Oral Presentation 51st International Society for the Study of the Lumbar Spine Annual Meeting 2025

Machine learning-driven clinical and imaging clustering of degenerative lumbar spondylolisthesis:  Implications for stratified surgical care (116215)

Karlo Pedro 1 2 , Aazad Abbas 2 , Christopher Bailey 3 , Jennifer Urquhart 4 , Gregory McIntosh 5 , Raymond Andrew Glennie 6 , Charles Fisher 7 , Phillipe Phan 8 , Y. Raja Rampersaud 9 10
  1. Institute of Medical Sciences, University of Toronto, Toronto, Ontario, Canada
  2. Department of Surgery, University of Toronto, Toronto, Ontario, Canada
  3. Combined Orthopaedic and Neurosurgical Spine Program, London Health Sciences, London, Ontario, Canada
  4. Combined Orthopaedic and Neurosurgical Spine Program, London Health Sciences, London, Ontario, Canada
  5. Canadian Spine Outcomes and Research Network, Toronto, Ontario, Canada
  6. Department of Surgery, Dalhousie University, Halifax, Nova Scotia, Canada
  7. Combined Neurosurgical and Orthopedic Spine Program, University of British Columbia, Vancouver, British Columbia, Canada
  8. Division of Orthopaedic Surgery, The Ottawa Hospital, Ottawa, Ontario, Canada
  9. University Health Network ,Toronto Western Hospital, Schroeder Arthritis Institute, Toronto, ONTARIO, Canada
  10. Schroeder Arthritis Institute, University Health Network, Toronto, Ontario, Canada

Introduction

 

Recent randomized trails for the surgical management of degenerative lumbar spondylolisthesis (DLS) report non-inferiority of decompression alone. Yet decompression and fusion remains the most common procedure for DLS patients. Previous work from the Canadian Spine Outcomes and Research Network (CSORN) including a modified Delphi process, suggest that the significant clinical and radiographic heterogeneity demonstrated by DLS patient creates substantial variability in surgeon reported decision-making. In this study, we applied unsupervised clustering techniques to identify distinct DLS groups, aiming to inform clinical decision-making in selecting the right patients for decompression or decompression and fusion.

 

Methodology

 

We performed a sub-analysis of an ongoing prospective observational CSORN DLS study from 2015 to 2022 at eight Canadian sites. The CSORN-DLS study aims to ultimately develop a consensus-based surgeon decision aid to guide surgical management of DLS in a stratified manner to optimize outcome and resource utilization (i.e. value). Consensus cluster analysis was used based on demographic, clinical, and radiological characteristics of patients enrolled in phase one (baseline current surgeon practice). We selected clinically relevant variables to guide feature selection and evaluated the optimal number of clusters. Euclidean distance metric was used to determine the cluster number and structure, with up to seven clusters considered and 50 sampling iterations. Key characteristics of each cluster were identified, and post-surgical outcomes were compared using the Oswestry Disability Index (ODI) with pairwise statistical tests.

 

Results

 

Consensus cluster analysis of 486 DLS patient (mean [SD] age, 55.4 [9.2] years, 180 men [37%]) identified four distinct clusters. Cluster 1 (n=175) included patients with moderate disability, low comorbidity, low depression, and radiographically balanced spine. Cluster 2 (n=85) comprised predominantly female patients with severe disability, high comorbidity, high depression, greatest slip % and imbalanced (mean SVA=76mm, PI-LL mismatch=23 degrees) spine. Cluster 3 (n=91) mainly consisted of male patients with moderate disability, low comorbidity, low depression, and imbalanced (mean SVA=57mm, PI-LL mismatch=18 degrees) spine; while Cluster 4 (n=135) included elderly patients with severe disability, high comorbidity, high depression, and balanced spine. Clusters 1 and 3 showed the most favorable final ODI outcomes at 1- and 2-years follow-up (Figure 1). Cluster 2 had the highest rate of fusion surgery (84%) and the greatest ODI change at 1 year (DODI mean [SD] 27.97 (19.32)). Decompression alone provided sustained benefit in Cluster 1 and 3, while improvement significantly regressed in Cluster 2 and to a lesser degree cluster 4 at two years (Figure 1b).

 

Discussion

 

Unsupervised machine learning identified four clinically distinct clusters of DLS patients with varying outcome and responses to surgical interventions, highlighting the importance of stratified care strategies to optimize patient outcomes. Furthermore, our results confirm the clinical and radiographic heterogeneity of the surgical DLS population and inform the need for stratified recruitment strategies in future trial design for the evaluation of surgical management of DLS.

 

674d00f5ef5c3-Screen+Shot+2024-12-01+at+7.06.59+PM.png