Having read the article and the abstract of the paper in question, it seems a little.. off. For a start, it's a meta-analysis pulling from a simply massive pool. And it asks people about areas of their life from a few years upto two and a half decades.

Do YOU know how often you slept under six hours a night in the last year? 2 years? 5? 10?

Did all these people keep detailed sleep diaries every single night?

This just raises too many questions. Meta-analysis is most useful when restricted to much smaller sample sizes where a greater amount of data can be obtained and analyzed.

-- Edit for More Info:

It seems that this paper is a meta-study of OTHER papers. And that 80% of the data used came from a single study by Daniel F. Kripke of UCSD.

The conclusions of Kripke study, which were meta-analysed, were notably different than Cappuccio's, namely:

"The best survival rates were found among those who slept 7 hours per night. The study showed that a group sleeping 8 hours were 12 percent more likely to die within the six-year period than those sleeping 7 hours, other factors being equal. Even those with as little as 5 hours sleep lived longer than participants with 8 hours or more per night."