Excess deaths is the most reliable one simply because it tracks the impact of a new disease. However, even that has problems because you then have to accommodate increased deaths due to the withdrawal of healthcare provision for other illnesses and disease. How many people could have lived if they could have seen a GP in person, or made that hospital appointment that was cancelled etc.
As I say, my best guess is the stats are generally around 15 to 20% off across the board, maybe as low as 10%, but I say that with absolutely no evidence because... Well, that's the problem isn't it?