Masks and COVID-19: a causal framework for imputing value to public-health interventions

Considerable analytical work during the COVID-19 pandemic has been devoted to developing predictive models to evaluate the merits of potential governmental interventions versus the cost of not carrying them out. These prospective modeling efforts are based on formal assumptions, like the choice of epidemiological mode, and on the extrapolation of past timecourse data to the future. As both these aspects carry substantial potential for error, trusted models used by decision-makers of public policy undergo nearly continuous revision cycles. Less methodologically developed is to assess retrospectively the effects these interventions had. Here we present a proposal for a model-free data-driven framework. We use a sparse regression method to fit the dynamic progression of the data (the Rt parameter) with the least number of changes, allowing us to attribute discrete change events to specific intervention events. Since the sparse fit yields a discrete jump between two constant Rt, we follow do-operator prescriptions and simulate the counterfactual case by forcing Rt to stay at the pre-jump value. We then attribute a nominal value to the intervention from the difference between true evolution and simulated counterfactual, for example, in terms of reducing the number of cases or accelerating the timecourse towards a milestone. As an example, we show that the recommendation to use facemasks for all activities would result in a putative total of 71000 (95% CI 64000 to 78000) fewer official cases in New York State.