A finding of reduced deaths from any cause, particularly by as much as they found, more likely indicates that the data is being fucked with in a substantial way. There is some sort of bias going on, perhaps there is a selection bias in the study (only selecting healthy people to be included), or there is some filtering being applied to exclude certain groups that would have a higher all cause death rate (smokers, drug users, etc). You would need to do a fairly detailed analysis of the underlying study, which would require access to ALL of their data that was used and the methods/tools that were used too. These, of course should be published with the study, such that a sufficiently trained person could repeat it, in reality, that is very unlikely.