Correct the effect size for publication bias/small-study effects

This function allows to add effect sizes estimates corrected for publication bias/ small-study effects to results of the runMetaAnalysis function.

Usage

correctPublicationBias(model, 
                       which.run = model$which.run[1],
                       lower.is.better = NULL,
                       selmodelSteps = 0.05,
                       ...)

Arguments

model: An object of class runMetaAnalysis, created by the runMetaAnalysis function.
which.run: The model in model that should be used for the publication bias analyses. Uses the first model in model if no value is specified by the user. Possible values are "overall", "combined", "lowest", "highest", "outliers", "influence" and "rob".
lower.is.better: Only used for the selection model. Should significant negative (TRUE) or significant positive (FALSE) effects be assumed to be favored for publication? This refers to the sign of the effect size in the data, not to the scale of the outcome measure. If set to NULL (default), the direction is determined by the sign of the pooled effect in the selected model. A warning is issued if the provided value contradicts the sign of the pooled effect.
selmodelSteps: Thresholds to be assumed for the step function in the selection model. Must be a vector of numbers referring to the cut-points in the selection models. If two-sided testing is assumed for the included studies, the cut-point must be doubled to obtain the assumed p-value (e.g. selmodelSteps = c(0.03, 0.05) means that p=0.06 and p=0.10 are assumed as selection thresholds). The default is 0.05.
...: Additional arguments. See trimfill.default and limitmeta.

Value

Returns an object of class "runMetaAnalysis" and "correctPublicationBias". This object includes all original objects included in model, but adds a list object with the name correctPublicationBias. This list object includes all three fitted publication bias analysis models, as well as the generated results.

Details

The correctPublicationBias function is a wrapper running three meta-analytic methods to control the pooled effect size for publication bias and/or small-study effects:

"trimfill". Applies Duval and Tweedie's (2000a, 2000b) trim-and-fill algorithm, using the trimfill method in the meta package.
"limitmeta". Runs a limit meta-analysis as described in Rücker et al. (2011), using the limitmeta function in the metasens package.
"selection". Runs a step function selection model using the selmodel function in metafor. For details see e.g. Vevea and Hedges (1995). The cut-points in selmodelSteps refer to one-sided p-values. Selection models are not estimated for event rates (es.measure = "EER" or "CER"), for which selection on significance is not a meaningful assumption.

If the selected model is a fixed-effect ("common effect") model (method.tau = "FE"), the trim-and-fill and selection model are also fitted assuming a fixed effect. The limit meta-analysis is based on a random-effects model by definition.

References

Duval S & Tweedie R (2000a): A nonparametric "Trim and Fill" method of accounting for publication bias in meta-analysis. Journal of the American Statistical Association, 95, 89–98

Duval S & Tweedie R (2000b): Trim and Fill: A simple funnel-plot-based method of testing and adjusting for publication bias in meta-analysis. Biometrics, 56, 455–63

Rücker G, Schwarzer G, Carpenter JR, Binder H, Schumacher M (2011): Treatment-effect estimates adjusted for small-study effects via a limit meta-analysis. Biostatistics, 12, 122–42

Vevea, J. L., & Hedges, L. V. (1995). A general linear model for estimating effect size in the presence of publication bias. Psychometrika, 60(3), 419–435. ⁠https://doi.org/10.1007/BF02294384⁠

Author

Mathias Harrer mathias.h.harrer@gmail.com, Paula Kuper paula.r.kuper@gmail.com, Pim Cuijpers p.cuijpers@vu.nl

Examples

if (FALSE) { # \dontrun{
data("depressionPsyCtr")
depressionPsyCtr %>%
  checkDataFormat() %>%
  checkConflicts() %>%
  calculateEffectSizes() %>%
  filterPoolingData(condition_arm1 %in% c("cbt", "pst")) %>%
  runMetaAnalysis() -> res

# Correct for small-study-effects/publication bias
res %>% correctPublicationBias()
# Use additional arguments to control settings of the trim-and-fill
# and limit meta-analysis
correctPublicationBias(res,
                       which.run = "combined",
                       type = "R",
                       method.adjust = "mulim") 
 
# Generate plots
correctPublicationBias(res) %>% plot("trimfill")
correctPublicationBias(res) %>% plot("limitmeta")
correctPublicationBias(res) %>% plot("selection")

# Returned object is of class "runMetaAnalysis"; therefore,
# all S3 methods are available:
res <- res %>% correctPublicationBias()
metaRegression(res$model.overall, ~country)
} # }