AI Tool Accurately Identifies Pediatric Patients at High Risk for Blood Clots

Blood clots in children are uncommon but can extend hospital stays and elevate the risks of complications and mortality after discharge. A new artificial intelligence (AI) tool is now available to accurately identify which pediatric patients are at high risk for developing blood clots.

Researchers conducting a clinical trial called CLOT, led by Vanderbilt University Medical Center (VUMC, Nashville, TN, USA), found the AI tool to be effective in identifying pediatric patients at a higher risk for blood clots. However, the outcomes didn't differ from the control group. This was partly because physicians accepted the AI's recommendation to start blood-thinning treatment in less than 26% of the high-risk cases. Doctors had concerns that the treatment might lead to significant bleeding, even though such complications were not noted during the study. Despite the unexpected results, the research confirmed the AI tool's safety and efficacy and offered key insights for its successful integration into healthcare practice.

To determine who was at higher risk, the research team examined over 110,000 admissions to a children's hospital from electronic medical records. They found 11 factors linked to a greater risk of blood clots, like specific lab results, diagnoses, and whether the patient had surgery or consultations with cardiology or infectious diseases. This data was used to create a predictive model that calculated daily risk scores for every child admitted to the hospital, enabling the team to rapidly assess more than 100 patients each day and zero in on those most likely to develop blood clots.

The 15-month trial, spanning from November 2020 to January 2022, included 17,000 hospital stays and had patients randomly assigned to two groups. For the study group, their risk scores and anti-clotting treatment recommendations were shared with their medical teams. The control group patients were deemed high risk by their doctors without the aid of the AI tool, and they also received blood-thinning medications. Neither group experienced bleeding complications from the treatment. Ultimately, the rate of blood clots did not differ between the two groups, and it was found that the recommendation to begin blood-thinning therapy was followed just 25.8% of the time in the study group.

The CLOT trial showed that it is possible to quickly get results without straining the medical team's time or resources, simply by using existing electronic medical record data to automatically include patients in clinical trials. It also proved that AI models have potential value in healthcare. However, there are still challenges to address. Future trials are being planned to delve into why healthcare providers are hesitant to follow AI recommendations for blood-thinning treatments in high-risk cases and how to overcome that reluctance.

“There is going to be more and more AI in healthcare. Having a system established where we can assess these (models) will allow us to provide safer and more effective care to our patients,” said Shannon Walker, MD.

“This study demonstrates that a pragmatic patient-level, randomized, controlled trial is the most ethical and effective way to assess whether AI tools are safe and effective,” added Daniel Byrne, MS.

Related Links:
VUMC