Building Large-Scale Registries from Unstructured Clinical Notes using a Low-Resource Natural Language Processing Pipeline

Abstract

Building clinical registries is an important step in improving the quality and safety of patient care. With the growing size of medical records, manual abstraction becomes more and more infeasible and impractical. On the other hand, Natural Language Processing Techniques have shown promising results in extracting valuable information from unstructured clinical notes. However, the structure and nature of clinical notes are very different from regular text that state-of-the- art NLP models are trained and tested on and they have their own set of challenges. In this study, we propose SE-K, an efficient and interpretable classification approach for extracting information from clinical notes, and show that it outperforms current state-of-the-art models in text classification. We use this approach to generate a 20- year comprehensive registry of anterior cruciate ligament reconstruction operations, one of the most common orthopedics operations among children and young adults. This registry can help us better understand the outcomes of this surgery and identify potential areas for improvement which can ultimately lead to better treatment outcomes.

Prevalence and Predictors of Concomitant Meniscal and Ligamentous Injuries Associated With ACL Surgery: An Analysis of 20 Years of ACL Reconstruction at a Tertiary Care Children’s Hospital

“…After institutional review board approval, a custom natural language processing (NLP) pipeline 47 was used to identify all patients treated surgically at a single institution for an ACL tear between January 1, 2000, and December 31, 2020, and extract pertinent clinical and operative data from the available unstructured electronic clinical notes to develop an institutional ACL surgery registry. Details on the NLP pipeline used along with model performance metrics and validation have been previously published.…”

“…Details on the NLP pipeline used along with model performance metrics and validation have been previously published. 47,48 Briefly, we developed an NLP model to identify ACL surgery cases from operative notes (accuracy, 1.00; sensitivity, 0.99; specificity, 1.00). 48 We then extracted relevant injury and surgical details from the unstructured clinical notes of the identified cases using NLP (accuracy, 0.98 ± 0.01; sensitivity, 0.97 ± 0.03; specificity, 0.98 ± 0.02).…”

“…48 We then extracted relevant injury and surgical details from the unstructured clinical notes of the identified cases using NLP (accuracy, 0.98 6 0.01; sensitivity, 0.97 6 0.03; specificity, 0.98 6 0.02). 47 The models were trained and tested on ground-truth labels extracted by a human examiner with adequate medical knowledge on ACL injuries on a subset of patients. During manual review, positive cases were labeled when the notes detailed an acute meniscal tear or an acute sprain or tear of the medial collateral ligament (MCL), PLC, or PCL.…”

Prevalence and Predictors of Concomitant Meniscal Surgery During Pediatric and Adolescent ACL Reconstruction: Analysis of 4729 Patients Over 20 Years at a Tertiary-Care Regional Children's Hospital

“…After obtaining institutional review board approval, we used a custom natural language processing (NLP) pipeline 37 to identify all patients who underwent ACLR at our institution between January 1, 2000, and December 31, 2020, and we extracted pertinent clinical and operative data from the available unstructured electronic clinical notes to develop an institutional ACLR registry. Details of the NLP pipeline used along with model performance metrics and validation have been previously published.…”

“…38 We then extracted relevant injury and surgical details from the operative notes of the identified patients using NLP (accuracy, 0.98 6 0.01; sensitivity, 0.97 6 0.03; specificity, 0.98 6 0.02). 37 Age, sex, height, weight, BMI, race, and insurance status were retrieved from the structured medical record. Mechanism of injury (ie, contact vs noncontact), ipsilateral ACL injury history, and sports participation were extracted from preoperative notes using NLP.…”