R/plot_ord_odds.R

In agdamsbo/daDoctoR: Functions For Health Research

#' Forrest plot from ordinal logistic regression.
#'
#' Heavily inspired by https://www.r-bloggers.com/plotting-odds-ratios-aka-a-forrestplot-with-ggplot2/
#' @param x input data.
#' @param title plot title
#' @param dec decimals for labels
#' @param lbls labels for variable names. Carefull, as the right order is not checked automatically!
#' @param hori labels the horizontal axis (this i the y axis as the plot is rotated)
#' @param vert labels the horizontal axis (this i the x axis as the plot is rotated)
#' @param short flag to half number of ticks on horizontal axis.
#' @param input can be either "model", which is a olr model (polr()), or "df", which is a dataframe whith three columns for OR, lower CI and upper CI.
#' @keywords forestplot
#' @export

plot_ord_odds<-function(x, title = NULL,dec=3,lbls=NULL,hori="OR (95 % CI)",vert="Variables",short=FALSE,input=c("model","df")){

  require(ggplot2)

  if (input=="model"){
    odds<-data.frame(cbind(exp(coef(x)), exp(confint(x))))
  }

  if (input=="df"){
    odds<-x
  }
  names(odds)<-c("or", "lo", "up")
  rodds<-round(odds,digits = dec)

  if (!is.null(lbls)){
    odds$vars<-paste0(lbls," \n",paste0(rodds$or," [",rodds$lo,":",rodds$up,"]"))
  }
  else {
    odds$vars<-paste0(row.names(odds)," \n",paste0(rodds$or," [",rodds$lo,":",rodds$up,"]"))
  }

  ticks<-c(seq(0, 1, by =.1), seq(1, 10, by =1), seq(10, 100, by =10))

  if (short==TRUE){
    ticks<-ticks[seq(1, length(ticks), 2)]
  }
  else {ticks<-ticks}

  odds$ord<-c(nrow(odds):1)

  ggplot(odds, aes(y= or, x = reorder(vars,ord))) +
    geom_point() +
    geom_errorbar(aes(ymin=lo, ymax=up), width=.2) +
    scale_y_log10(breaks=ticks, labels = ticks) +
    geom_hline(yintercept = 1, linetype=2) +
    coord_flip() +
    labs(title = title, x = vert, y = hori) +
    theme_bw()
}