
ABM-SIRTEM combines SIRTEM with an agent-based model representing occupations, welfare, restrictions, and dynamically evolving compliance.
Results
The model is calibrated against historical positive and negative test counts from four U.S. states, with positive-test trajectories broadly reproducing the timing of larger observed waves.
Takeaways & Limitations
ABM-SIRTEM provides a framework for studying socioeconomic and epidemiological impacts and feedback between compliance and disease dynamics in pandemic response.
Takeaways & Limitations
Because observed data within the lookahead horizon estimate transmission parameters, the comparisons assess calibration fit rather than held-out performance, and the historical restriction index does not establish causal policy effects.
Abstract
from arXiv · showhide
The COVID-19 pandemic has had profound impacts on global health, social structures, and economies. It disproportionately affected lower socioeconomic groups and those reliant on interaction-based jobs. Regulatory bodies faced the challenge of designing policies that preserve public health while limiting disruption to economic stability and productivity. Epidemiological models such as SIR and agent-based models (ABMs) have been used to study disease dynamics and the socioeconomic impacts of disease and interventions. Population-level models often simplify individual heterogeneity, while detailed ABMs can become computationally expensive as the numbers of agents and interactions increase. We propose ABM-SIRTEM, a hybrid model that incorporates occupation categories, economic productivity, and welfare at the individual level while dynamically modeling compliance with government interventions. We calibrate the model against historical positive and negative test counts from four U.S. states and examine the resulting compliance dynamics. This framework provides a basis for studying the interaction between disease spread and socioeconomic behavior in pandemic-response planning.
1 Introduction
ABM-SIRTEM addresses limitations of population-level epidemiological models and computationally intensive ABMs by linking disease dynamics with heterogeneous individual behavior, occupations, welfare, and compliance. The hybrid model calibrates transmission against historical positive and negative test counts from four U.S. states.
- Motivation: Population-level models can simplify individual differences, whereas detailed ABMs capture heterogeneous behavior and contacts but require substantial data and computation.These limitations motivate a hybrid modeling approach for pandemic-response analysis.
- Compliance dynamics: Government restrictions affect agent interactions and mobility, while compliance evolves with economic incentives, social interactions, and perceived health risks.The model uses an evolutionary-game-theory-motivated update rule to model changing compliance within occupation and health-state groups.
- Model overview: ABM-SIRTEM combines the population-level SIRTEM epidemiological model with an agent-based model that simulates individual interactions.SIRTEM supplies health-state distributions, and agents are assigned susceptible, exposed, infected, tested, hospitalized, or recovered states.
- Individual heterogeneity: Agents are represented by occupation, health status, and compliance, with interactions determined by contact-frequency mixing rates and welfare combining productivity, mental well-being, and health risk.Occupation categories encode differences in work, interaction, and compliance behavior.
- Model integration and calibration: The ABM feeds mean compliance back into SIRTEM, changing infection rates and simulated daily positive and negative test counts.The resulting test-count trajectories are compared with historical COVID-19 data from four U.S. states.
2 Agent-Based Model
The ABM represents heterogeneous agents through occupation, health, productivity, mental well-being, health risk, and compliance attributes. It couples simulated interactions and welfare with government restrictions and group-specific compliance dynamics.
- Agent Description: Agents are assigned occupations, health states, economic productivity, mental well-being, health risk, and compliance attributes that shape their behavior and interactions.Occupations are Essential, Remote, or Non-essential non-remote; health states include susceptible, exposed, tested, infected, recovered, and hospitalized.
- Restriction Policy: Government restrictions are represented by an index from 0 for no restrictions to 1 for full lockdown, influencing voluntary interactions.Compliance ranges from 0 to 1, with higher compliance reducing voluntary interaction counts.
- Compliance and Health Dynamics: The ABM couples daily SIRTEM health-state dynamics with compliance updates based on welfare within occupation–health groups.Agents imitate higher-welfare compliance behavior within their own occupation–health group, and mean compliance is fed back to SIRTEM.
- Interaction Simulation: Necessary interactions depend on occupation and home or work mixing rates, whereas voluntary interactions depend on compliance and the restriction index.Essential workers have both home and work interactions; remote and non-essential non-remote workers use home mixing for necessary interactions.
- Interaction Simulation: Interactions determine economic productivity, mental well-being, and health risk, with non-essential non-remote workers gaining productivity from voluntary interactions with non-hospitalized agents.Essential and remote workers receive fixed productivity payoffs, while each voluntary interaction contributes to mental well-being.
3 Integration with SIRTEM
ABM-SIRTEM couples SIRTEM’s detailed epidemic states with an agent-based model of interactions, welfare, and compliance, updating disease dynamics and behavior in daily feedback cycles. Transmission parameters are calibrated by matching simulated positive and negative test counts to observations over receding weekly horizons.
- SIRTEM: SIRTEM extends compartmental modeling with testing, quarantine, hospitalization, immunity, symptom status, false test results, and spatially varying mixing.Its subcompartments are aggregated into the health states used by the ABM.
- Transmission coupling: Weekly transmission parameters capture transmission together with mixing and restrictions, while daily compliance determines the corresponding daily infection rate.The model uses the transmission parameter for each day’s week and updates compliance daily.
- Integration with SIRTEM: The integrated model alternates daily between SIRTEM health-state updates and ABM compliance updates, feeding mean compliance back into the infection rate.Health states affect interaction payoffs, payoffs affect compliance, and compliance changes SIRTEM’s infection rate.
- Bayesian optimization: Bayesian optimization selects transmission parameters by minimizing normalized mean squared error between simulated and observed daily positive and negative test counts.A receding horizon evaluates candidate weekly rates, fits a surrogate objective, and advances one week after selecting the lowest evaluated objective.
4 Experiments
The experiments calibrate ABM-SIRTEM for Arizona, Florida, Minnesota, and Wisconsin using state-specific occupation, telework, policy, and contact inputs. Figure 4 compares fitted test counts with historical data and displays simulated compliance alongside observed restrictions.
- Calibration: The model is calibrated for Arizona, Florida, Minnesota, and Wisconsin using transmission parameters estimated from observed positive and negative test counts.The calibration procedure is applied to four U.S. states.
- Inputs: State occupation distributions are based on May 2019 BLS employment estimates, with telework capability estimated from October 2022 BLS survey responses.Occupational groups are also classified as essential or non-essential using Arizona Executive Order 2020-12.
- Inputs: The experiments use the OxCGRT stringency index, rescaled to [0, 1], as the government restriction index.The index summarizes containment and closure policies, including workplace and school closures, gathering and movement restrictions, and travel controls.
- Inputs: Daily at-home and at-work mixing rates are derived from contact surveys and normalized using each state’s population.Arizona contact averages are listed by wave and applied over the corresponding periods.
- Calibration results: Figure 4 reports simulated and historical daily positive and negative test counts, plus simulated mean compliance and the observed government restriction index.The same calibration procedure produces results for each state.
5 Results
The calibrated model broadly reproduces the timing of larger observed positive-test waves across four states, while negative-test fits vary and compliance changes alongside restrictions and test counts. These comparisons assess calibration fit rather than held-out forecasting performance.
- Positive-test trajectories broadly reproduce the timing of larger observed waves across Arizona, Florida, Minnesota, and Wisconsin.
- Negative-test fits vary across states, with substantial underestimation during parts of the Minnesota and Wisconsin simulations.
- Because observations within the lookahead horizon estimate transmission parameters, the comparisons evaluate calibration fit rather than held-out-data performance.
- Compliance and restrictions: Simulated mean compliance changes alongside the historical government restriction index and test counts.
6 Discussion
The discussion situates ABM-SIRTEM among work on pandemic modeling, economic impacts, and policy optimization. It concludes that the hybrid model represents coupled behavioral and epidemiological dynamics, while forecasting and policy-optimization value still require further validation.
- Related Work: Related research uses reinforcement learning, individual-agent contact networks, and economics–epidemiology reviews to address pandemic interventions and policy design.
- Conclusions: ABM-SIRTEM combines agent-based and population-level models to represent individual behavior, occupation-dependent incentives, and feedback between compliance and disease dynamics.
- Conclusions: The model’s value for forecasting and policy optimization remains subject to further validation.
- Future Work: Future work will examine forecasting under varying conditions and optimize restrictions against welfare, deaths, hospitalizations, and infections.