Machine Learning in Adult Cervical Deformity Surgery
A retrospective subgroup analysis utilizing machine learning algorithms developed models for preoperative predictors of clinical improvement after decompression surgery in patients with mild degenerative cervical myelopathy.
Tyler K. Williamson, MS, and Peter G. Passias, MD
Khan O, Badhiwala JH, Witiw CD, Wilson JR, Fehlings MG. Machine learning algorithms for prediction of health-related quality-of-life after surgery for mild degenerative cervical myelopathy. Spine J. 2021 Oct;21(10):1659-1669. doi: 10.1016/j.spinee.2020.02.003. Epub 2020 Feb 8. PMID: 32045708.
Patient-reported outcomes are used throughout orthopaedic surgery to assess quality of life preoperatively and improvements in health and functionality postoperatively. In many areas of orthopaedic surgery, preoperative factors correlating with postoperative improvements have been identified, changing the focus of management towards these impactful predictors.
Researchers have had less success with developing similar predictors for adult cervical deformity. Khan et al sought to rectify this by conducting a retrospective subgroup analysis of adult patients who had undergone decompression surgery for mild degenerative cervical myelopathy and who were participating in the AOSpine cervical spondylotic myelopathy clinical trial.  After identifying inclusion criteria, a total of 173 patients from the AOSpine trial were available for analysis.
The study’s primary outcome was development of machine learning (ML) algorithms incorporating preoperative variables that could predict which patients would benefit most from decompression surgery. The researchers used data from the patients’ Short Form-36 (SF-36) Physical Component Score (PCS) and Mental Component Score (MCS) to develop the algorithms.
Utilizing 7 ML techniques, the researchers identified useful predictive models for health-related quality-of-life measurements. The diagnostic ability of the model to predict whether a patient will improve in MCS at 1 year after decompression surgery generated an area-under-the-curve (AUC) of 0.77, while the model for PCS generated an AUC of 0.78.
The study findings demonstrate the utility of ML algorithms to see “hidden” connections between preoperative variables and postoperative outputs. Although baseline MCS and PCS would intuitively be predictive of their postoperative scores, the generalized boosted model (GBM) and earth algorithm established associations with other characteristics, such as the gender difference in outcomes, along with use of corpectomies, baseline spasticity, and involvement of C3 in the operation. Previous literature has also correlated the female gender with poor myelopathy and disability improvement after decompression surgery, and this study concluded the association between the mental health component and the female gender as well.
Degenerative cervical myelopathy is a major source of neurologic dysfunction in the adult population, deemed the leading cause of myelopathy worldwide. Prior literature has tried, unsuccessfully, to answer these questions: Should we be operating on patients with mild myelopathy? And if so, who would benefit the most?
In this study, Khan et al showcased a simple application of ML algorithms that can be used throughout orthopaedic surgery to determine helpful predictors of successful outcomes. These innovative techniques also alleviate the time and energy needed, providing a logistic convenience. Tasks that may take a human researcher hour upon hour to perform are almost instantaneously computed by these algorithms. This process not only offers a better solution for deriving predictive models than repeated trial-and-error using traditional methods, but it also minimizes a vast amount of human error by automating tasks.
Some of the predictors identified in this study have not been evidenced by previous literature, but that is an important component of ML algorithms: They can capture new, complex associations that were previously unidentified by traditional methods. It may be unclear how these predictors relate to the outcome in question. However, continued utilization and calibration of ML algorithms, in spine surgery in particular and orthopaedic surgery in general, will help bolster our understanding of the factors correlated with successful outcomes.
Specifically for degenerative cervical myelopathy, the authors were able to pinpoint predictors of improvement in both physical and mental health components following cervical decompression surgery. This understanding will help us better fit surgical planning and clinical management to each of our patients, in addition to possibly optimizing modifiable risk factors to increase the likelihood of a successful outcome.
Tyler K. Williamson, MS, is an orthopaedic surgery clinical research fellow, NYU Langone Medical Center – Langone Orthopedic Hospital, New York, New York. Peter G. Passias, MD, is a Clinical Associate Professor of Orthopaedic Surgery, Division of Spine Surgery, Department of Orthopaedic Surgery, at NYU Langone Medical Center – Langone Orthopedic Hospital, New York, New York.
Disclosures: The authors have no disclosures relevant to this article.
- AOSpine North America Research Network. Assessment of Surgical Techniques for Treating Cervical Spondylotic Myelopathy (CSM). ClinicalTrials.gov Identifier: NCT00285337. Available at https://clinicaltrials.gov/ct2/show/NCT00285337. Accessed December 6, 2021.