OHDSI Home | Forums | Wiki | Github

Physiologically Plausible Measurement Values

Researchers
#1

I suspect that someone in the community (@Vojtech_Huser?) has already done this so I wanted to check before writing any code…

I’m looking to generate a list of plausible upper and lower limits for measurement values. Not normal ranges, but rather the 99+% distribution of all values seen, with the goal of tossing out clearly erroneous values (eg Potassium 0.3, Weight 12000) that are probably due to data entry.

Has someone generated such a list from Achilles Heel or other data quality queries?

Thanks!

Jon

#2

This would be useful for a lot of people. One possible solution is to use NHANES to identify extreme value cutoffs. The QC on their labs is great, and it might be possible to use their data to identify cutoff values. I have not looked into this, but it might be possible to get into the right ballpark for a good number of common lab values.

It is a tricky topic though since there are people who are missing limbs (height, weight), are on dialysis (almost any lab measure), have genetic diseases (familial hypercholesterolemia, etc) and it is hard to know what “wrong” is sometimes.

#3

Thanks @Mark_Danese. NHANES is a good thought and I will go ahead and put together some queries since it doesn’t sound like its duplicating what others have already done.

I agree the definition of “wrong” has to be very loose. Hopefully if multiple sites can run then we will be able to triangulate towards including the extreme while excluding the implausible.

t