make some sort of grouping variable:

yourdata$groupmedian(Survival),1,0)

you could use lattice and barchart

library(lattice)

barchart(Bloodvalue~factor(group),data=yourdata,groups=PatientID)

This should normally give you bars for each patient and grouped by prognosis.

Hi Gregory,

Thanks so much. This is awesome!

I actually think, the grouping should be different:

yourdata$groupmedian(Survival),1,0)

The condition you gave group actually means, if the BloodValue is over the median of Survival, then assign 1.

But it works really nicely.

What I was wondering now....is how do you get rid of the spacing between the bars? I've attached the output, so you've got an idea of what happens when I do this with my data.

Thanks so much. This is excellent. I didn't know about the ifelse function and I didn't think of defining a new group and exploiting lattice to plot it! Great stuff! Thanks.

oops, yeah you're right I made a mistake there

see ?panel.barchart

arguments box.ratio and box.width

box.width deals with thickness of bars, so that could reduce space

you could try something like:

barchart(Bloodvalue~factor(group),data=yourdata,groups=PatientID, box.width=1)

or whatever width that would be needed to get rid of the spaces

The grouping problem is solved, I think. But just a completely other Idea to possibly present the data better:

Barcharts are not so nice and especially the one shown above is hard to read (too many bars, so many colors; additionally it is a bad practice to cut the bars).

If there are values for the same blood parameters in both groups, you could show the values in a 2D scatterplot. You can then very quickly see if the values are around the diagonal (45°-line) showing no differences or if they are above or below this line. Improtant parameters (e.g. those with a big difference between the groups) could be labelled. Different symbols and colors could be used, too (but I would use such things sparingly).

Alternatively, you could show the results in a Bland-Altman plot (average value in the two groups versus difference of the value between the groups). This essentially turns the cloud of points show in the "normal" scatterplot by 45°, so that the "indentity line" falls on the horizontal axis.

Finally: When the values of the different parameters are quite differen it might be useful to plot the logarithms instead (or on logarithmic scaled axes).

I'll answer both sequentially.

First @ Gregory:

Changing the box.width won't help. This is because as far as I can tell the spacing between the bars is because of the StudyID variable. R has taken the order of the StudID and rows them (from left to right) as they appear in the data.frame. If one StudyID falls into the group "Dead at FU", then it will leave a space for that StudyID in the "Alive at FU" group and vice a versa. I was wondering if there's any function within lattice, that allows you to group your variables together without that spacing.

Second @ Jochen.

That's actually a very good idea, I didn't think of bringing in a second dimension to allow for potential overlapping of values to be accounted for.

However, I'll be honest with you, that I actually have no idea, how to either make the scatterplot you described (I know how to make scatterplots...) or the Bland-Altman plot. I'll probably have a look around Stackoveflow and Youtube to find out how to do what you proposed.

Regarding the colors: yeah I changed it to grey. Makes it easier for the eye.

Thanks guys! This is a really helpful Q&A.

Cheers,

Oliver

Oliver,

a scatterplot in R is really simple: it's just the plot() function. plot (x,y) will produce a scatterplot of the y-values versus the x-values. Both variables (x and y) are numeric vectors of the same length. In your case they should hold the mean values for the blood parameters. One variable is calculated from the values of the "low survivors", the other from the "long survivors" So if you have for instance 15 different blood parameters, each variable will have a length of 15.

It follows some example code to show you how it can be solved. I assume that you have your data in a data.frame, with a column for each blood parameter (plus some other columns for other patient characteristics, for instance the survival time!) and a row for each patient. For this example I first have to generate some random nonsense data to show you the results and test if it is working (this is the longest part of the code). You would simply use your real data for plotting, so actually only the last fwe lines of the code are really relevant for you:

# ---- generate some data ----

# some blood params strongly increased for low survival

nPatients = 100

nBloodParams = 15

BloodParams = LETTERS[1:nBloodParams]

x = data.frame( Surv=rgamma(nPatients,2.277,1/1000 ) )

longSurv = x$Surv>=1277

params = sample(BloodParams, 5)

for (BloodParam in BloodParams)

x[,BloodParam] = rnorm(nPatients, m=runif(1,300,2000)) +

ifelse(longSurv,rnorm(1,m=0,sd=100), ifelse(BloodParam %in% params, rnorm(1,m=200,sd=500), 0))

# have a look at the data

dim(x)

head(x)

# ---- plotting: get the group means for each blood param ----

dataCols = 2:ncol(x)

group = factor(x$Surv>=1277, labels=c("low","high"))

groupmeans = apply(x[,dataCols],2,tapply,group,mean,na.rm=TRUE)

# ---- 2D scatterplot ----

plot(groupmeans["low",], groupmeans["high",], pch=16, xlab="low survival", ylab="high survival")

abline(0,1,lty=3)

# ---- Bland-Altman ----

averages = apply(groupmeans,2,mean)

differences = apply(groupmeans,2,diff)

plot(averages, differences, pch=16, xlab="average value", ylab="difference in\nhigh versus low survival")

abline(h=0,lty=3)

@Oliver

I see. The thing is the data that I have and tried it on always has ID-numbers that are ordered and follow each other nicely and thus didn't really have spaces.

Could it be that that your ID-values take a few jumps here and there? If so you could make a different ID-variable that is completely ordered from 1 to number of patients. That way R should only make a nice gap between the groups and have no spaces between the bars.

@ Jochen.

Thanks for the example. I knew how to make scatterplots etc, but wasn't sure what you meant regarding plotting them. A very interesting approach indeed! Thank you very much.

@ Gregory.

I tried assigning a new StudyID as there are infact jumps inbetween my IDs, where it was simple 1-30 for number of patients. The plot is still identical, as lattice (and ggplot2 for that matter) still seem to group the bars with breaks inbetween them, depending on which StudyID gets allocated to which category (if that makes any sense). Sadly enough, I resorted to prism to make this plot...which worked well, but kind of defeats the point of my conitnously wanting to work with R :o).

If you come up with a solution to this, that'd be great. I'll also post it as a question on Stackoverflow and if someone has a good idea, I'll let you know (unless you beat me to it) :o).

Thanks again for your help guys, really great stuff!

Cheers,

Oliver

So here's the solution using ggplot. I gather there will be a similar solution to that using lattice (but couldn't find it straight away):

http://stackoverflow.com/questions/22292287/removing-empty-group-spaces-in-barcharts-of-ggplot2-lattice/22292849?noredirect=1#22292849

In summary:

a) when defining the groups use factor(Study.ID)

b) in ggplot using the facet_grid() type in scales="free" like this facet_grid(scales="free").

Works a charm. I'm still trying to make up my mind if I could use lattice for anything else than the excellent xyplot...otherwise it seems as if ggplot has more power to it. Any views?

Hi

So you found the grouping solution, personally, I would have advise you the bargraph.CI {sciplot} function, just have a try ;)

Hm. You've got me. My R skills haven't progressed to using functions yet....How would you program the code for that function?

Well… So from an old script I did:

#Bibliothèque

library(sciplot)

#Réglages

CEXT

Thanks benjamin....sad to say however, that I can't find your package "sciplot"....

Hi Benjamin,

If I understand your problem correctly you want to look at High vs Low survival.

I find the easiest way to do this is to create another column in your dataframe and then fill it in based on the values you want to subset. Here's some code that should help you:

# let's assume your data frame is called 'data'

# create a new column and fill values with NA and then fill

# in the column based on your criteria

data$prognosis = NA

data$prognosis[data$Survival>=1277] = "High"

data$prognosis[data$Survival=median(data$Survival)] = "High"

# data$prognosis[data$Survival