Lancet Digital Health: Is the future of medical diagnosis in computer algorithms?

5 May, 2019

Dear HIFA colleagues,

Below are the citation and extracts from a news item in the new OA journal The Lancet Digital Health, and a comment from me below.

CITATION: Karl Gruber. Is the future of medical diagnosis in computer algorithms?

News, volume 1, issue 1, pe15-e16, may 01, 2019

Published: May, 2019 DOI:https://doi.org/10.1016/S2589-7500(19)30011-1

--

A 2014 study led by Singh found that at least one in 20 adults in the USA left the doctor's office with a misdiagnosis, which equates to 12 million people per year. Half of these misdiagnoses, the authors estimated, are potentially harmful.

One way to help physicians improve their diagnostic accuracy is by helping them analyse data more efficiently. With this goal in mind, and as computers become more powerful, scientists are looking at AI [artificial intelligence] to design systems that can think like humans, only faster and more efficiently...

Babylon's AI system, for example, is currently being used by over 3 000 000 users from the UK and Rwanda. The system provides quick and easy-to-access advice for common ailments. But, how accurate is this advice? In a non-peer-reviewed 2018 study, Babylon researchers compared the performance of human physicians with Babylon's AI system at evaluating identical medical cases. According to their results, the AI system outperformed the average human doctor.

However, in a 2018 correspondence in The Lancet, Hamish Fraser, Associate Professor of Medical Science at Brown University, Providence, RI, USA, argued that Babylon's AI system still needed more testing before being able to prove it can perform well in real-world settings. For instance, Babylon's study did not use real patient data, but so-called patient vignettes, clinical examples of patient-related cases that are normally used for educational purposes...

“I would envision AI can be used as a primary deployment to diagnose and triage diseases in rural or resource poor areas, for example, as an online system”, Zhang says. But even under these circumstances, Zhang [University of California, San Diego] argues that human doctors should still be involved and that “AI should be used primarily as a physician assistant to reduce overall burdens of physicians and it should not replace physicians”.

--

Comment (NPW): The potential role of AI 'in rural or resource poor areas' is hugely important to the progressive realisation of healthcare information for all (and the ability to interpret it). Are any of HIFA's 64 members in Rwanda aware of or involved in the work of Babylon? It would be fascinating to know more about how this AI system is deployed.

Best wishes, Neil

Let's build a future where people are no longer dying for lack of healthcare information - Join HIFA: www.hifa.org

HIFA profile: Neil Pakenham-Walsh is coordinator of the HIFA global health campaign (Healthcare Information For All - www.hifa.org ), a global community with more than 19,000 members in 177 countries, interacting on six global forums in four languages. Twitter: @hifa_org FB: facebook.com/HIFAdotORG neil@hifa.org