77. Development of predictive models for 90-day and 1-year mortality in spinal metastatic disease

Abstract

BACKGROUND CONTEXT Increasing prevalence of metastatic disease has been accompanied by increasing rates of surgical intervention. Current tools have poor to fair predictive performance for intermediate (90-day) and long-term (one-year) mortality. PURPOSE The purpose of this study was to develop predictive algorithms for spinal metastatic disease at these time points and to provide patient-specific explanations of the predictions generated by these algorithms. STUDY DESIGN/SETTING Retrospective review was conducted at two large academic medical centers. PATIENT SAMPLE Patients undergoing initial operative management for spinal metastatic disease between January 2000 and December 2016. OUTCOME MEASURES Ninety-day and one-year overall survival. METHODS Five models (penalized logistic regression, random forest, stochastic gradient boosting, neural network, and support vector machine) were developed to predict ninety-day and one-year mortality. RESULTS Overall, 732 patients were identified with ninety-day and one-year mortality rates of 181 (25.1%) and 385 (54.3%), respectively. The stochastic gradient boosting algorithm had the best performance for 90-day mortality and one-year mortality. On global variable importance assessment, albumin, primary tumor histology, and performance status were the three most important predictors of 90-day mortality. The final models were incorporated into an open access web application able to provide predictions as well as patient-specific explanations of the results generated by the algorithms. The application can be found at: https://sorg-apps.shinyapps.io/spinemetssurvival/ CONCLUSIONS Preoperative estimation of 90-day and one-year mortality was achieved with assessment of more flexible modeling techniques such as machine learning. Integration of these models into applications and patient-centered explanations of predictions represent opportunities for incorporation into health care systems as decision tools in the future. FDA DEVICE/DRUG STATUS This abstract does not discuss or include any applicable devices or drugs.