Just in relation to point three... the algorithm can't learn unless it's taught. And the way it gets taught is by having somebody enter the outcome of each consultation. You have two stages:
1) Prediction.
a) Algorithm (listens) to cough
b) Algorithm relates what it heard to what it has learned in the past from previous patients.
c) Algorithm makes a prediction.
So far, no learning. That happens in:
2) Training
a) Somebody tells algorithm what the actual outcome was
b) Algorithm smiles if correct and "reinforces" the selected pattern it used before.
c) Algorithm frowns if incorrect and tries to find a way of predicting the correct answer from what it heard.
If the training step is not taken then the predictions won't improved over the course of the trial. This could be a problem if somebody coughing with an American accent makes it harder for the algorithm to operated. But otherwise, with the amount of training the algorithm has already received, there will be issue at all! How much better do you think the results should be?
There's nothing to stop the results of the trials being entered after the end of the trial, so the learning opportunity isn't lost as long as it's possible to connect the outcome with the original cough.
And finally given that some of these conditions are difficult to diagnose without tests and a proficient clinician, there will be a percentage of incorrect outcomes to screw up the knowledge.
RAP Price at posting:
43.8¢ Sentiment: Buy Disclosure: Held
(20min delay)