Create a Forest Plot from Univariate Regression Tables

This function generates a forest plot from the results of MakeUnivariateRegressionTable().

plotForestFromTable() was renamed to PlotForestFromTable() in SciDataReportR 20.5.0 to match the package's Plot* naming convention. It remains available as a backwards-compatible synonym.

Usage

PlotForestFromTable(UnivariateRegressionTables, pSize = 2, Flip = FALSE)

plotForestFromTable(UnivariateRegressionTables, pSize = 2, Flip = FALSE)

Arguments

UnivariateRegressionTables

Either the full list returned by MakeUnivariateRegressionTable() (its Results dataframe is used directly), or a dataframe with the Results columns. Passing a dataframe lets you filter, reorder, or relabel Results before plotting; required columns are OutcomeLabel, TermLabel, Estimate, ConfLow, ConfHigh, and PValue (Significant and ReferenceValue are recomputed if absent). Lists created by older package versions (without a Results element) are still supported.

pSize

Numeric. Size of the points in the plot. Default is 2.

Flip

Logical. If FALSE, outcomes are facets and predictors/terms are rows. If TRUE, predictors/terms are facets and outcomes are rows.

Value

A ggplot object representing the forest plot.

Reading the plot

The forest plot earns its place when several predictors are screened against several outcomes at once: one panel per outcome, the same predictors down every panel, so a predictor that matters for one outcome and not the others is visible in a single glance.

Flip = TRUE swaps the roles - one panel per predictor, outcomes down the rows. Use it to ask "what does age predict?" rather than "what predicts tau?": the same estimates, organized around the other question.

Passing the Results dataframe instead of the whole object means the plot can be filtered first, for instance to the associations that survive FDR correction across the whole screen. Standardized estimates put every predictor on the same axis, which is what makes the widths of the intervals comparable across rows. A binary outcome gives odds ratios, and the reference line moves from 0 to 1.

Examples

# \donttest{
data(SampleData)
data(SampleVariableTypes)

Labelled <- RevalueData(SampleData, SampleVariableTypes)$RevaluedData

# Several predictors screened against several outcomes
urt <- MakeUnivariateRegressionTable(
  data = Labelled,
  outcome_vars = c("AXL", "tau", "p_tau", "Ferritin"),
  predictor_vars = c("age", "sex", "Adiponectin", "Cortisol", "Insulin")
)

PlotForestFromTable(urt)


# One panel per predictor instead
PlotForestFromTable(urt, Flip = TRUE)


# Filtered to associations surviving FDR correction
urt$Results$PValueFDR <- ApplyFDRCorrection(urt$Results$PValue)
PlotForestFromTable(urt$Results[urt$Results$PValueFDR < 0.05, ])


# Standardized estimates
urt_Std <- MakeUnivariateRegressionTable(
  data = Labelled,
  outcome_vars = c("AXL", "tau", "p_tau", "Ferritin"),
  predictor_vars = c("age", "sex", "Adiponectin", "Cortisol", "Insulin"),
  Standardize = TRUE
)
PlotForestFromTable(urt_Std)


# A binary outcome: odds ratios
urt_Logistic <- MakeUnivariateRegressionTable(
  data = Labelled,
  outcome_vars = "Diagnosis",
  predictor_vars = c("age", "sex", "AXL", "tau", "p_tau")
)
PlotForestFromTable(urt_Logistic)

# }