Trim-and-fill method

Definition:

A method used in meta-analysis to assess how publication bias might affect the pooled effect estimate. It assumes that missing studies on one side of a funnel plot account for the plot’s asymmetry. The procedure temporarily removes (“trims”) extreme studies from the more populated side and repeatedly re-estimates the centre of the plot. It then restores the trimmed studies and adds hypothetical studies that mirror them on the opposite side (“fills”). The pooled effect and its confidence interval are recalculated using the observed and imputed studies (Duval & Tweedie, 2000).

Funnel-plot asymmetry can arise for reasons other than publication bias, including differences in the underlying effects across studies. Trim-and-fill therefore provides a sensitivity analysis under particular assumptions about missing studies; it does not establish that publication bias occurred or guarantee an unbiased effect estimate (Shi & Lin, 2019; Viechtbauer, n.d.).

Several methodologists recommend using other approaches, such as selection models or suitable regression-based methods, instead of trim-and-fill because it can leave substantial bias uncorrected and perform poorly when effects differ across studies (Moreno et al., 2009; Carter et al., 2019).