7 ms·

I'm speechless. I'm going to read the publications in the about page to see what went into this because, at first glance, it appears to be very poor information

by asolove 13y ago

I'm speechless. I'm going to read the publications in the about page to see what went into this because, at first glance, it appears to be very poor information design: - The use of a circular arrangement is questionable. Judging the distance of items from the center along radial axes is not something humans are good at. And drawing lines between the points adds uninformative visual noise. Better to just have a little bar chart with consistent axes. - All the 'data' is presented without scales, numeric values, or trends over time. The one overall numeric value is presented without any context. We need a scale, a min and max, and a population average for this to be remotely meaningful. My first assumption is that this is like school grades in the US, where 85 is perhaps normal and 60 is abysmal. Is that right? Will that make sense to those from countries without grade inflation? - The directionality is all messed up. Sleep is bad when it's low, that makes sense. Nutrition is bad when it's too high? But 'nutrition' is good, right? It's not clear what 'conditions' and 'environment' are, or how they could be too low. - I would guess that the relative importance of these pieces of information differs by individual, gender, race, etc. Someone with diagnosed cancer should be marked as such rather than just being high on 'conditions.' And individuals with specific diagnosed ailments or family histories should have specific measures listed more prominently as important predictors. - Some information that isn't good or bad but just descriptive belongs on here, like age, occupation, and, for women, period/pregnancy/menopause status.

jsonin 13y ago
Maybe this takes a second glance to understand, I agree. There are a few dozen ways to present and graph data, that humans can quickly lock onto. The circular arrangement may or may not be the penultimate method to visualize health data, but it does afford a few good byproducts like seeing outliers quickly, identifying potential patterns (if you're looking across people), etc. The high-level hGraph with ~18 metrics (we're pretty sure this isn't The Set) is the galactic view, just to give you an instant sense of your health. Once you dive into the data, the level of fidelity increases. Think of the Eames 10x10 flick of seeing the universe and then zooming in to get to the atomic levels that reveal trends, history, etc. Directionality is how you construct the model (and then interpret it). You can eat too much. You can live in a hazardous environment (with too little X and too much Y). You're right that demographics will have a severe impact on metrics. That bias should be represented in the viz/algorithm. hGraph needs some TLC from clinicians, policy-makers, and we the patients in order to get it to the next level. Check out the screenshots of the health detail displays. If you have questions, fire away. -Juhan, juhan@mit.edu, hGraph co-author
- asolove 13y ago
  First let me back up and say, if you have read the research available on the items below and come to the current design after very careful thought, then I am just a jerk giving immediate negative feedback on your hard work and I hope you will forgive me. But if, as I suspect, you have given a lot of thought to the medical, statistical, and visual side of this, but are not well-read in the research on information display, I hope my suggestions will be helpful. - I would recommend being very careful with the wording. One can have too much "food" but "nutrition" has positive connotations and no lay person would speak of "too much nutrition" or "too much environment". - Please read the experimental material on information display. People study these things and polar display is very poor for picking outliers and recognizing patterns relative to almost any other option, including just printing numbers next to labels with no graph. - For building "interactive" data displays, please read Bret Victor's "Magic Ink", specifically the section on the evil of requiring interaction to display information. [1] One of the problems of the circular layout is that 18 is about as many items as can fit, thus you need interactions to see more data. With most other displays of this data, you could easily fit far more information into one, interaction-free display, which could even be printed out for use in a wider variety of contexts. - You could also use the clearer display to show actual numerical labels or trends over time. (Surely someone with high but declining blood pressure should appear different than someone with high and increasing blood pressure.) - Finally, I am very curious about the decision to include the same set of metrics for everyone and to display them as a single graph. Some of these metrics belong together on one scale, some belong on separate scales, some belong only on people with specific conditions, etc. - Are you planning to do a controlled test of this against other high-level patient summaries? Obviously my guess is that this graph and its interactions would fare worse than a one-sentence textual description provided by a past doctor. "Patient is a well-dressed and intelligent elderly man, with a successful history of controlling his diabetes through diet and exercise, but still showing signs of hypertension" seems to me much better than a graph with conditions: red that requires zooming in to see the word "diabetes" and LDL right on the edge that requires zooming in to see a declining trend over time. [1] http://worrydream.com/MagicInk/#interactivity_considered_harmful http://worrydream.com/MagicInk/#interactivity_considered_har...
  - 13y ago
frogpelt 13y ago
Turns out you weren't speechless.

by asolove 13y ago

jsonin 13y ago

Maybe this takes a second glance to understand, I agree. There are a few dozen ways to present and graph data, that humans can quickly lock onto. The circular arrangement may or may not be the penultimate method to visualize health data, but it does afford a few good byproducts like seeing outliers quickly, identifying potential patterns (if you're looking across people), etc. The high-level hGraph with ~18 metrics (we're pretty sure this isn't The Set) is the galactic view, just to give you an instant sense of your health. Once you dive into the data, the level of fidelity increases. Think of the Eames 10x10 flick of seeing the universe and then zooming in to get to the atomic levels that reveal trends, history, etc. Directionality is how you construct the model (and then interpret it). You can eat too much. You can live in a hazardous environment (with too little X and too much Y). You're right that demographics will have a severe impact on metrics. That bias should be represented in the viz/algorithm. hGraph needs some TLC from clinicians, policy-makers, and we the patients in order to get it to the next level. Check out the screenshots of the health detail displays. If you have questions, fire away. -Juhan, juhan@mit.edu, hGraph co-author

asolove 13y ago
First let me back up and say, if you have read the research available on the items below and come to the current design after very careful thought, then I am just a jerk giving immediate negative feedback on your hard work and I hope you will forgive me. But if, as I suspect, you have given a lot of thought to the medical, statistical, and visual side of this, but are not well-read in the research on information display, I hope my suggestions will be helpful. - I would recommend being very careful with the wording. One can have too much "food" but "nutrition" has positive connotations and no lay person would speak of "too much nutrition" or "too much environment". - Please read the experimental material on information display. People study these things and polar display is very poor for picking outliers and recognizing patterns relative to almost any other option, including just printing numbers next to labels with no graph. - For building "interactive" data displays, please read Bret Victor's "Magic Ink", specifically the section on the evil of requiring interaction to display information. [1] One of the problems of the circular layout is that 18 is about as many items as can fit, thus you need interactions to see more data. With most other displays of this data, you could easily fit far more information into one, interaction-free display, which could even be printed out for use in a wider variety of contexts. - You could also use the clearer display to show actual numerical labels or trends over time. (Surely someone with high but declining blood pressure should appear different than someone with high and increasing blood pressure.) - Finally, I am very curious about the decision to include the same set of metrics for everyone and to display them as a single graph. Some of these metrics belong together on one scale, some belong on separate scales, some belong only on people with specific conditions, etc. - Are you planning to do a controlled test of this against other high-level patient summaries? Obviously my guess is that this graph and its interactions would fare worse than a one-sentence textual description provided by a past doctor. "Patient is a well-dressed and intelligent elderly man, with a successful history of controlling his diabetes through diet and exercise, but still showing signs of hypertension" seems to me much better than a graph with conditions: red that requires zooming in to see the word "diabetes" and LDL right on the edge that requires zooming in to see a declining trend over time. [1] http://worrydream.com/MagicInk/#interactivity_considered_harmful http://worrydream.com/MagicInk/#interactivity_considered_har...
- 13y ago

frogpelt 13y ago

Turns out you weren't speechless.