This is a super, super Twitter thread that nicely illustrates the problem of using survival analysis to predict positive outcomes (e.g., time to recovery). He does a great job in the first few tweets of reminding you what the data structure of a time-to-event data set looks like, and then promptly illustrates the problem when applied to something like “time to recovery” from COVID. 

I know you’re not in my class anymore, and I won’t keep sending messages forever, but I also am unable to just turn off my teacher brain :-).

Michael

 
 
Andrew Althouse
⁦‪@ADAlthousePhD‬⁩
The NEJM compassionate-use remdesivir data was always sort of a "passing curiosity while we await trial data" but it's disappointing that the analysis was done wrong.

But, it offers an opportunity for a teachable moment about "time-to-event" analysis of non-mortality outcomes.
 
5/21/20, 9:13 AM
 
 

Download the Twitter app


Michael Marsiske 
Professor
Department of Clinical and Health Psychology
University of Florida
PO Box 100165
1225 Center Dr, Room 3150
Gainesville, FL 32610-0165
Phone: (352) 273-5097
Fax: (801) 720-5897
Email: [log in to unmask]
Website: http://marsiskelab.phhp.ufl.edu
Twitter: @MMarsiske