I believe what you're asking about is covered in table 1 and figure 2.

They're reporting hazard ratios after propensity score matching with people who had the flu during the same time period.

This is not a 1 in a million event. There are studies like my comment above that back this statement whole heartedly. This is a virus with proven neuroinvasive potential that affects large swaths of otherwise incredibly healthy people.

You can't compare these supposed 1 in a million events to something like doing perf testing on a 64 node raspberry pi kubernetes cluster.