Fig. 2
Parameter map for sensitivity analyses accounting for potential unmeasured confounding. Literature-derived input parameters: RRCXY: unadjusted meta-analysis-derived risk ratio (RR), presumed to have unmeasured confounding; RRTXY: true RR estimate, corrected for presumed unmeasured confounding; RRXU: RR between the exposure and U; RRUY: RR between U and the outcome; τ²: heterogeneity between studies. User-specified input parameters: q: minimum threshold (RR scale) of a meaningfully strong effect; r: the minimum proportion of constituent studies with true effects above q required to indicate moderate to strong evidence of an effect; p: proportion of heterogeneity (τ²) due to variation in confounding bias. Output parameters: E-value: Minimum confounding strength (risk ratio (RR) scale) by an unmeasured confounder that could explain RRCXY); Bias factor: maximum amount the unmeasured confounder could impact RRCXY, derived from RRXU and RRUY, applied to RRCXY to calculate RRTXY; p̂(q): the proportion of studies that have scientifically meaningful strong effects (RR > q); Ĝ(r, q): the minimum confounding strength required to lower p̂ to < r; \({\rm{\hat T}}\)(r, q): the minimum bias factor required to lower p̂ to < r