Validation of Clinical Prediction Models and Study of the Role of Modern Therapies Using Real-World Data in Spine Oncology

Abstract

With a rising incidence of cancer and improved life expectancy, more patients are being diagnosed and surviving longer with spinal metastases. Spinal metastases can significantly worsen patients' quality of life because they can cause pain, vertebral compression fractures (VCF), and malignant epidural spinal cord compression, leading to loss of motor and sensory functions and fecal and/or urinary incontinence. There has been increasing interest in prediction modeling to determine patients' prognosis with spine metastases. Because the landscape of local and systemic therapy is rapidly changing, it is vital to understand how the risk captured by prediction models relates to treatment. In this thesis, we leverage machine learning methods to predict the prognosis of patients selected for surgery to treat spinal metastases. Also, we emulated a target trial using real-world data to determine the average treatment effect of receiving systemic therapy after spine surgery. Lastly, we discussed the role of machine learning and causal inference to predict prognosis when systemic therapy is considered after spine surgery.