Case Study: Understanding Geographic Impact on Infectious Disease Modelling with Advanced Sensitivity Analysis

Unpacking the Pandemic Puzzle: How Geography Shapes Disease Spread – A Breakthrough Powered by SEAVEA's FabSim3 and EasyVVUQ

The COVID-19 pandemic underscored the critical need for accurate infectious disease models to guide public health decisions. But how do we ensure these models truly reflect the intricate, often messy, realities of diverse geographical landscapes, and how certain can we be of their predictions?

A fascinating paper by Arindam Saha and his colleagues – "Impact of geography on the importance of parameters in infectious disease models" (2023) – tackled this very question. Their groundbreaking work, which revealed how geographical variations fundamentally alter which disease parameters matter most, was made possible by advanced computational tools, including FabSim3 and EasyVVUQ from the SEAVEA toolkit.

The Geographic Conundrum: Why Location Matters

It's intuitive that disease spread differs between a dense, interconnected city and a sparsely populated rural area. But to what extent? And how does this geographical influence affect the effectiveness of public health interventions?

The challenge lies in the complex nature of real-world geography, which can't be easily reduced to simple model inputs. Saha and his team aimed to quantify this intricate relationship, exploring how geographical variations impact the sensitivity of input parameters in agent-based models (ABMs) of infectious disease. ABMs, which simulate individual behaviours and interactions, are ideal for capturing these nuanced, granular effects.

Powering Discovery: The Indispensable Role of FabSim3 and EasyVVUQ (from SEAVEA)

To unravel this complex web of interactions and quantify the uncertainties, the researchers embarked on a massive computational undertaking. They performed a grouped Sobol's sensitivity analysis on COVID-19 spread simulations across three distinct real-world geographical representations. This advanced statistical method is paramount for understanding how multiple input parameters, even those that interact, affect model outcomes.

Such an extensive analysis, involving thousands, potentially millions, of individual simulations across high-fidelity geographical models, demands a robust and efficient automation framework alongside sophisticated UQ/SA capabilities. This is precisely where FabSim3 and EasyVVUQ, core components of the SEAVEA toolkit, proved to be indispensable.

• FabSim3 served as the automation toolkit, designed for managing, submitting, executing, retrieving, and analysing large-scale simulation workflows on high-performance computing (HPC) clusters. For this research, FabSim3 was instrumental in:

  • Orchestrating Complex Simulations: Handling the intricate setup and execution of numerous agent-based disease spread simulations across varied geographical datasets.

  • Automating Computational Campaigns: Streamlining the process of running thousands of Sobol's sensitivity analysis samples, which would be prohibitively time-consuming and error-prone if done manually.

  • Managing HPC Resources: Efficiently submitting and managing jobs on supercomputing environments, ensuring the computational power was effectively harnessed for the analysis.

• EasyVVUQ was then deployed for the Uncertainty Quantification (UQ) and Sensitivity Analysis (SA) work. This Python package seamlessly integrates with simulation codes to enable efficient and robust UQ and SA. For this research, EasyVVUQ provided the methods to:

  • Perform Sobol's analysis: Quantitatively determine the contribution of each input parameter (and their interactions) to the variance in the model's output.

  • Manage Samplers and Encoders: Facilitate the systematic generation of input variations for the simulations and the processing of their outputs for sensitivity analysis.

  • Provide Confidence in Results:  EasyVVUQ ensures that the findings on parameter sensitivities are rigorously quantified and reliable, directly supporting the "actionable" aspect of the SEAVEA toolkit's mission.

The SEAVEA (Software Environment for Actionable & VVUQ-evaluated Exascale Applications) toolkit focuses on providing comprehensive Verification, Validation, and Uncertainty Quantification (VVUQ) capabilities for large-scale applications. The combined and successful application of FabSim3 and EasyVVUQ in this research is a powerful demonstration of how SEAVEA's tools contribute directly to making complex scientific simulations more robust, reliable, and actionable.

Key Results and Wider Impact

The findings from this computationally-driven research provide crucial insights:

• Geographical Impact on Parameter Sensitivity: The study found compelling evidence that the infection rate can be significantly more sensitive (important) in regions with segregated populations. This suggests that in highly compartmentalised communities, interventions targeting transmission rates are even more critical.

• Varying Importance of Recovery Time: Conversely, parameters like the recovery period of mild cases showed greater sensitivity in regions with mixed populations. In areas with high inter-group interaction, how long an individual remains infectious plays a more pronounced role in overall disease spread.

• Dynamic Sensitivity: The research also highlighted how geographical structure influences how parameter sensitivities change over time during an epidemic.

These results have a profound and practical impact:

• Smarter Model Design: The findings guide modellers in building more accurate, context-specific predictive models, ensuring the correct parameters are weighted appropriately for different geographical settings.

• Tailored Public Health Strategies: Armed with this granular understanding, public health officials can develop highly targeted interventions. For example, resource allocation for rapid contact tracing might be prioritised in segregated areas, while public awareness campaigns about recovery periods might be more impactful in mixed populations.

• Validation for Complex Models: The study provides strong validation for using detailed, high-fidelity models that incorporate real-world geographical data, moving beyond simpler, stylised representations. This, in turn, underscores the continuing need for advanced computational tools like FabSim3 and EasyVVUQ.

The Future of Informed Decision-Making

The "Impact of geography on the importance of parameters in infectious disease models" paper is a stellar example of how cutting-edge scientific questions, when paired with powerful computational tools like FabSim3 and EasyVVUQ from the SEAVEA toolkit, can lead to transformative insights. By enabling the rigorous exploration of complex scenarios and robust uncertainty quantification, SEAVEA is directly contributing to a future where our responses to global challenges are more precise, effective, and informed by the best available science.