Table 3 How trial sequential analysis can supplement the assessment of GRADE for ‘imprecision’[77]

If none of the sequential boundaries for benefit, harm, or futility are crossed and the anticipated intervention effect is realistic.

The evidence should be downgraded two levels of quality according to imprecision (see Table 1).

If one of the boundaries for benefit, harm, or futility are crossed and the anticipated intervention effect is realistic.

The evidence should not be downgraded according to imprecision (see Table 1).

If the anticipated intervention effect is considered unrealistic.

The trial sequential analysis should be repeated using the limit of the confidence interval, closest to zero effect from the traditional meta-analysis as the anticipated intervention effect. If the sequential boundaries are crossed then the level of evidence should not be downgraded (see Table 1). If the sequential boundaries are not crossed, the trial sequential analysis should be repeated, this time, using the intervention effect estimate from the meta-analysis as the anticipated intervention effect. If the sequential boundaries are crossed, then the evidence should be downgraded one level of quality; if the sequential boundaries are not crossed, then the evidence should be downgraded two levels of quality (see Table 1).