AI MedTech

JPM 2020 Insights

Released on January 17, 2020 | Written by Joe Mullings of The Mullings Group

There is a “dumb dumb” punch coming in medtech as it applies to data and the validity and reliability of algorithms for AI in medtech.

The access to historical data as a fast track to success can be a liability.

A percentage of the industry is going to get dinged around the subject of retrospective data versus prospective data.

Prospective data usually has fewer biases in it than retrospective data.

Retrospective studies may take less time and be less expensive because the data in them has already been measured.

However, retrospective studies are open to a higher level of bias in data selection and analysis.

AI algorithms are slaves to the data from which they learn.

The data that is used to train these systems are critical to ensure accuracy.

A majority of retrospective data sets available for AI are notoriously biased.

Especially in the US, retrospective data is over-indexed towards males and whites and that will be training an algorithm that will have real-world impact.

Historically medtech companies and the FDA have counted on STEM professionals to be sure the patient and public are as safe as possible.

I expect ethicists, philosophers, and anthropologists will be joining the ranks of medtech companies and regulatory agencies as AI and ML continue to drive forward.

Written by

The Mullings Group
s