Machine learning predictive model for lumbar disc reherniation following microsurgical discectomy.

Abstract

Introduction: The integration of machine learning (ML) algorithms into the field of neurosurgery has the potential to facilitate the decision-making process for the surgeons, improve the surgical outcomes and the overall patient satisfaction rates. Reoperations for same level lumbar disc reherniation are associated with poorer outcomes and greater rate of complications.

Research question: Proper preoperative patient evaluation could reveal the individuals at higher risk of reherniation. A novel machine learning algorithm was used for the creation of a predictive scoring system for lumbar disc reherniation for patients requiring microdiscectomy without fusion.

Material and methods: Retrospective chart review was completed of all adult patients that underwent microdiscectomy without fusion for symptomatic single level LDH, in a single center, over the last 3 years. 230 patients met the inclusion criteria. 19 of them required a second surgical intervention due to same level reherniation.

Results: Utilizing the Risk-SLIM model, the Lumbar Reherniation Score (LRS) was created. The score's accuracy was tested against other model architectures, and a standard five-fold cross-validation was performed. The LRS has AUC of 0.87, confusion matrix accuracy of 0.74, Matthews correlation coefficient of 0.36 and informedness of 0.62. The LRS individual reherniation risk probability ranges from 0% to 88.1%.

Discussion and conclusion: The LRS is a novel, easy-to-use, patient-specific tool for preoperative prediction of the individual patient-specific risk of same level symptomatic reherniation following microdiscectomy. Further validation and testing of the model is needed before it can be used in real-life patient treatment.