XAI_design_prototype.txt

The overall title for this prototype is "AI for Explainable Healthcare Adverse Event Risk Prediction" 

On the top of the prototype is the (fictional) patient's information: His name is Steve Rogers, he is 78 years old, his race is Black, and his Charlson Comorbidity Index conditions are COPD, PVD, and Type 2 DM with a 2% 10-year survival

On the left hand side of the image are questions grouped together within the larger user questions. 

At the top are "why" questions: 1. why is this patient predicted of this risk and 2. what are his risk factors?
Below the "why" questions are the "how to be that" questions: 1. what can be doen to reduce the patient's risk? and 2. what worked for other patients with similar profiles?
After the "how to be that" questions are the "performance" questions: 1. on what types of patient might it work worse? and 2. how well does it work? 
After the "performance" questions, and at the bottom of the left hand side are the "data" questions: 1. is the training data similar to my patients? 2. What is the population of the training data?

Within a larger panel, there are smaller panels. 

One of these smaller panels is titled "History". It shows the last 12 months of patient history. There has been 1 Admission, 0 Emergency Department visits, and zero Hospital Aquired Conditions

One of the smaller panels is titled "30 day risk of all cause admission" with smaller pieces of information and there is a label of "performance" on it.
On the upper left hand section of this panel is a link to "View data sources" and it has a label of "data" on it.
there is a bar showing there is 0% chance of low, 15% chance of moderate, and 25% chance of high, with a risk score confidence of good in the low category of plus or minus 2%
The patient's 30 day admission risk is stated to be 5% or 1 in 20 chance
The Medicare average is 16%
Another average that is obscured by a label is 13%

Another smaller panel is titled "factors that contribute to the risk of admission" and this is labeled "why" 
There is a chart with three measures on each side of a central line. One side is labeled "decrease" and the other side is labeled "increase"
The patient's age being less than 80 decreases his risk of admission to almost one measure.
The patient having COPD increases his risk of admission to over two measures
the patient's number of ED visits increases his risk of admission a full three measures
the patient's positive indication of mood disorders increases his risk of admission beyond the three measures
the patients Chraleson Comorbidity Index of 6 points or 13% increases his risk of admission beyond the three measures.

Another panel has the title "Action impact" and is labeled "how to change to be that"
It shows that Steve has had a pneumonia vaccine and this gives him a 3 percent point lower risk.
It shows that people like Steve who don't smoke have a 1 percent point lower risk.

A final side panel has the title "Medicare Claims data (2008-2011)"
The text states "characteristics of 212,236 medicare beneficiaries randomly selected and shared by CMS"
A chart with the title "age" shows the following: 
people under age sixty have a five percent chance
people between the ages of sixty and sixty-nine have a thirty-five percent chance
people between the ages of seventy and seventy-nine have a forty-five percent chance
people the age of eighty and over have a fifteen percent chance
a chart with the title "gender" shows the following: 
people labeled "male" have a fifty-one percent chance
people labeled "female" have a forty-nine percent chance
a chart labeled "race" shows the following: 
people labeled "caucasian" have a forty-one percent chance
people labeled "Black" have a twenty-two percent chance
peopled labeled "Hispanic" have an eighteen percent chance
people labeled "other or unidentified" have a twenty percent chance
The panel indicates that there are no Medicare part D (medications) data or any EHR data (labs, physiological data, notes) used in the prediction