An Artificial Intelligence-Based Approach for Clustering of Patient Characteristics and Outcomes in Oncologic Surgery for Metastatic Tumors of The Spine

Introduction: Classification tools have been developed to stratify patients undergoing oncologic spine surgery based on perioperative risk and expected survival. An unsupervised machine learning approach may help identify patients at highest risk for perioperative morbidity or mortality. The objective of this study was to evaluate an artificial intelligence-based unsupervised machine learning approach to group patients’ characteristics and outcomes in oncologic spine surgery.

Methods: This was a retrospective cohort study. The records of 142 patients who underwent surgery for metastatic spinal tumors at our institution were reviewed for perioperative characteristics including demographics, comorbidities, oncologic history, and intraoperative details. A hierarchical clustering algorithm using Ward distances was employed and a dendrogram was generated. The largest vertical drop was used to define the number of clusters. Our primary outcome was overall survival (OS). Secondary outcomes included any complication, major complication, in-hospital mortality, ambulatory status, and length of stay postoperatively.

Results: Three distinct patient clusters were identified with significant differences in sex, race, chemotherapy history, pathologic fracture or instability on presentation, muscle mass, ambulatory status, primary tumor histology, tumor location, preoperative embolization status, surgical approach, and surgical extent. Clusters 1, 2, and 3 were found to be associated with significant differences in median OS (46 months, 27 months, 7 months; p=0.025) and in in-hospital mortality incidence (5%, 3%, 16%; p=0.037). No differences were found between clusters for any complication, major complication, ambulatory status, and length of stay.

Conclusion : Artificial intelligence-based prognostication models may be useful to uncover patterns and important covariates amongst patients undergoing surgery for spinal metastases who have poor postoperative outcomes.