One last quick comment on reporting mortality generally, as this recent NY paper is one of many. We've been reporting mortality for many decades, and we have some conventions. First, and this will seem obvious, but it's important: mortality will vary between 0 and 100% based
on the time frame. If your time horizon is one nanosecond, mortality is 0%. If your time horizon is 200 years, mortality is 100%. Generally we don't want one of those answers, we either want mortality over a specific human-relevant time horizon or, commonly we want the
"attributable" mortality, by which we mean the mortality that would NOT have occurred in the absence of the disease or exposure of interest. Bracketing the complex problem of "attributable" mortality, we can focus just on the observed mortality. But to do that, we have to
measure at a given moment in time. Call it 30 days, call it 90 days, call it 1 year, that's fine. If you don't mind a bit of imprecision, maybe you could measure "inhospital" mortality. But in any case, at that moment in time, the mortality is the ratio of dead to (dead + living)
To calculate that you have to know both of those numbers. During COVID19, we want to know the chances that people will die before they recover from COVID. To get that answer will take 6 months. But we're impatient. We want to know _now_. But we can't have answers in less than
6 months. So we have two choices: use a shorter-term time point and be honest about it, or find a surrogate or predictive model for 6-month mortality and admit that this introduces substantial noise. This is not hard to do, even if it doesn't provide us a satisfying answer right
away. But please never make this mistake again--do not exclude living people from your calculation of mortality. It makes no sense, and it leads to serious misinterpretation.
