Hacker News Reader

Comment by ckastner | Hacker News Reader

As far as I understand it, they are comparing the statistics of the 5,683 sample to the 17M in order to determine whether anything stands out. If Covid-19 deaths had no associated factors, then we shouldn't see much of a difference between the two.

However, we are seeing differences, for example with diabetes.

View on HN

Yes that's true, but the 17M figure isn't the interesting / groundbreaking thing. At least it shouldn't be made out to be as if that's the significant thing.

17M just provides the baseline of "normal". Tons more data isn't needed to get that right. 170M wouldn't make a difference. I bet 1.7M wouldn't make a big difference either.

What we really need to know right now is how the Covid deaths behave. All that this offers is the 5,683 sample, but the title inflates it to emphasize the 17M. We need the 5,683 sample to be merged with other samples to reach 56,830, etc. That is the significant thing.

I have to admit, the title does sound misleading in this regard, and (to me) even more so the tweets advertising the paper as per the submission.

You misunderstand. The only reason the study has enough power to make statements about the relative risk of, say, obesity independently of age, sex, and ethnicity is because there are 17 million patients to use as "controls".

The 17M is the data set from detailed primary care records linked to 5,683 of the deaths recorded in English hospitals in the time period. The data could be improved by linking to more primary care records - but the population of England is 56 million so they already captured almost a third.

Sadly there will be more data as there are more deaths, and this team will be able to pull more analytics out. However 5,683 is a big enough sample size to find some interesting information on hazard rates.

For the team doing the analysis, plugging data analytics into 17 million primary care records is definitely a significant achievement. Medical records are intended for use by doctors and letting researchers into them has confidentiality issues. The process they followed was covered in the paper and seemed thorough.

Reply on news.ycombinator.com