Medical Student University of Michigan Medical School Ypsilanti, Michigan, United States
Introduction: Motor evoked potentials (MEPs) give real-time feedback on the integrity of the corticospinal tract. This modality is invaluable during intradural tumor resection by informing intraoperative decisions regarding extent of resection. Preoperative knowledge of the reliability of intraoperative baseline MEPs permits accurate risk stratification of patients and surgical planning. We aim to characterize preoperative risk factors for unmonitorable baseline MEPs and associated clinical outcomes. We also produced a machine learning based risk stratification tool.
Methods: We conducted a retrospective review of 72 patients who underwent resection of intradural spinal tumors at a single tertiary medical center. Variables of interest included pertinent medical history, neurologic exam, tumor histology, degree of canal tumor infiltration, intraoperative outcomes, and 1-month neurologic function. Various machine learning models were then trained and compared on their ability to predict baseline MEP monitorability.
Results: Patients with unmonitorable baseline MEPs were significantly more likely to have preoperative lower-extremity motor deficits (68% vs. 37%, p = 0.01), peripheral nerve disease (38% vs. 15%, p=0.02), and higher canal tumor infiltration (66% vs. 54%, p< 0.01). Histologic distribution was similar except for a higher rate of schwannomas in the monitorable group (p < 0.01). Unmonitorable patients had significantly higher EBL (385 vs. 201cc, p =0.02) and rates of worsened/new lower-motor deficits (27% vs. 9%, p< 0.05). Logistic regression outperformed Random Forest and Support Vector Machine predictions of MEP monitorability with 70% sensitivity and 75% specificity, emphasizing pre-op lower motor deficits (3.3x higher odds) and peripheral nerve disease (2.1x higher odds).
Conclusion : Preoperative motor deficits, peripheral nerve disease, and tumor infiltration within the canal were associated with unmonitorable baseline MEPs. Furthermore, unmonitorable MEPs during intradural tumor resection was associated with higher intraoperative bleeding and postoperative motor deficits. The machine learning model showed promise as a preoperative risk stratification tool by predicting the reliability of intraoperative baseline MEPs.