Table 1.

overview of adaptions

GRADE criteria	Original	Challenge	Specifications or adaptions
Study type/methodological quality	NRS start as “low quality” of evidencea	Data on harms from RCTs is often insufficient and thus it is advisable to consider NRS	NRS start as “high quality” of evidence if rated as low risk for confounding and selection bias
Imprecision (binary outcomes)	Usually, 95% of CIs of relative effects are used 95%CI overlaps decision threshold (e.g. null effect) → rating down one level 95%CI includes appreciable harm and benefit → rating down two levels	Harms are often rare events and rare in the included studies despite large sample sizes. In the case of rare events 95% CIs of relative effects can be misleading.	Imprecision is assessed based on absolute effects
Publication bias/missing results in the synthesis	Rating down for publication bias	For harms selective dissemination would result in underestimation of harms	Rating up for publication bias
Large magnitude of effect	Rating up if RR >2 (<0.5)	Harms are usually less affected by confounding by indication	Rating up if RR >1.67 (<0.60)
Originally not used	Subgroup effects	Harms are often subgroup-specific but analysis within subgroups is underpowered	Rating up if there is a statistically significant subgroup effect from a well-designed subgroup analysis

^aWhen using the ROBINS-I tool NRS usually start as high quality of evidence and an adaption is not necessary