First, using an open-source NLP “engine,” the researchers developed NLP algorithms focused on 3 elements of >1,500 total hip arthroplasty (THA) procedures captured in the Mayo Total Joint Registry: (1) operative approach, (2) fixation technique, and (3) bearing surface. They then applied the algorithm to operative notes from THAs performed at Mayo and to THA-specific EHR data from outside facilities to determine external validity.
Relative to the current “gold-standard” of manual chart reviews, the algorithm had an accuracy of 99.2% in identifying the operative approach, 90.7% in identifying the fixation technique, and 95.8% in identifying the bearing surface. The researchers found similar accuracy rates when they applied the algorithm to external operative notes.
The findings from this study strongly suggest that properly “trained” NLP algorithms may someday eliminate the need for manual data extraction. That, in turn, could substantially streamline future research, policy, and surveillance tasks within orthopaedics. As Gwo-Chin Lee, MD predicts in his Commentary on this study, “When perfected, NLP will become the gold standard in the initial data mining of patient records for research, billing, and quality-improvement initiatives.” Dr. Lee is quick to add, however, that “no machine learning can occur…without the integral and indispensable input of the human element.”
Orthopaedic surgeons are already using robots to assist them in performing total joint arthroplasties. Wyles et al. show how we can use technology to reliably expedite research on that same subject. I believe the future holds much promise for the use of ever-advancing technologies in orthopaedic surgery and research.
Matthew R. Schmitz, MD
JBJS Deputy Editor for Social Media