Identifying Acute Low Back Pain Episodes in Primary Care Practice From Clinical Notes: Observational Study

Abstract

Background Acute and chronic low back pain (LBP) are different conditions with different treatments. However, they are coded in electronic health records with the same International Classification of Diseases, 10th revision (ICD-10) code (M54.5) and can be differentiated only by retrospective chart reviews. This prevents an efficient definition of data-driven guidelines for billing and therapy recommendations, such as return-to-work options. Objective The objective of this study was to evaluate the feasibility of automatically distinguishing acute LBP episodes by analyzing free-text clinical notes. Methods We used a dataset of 17,409 clinical notes from different primary care practices; of these, 891 documents were manually annotated as acute LBP and 2973 were generally associated with LBP via the recorded ICD-10 code. We compared different supervised and unsupervised strategies for automated identification: keyword search, topic modeling, logistic regression with bag of n-grams and manual features, and deep learning (a convolutional neural network-based architecture [ConvNet]). We trained the supervised models using either manual annotations or ICD-10 codes as positive labels. Results ConvNet trained using manual annotations obtained the best results with an area under the receiver operating characteristic curve of 0.98 and an F score of 0.70. ConvNet’s results were also robust to reduction of the number of manually annotated documents. In the absence of manual annotations, topic models performed better than methods trained using ICD-10 codes, which were unsatisfactory for identifying LBP acuity. Conclusions This study uses clinical notes to delineate a potential path toward systematic learning of therapeutic strategies, billing guidelines, and management options for acute LBP at the point of care.

Natural language processing in low back pain and spine diseases: A systematic review

“…Thus, different conditions lead to different treatment recommendations, leading to different costs to the healthcare systems. Miotto et al [ 20 ] faced this task.…”

“…To identify (and then remove) negated occurrences, authors usually exploits algorithms such as NegEx [ 39 ]. This approach was implemented in [ 20 ] to identify acuity in LBP, and in [ 28 ] to identify Type 1 Modic changes, while in [ 18 , 25 , 26 ] to identify several findings related to LBP and stenosis from MRI and/or x-ray reports.…”

Improving appropriate imaging for non-specific low back pain

“…Accurate documentation with associated ICD-10 diagnoses is essential for correct claims data to determine appropriate utilisation of imaging for LBP. 36 Furthermore, ongoing patient education is necessary to minimise inappropriate imaging without affecting quality of care, safety or patient satisfaction. As others have found, our focus group discussants shared expectations for imaging to diagnose the cause of their LBP, even when it was not indicated.…”

Modern Clinical Text Mining: A Guide and Review

“…One of their key advantages is a reduced need for feature engineering; representations of words, phrases, and higher-order text structures can be learned as part of the overall training process or incorporated via transfer learning from other pre-trained models. Several studies have deployed convolutional neural networks (CNNs) with high success on a variety of clinical text classification tasks: assigning diagnosis codes (104,105), classifying radiology reports (19,106), subtyping diseases (91), and determining the presence or absence of comorbidities (107). Alternative neural network architectures, such as LSTMs and attention networks, are commonly used in text classification tasks in the general NLP domain, although as of this writing, CNNs have been the dominant architecture in clinical text classification (23,108).…”