We enjoyed the first one so much, we thought we'd go again. Our second #WorldEBHCDay #Tweetorial we want to showcase is the infrequent times when you may need to upgrade the certainty you have in your evidence when following the #GRADE approach @JBIEBHC @WorldEBHCDay (1/8)
Evidence from observational studies and non-randomised experimental studies is (under most-circumstances) initially classified as low. However, there are rare occasions when these studies may provide higher than low certainty in the evidence #Tweetorial #WorldEBHCDay (2/8)
There are 3 determinants to consider when raising the certainty in the evidence. A large magnitude of effect, a dose response gradient and the effect of plausible residual confounding #Tweetorial #WorldEBHCDay (3/8)
Large Magnitude of Effect. When the effect estimate is large (RR of 2/0.5) or v. large (RR of 5/0.2) we could upgrade the certainty we have in the evidence. OR of babies sleeping on stomachs of 4.1 (95%CI 3.1-5.5) for SIDS compared to back sleeping #Tweetorial #WorldEBHCDay (4/8)
Dose Response Gradient. The example below would represent another opportunity to upgrade the certainty we have in the evidence. As time of antimicrobial initiation following hypotension onset increases, survival decreases #Tweetorial #WorldEBHCDay (5/8) 
Plausible Residual Confounding. When confounders unaccounted for in adjusted analysis may underestimate a treatment effect (sickest patients exposed to factor, yet still fare better) it’s likely that the true effect is larger than that estimated #Tweetorial #WorldEBHCDay (6/8)
A few disclaimers to close. We would only upgrade the certainty we have in the evidence from these studies if we have not already downgraded the certainty based on our 5 standard determinants #Tweetorial #WorldEBHCDay (7/8)
When upgrading the quality of the evidence for large effect sizes, do so cautiously and consider the precision of the CIs. The decision to rate up should only be made rarely. More detailed guidance here: shorturl.at/ipGJ2 #tweetorial #WorldEBHCDay (8/8)
• • •
Missing some Tweet in this thread? You can try to
force a refresh
