Summarising and Visualising three attributes in R

Question

I am trying to summarise and visualise three attributes in R:

Patient_Age, Patient_Deprivation and Hospital_Time.

I am trying to summarise the time patients spent in hospital by deprivation(scale 1-5) & Age. I also calculated the covariance and correlation. If I want to summarise the total time (diffr_days) by Age and Deprivation. What would be the best way of visualisation because only violin plot and correlation matrix make a bit of sense to understand the data? Other plots are not making any sense of the data. What would be the best plot and descriptive method to explain such data scenario?

Age Cov: 28.42936 Corr: 0.2497208

Deprivation Cov: 0.3389552 Corr: 0.07772134

Below is the screenshot of my data frame and some visualisations

do.call(cbind.data.frame,list(id = c( 
1011L, 1012L, 1012L, 1014L, 1015L, 1015L, 1018L, 1018L, 1021L, 
1022L, 1028L, 1029L, 1036L, 1037L, 1042L, 1044L, 1045L, 1048L, 
1049L, 1050L, 1050L, 1051L, 1051L, 1052L, 1054L, 1057L, 1061L, 
1064L, 1064L, 1065L, 1066L, 1067L, 1067L, 1069L, 1072L, 1073L, 
1078L, 1079L, 1082L, 1083L, 1084L, 1086L, 1087L, 1089L, 1090L, 
1090L, 1091L, 1092L, 1095L, 1096L),
age = c(23L, 50L, 
6L, 92L, 70L, 70L, 53L, 16L, 16L, 70L, 58L, 58L, 30L, 30L, 30L, 
10L, 34L, 69L, 22L, 16L, 38L, 71L, 24L, 57L, 5L, 79L, 79L, 37L, 
37L, 35L, 40L, 45L, 72L, 97L, 97L, 34L, 28L, 29L, 29L, 78L, 22L, 
25L, 31L, 36L, 53L, 49L, 17L, 48L, 56L, 32L),
deprivation = c(1L, 2L, 3L, 2L, 3L, 3L, 1L, 2L, 
2L, 2L, 4L, 4L, 2L, 2L, 2L, 2L, 3L, 2L, 2L, 2L, 3L, 2L, 2L, 2L, 
3L, 2L, 2L, 4L, 4L, 2L, 4L, 1L, 1L, 3L, 3L, 4L, 3L, 3L, 3L, 2L, 
2L, 1L, 2L, 3L, 3L, 3L, 1L, 2L, 2L, 2L),
diffr_days = c(8L, 2L, 4L, 12L, 12L, 
6L, 7L, 6L, 9L, 7L, 8L, 4L, 11L, 5L, 1L, 10L, 7L, 9L, 2L, 5L, 
10L, 6L, 3L, 2L, 9L, 1L, 2L, 9L, 0L, 9L, 4L, 1L, 10L, 2L, 5L, 
9L, 14L, 0L, 5L, 0L, 0L, 1L, 10L, 8L, 3L, 4L, 2L, 1L, 0L, 2L
)))

Answer 1

Here is a possibility: plot hospital stay against age, with five different loess fit lines, one per deprivation level. In the plot below, I used your example data (actually, I replicated it five times, because loess needs enough data to calculate the smoother - with your original data, this should not be a problem). I also plotted the data from the example. With your original data, this will likely not be very helpful, and if you really want to include the raw data, you should make the dots a lot smaller.

I used a black body radiation palette as per here.

As an alternative for your first plot, I would suggest a beanplot. I overlaid it with boxplots and again added the example data, with some horizontal jittering to avoid overplotting. You could even indicate ages using some kind of color coding, e.g., a black body radiation palette as above, but with your original sample size, I don't think that would be very helpful.

I personally prefer base graphics. You should be able to build the first plot with ggplot2, but I don't know whether beanplots are available there.

R code:

dataset <- 
    do.call(cbind.data.frame,list(id = c( 
    1011L, 1012L, 1012L, 1014L, 1015L, 1015L, 1018L, 1018L, 1021L, 
    1022L, 1028L, 1029L, 1036L, 1037L, 1042L, 1044L, 1045L, 1048L, 
    1049L, 1050L, 1050L, 1051L, 1051L, 1052L, 1054L, 1057L, 1061L, 
    1064L, 1064L, 1065L, 1066L, 1067L, 1067L, 1069L, 1072L, 1073L, 
    1078L, 1079L, 1082L, 1083L, 1084L, 1086L, 1087L, 1089L, 1090L, 
    1090L, 1091L, 1092L, 1095L, 1096L),
age = c(23L, 50L, 
6L, 92L, 70L, 70L, 53L, 16L, 16L, 70L, 58L, 58L, 30L, 30L, 30L, 
10L, 34L, 69L, 22L, 16L, 38L, 71L, 24L, 57L, 5L, 79L, 79L, 37L, 
37L, 35L, 40L, 45L, 72L, 97L, 97L, 34L, 28L, 29L, 29L, 78L, 22L, 
25L, 31L, 36L, 53L, 49L, 17L, 48L, 56L, 32L),

deprivation = c(1L, 2L, 3L, 2L, 3L, 3L, 1L, 2L, 
2L, 2L, 4L, 4L, 2L, 2L, 2L, 2L, 3L, 2L, 2L, 2L, 3L, 2L, 2L, 2L, 
3L, 2L, 2L, 4L, 4L, 2L, 4L, 1L, 1L, 3L, 3L, 4L, 3L, 3L, 3L, 2L, 
2L, 1L, 2L, 3L, 3L, 3L, 1L, 2L, 2L, 2L), 

diffr_days = c(8L, 2L, 4L, 12L, 12L, 
6L, 7L, 6L, 9L, 7L, 8L, 4L, 11L, 5L, 1L, 10L, 7L, 9L, 2L, 5L, 
10L, 6L, 3L, 2L, 9L, 1L, 2L, 9L, 0L, 9L, 4L, 1L, 10L, 2L, 5L, 
9L, 14L, 0L, 5L, 0L, 0L, 1L, 10L, 8L, 3L, 4L, 2L, 1L, 0L, 2L
)))


dataset <- do.call(rbind.data.frame,list(dataset,dataset,dataset,dataset,dataset))
blackBodyRadiationColors <- function(x, max_value=1) {
    # x should be between 0 (black) and 1 (white)
    # if large x come out too bright, constrain the bright end of the palette
    #     by setting max_value lower than 1
    foo <- colorRamp(c(rgb(0,0,0),rgb(1,0,0),rgb(1,1,0),rgb(1,1,1)))(x*max_value)/255
    apply(foo,1,function(bar)rgb(bar[1],bar[2],bar[3]))
}
n.colors <- length(unique(dataset$deprivation))
colors.blackBody <- blackBodyRadiationColors(seq(0,0.6,length.out=n.colors))
par(mai=c(.8,.8,.1,.1))
with(dataset,plot(range(age),range(diffr_days),type="n",las=1,xlab="Age",ylab="Time in hospital (days)"))
with(dataset,points(age,diffr_days,pch=19,col=colors.blackBody[deprivation]))
legend("topleft",pch=19,lwd=1,col=colors.blackBody,legend=unique(dataset$deprivation),title="Deprivation")
for ( ii in unique(dataset$deprivation) ) {
    loess_model <- loess(diffr_days~age,data=subset(dataset,deprivation==ii))
    xx <- seq(min(subset(dataset,deprivation==ii)["age"]),max(subset(dataset,deprivation==ii)["age"]))
    loess_fit <- predict(loess_model,newdata=data.frame(age=xx))
    lines(loess_fit,col=colors.blackBody[ii])
}
library(beanplot)
with(dataset,beanplot(diffr_days~deprivation,las=1,what=c(0,1,0,0),col="lightgray",border=NA,
    xlab="Deprivation",ylab="Time in hospital (days)"))
with(dataset,boxplot(diffr_days~deprivation,add=TRUE,outline=FALSE,col=NA,yaxt="n"))
with(dataset,points(deprivation+runif(nrow(dataset),-0.2,0.2),diffr_days,pch=19))

Answer 2

A Wilkinson dot plot is a handy graphics for visualizing counts. It's a histogram of stacked dots.

Here is a Wilkinson dot plot of the sample data, split by deprivation. It's easy to notice for example that all counts are multiples of 5.

With the sample data, one dot represents one patient. With the full dataset, it would probably be necessary to summarize so that the chart doesn't get cluttered, eg. one dot can represent 10 patients.

PS: Color is under-used in this dot plot. It will be great to color by age (or by average age, if dots represent multiple patients of similar age). Unfortunately, I hit the limits of ggplot2, so in R at least it would have to be a custom-made chart.

---

R code to make a Wilkinson dot plot.

# Adapted from:
# https://stackoverflow.com/questions/49330742/change-y-axis-of-dot-plot-to-reflect-actual-count-using-geom-dotplot
library("tidyverse")
dataset <-
  as_tibble(dataset) %>%
  mutate(
    deprivation = factor(deprivation)
  )
counts <- dataset %>%
  count(
    deprivation, diffr_days
  )
binwidth <- 0.2
dotsize <- 1
yheight <- max(counts$n)
dataset %>%
  ggplot(
    aes(x = diffr_days, fill = deprivation)
  ) +
  geom_dotplot(
    binwidth = binwidth,
    dotsize = dotsize,
    method = "histodot"
  ) +
  coord_fixed(
    ratio = binwidth * dotsize * yheight
  ) +
  facet_wrap(
    ~ "deprivation" + deprivation,
    ncol = 2
  ) +
  scale_x_continuous(
    limits = c(0, 15)
  ) +
  scale_y_continuous(
    limits = c(0, 1),
    expand = c(0, 0),
    breaks = seq(0, 1, 5 * 1 / yheight),
    labels = seq(0, yheight, by = 5)
  ) +
  labs(
    x = "Time in hospital (days)",
    y = "Number of patients"
  ) +
  theme(legend.position = "none")