Although they had statistical significance with a p=0.001, it doesn't matter to the subjects. They're only falling asleep 5 minutes faster, and are personally not reporting an improvement in quality of life. So, clinically, this medication doesn't matter to the subjects because 5 minutes faster might not be that big of a deal.

It's not attrition bias because the threshold there is 5%. Here, 20/2000 subjects (1%) are lost, having little effect on attrition. Additionally, the acceptable range for bias is between 5-20%, which this doesn't approach. https://catalogofbias.org/biases/attrition-bias/

All I can contribute is: you always include everyone in the analysis even if they are not adherent

This is because clinical research follows intention-to-treat protocol

Dr. Jason Ryan from BB emphasized that just because something is statistically significant, does not automatically mean it is clinically significant!