Plot diagnostic likelihood ratios as a forest plot

Visualizes likelihood ratios calculated by DiagnosticLikelihoodRatioTable() without recalculating diagnostic statistics. Each point is a diagnostic test result and its horizontal interval is the likelihood-ratio confidence interval. The logarithmic scale makes reciprocal likelihood ratios equally distant from the neutral value of one.

Usage

PlotDiagnosticLRForest(
  x,
  result = c("all", "positive", "negative"),
  predictor_order = c("original", "alphabetical", "strength"),
  outcome_order = c("original", "alphabetical", "strength"),
  facet_by = c("outcome", "predictor"),
  facet_strata = TRUE,
  limits = NULL,
  p_size = 2
)

Arguments

x

An object returned by DiagnosticLikelihoodRatioTable(), or a tidy data frame compatible with its Results element.

result

Diagnostic result levels to display: "all", "positive", or "negative". Positive and negative selections require the result object.

predictor_order

Predictor ordering: "original", "alphabetical", or "strength". Strength orders predictors by their largest finite absolute log2 likelihood ratio.

outcome_order

Outcome ordering with the same choices.

facet_by

Plot panels by "outcome" (the default) or "predictor".

facet_strata

Logical; add strata as facet row groups when present.

limits

Optional numeric vector of two positive, increasing likelihood ratio limits. By default, limits are chosen from finite estimates and confidence intervals while always including one.

p_size

Numeric point size.

Value

A ggplot object. Its DiagnosticLRForestData, DiagnosticLRForestLimits, and DiagnosticLRForestFacetBy attributes retain the prepared data and resolved plotting settings.

Reading the plot

The dashed vertical line marks LR = 1. Dark black estimates have an unadjusted likelihood-ratio confidence interval that excludes one; gray estimates do not. This visual cue is not an FDR-adjusted significance test. Uncorrected zero and infinite likelihood ratios are shown as boundary arrows labelled 0 and Inf, because a finite confidence interval is unavailable.

See also

DiagnosticLikelihoodRatioTable() to calculate diagnostic LRs, and PlotDiagnosticLRHeatmap() for matrix and count-matrix views.

Examples

data(SampleData)
data(SampleVariableTypes)
df_Labelled <- RevalueData(SampleData, SampleVariableTypes)$RevaluedData
df_Labelled$DiagnosisBinary <- factor(df_Labelled$Diagnosis,
  levels = c("Control", "Impaired"))
lr <- DiagnosticLikelihoodRatioTable(df_Labelled, "DiagnosisBinary",
  c("sex", "Genotype"))
#> DiagnosisBinary: 'Impaired' treated as outcome-positive.
PlotDiagnosticLRForest(lr)

PlotDiagnosticLRForest(lr, result = "positive")