forestplotRegrObj {Greg}

Forest plot for multiple models

Description

Plot different model fits with similar variables in order to compare the model's estimates and confidence intervals. Each model is represented by a separate line on top of eachother and are therefore ideal for comparing different models. This extra appealing when you have lots of variables included in the models.

Usage

forestplotRegrObj(
  regr.obj,
  postprocess_estimates.fn = function(x) x,
  rowname = "Variable",
  ci.txt = "CI",
  ci.glue = "{lower} to {higher}",
  digits = 1,
  get_box_size = fpBoxSize,
  ...
)

## Default S3 method:
forestplotRegrObj(
  regr.obj,
  postprocess_estimates.fn = function(x) x,
  rowname = "Variable",
  ci.txt = "CI",
  ci.glue = "{lower} to {higher}",
  digits = 1,
  get_box_size = fpBoxSize,
  ...
)

## S3 method for class 'coxph'
forestplotRegrObj(
  regr.obj,
  postprocess_estimates.fn = function(x) x,
  rowname = "Variable",
  ci.txt = "CI",
  ci.glue = "{lower} to {higher}",
  digits = 1,
  get_box_size = fpBoxSize,
  xlab = "Hazard Ratio",
  estimate.txt = "HR",
  xlog = TRUE,
  zero = 1,
  exp = TRUE,
  ...
)

## S3 method for class 'lrm'
forestplotRegrObj(
  regr.obj,
  postprocess_estimates.fn = function(x) x,
  rowname = "Variable",
  ci.txt = "CI",
  ci.glue = "{lower} to {higher}",
  digits = 1,
  get_box_size = fpBoxSize,
  xlab = "Odds ratio",
  estimate.txt = "HR",
  xlog = TRUE,
  zero = 1,
  exp = TRUE,
  ...
)

## S3 method for class 'lm'
forestplotRegrObj(
  regr.obj,
  postprocess_estimates.fn = function(x) x,
  rowname = "Variable",
  ci.txt = "CI",
  ci.glue = "{lower} to {higher}",
  digits = 1,
  get_box_size = fpBoxSize,
  xlab = "Effect",
  estimate.txt = "Coef",
  xlog = FALSE,
  zero = 0,
  exp = FALSE,
  ...
)

## S3 method for class 'glm'
forestplotRegrObj(
  regr.obj,
  postprocess_estimates.fn = function(x) x,
  rowname = "Variable",
  ci.txt = "CI",
  ci.glue = "{lower} to {higher}",
  digits = 1,
  get_box_size = fpBoxSize,
  xlab = NULL,
  xlog = NULL,
  zero = NULL,
  estimate.txt = NULL,
  exp = NULL,
  ...
)

## S3 method for class 'list'
forestplotRegrObj(
  regr.obj,
  postprocess_estimates.fn = function(x) x,
  rowname = "Variable",
  ci.txt = "CI",
  ci.glue = "{lower} to {higher}",
  digits = 1,
  get_box_size = fpBoxSize,
  xlab = NULL,
  xlog = NULL,
  zero = NULL,
  estimate.txt = NULL,
  exp = NULL,
  ...
)

fpBoxSize(p_values, variable_count, boxsize, significant = 0.05)

Arguments

regr.obj

A regression model object. It should be of coxph, crr or glm class. Warning: The glm is not fully tested.

postprocess_estimates.fn

A function that takes the regression outputs and returns the same data with modifications. The input columns are:

* 'Rowname' * 'Coef' * 'Lower' * 'Upper' * 'Sort'

rowname

The name of the variables

ci.txt

The text above the confidence interval, defaults to '"CI"'

ci.glue

The string used for [glue::glue()] the 'lower' and 'higher' confidence intervals together.

digits

The number of digits to round presented values to

get_box_size

A function for extracting the box sizes

...

Passed to forestplot()

xlab

x-axis label

estimate.txt

The text above the estimate, e.g. Est, HR

xlog

If TRUE, x-axis tick marks are to follow a logarithmic scale, e.g. for logistic regression (OR), survival estimates (HR), Poisson regression etc. Note: This is an intentional break with the original forestplot function as I've found that exponentiated ticks/clips/zero effect are more difficult to for non-statisticians and there are sometimes issues with rounding the tick marks properly.

zero

Indicates what is zero effect. For survival/logistic fits the zero is 1 while in most other cases it's 0.

exp

Report in exponential form. Default true since the function was built for use with survival models.

p_values

The p-values that will work as the foundation for the box size

variable_count

The number of variables

boxsize

The default box size

significant

Level of significance .05

See Also

Other forestplot wrappers: forestplotCombineRegrObj()

Examples

org.par <- par("ask" = TRUE)

library(tidyverse)
# simulated data to test
set.seed(102)
cov <- tibble(ftime = rexp(200)) |> 
  mutate(x1 = runif(n()),
         x2 = runif(n()),
         x3 = runif(n()),
         fstatus1 = if_else(x1 * 1 + 
                              x2 * 0.2 + 
                              x3 * 0.5 + 
                              runif(n()) * 0.5 > 1, 
                            1, 0),
         fstatus2 = if_else(x1 * 0.2 + 
                              x2 * 0.5 + 
                              x3 * 0.1 + 
                              runif(n()) * 2 > 1, 
                            1, 0)) |> 
  # Add some column labels
  Gmisc::set_column_labels(x1 = "First variable",
                           x2 = "Second variable")

library(rms)
dd <- datadist(cov)
options(datadist = "dd")

fit1 <- cph(Surv(ftime, fstatus1 == 1) ~ x1 + x2 + x3, data = cov)

fit1 |> 
  forestplotRegrObj() |> 
  fp_set_zebra_style("#f0f0f0")

fit2 <- update(fit1, Surv(ftime, fstatus2 == 1) ~ .)
list("Frist model" = fit1, "Second model"  = fit2) |> 
  forestplotRegrObj(legend_args = fpLegend(title = "Type of regression"),
                    postprocess_estimates.fn = function(x) {
                      x |> 
                        filter(str_detect(column_term, "(x2|x3)"))
                    }) |> 
  fp_set_style(box = rep(c("darkblue", "darkred"), each = 3))


par(org.par)