Table 2.
Algorithm for determining “downgrades” to levels of evidence in reviews
| Area assessed |
Imprecision |
Risk of bias (trial quality) |
Inconsistency |
Risk of bias (review quality) |
|---|---|---|---|---|
| Method of assessment | Number of participants within pooled analysis | Proportion of participants included in the pooled analysis judged to have low ROB for randomization and observer blinding | Heterogeneity, assessed by I2 statistic | Responses to AMSTAR questions 1–4 (covering a priori research design, search characteristics, independence of study selection, and data extraction) |
| No downgrade (no serious limitations) | ≥200 | ≥75% of participants have low ROB | I2 ≤ 75% | 4/4 are all “yes” (i.e., low ROB) |
| Downgrade 1 level (serious limitations) | 100–199 | <75% of participants have low ROB | I2 > 75% | 3/4 are “yes” and 1 is “unclear” or “no” on AMSTAR |
| Downgrade 2 levels (very serious limitations) | 1–99 | < 3/4 are “yes” and remainder are “unclear” or “no” on AMSTAR | ||
| Notes | If ROB for individual trials was not reported within the review, we were conservative and assumed that less than 75% of participants had low ROB. | If only one trial contributed to analysis, no downgrade; if I2 not reported, assumed to be greater than 75%. |