Healthcare AI Looks for Patterns

The easiest thing to do with AI (or other kinds of analytics) is to pattern match.  Not the only thing,  but if you have enough data to train it, you can find a likelihood that you have detected the same thing.    You also have to consider the risk of false positives and negatives in context.  There is no downside in testing the idea with just the data you have.  I am then often asked then, how much data?

  The answer is really any amount of data,  say hundreds of examples, to get started to test some ideas, set up the methods.    But then to deliver real results your are talking thousands to many thousands of examples.    And then also consider re-testing of other contexts, and testing over time.   And reworking the risk analysis as well.   Below a real example, note the variants in setting up the idea, and the context of alerts.  Note that the data is plentiful, generated by reliable sensors, a good sign,

Hospital System Uses AI to Predict Deadly Condition
By The Wall Street Journal 

The app predicts the likelihood of acute respiratory failure.  ... 
The Montefiore Health System and its affiliated medical school in New York are employing artificial intelligence to predict the likelihood of a patient suffering from a common respiratory malady.  ... 

The Montefiore Health System and its affiliated medical school in New York are employing artificial intelligence (AI) to predict a common respiratory malady among patients.

Parsa Mirhaji at Montefiore's Center for Health Data Innovations said his team uses three live AI applications, with four more in development; three hospitals use the app to predict acute respiratory failure, which can require intubation.

The app analyzes more than 40 data points for each patient every several hours, including drug intake and blood pressure, with patients at risk of intubation or death assigned a risk score.

An intervention alert is activated, and sent to doctors, when patients' scores surpass a specific threshold.

Tests indicated the AI app yielded lower false-positive rates than traditional hospital alert systems.

From The Wall Street Journal
View Full Article - May Require Paid Subscription  ... "