Natural Language Processing Useful for Surveillance of Infections After Spine Surgery
- In this proof-of-concept study, a natural language processing (NLP) algorithm based on machine learning was examined for its utility in automated surveillance of wound infections requiring reoperation over the 90 days following lumbar discectomy
- The NLP was more accurate in identifying patients requiring reoperation than conventional methods such as administrative billing and procedure codes
- NLP could be used to improve quality and safety efforts by automating surveillance of postoperative complications.
Subscribe to the latest updates from Orthopaedics Advances in Motion
Surgical site infections are a major driver of morbidity and cost after spine surgery, and value-based payment models and risk-based reimbursement make providers liable for these and other complications. Currently, surveillance for adverse events depends on labor-intensive methods—manual chart review, dedicated prospective clinical documentation or retrospective queries based on procedural and diagnosis codes.
Natural language processing (NLP), a form of computer-assisted abstraction, is a potential alternative. This methodology makes use of a computer algorithm to filter through the unstructured narrative portion of electronic health records and learn patterns that match the outcome of interest, such as postoperative infection.
In a proof-of-concept study, Aditya V. Karhade, MD, MBA, orthopaedic surgery resident, Joseph H. Schwab MD, chief of the Orthopaedic Spine Center at Massachusetts General Hospital, and colleagues demonstrated that an NLP algorithm was useful for automated surveillance of surgical-site infections after lumbar discectomy. They published their findings in The Spine Journal.
Using a spine surgery registry for two academic medical centers and three community medical centers, the researchers identified 5,860 adults who underwent lumbar discectomy for lumbar disc herniation between January 1, 2000, and July 31, 2019. Records of 4,483 patients who had surgery before December 31, 2015, were used for algorithm training. Then 1,377 patients who had surgery after January 1, 2016, were used for algorithm testing.
A supervised machine learning method was used to develop the NLP algorithm—meaning that the NLP algorithm identified patterns from the training set without being explicitly programmed.
Of all 5,860 patients, 62 (1.1%), including 16 of the patients in the testing set, required reoperation for surgical-site infection within 90 days of surgery.
At a threshold of 0.05, the performance of the NLP algorithm was:
- Sensitivity 94% (detected records for 15 of 16 patients)
- Specificity 99.8%
- Positive predictive value (PPV) 83%
- Negative predictive value (NPV) 99.9%
When current procedural terminology and international classification of disease codes were used, 12 of the 16 patients were detected (sensitivity of 75%). The NPV was 99.7% and the specificity and PPV were 100%.
These results show that machine learning–based NLP is promising for automated surveillance of adverse events after spine surgery. The same approach could be adapted to other postoperative outcomes and procedures.
The study authors have similarly developed NLP algorithms for automated detection of intraoperative complications in spine surgery. These algorithms have been validated by independent external institutions, as published in The Journal of Neurosurgery.
With further development, NLP algorithms should be useful for hospital quality and safety reporting, automated administrative coding, national surgical registries and clinical outcomes research.
view original journal article Subscription may be required
Learn more about the Orthopaedic Spine Center at Mass General
Refer a patient to the Department of Orthopaedics at Mass General