Abstract PR622

Abstract

Background & Objectives: A patient’s health status often determines whether surgery is recommended for cancer therapy. The risk of complications may outweigh the perceived benefit. Consequently, measures that predict surgical outcomes may be helpful in determining whether or not to operate. Materials & Methods: We retrospectively apply four mortality predictors to 62,763 adult surgical cancer patients from the University of Texas MD Anderson Cancer Center during January 2007 - March 2014. We use the first surgery for each patient that is over 60 minutes in duration and compare the following indexes: Charlson Comorbidity Index1 as implemented by Deyo et al.2; Dalton’s Risk Quantification Index (RQI)3; Sessler’s Risk Stratification Index (RSI)4 and the Surgical Apgar Score5. We implement each metric as described by the authors and evaluate each score’s ability to predict 30-day mortality using logistic regression, randomly selecting two-thirds of the cases for learning the model and using the remaining cases for evaluation. We report the C-Statistic for each metric in isolation and in combination with the other measures. Results: We obtain individual C-Statistics from 0.639 to 0.784 and a combined C-Statistic of 0.817. The least predictive index is the RQI and the most predictive is the RSI.Conclusion: Despite not being designed specifically for cancer patients, or even necessarily for 30-day mortality, these metrics provide accurate predictions for our patient body. Furthermore, by combining the scores together, we are able to improve predictions, as each index is based on a different set of patient characteristics. RQI was the least useful metric, contributing only 0.01 to the combined C-Statistic. These indexes can be used to calculate and communicate the likelihood of 30-day mortality and may aid in clinical decision-making.