The Cost of Getting Sick

GE and Ben Fry (now the director of SEED visualization), show the cost of getting sick, from the individual’s and insurer’s perspective. The data: 500k records from the Medical Expenditure Panel Survey from GE’s proprietary database. The visualization: a polar area pie chart.

Wedges are colored by chronic condition and wedge size (angle), represents the percentage of patients with the condition. Wedge length from center to edge is cost. The lightly colored portions are personal cost and the darkly colored are insurer cost.

So we’re sort of seeing cost by area of a wedge, but not quite.

You can also use the scroll at the bottom right to adjust age, which works as a good filter.

Update: Here’s Ben Fry talking a little bit about his work as director of SEED visualization.


  • Reminds me to the Sunburst visualization implementation I did for the JIT some time ago: .
    There are a couple of artifacts in this flash version (at least when rendered with my browser), like for example the graph disappears when scrolling, etc.

  • This is a little weird, since people usually read pie charts as having mutually exclusive categories… but in this case people could be part of more than one wedge.

    It’s also kind of interesting to see how he’s using area/length confusion to subtly emphasize the personal cost of getting sick in proportion to the insurer cost.

  • It’s a terrible chart, with lots of visualization data problems. I wrote an in-depth critique on my site: The Cost of a Sick Chart

  • I find it ineffective.

    1. It rotates when I click on a wedge, destroying the angular relationship I had built with a previous wedge.

    2. By the time I look at the blue wedge, the red wedge, and the purple wedge, I’ve forgotten what the blue one represented.

    3. It’s a good thing the costs are shown numerically when you click on a wedge, because this inner and outer radii thing isn’t working for mw.

    4. I can’t look at a condition in the last dimension, to see for example how the incidence and cost of depression varies with age.

    Is the compiled data available for dabbling?

  • Here’s a huge distortion. Drag the slider to 25 years old. Diabetes has a total cost of $4513 per patient. Now drag the slider to 50 years old. The cost is $6607 and the radius is 49. So the cost goes up by 46% while the radius, which is supposed to encode cost, goes up by 123%.

    I also have no feel anymore for what the wedges represent. They can’t be something that adds to 100%, because a diabetic patient may well be depressed as well.

    Yeah, it’s eye candy, the bad kind. It captures your attention, and obscures the meaningful information behind gratuitous animation and horrible graphical metaphors.

  • I looked at this and couldn’t understand what or how I was supposed to be comparing.

    From my beginner’s perspective, I question the choice of using a pie shape and I question the usage of animation.

    The cost of healthcare is too important for this to be the only visualization attempt with this data. It would be wonderful to see how others in the visualization community would re-invent this.

  • constantnormal November 23, 2009 at 11:44 am

    Looking at the numbers and bouncing them off one’s own “reality filter” reveals that either GE is being subjected to a huge health insurance scam, or the claims processing costs at the health insurers exceeds benefit payments by a factor of several hundred percent.

    I guess those two things really collapse down to the same thing …

  • I have never liked these charts but there is a variant called spie charts that do seem useful in certain settings, specifically, for comparing two pie charts together. An example would be visualizing changes in proportion of budget spending one year versus another. See the bottom of this website for a nice example (and link to an academic paper) showing how to use this type of visualization.

  • The spie chart is just another cutesie visualization that can’t display information comprehensibly. A bar chart or dot plot would be a better way to compare two data sets like that.

  • I think this does work out to 100% if we treat the case as the illness, not the person.

  • Mary Jane Rutkowski December 2, 2009 at 4:37 pm

    I would like to see the programming behind this. I have a project where we would like to consider economic sectors (analogous to the illnesses), emissions benefits and costs ranging over a period of about 30 years. This seems like a good visualization approach. Thanks for the inspiration!