OC-019 LEVERAGING NATURAL LANGUAGE PROCESSING AND ARTIFICIAL INTELLIGENCE TO LABEL UNSTRUCTURED DATA FOR RISK PREDICTION

Abstract Aim Majority of clinical data is categorized into unstructured data, including operative notes containing unexplored key details yet to be incorporated a surgical risk model. We sought determine whether Natural Language Processing (NLP) methods could implemented systematically capture note features that can considered for patient-specific modeling. Methods Operative describing abdominal wall closure 8233 patients undergoing colorectal operations between 2013–2019 were obtained. A development dataset was manually generated 600 skin and fascia suture material, size, pattern. These analyzed train/test various NLP approaches assess their corresponding accuracy. Results Of the remaining 7633 notes, assessment type, pattern by keyword search, regular expression, RoBERTa 5-fold Cross Validation(CV). Overall, CV demonstrated superiority in all feature extraction tasks. For fascial achieved 92.1% accuracy, expression 83.5%, search 9.2%. 91.6% 89.4%, 81%. pattern, 91.4% 80.8%, 24.7%. Skin analysis also while assessing with 93.3%, 84.9% respectively. Conclusions reliably used label contained within datasets, enabling use predictive modeling medicine.