Related Experiment Video
Updated: Oct 25, 2026
Diffusion Tensor Magnetic Resonance Imaging in Chronic Spinal Cord Compression
Published on: May 7, 2019
Neurologic Recovery After Metastatic Spine Tumor Surgery for Spinal Cord Compression: Can Unsupervised Machine
Clayton R Baker1, Harsh Jain2, Nick De Oliveira1
1Vanderbilt University School of Medicine, Nashville, TN.
Study Design:
Retrospective Cohort Study.
Objective:
To use unsupervised machine learning to identify preoperative phenotypes among patients undergoing surgery for metastatic epidural spinal cord compression (MESCC) and evaluate whether these phenotypes stratify postoperative neurologic recovery.
Summary Of Background Data:
MESCC frequently presents with neurologic deficits, but recovery after separation surgery is variable. Most prior work has evaluated individual predictors in isolation.
Methods:
Adults undergoing surgery (2010-21) for MESCC with preoperative motor deficits were included. Latent profile analysis clustered patients on preoperative upper and lower motor scores, deficit duration, and sensory deficits. The primary outcome was postoperative motor improvement, reported as any improvement and the proportion achieving ≥25%, ≥50%, ≥75% and 100% recovery and absolute motor change. Groups were compared with ANOVA and chi-square tests, and with multivariable models adjusted for age, sex, and tumor location.
Results:
Among 101 patients (mean age: 60.7±10.9 years, 61.4% males), 72 (71.3%) patients had any postoperative improvement. Four clinically relevant phenotypes were identified: - Grade-1: acute, mild deficit without sensory loss, n=36. - Grade-2: acute, mild deficit with sensory loss, n=26. - Grade-3: chronic, mild deficit without sensory loss, n=20. Grade-4: chronic, severe deficit with sensory loss, n=19. Rates of any postoperative improvement were similar between Grades (P=0.479), as was percent of motor recovery (P=0.109). Grades differed in absolute motor change (P=0.028) and rates of patients achieving ≥50% (P=0.004), ≥75% (P=0.014), and 100% (P=0.004) recovery. Rates of ≥75% and 100% recovery were lowest in Grade-3 (30.0%, 30.0%) and Grade-4 (21.1%,5.3%), and Grade-4 independently predicted lower recovery after adjustment. Mortality did not differ between Grades.
Conclusion:
Unsupervised clustering identified four preoperative phenotypes of MESCC patients that stratified clinically meaningful motor recovery and summarized neurologic presentation. This framework supports tailored, multidisciplinary counseling.
Level Of Evidence:
3.