Meniscus Tear Morphology and Patient Demographics as Predictors of Treatment for Meniscal Tears: A Natural Language Processing Study

TL;DR

This study uses big data and natural language processing to find patterns in MRI reports that predict treatment choices for meniscus tears, showing that tear type and patient demographics influence decisions.

Contribution

The novel use of NLP on MRI reports to identify predictors of treatment decisions for meniscus tears in a large patient cohort.

Findings

Bucket-handle tears are the strongest predictor for surgical treatment.

Opioid prescriptions were given to 25.6% of patients with meniscus tears.

Demographics and degenerative changes independently predict treatment type.

Abstract

Meniscus tears occur in tandem with other degenerative changes at the knee joint, making interpretation of imaging findings challenging. Meniscus tears, commonly treated by general physicians and musculoskeletal specialists, have significant treatment variability.

The purpose of this study was to investigate whether big data analytical tools and natural language processing (NLP) of magnetic resonance imaging reports could identify factors that are predictive of treatment decisions in a large cohort of patients. It was hypothesized that surgical treatment would be associated with specific meniscus tear patterns.

Cross-sectional study; Level of evidence, 3.

Deidentified electronic health records from approximately 5 million patients, available at Information Commons, were analyzed to identify patients with meniscus tears on knee MRI reports. NLP was used to extract descriptive features from MRI reports. Demographics and clinical and treatment data were recorded. Univariate and multivariate analyses were performed.

For 4013 patients with meniscus tears on MRI, physical therapy was the most common treatment (62.4% of patients). Opioid medications were prescribed to 25.6% of patients diagnosed with a meniscus tear. Surgery was performed for 19.6% of patients. A bucket-handle tear was the strongest predictor for receiving surgery. Meniscus tear patterns, the presence of degenerative changes, and demographic factors were independent predictors of treatment received.

Meniscus tear types and other findings at the knee joint, along with demographic variables, are associated with treatment for meniscus tears. Bucket-handle and root tears were most likely to undergo surgical treatment, while patients on Medicare and those with arthrosis were more likely to be treated with nonoperative modalities. NLP of big data, including radiology findings, may have a role in providing clinical decision support for interpreting findings on knee MRIs. Prescription of opioid medications in meniscal tears was observed at a higher-than-expected rate.

Figures