You can't use 2020 vs 2021 as annual death rate cut off. It created biased variation due to vaccine was not widely available until May. You should use rolling forward looking projection based on current run rate. Take Santa Clara county for example, it's close to herd immunity now with 72% fully vaccinated. Everything is open and no mask requirement. The current death rate is 2 per day due to unvaccinated people. You can redo your model now -:)