Presentation (CTI)

OBJECTIVES: Rotator cuff (RC) tears present significant clinical challenges, with treatment decisions, ranging from conservative management to surgical intervention, relying heavily on accurate tear severity assessment. However, International Classification of Diseases (ICD) codes often lack granularity to capture RC tear severity, complicating the use of real-world data (RWD) for research and clinical decision-making. This study aimed to utilize machine learning (ML) Natural Language Processing (NLP) to extract and classify RC tear severity from radiology reports, enabling scalable and precise identification of massive RC tears in RWD sources.
METHODS: We analyzed MRI radiology reports of patients with an ICD-coded diagnosis of RC tear from a state-mandated medical center in Israel. A board-certified radiologist labeled the criteria for massive RC tears to create a ground truth dataset. Using the Briya© computational platform, a NLP classification model was developed and trained to identify tear severity. Model performance was evaluated against the ground truth using metrics such as area under the curve (AUC), sensitivity, specificity, and accuracy.
RESULTS: Radiology reports of 406 patients were analyzed, with a mean age of 58.8 years (standard deviation=8.78), and 61.6% of the patients were male. Based on expert radiologist tagging, 70 cases (17.2%) were identified as massive rotator cuff tears. The NLP model demonstrated strong performance in classifying tear severity (detailed metrics to be provided). The use of NLP significantly enhanced the identification of massive rotator cuff tears compared to relying on ICD codes alone.
CONCLUSIONS: This RWD-driven NLP approach offers a robust, scalable solution for identifying and classifying massive RC tears from radiology reports, addressing the limitations of ICD coding in real-world healthcare settings. By enabling enhanced data extraction and improving the precision of RC tear diagnosis, this method has the potential to support both clinical decision-making and large-scale RWD research on RC tears.