How can we answer questions of causation (“what if?” questions) when we can’t perform experiments directly. In the case of COVID-19, a typical causal question would look like this: what would happen if we prescribed people anti-inflammatory medications? Would it improve outcomes? While we can answer this question definitively by performing a randomized trial, this approach is slow. With this outbreak, time is of the essence and there are many potential treatments to evaluate. Instead of performing the randomization ourselves, our approach is to exploit a randomization that nature has provided for us in genetics.
The intrinsically random nature of genetic transmission from parent to child provides us with a natural experiment. In our most recent work, we consider genetic variants that mimic the effect of angiotensin-converting enzyme inhibitors (ACE-inhibitors). ACE-inhibitors are blood pressure lowering drugs that have anecdotally been linked to risk of COVID-19 infection. We compare individuals with genetic variants that predispose them to higher or lower levels of ACE in a similar way to how in a randomized trial, we would compare individuals allocated at random to ACE-inhibitors or to a placebo.
Our approach relies on several assumptions about the nature of genetic inheritance, but we can assess the plausibility of these assumptions in our data. For example, are the genetic variants correlated with diabetes status? If the genetic variants aren’t associated with competing risk factors, then we have a fair test – simplifying the details somewhat, if the genetic group of individuals with the ACE lowering variant and the genetic group of individuals without the ACE lowering variant only differ with respect to their levels of ACE, then any difference in outcomes between the genetic groups must be due to the ACE pathway – and hence ACE inhibition influences the outcome.