,
4 tweets,
4 min read
Read on Twitter
Excited about real-world data (ie, electronic health records) for causal inference? a cautionary tale👇🏾
in @uk_biobank analysis LDL extremes & risk for future coronary disease (one of surest causal relationships in medicine) by @jpirruccello yields:
WRONG ANSWER
What happened?
in @uk_biobank analysis LDL extremes & risk for future coronary disease (one of surest causal relationships in medicine) by @jpirruccello yields:
WRONG ANSWER
What happened?
Answer: confounding based on LDL-lowering treatment at entry into @uk_biobank
(L) plot without accounting for lipid-lowering therapy
(R) plot after accounting for lipid-lowering therapy at baseline (about 17% of participants on lipid-lowering Rx at entry)
(L) plot without accounting for lipid-lowering therapy
(R) plot after accounting for lipid-lowering therapy at baseline (about 17% of participants on lipid-lowering Rx at entry)
Large scale data-analyses with millions of real world evidence data points,
just imagine all the false causal inference claims that we can look forward to!!
😳
just imagine all the false causal inference claims that we can look forward to!!
😳
By the way, here are the distributions of 29 biomarkers measured in nearly all 500K @uk_biobank (plots care of @jpirruccello)
Just released to us.
Thank you @uk_biobank
Just released to us.
Thank you @uk_biobank
