,
5 tweets,
3 min read
Read on Twitter
A new study in JAMA-IM suggests that a care bundle for pneumonia patients including oral steroid, rapid transition to PO antibiotics, and early mobilization caused harm (increased GI ulceration). Yikes😱 Run for the hills! (#rantorial 1/5)
jamanetwork.com/journals/jamai…
jamanetwork.com/journals/jamai…
Here are the main results. Every outcome was negative except for GI bleeding. Due to the rarity of GI bleeding, this was analyzed using adjusted analysis with a regression model (#rantorial 2/5)
However, the fragility of this effect on GI bleeding is zero. If we perform an unadjusted Fisher exact test, the p-value is 0.09 (#rantorial 3/5)
Furthermore this table analyzes 10 outcomes. The likelihood of at least one being “significant” at a p-value cutoff <0.05 is equal to 1-(.95)^10 = 40%. So the appropriate p-value cutoff for GI bleeding should be <0.05 to account for multiple comparisons (#rantorial 4/5).
Overall, this paper seems to suggest that we should change practice on the basis of a *secondary* endpoint with an unadjusted p-value of 0.09. Especially considering the issue of multiple statistical comparisons, this just doesn’t work (#rantorial 5/5).
