Observational causal inference with infectious disease outcomes

Project Summary Infectious disease is a leading cause of global morbidity and mortality. Transmission dynamic models have played a critical role in guiding interventions related to many infectious pathogens, including HIV, influenza, SARS-CoV-1, ebolaviruses, SARS-CoV-2, and mpox. Models project how potential interventions (e.g., non- pharmaceutical measures, therapeutics, and vaccines) may affect disease future transmission. However, they often rely on small scale studies to project effects, and there have been growing concerns that models may produce inaccurate, overly optimistic estimates of population-level intervention effectiveness. Observational causal inference models, which measure intervention effectiveness in real-world settings, could help address this limitation, but applying these to infectious disease is not straightforward. Observational approaches, such as difference-in-differences and synthetic control methods, estimate the impact of an intervention based on empirical counterfactuals: comparing outcomes of interest between treated units or places and similar untreated units. While well-understood with linear outcomes, they can produce biased and misleading results in the context of nonlinear transmission dynamics. Even where observational models perform well, it further remains challenging to transport estimates to new settings to project the impact of future interventions. To address these issues, this project will develop comprehensive theoretical architecture for synthesizing transmission dynamic models with observational causal inference models - employing empirical counterfactuals while accounting for complex biological and population dynamics. In the retrospective workstream, I will propose robust specifications for observational causal inference models that can estimate unbiased treatment effects in policy evaluations using infectious disease outcomes. I will also develop model selection and decision-analytic methods to address potentially significant parameter uncertainty. In the prospective workstream, I will develop approaches to generalize estimates from observational causal inference models to new settings using transmission dynamic models and update projected effects in real-time based on local surveillance indicators. I will illustrate the implications of our methods by re-analyzing prior studies on COVID-19 as well as applying them to answer new questions about respiratory illness control, in collaboration with partners in state and local public health institutions. Across both workstreams, I will develop and disseminate open-source public tools and software to facilitate adoption of these methods. Overall, this work will produce a suite of methodological innovations to improve understanding of the impact of past policies and the accuracy of future projections, while also supporting their implementation in public health institutions to guide planning and priority setting.