Artificial intelligence may predict progression of elective lumbar spine complaints to surgery

Use of an artificial neural network may accurately predict surgical progression of patients with elective lumbar spine complaints and improve the personalization of referrals to neuro and orthopedic spine surgeons, according to results presented at the American Association of Neurological Surgeons Annual Scientific Meeting.

“Machine learning could help address long wait-lists and potentially improve the personalization of referrals to neuro and orthopedic spine surgeons,” Nathan Xie, an MD candidate at the University of New South Wales, Sydney, Australia, told Healio.com/Orthopedics. “It is also clear that such technology is not only limited to this circumstance. It could help resolve many decision-making challenges in the surgical field.”

Winner of the Sanford J. Larson, MD, PhD, Award, Xie and colleagues identified 55 factors in the literature associated with surgical progression and reviewed the medical records of 326 patients who presented with an elective lumbar spine complaint between 2013 to 2018 at a single Australian tertiary hospital. Researchers constructed an artificial neural network with the outcome being progression to spinal surgery and compared it to a logistic regression model with the same data.

With 10 clinical and imaging predictive variables included as input, researchers found a 94.2% accuracy of surgical prediction with the artificial neural network. Results also showed the artificial neural network exhibited excellent discriminative ability with good fit of the data. When compared with the logistic regression model, the artificial neural network was found to be superior, according to results.

To ensure generalizability, Xie noted that these results need to be prospectively and externally validated. He added that the ultimate goal is to have an outcome-based referral system that can recommend particular treatments to patients based on their expected response to those treatments.

“Such a system is not and probably will not be ready for some time. In the interim, models like this one created by our study can help to bridge the gap between the current system and what we envision will be the future system,” Xie said. “Therefore, exploring how our model and the information provided by it can be integrated into an existing multidisciplinary triage pathway represents important further work.” – by Casey Tingle

Reference:

Xie N, et al. Abstract 227. Presented at: American Association of Neurological Surgeons Annual Scientific Meeting; April 13-17, 2019; San Diego.

Disclosure: Xie reports no relevant financial disclosures.