Discussion


Note: Images may be inserted into your messages by uploading a file attachment (see "Manage Attachments"). Even though it doesn't appear when previewed, the image will appear at the end of your message once it is posted.
Register Latest Topics
 
 
 


Reply
  Author   Comment  
Parcellus

Registered:
Posts: 3
Reply with quote  #1 
I am a Consultant Biochemist working for the NHS in the UK and have a major interest in the way we present pathology blood test results.
 
The current view of laboratory data is a display form known as “separable”, where each result is shown as a single output. Currently, we attempt integration or higher level relations to hopefully emerge by grouping results into profiles e.g. Liver Function Tests, as an aid to interpretation.
 
These profiiles typically have no more than 5 separate analyte results.
 
I am working on configural displays which arrange low level data into a meaningful form which  is greater than the sum of the parts and is in part, based on principles of gestalt psychology. Separable display generally makes it easier to extract low level information and although harder to integrate does afford some benefits in test interpretation. In contrast configural display makes it harder to extract low level information. I have come to a compromise which I have called "diagnostic schema maps" which preserve the separable view of analytes whilst still attaining a configurable view making the data easier to integrate.

 

Diagnostic schema maps are polar style N polygon configural displays. They, typically form an asymmetric geometric shape. The emergent property is the shape of this object. They rely on human pattern recognition and visual cognition to identify a match with a diagnostic category. The schema maps can be thought of as an abstraction hierarchy at the lowest level it contains a graphic depicting range, above this are plotted the analytes that form the profile in the context.

 
The example below is to help junior doctors diagnose the cause of having a low blood sodium (hyponatraemia). The most important variables are arranged around the perimeter of the polygon. The configurable displays which differ in shape and colour represent the major disease states that cause a low blood sodium with the points on the axis correlating with their evoking strength. Overlayed are the patients results, represented by a thicker line. By comparing the shapes, junior doctors should hopefully grasp the most likely cause of their patients problem, thus reducing diagnostic uncertainty.
 
Do you think I am on the right track with this or is there a better way of approaching this.
 
Diagram below.
 
BW John

Attached Images
Name: radar.jpg, Views: 387, Size: 54.40 KB


wd

Registered:
Posts: 167
Reply with quote  #2 

A different track would be more enlightening.  Take a look at http://www.perceptualedge.com/articles/dmreview/radar_graphs.pdf first and then also http://www.perceptualedge.com/articles/visual_business_intelligence/our_fascination_with_all_things_circular.pdf

If you search the discussion forums for 'spider', you might find some application ideas as well.


__________________
Bill Droogendyk
sfew

Moderator
Registered:
Posts: 805
Reply with quote  #3 
John,

If you provide the data set, we'll be able to put together alternative designs for you to consider. If there's no easy way to provide the data here in the discussion forum, you can email a copy to info@perceptualedge.com and I'll make it available.

Thanks,

Steve

__________________
Stephen Few
sfew

Moderator
Registered:
Posts: 805
Reply with quote  #4 
John,

I assume that a doctor would use this display to compare the patient's multivariate profile to that of those disease states that can contribute to low blood sugar. To serve this purpose, it must be easy to compare the multivariate profiles of the patient to each disease state, but when they are all combined in a radar chart, they cannot be easily discriminated and compared. This could be alleviated by placing each profile in a separate radar chart. This would allow the shapes of the multivariate profiles to be compared, but a series of bar graphs could provide the same functionality in a way that is easier to label. The following example illustrates what I have in mind.

Attached Images
Name: Small_Multiples_of_Parallel_Coordinates_#1.png, Views: 354, Size: 18.90 KB



__________________
Stephen Few

Parcellus

Registered:
Posts: 3
Reply with quote  #5 

Dear Steve

 

Firstly, thank you for this :)

 

I agree the bar chart approach is better. The ability to instantly pattern match those shapes seems intuitively easier than using radar charts. I guess that visually there must be an optimum number of alternatives to be displayed in order to maximise the ability to discriminate between them. A lot of what we deal with in clinical biochemistry as part of a differential diagnosis are quite large dimensional lists. We can employ statistical procedures like categorical principal component analysis to reduce the number of dimensions. Or alternatively create a hierarchy to break the problem up and then tackle those in a stepwise manner.

 

I really appreciate your feedback.

 

Thanks again

BW John


Anders

Registered:
Posts: 18
Reply with quote  #6 
Hi John,

being a chemist working in the field of clinical biochemistry myself, I found this thread really interesting.

I have one tip for you, read the article "Grapical Summary of Patient Status" by Seth Powesner and Edward Tufte (The Lancet, Volume 344, Issue 8919, P 386-389). This is also discussed in one of Tufte's books (can't remember which one right now).

Regards,
Anders

laust

Registered:
Posts: 6
Reply with quote  #7 
You could also just show them the deviation between the Patient data and the disease indication. This makes it even faster to see how well the profile matches.

Attached Images
Name: DiagnosticVisualisation.JPG, Views: 268, Size: 128.97 KB


Parcellus

Registered:
Posts: 3
Reply with quote  #8 
Thanks Laust

Interesting idea using the standard deviation approach. You right, the display not only demonstrates which variables are useful, but also conveys the goodness of fit with the outcome

BW John
laust

Registered:
Posts: 6
Reply with quote  #9 
Hi John,

Just to make sure we agree on terms, the display I suggested does not make use of standard deviation. It simply calculates the difference between the customers profile and the disease profile.

Also, I would advice you to consider two things before employing this display.
1) How large are the uncertainties of the measures / disease profiles. You may wish to include these in the display.

2) Right now the display adds equal weight to all measures. Is this true in real life? Or are some measures more important than others?
Previous Topic | Next Topic
Print
Reply

Quick Navigation: