Вештачка интелигенција

Tackling High-Risk Thrombosis in Pediatric Patients with Artificial Intelligence

Summary

Researchers at Vanderbilt University Medical Center (VUMC) have developed an artificial intelligence (AI) tool capable of accurately identifying the risk of thrombosis in pediatric patients. Thrombosis is rare in children, but patients who develop it may experience longer hospital stays […]

Tackling High-Risk Thrombosis in Pediatric Patients with Artificial Intelligence

Researchers at Vanderbilt University Medical Center (VUMC) have developed an artificial intelligence (AI) tool capable of accurately identifying the risk of thrombosis in pediatric patients. Thrombosis is rare in children, but patients who develop it may experience longer hospital stays and an increased risk of adverse outcomes.

To address this issue, the researchers embarked on developing and validating an algorithm to identify high-risk patients, using electronic medical record (EMR) data from 110,000 patients at Monroe Carell Jr. Children’s Hospital at Vanderbilt. Based on this data, the team identified 11 risk factors associated with thrombosis, including whether the patient had undergone surgery, had consultations with infectious disease specialists or cardiologists, or had received certain diagnoses.

Using this information, the researchers developed a predictive model that automatically assesses EMR data and calculates the risk for each pediatric hospital admission on a daily basis. This allowed the research team to prioritize patients with the highest risk.

The risk prediction tool was used during a 15-month clinical trial called “Children’s Likelihood of Thrombosis” (CLOT), involving 17,000 patients. The patient cohort was divided into a study group and a control group, with the risk assessed using the AI tool in both groups. However, the risk was communicated to the care teams of the study group patients, while the control group remained unaware.

Patients in the study group were provided with recommendations to initiate thrombosis prevention therapy based on their risk. The control group patients were informed by their physicians that they were at high risk and also received preventive therapy.

No bleeding occurred in either group of patients receiving anticoagulant therapy as part of thrombosis prevention. Additionally, there was no significant difference in the rate of thrombosis formation between the two groups at the end of the study.

However, the researchers also discovered that recommendations to initiate thrombosis prevention therapy were followed by physicians in less than 26% of cases in the study group. Physicians expressed concerns that adhering to the recommendations could lead to greater bleeding, though this did not occur during the study.

The researchers explained that these insights could be used to inform the implementation of AI in healthcare. “There will be increasingly more AI in healthcare. Establishing systems in which we can assess these models will allow us to provide safer and more efficient care for our patients,” said Dr. Shannon Walker, Assistant Professor of Pathology, Microbiology, Immunology, and Pediatrics, and the first author of the study.

“This study demonstrates that pragmatic, patient-level, randomized controlled trials are the most ethical and effective way to assess whether AI tools are safe and effective,” noted co-author Daniel Byrne, Director of AI Research at the Advanced Vanderbilt AI Laboratory (AVAIL) and the Department of Biostatistics.

The research team emphasized that the clinical trial was necessary to determine why the implementation of the model was unsuccessful, highlighting that it was likely due to mistrust towards the tool’s recommendations rather than the failure of the model itself.

To further examine the potential value of predictive AI models in healthcare, the researchers plan to conduct another clinical trial that will explore physician hesitancy in following AI recommendations and how to overcome these barriers.

“We need to ensure that these models perform as expected,” said Walker. “The infrastructure from this trial will enable the study of large populations and determine whether interventions that utilize artificial intelligence are safe and effective, helping us identify patients who will benefit the most.”

FAQs: