Using causal machine learning to disentangle the demographic, social, and medical drivers of COVID-19 outcomes
Approved Research ID: 66246
Approval date: November 25th 2020
A myriad of demographic, economic, behavioural and medical factors have been suggested as drivers of differential COVID-19 disease outcomes. Many of these, such as race and socioeconomic status, have long been associated with institutional disparities, that themselves have been suggested as potential causes for health disparities. While these factors are apparently correlated with worse COVID-19 outcomes, their causal pathways are much less understood because they are often confounded with each other. Without a true understanding of their causal relationships, however, an health interventional policy is like shooting targets in the dark: at best an imprecise approach, and at worst a waste of resources for ineffective outcomes.Relying solely on predictive models-- especially in health outcomes-- risks serious consequences when correlations are mistaken for causation.
This project aims to use causal machine learning techniques to build a better understanding of these variables' intertwined roles in COVID-19 outcome disparities. We will use machine learning techniques to gain a thorough understanding of the causal pathways behind differential COVID-19 outcomes.
Since the pandemic is ongoing, we can only estimate that the project will last 6 to 12 months after the global pandemic has subsided to complete near-real time and subsequent analyses. We estimate that this project will last at least 24 months.
Our research will provide a better understanding of which factors are most important in determining COVID-19 infections, hospitalizations, and deaths. Understanding how different risk factors are related to one another and to COVID outcomes will provide insight into the most effective public health interventions. Our machine learning analysis will provide insight into which groups or behaviors to target to reduce adverse COVID outcomes.