That's the general idea but some care is needed to account for any trends and to account for the natural variance in the numbers. For example the 2019 numbers

That's the general idea but some care is needed to account for any trends and to account for the natural variance in the numbers. For example the 2019 numbers subtracted from 2018 numbers wouldnt be zero. There will be some systematic trends, seasonal variations and some unexplained fluctuations. This requires some model fitting that uses not only past mortality data but also the usual macroeconomic data. Once you are at a point where you see that there are no systematic pattern in the error between the predicted number and the actual numbers, that this error is equally likely to be an over-estimate as an under estimate, then you would be in a position to use the model.