Better ways to use — and measure — AI in medicine

A group of newly published studies outlines how artificial intelligence can be used to improve care in hospitals and enhance clinical trials.

Why it matters: Patients stand to benefit hugely from the use of AI in medicine, but only if there is solid evidence the interventions work — and it can be done without introducing errors or compromising privacy.

What's happening: In a paper published Wednesday in Nature, researchers from Stanford University reviewed the field of "ambient intelligence" — the use of hospitals and homes that employ sensors and AI to improve patient care.

• They note that as many as 400,000 Americans die each year because of medical errors, many of which could be prevented with smart sensors that alert doctors and caregivers when things are going wrong — or when they're making mistakes.
• As infrared sensors get cheaper and more ubiquitous, they can be used for everything from detecting whether visitors have washed their hands upon entering a hospital room to alerting doctors when patients are writhing beneath bedsheets.

Yes, but: One challenge in determining the effectiveness of AI in medicine has been the quality of the research itself.

• A paper published last year found less than 1% of 20,500 studies of AI in health care were good enough for independent readers to have confidence in their conclusions.
• Now in a pair of studies published Wednesday, researchers laid out the first international standards for the reporting of clinical trials for AI.
• The rules specifically call on researchers to indicate how AI studies handle poor-quality or unavailable data, as well as how involved humans are in supposedly AI solutions.

The bottom line: We need to know AI can work in medicine before we put it to work.

Go deeper: Artificial journalism gets a trial run