Clinical Researcher here. I don't intend to mean anyone any personal offence by stating what I'll state. Please choose to ignore me if you're going to attack me personally for having an opinion different to yours, especially in a public forum.
This is purely for the ones who are genuinely trying to understand why " in 33% cases/presentations, the 2 doctors disagreed".
It is extremely incorrect to interpret from THAT statement that doctors are unreliable. That does also not mean that they are correct 50% of the time. I was shocked when I read that here somewhere. For the intents and purposes of RAP - most likely it means there was insufficient data to interpret.
As I said, the disagreement in medical opinions often arise in complex cases, rather than simple ones. Additionally, the amount of disagreement depends on
a) the history and examination given to each individual doctor. "Patient factors"
b) the specificity of the pathology and radiology tests used. "Test factors"
c) personal professional experience, knowledge of patient populations in the area of practice etc. "Doctor factors"
Some may choose to think of (c) as a subjective, wishy-washy thing and some may interpret it as customized medicine. Don't bother debating it - because in the end you cannot remove that human factor from a human doctor and we will still need that doctor to treat us AFTER this wonderful app diagnoses us. We will also need that doctor to deal with side effects.
Consider this. You chose to treat your child with an antibiotic because the app objectively said so. Your child has an anaphylactic attack and dies from that possibly unnecessary antibiotic. Who will be to blame? The prescriber? The app? You? *** topic for another day lol***
Imagine a child with a history of allergies and asthma comes in to the ED with a cough. Now those 2 symptoms could represent any of the diagnoses that the app is able to distinguish. The doctors will read the same notes and will have the same data provided to them. One doctor may have seen the child before, or someone very similar. The other may say I would like x and y more prior to making a decision. This is where the disagreements may begin. There is an intrinsic lack of sensitivity and specificity in EVERY MEDICAL TEST. Very few medical tests are 100% predictive of anything, 100% sensitive or 100% specific.
Now when a person is basing their clinical decision on (3-4 or more) such tests which in and of themselves have an intrinsic lack of accuracy, then you cannot expect 100% correctness from a compiled clinical diagnosis.
For example:
1. Not every pneumonia shows up on a chest X ray. Some show up later. Some have poor quality films because the child didn't breath deep enough. Some may have an underlying disease that makes interpretation of a x ray difficult. Not every asthmatic wheezes. Not every patient mounts a fever.
2. Not all pathology results that show a bacteria mean that the bacteria is causing a pneumonia.
3. Conditions progress and a test done at time A does not necessarily depict the situation at time B when the clinician evaluates or reevaluates a patient.
4. Instruments can record or measure incorrectly or insufficiently. Eg. Thermometers and saturation monitors can provide unreliable readings/have a margin of error causing difficulty interpreting borderline ranges.
If one truly believes that all the information about each and every differential diagnosis of a cough can be gleaned from the frequency/vibrations/depth/sound etc of the cough itself WITHOUT relying on additional data that the doctors use(thermometers, oxygen saturation monitors, age, family history, previous history) then I would expect the app to perform better than doctors, despite the doctors disagreeing. And that's assuming the algorithm works.
If the app uses some or most of those parameters in addition to the cough sounds, then there is an inherent subjectivity in the app itself. It provides an objective answer, but that does NOT translate to accurate answers all the time. The subjectivity that is eliminated by the app, does have some meaningful use.
Yes please understand this.
The app may shine a bright blue screen and say "pneumonia". Sure it's objective. But it is not accurate and that is the issue resapp faces in the future.
TLDR: Doctors disagree because there are cases,subjective situations that cannot always be fully objectified. And because the tools they have, are inherently flawed. Their subjective experiences and expertise actually works in the patient's favour - again NOT always but veryvery often.
If the app is able to gain an ability to learn further from the objective data(sound frequencies) to the point that it eliminates the subjective benefits (yes, benefits!) a doctor has, then and only then it will perform better than doctors.
I probably have wasted my time writing all this but hopefully it clarifies some (not all)of the stuff.