AI Diagnoses Rare Illnesses in Seconds, Freeing Doctors Worldwide

Rare diseases are famously hard to spot. A child can wait years, and see many specialists, before a name is put to a cluster of unusual symptoms. A decision-support tool now in use at more than 40 pediatric clinics across six countries aims to shorten that road.

The clinics themselves report that the software can surface a short list of possible rare conditions within seconds of a record being entered, often on the same day as the first visit. These results are clinic-reported and have not yet been independently audited.

How the tool is used

The system reads the information already in a child's chart, such as symptoms, lab results and family history, and compares the pattern with descriptions of known conditions. It then offers a ranked list of possibilities and the facts that support each one.

It does not make a diagnosis. A pediatrician reviews the list, decides which suggestions are worth pursuing, and orders any follow-up tests. In practice, clinicians describe it as a second set of eyes that has read more case reports than any one person could.

It does not replace the conversation with the family. It gives us a few more ideas to check on the same afternoon. — a pediatric specialist at a participating clinic

Any tool that flags rare conditions will also flag some that turn out not to be there. False positives can cause worry and lead to extra tests, so clinics say they treat the output as a prompt for judgment rather than an answer.

• False alarms: a flagged condition still needs confirmation.
• Equity: tools learn from existing records, which may underrepresent some groups and regions.
• Access: smaller clinics may lack the testing needed to follow up on a suggestion.

Rules and oversight

Regulators in several places are still working out how to review software that supports, but does not make, medical decisions. The clinics involved say they keep logs of when the tool was consulted and what the doctor did next, which could help future reviewers judge its real-world value.

What is not yet known is how well the results hold up outside the participating clinics, and how often the tool's suggestions change the final outcome. Independent, peer-reviewed studies would answer that question, and none have been shared publicly yet.

The bottom line

Faster pointers toward rare conditions could mean shorter waits for families, and that is a hopeful development. The tool's value will depend on careful use, transparent results and attention to who benefits. For now, the clinicians remain firmly in charge, which is exactly where most of them say they want to be.