Two new functions: rep_olr and plot_ord_odds

DESCRIPTION

@@ -9,7 +9,8 @@ Description: Tools for Danish health research. A collection of
 Depends: R (>= 3.4.4)
 Imports: broom,
   dplyr,
-  epiR
+  epiR,
+  ggplot2
 License: GPL (>= 2)
 Encoding: UTF-8
 LazyData: true

NAMESPACE

@@ -10,10 +10,12 @@ export(dob_extract_cpr)
 export(hwe_allele)
 export(hwe_geno)
 export(hwe_sum)
+export(plot_ord_odds)
 export(rep_biv)
 export(rep_epi_tests)
 export(rep_glm)
 export(rep_lm)
+export(rep_olr)
 export(rep_reg_cie)
 export(strobe_diff_bygroup)
 export(strobe_diff_byvar)

R/plot_ord_odds.R

@@ -0,0 +1,39 @@
+#' Forrest plot from ordinal logistic regression
+#'
+#' Heavily inspired by https://www.r-bloggers.com/plotting-odds-ratios-aka-a-forrestplot-with-ggplot2/
+#' @param x ordinal logistic regression model.
+#' @param title plot title
+#' @param dec decimals for labels
+#' @param lbls labels for variable names. Carefull, as the right order is not checked automatically!
+#' @keywords forestplot
+#' @export
+#' @examples
+#' plot_ord_odds()
+
+plot_ord_odds<-function(x, title = NULL,dec=3,lbls=NULL){
+
+  require(ggplot2)
+
+  odds<-data.frame(cbind(exp(coef(x)), exp(confint(x))))
+  names(odds)<-c("or", "lo", "up")
+  rodds<-round(odds,digits = dec)
+
+  if (!is.null(lbls)){
+    odds$vars<-paste0(v.names," \n",paste0(rodds$or," [",rodds$lo,":",rodds$up,"]"))
+  }
+  else {
+    odds$vars<-paste0(row.names(odds)," \n",paste0(rodds$or," [",rodds$lo,":",rodds$up,"]"))
+  }
+
+  ticks<-c(seq(.1, 1, by =.1), seq(0, 10, by =1), seq(10, 100, by =10))
+  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 = "Variables", y = "OR (95 % CI)") +
+    theme_bw()
+}

R/rep_glm.R

@@ -81,11 +81,10 @@ rep_glm<-function(meas,vars,string,ci=FALSE,data){
 
     }}
 
-  pa<-as.numeric(df[,3])
+  pa<-as.numeric(df[,"pv"])
+  t <- ifelse(pa<=0.1,"include","drop")
+
   pa<-ifelse(pa<0.001,"<0.001",pa)
-
-  t <- ifelse(pa<=0.1|pa=="<0.001","include","drop")
-
   pa <- ifelse(pa<=0.05|pa=="<0.001",paste0("*",pa),
                ifelse(pa>0.05&pa<=0.1,paste0(".",pa),pa))
 

R/rep_olr.R

@@ -0,0 +1,99 @@
+#' A repeated ordinal logistic regression function
+#'
+#' @description For bivariate analyses. The confint() function is rather slow, causing the whole function to hang when including many predictors and calculating the ORs with CI.
+#' @param meas Effect meassure. Input as c() of columnnames, use dput().
+#' @param vars variables in model. Input as c() of columnnames, use dput().
+#' @param str variables to test. Input as c() of columnnames, use dput().
+#' @param ci flag to get results as OR with 95% confidence interval.
+#' @param dta data frame to pull variables from.
+#' @keywords olr ordinal logistic regression
+#' @export
+#' @examples
+#'   rep_olr()
+
+
+rep_olr<-function(meas,vars,string,ci=FALSE,data){
+
+  require(broom)
+  require(MASS)
+
+  d<-data
+  x<-data.frame(d[,c(string)])
+  v<-data.frame(d[,c(vars)])
+  names(v)<-c(vars)
+  y<-d[,c(meas)]
+  dt<-cbind(y,v)
+  m1<-length(coef(polr(y~.,data = dt,Hess=TRUE)))
+
+  if (!is.factor(y)){stop("y should be a factor!")}
+
+  if (ci==TRUE){
+
+    df<-data.frame(matrix(ncol = 3))
+    names(df)<-c("pred","or_ci","pv")
+
+    for(i in 1:ncol(x)){
+      dat<-cbind(dt,x[,i])
+      m<-polr(y~.,data=dat,Hess=TRUE)
+
+      ctable <- coef(summary(m))
+
+      l<-suppressMessages(round(exp(confint(m))[-c(1:m1),1],2))
+      u<-suppressMessages(round(exp(confint(m))[-c(1:m1),2],2))
+      or<-round(exp(coef(m))[-c(1:m1)],2)
+      or_ci<-paste0(or," (",l," to ",u,")")
+
+      p <- (pnorm(abs(ctable[, "t value"]), lower.tail = FALSE) * 2)[1:length(coef(m))]
+      pv<-round(p[-c(1:m1)],3)
+
+
+      x1<-x[,i]
+
+      if (is.factor(x1)){
+        pred<-paste(names(x)[i],levels(x1)[-1],sep = "_")}
+
+      else {pred<-names(x)[i]}
+
+      df<-rbind(df,cbind(pred,or_ci,pv))
+    }}
+
+  if (ci==FALSE){
+
+    df<-data.frame(matrix(ncol = 3))
+    names(df)<-c("pred","b","pv")
+
+    for(i in 1:ncol(x)){
+      dat<-cbind(dt,x[,i])
+      m<-polr(y~.,data=dat,Hess=TRUE)
+
+      b<-round(coef(m)[-c(1:m1)],2)
+
+      p <- (pnorm(abs(ctable[, "t value"]), lower.tail = FALSE) * 2)[1:length(coef(m))]
+      pv<-round(p[-c(1:m1)],3)
+
+      x1<-x[,i]
+
+      if (is.factor(x1)){
+        pred<-paste(names(x)[i],levels(x1)[-1],sep = "_")
+      }
+
+      else {pred<-names(x)[i]}
+
+      df<-rbind(df,cbind(pred,b,pv))
+
+    }}
+
+  pa<-as.numeric(df[,c("pv")])
+  t <- ifelse(pa<=0.1,"include","drop")
+  pa<-ifelse(pa<0.001,"<0.001",pa)
+  pa <- ifelse(pa<=0.05|pa=="<0.001",paste0("*",pa),
+               ifelse(pa>0.05&pa<=0.1,paste0(".",pa),pa))
+
+  r<-data.frame(df[,1:2],pa,t)[-1,]
+
+  return(r)
+}
+
+
+
+

man/plot_ord_odds.Rd

@@ -0,0 +1,24 @@
+% Generated by roxygen2: do not edit by hand
+% Please edit documentation in R/plot_ord_odds.R
+\name{plot_ord_odds}
+\alias{plot_ord_odds}
+\title{Forrest plot from ordinal logistic regression}
+\usage{
+plot_ord_odds(x, title = NULL, dec = 3, lbls = NULL)
+}
+\arguments{
+\item{x}{ordinal logistic regression model.}
+
+\item{title}{plot title}
+
+\item{dec}{decimals for labels}
+
+\item{lbls}{labels for variable names. Carefull, as the right order is not checked automatically!}
+}
+\description{
+Heavily inspired by https://www.r-bloggers.com/plotting-odds-ratios-aka-a-forrestplot-with-ggplot2/
+}
+\examples{
+plot_ord_odds()
+}
+\keyword{forestplot}

man/rep_olr.Rd

@@ -0,0 +1,29 @@
+% Generated by roxygen2: do not edit by hand
+% Please edit documentation in R/rep_olr.R
+\name{rep_olr}
+\alias{rep_olr}
+\title{A repeated ordinal logistic regression function}
+\usage{
+rep_olr(meas, vars, string, ci = FALSE, data)
+}
+\arguments{
+\item{meas}{Effect meassure. Input as c() of columnnames, use dput().}
+
+\item{vars}{variables in model. Input as c() of columnnames, use dput().}
+
+\item{ci}{flag to get results as OR with 95% confidence interval.}
+
+\item{str}{variables to test. Input as c() of columnnames, use dput().}
+
+\item{dta}{data frame to pull variables from.}
+}
+\description{
+For bivariate analyses. The confint() function is rather slow, causing the whole function to hang when including many predictors and calculating the ORs with CI.
+}
+\examples{
+  rep_olr()
+}
+\keyword{logistic}
+\keyword{olr}
+\keyword{ordinal}
+\keyword{regression}