Andrew T Bates, Jie Yang, Maria Altieri, Mark Talamini, Aurora Pryor, Dana Telem. Stony Brook Medical Center
Introduction: Based on data previously published by our institution, this study aims to develop a long-term mortality probability calculator for bariatric patients.
Methods: Using SPARCS NY longitudinal administrative database, 7,862 adult patients who underwent bariatric surgery from 1999-2005 were analyzed. A multiple Cox proportional hazard model was used to build a prediction model for post procedure long-term mortality. The baseline survival function was estimated by exponentiating the negative empirical cumulative hazard function. Internal validation of the model was based on bootstrapping and C-index was calculated to evaluate discrimination accuracy of the model and shrinkage estimates were used for further calibration. External validation was based on SPARCS data from 2006-2010.
Results: The mean bariatric mortality rate was 2.5% with 8-14 years of follow-up. The long-term mortality risk stratification model included the following identified risk factors associated with an earlier time-to-death: age, male gender, Medicare/Medicaid insurance, congestive heart failure, rheumatoid arthritis, pulmonary circulation disorders and diabetes. The final model was predictive (c-index=0.74 for internal validation, c-index=0.70-0.73 for external validation years 2006-2010) for long term mortality. The hazard ratio prediction model was described below. The shrinkage used in calibration step was 0.879. The attached table details the c-index of the model and external validation results.
.jpg)
.jpg)
Conclusion: The model effectively predicts long-term mortality following bariatric surgery through both internal and external validation. A computerized application allowing easy input of patient specific information and output of individual long-term mortality risk as compared to the general and obese population is under development.