The argument wasn't specific to Covid. The people who get "severe viral infections" tend to be people who are old and unhealthy to begin with. When you start with a population of medical records (as this study did) you unavoidably bias your sample toward sicker people.

Observational studies don't/can't perfectly correct for this residual confounding. This exact error has been repeatedly made by "long Covid" research as well (most notably by Ziyad Al-Aly, who has published a half dozen different articles on Long Covid using the same confounded, observational VA medical record dataset, and claimed associations with a huge number of different illnesses. This research is garbage, but it has scared people out of their wits.)