Medical AI accuracy declines when processing new patient data

Personalized medicine moves away from one-size-fits-all prevention and care
The algorithms that predict effective care are accurate with familiar data
That accuracy falters when models are confronted with new patients

Updated: Jan 17, 2024 / 01:18 PM CST

(Getty Images)

(NewsNation) — An analysis of the algorithms doctors use to plan personalized medical treatments found the tools don’t perform as well when confronted with new data.

The research, published on Jan. 11 in the journal Science, found that AI-reliant tools that predict how patients will respond to certain treatments are highly accurate for people within a sample they were trained on.

That accuracy declines to “little better than chance” when the algorithms are fed information about an original sample or different data sets, a summary published in the journal Nature explained.

Personalized medicine, still an emerging practice, aims to usher medical care away from a one-size-fits-all approach. Instead, it relies on a patient’s unique genetic profile to guide decisions about the prevention, diagnosis, and treatment of disease, according to the National Human Genome Research Institute.

Researchers hope that new technologies using machine learning and AI could produce better models for predicting treatments will be most effective for different patients, Medical Xpress reported.