Utilizing Machine Learning Models to Guide and Optimize Clinical Management of Cancer-Associated Compression Fractures

Introduction: Spinal metastases (SM) occur in an astounding 60-70% of patients with systemic cancer and can cause pain and instability. Treatment of SM involves an increasing role for minimally-invasive procedures such as percutaneous kyphoplasty to allow for faster initiation of chemoradiation. Current clinical tools, like the Spinal Instability Neoplastic Score (SINS), are limited in combining the complex patient factors that determine the optimal treatment approach.

Methods: This study retrospectively evaluates a prospective database of 768 SM patients treated with kyphoplasty at a single Comprehensive Cancer Center from 2009-2020. Patient demographics, tumor type, pain score, KPS and radiographic parameters were collected. Patients with poor response to kyphoplasty were considered those with non-improved pain scores at latest follow-up or those requiring open surgery at sites of prior kyphoplasty. Using machine-learning (ML) models, we further aim to evaluate patient predictors of treatment outcome. A 60:20:20 distribution will be used for training, validation, and testing, respectively. Predictive models with the highest performance will be indicated by the area under the receiver operating characteristic (AUROC).

Results: Multiple myeloma (254) and breast cancer (103) were the most commonly treated SM patients. The 23 prostate cancer patients showed a nonsignificant trend for improved outcome from kyphoplasty with 13% having non-improved pain and 0 requiring open surgery. The 36 patients with renal cell carcinoma showed a nonsignificant trend for worse outcome with approximately 20% having non-improved pain and nearly 20% requiring open surgery. ML models are being created to evaluate a combination of complex parameters that correlate with kyphoplasty outcome.

Conclusion : Developing superior assessment tools for SM requires detecting subtle relationships between numerous complex clinical and radiographic variables. This project leverages ML techniques to detect non-linear relationships and interactions between heterogenous variables that we aim to share with spine surgeons for improved SM management.