Their success metric may just be reversion to the mean; since they started by selecting the people who had made the most calls, you would expect the number of calls to come from those specific people to fall in the coming years. Since there was also likely some underlying health condition, you would expect the between-year variance to be large. It might well be that their program has no effect whatsoever.
If no other group is calling at a similar level, then it's not reversion to the mean. The mean has actually been shifted downwards.
> Since there was also likely some underlying health condition, you would expect the between-year variance to be large. It might well be that their program has no effect whatsoever.
The rate dropped significantly each year from 2015 through 2017 -- 3 years in row. It's possible this is pure chance, but it seems rather unlikely.
If they're really tracking the same 100 identified people, with no replacement, then you'd expect the same result even without regression to the mean, simply because very sick people tend to die, and dead people tend not to call 911.
One would hope that they're not doing that, and it's just that the article is imperfectly worded.
