#AI #ML in healthcare: Will the twain ever meet?

Incorporating the Artificial Intelligence in healthcare has several issues. I stumbled on an excellent post in HBR that lists the key deficiency of AI- it is a black box. For all its deficiencies, AI and machine learning will never explain a process of “inclusion and exclusion” or why it came to a decision the way it did.

The author speaks about having “regulatory frameworks” to leave it at the FDA’s doorstep. I’d be genuinely surprised if the regulator is going to press forward and steam roll its bureaucratic lethargy to respond to lightning fast changes.

For example, if an algorithm predicts the diagnosis of a heart attack incorrectly and requires an urgent update, who will push and rectify that update is safe to be deployed?

The FDA has also been enrolling select software-as-a-medical-device (SaMD) developers in its Digital Health Software Precertification (Pre-Cert) Pilot Program. The goal of the Pre-Cert pilot is to help the FDA determine the key metrics and performance indicators required for product precertification, while also identifying ways to make the approval process easier for developers and help advance healthcare innovation.

Clinical Decision Support systems have been proposed as “higher-risk software function”.

In a related statement from the FDA, Amy Abernethy, its principal deputy commissioner, the agency plans to focus regulatory oversight on “higher-risk software functions,” including those used for more serious or critical health circumstances. This also includes software that utilises machine learning-based algorithms, where users might not readily understand the program’s “logic and inputs” without further explanation.

An example of CDS software that would fall under the FDA’s “higher-risk” oversight category would be one that identifies a patient at risk for a potentially serious medical condition — including a postoperative cardiovascular event — but does not explain why the software made identification.

I cannot but argue while it is required for the safety of the patients, the classical “guidelines based approaches” are paramount. The author makes a case for bottom up approach- using the AI in coding but offers no concrete examples of how it can be implemented.

Here’s some naivete too:

Similarly, investors must also have a clear understanding of a company’s product development plans and intended approach for continual FDA approval as this can provide clear differentiation over other competitors in the same space.

Investors don’t understand any specialised niche. Most of the money that’s floating in the market is primarily  punts made on an “opportunity for exit”. I am not entirely convinced. Investors may also come on board if they believe they can change the regulations in their favour.

Read more.