The Design of Software (CLOSED)

A public forum for discussing the design of software, from the user interface to the code architecture. Now closed.

The "Design of Software" discussion group has been merged with the main Joel on Software discussion group.

The archives will remain online indefinitely.

Efficient presentation of complex data

I would like to present graphically complex data:
We are measuring the levels of a clinical marker in the blood  during a period of time. During this period of time, the patient is given different doses of a drug. I want to show the dynamics of the biomarker together with the treatment course. One option is shown on pane A of the following figure:
Using this option we plot the biomarker on the left Y axis and the drug doses (as a bar plot) using the Y axis.

Another option (pane B, the same figure) is to use only one Y axis and to mark the doses using arrow below the X axis.

Which option is more "readable" and intuitive? Are there any better ways to present such an information?
Wednesday, May 28, 2008
For expert opinions on information display I would browse the forum or post a question at the Edward Tufte site:

Alternately, read his books and writings.
IT guy
Wednesday, May 28, 2008
Frankly neither. Better might be vertical bars in 'a' with the data labels in 'b'

Go and buy Tufte's book on visualising data - it will be worth it.
Martin Send private email
Wednesday, May 28, 2008
Why does fogbugz do that?
I post the first reply and then an hour later something else is posted before me.
Martin Send private email
Wednesday, May 28, 2008
Dude, I posted first.
IT guy
Wednesday, May 28, 2008
'A' was quite readable to me.  I agree that making the doses vertical bars might help as well (then just labeling those).

The data is fairly sparse, but I'd like to see/know:
1. Is the dose applied before or after the measurement on the same day?
2. I'd be interested in seeing a graph with 4 lines all starting at the time a dose is applied.  These 4 would all start stacked above each other based on their initial marker value, tick upward and then perhaps decline.  Seeing them stacked might be a better way to compare them. (Or possibly even an overlay normalized for the initial marker level.

Note that #2 will have its inaccuracies as the various lines are not in isolation, but taken one after the other.  After the first dose, one dose is wearing off, where after the second dose the first two are wearing off, possibly making the curves different, etc...
Lance Hampton Send private email
Wednesday, May 28, 2008
Figure A is definitely superior, although I do like the idea of putting arrowheads on the vertical lines as you did in Figure B because it is consistent with the notation used in signal processing (the administered dose is effectively a scaled Dirac delta function).  The main thing that would make the figure more clear is to come up with better labels for the two vertical axes.  Right now, you have "levels" and "Dose level" and the distinction between the two is not at all clear to the reader.  Changing the labels to something like "Administered Dose (mg)" and "Serum concentration of XYZ (units)" (where XYZ is whatever you're measuring the "levels" of) would greatly improve the clarity of the information.  It would also help to use color to identify which information matches which axis.  For example, if you made the vertical impulses red and made the numbers on the Administered Dose axis red also, that would help.  Also, in your graph there is ambiguity concerning the temporal relationship between administration of the drug and measurement of the "levels".  For example, looking at the data point around day 23 it isn't clear whether that "level" was measured before or after the dose that was given on that day.  The same ambiguity is present at day 51 and day 72.

Other ideas:
1) Depending on what it is that you're trying to demonstrate, it might be appropriate to plot total administered dose instead of showing the instantaneous administered dose.
2) Depending on what the physical relationship between the administered dose and the measured quantity is, it may potentially be meaningful to model the change in the measured quantity as an impulse response function.
3) It may be helpful to show some baseline information about the measured quantity before the initial administration of the drug.  How do we know what the effect of the drug is on the "levels" if we don't know what the "levels" were prior to the initial administration of the drug?  If the "levels" are simply a measurement of the serum concentration of the drug then, of course, this wouldn't be meaningful (thus underscoring the need to define what the "levels" being measured are).
Wednesday, May 28, 2008
Thank you all for the responses.

Several clarifications:
 * The baseline is known to the potential auditory of these graphs (clinicians)
 * Measurement dates and drug administration dates do not always overlap.
 * Reporting a total dose doesn't deliver the information I want to show, since the time relationship between drug delivery and biomarker changes is crucial
 * It is very important to show the actual biomarker levels and not relative changes etc, since their absolute values are very important for the clinicians

Anywhow, thank you for the link for Tufte's forum (and the book). I will post this question over there.
Thursday, May 29, 2008
>Anywhow, thank you for the link for Tufte's forum (and the
>book). I will post this question over there.

If only I could find out how to post a new question over there ....
Thursday, May 29, 2008
Post a message at  Stephen Few is basically a down to earth Edward Tufte.
falcon Send private email
Friday, June 06, 2008

This topic is archived. No further replies will be accepted.

Other recent topics Other recent topics
Powered by FogBugz