Prediction of Complications in Radical Prostatectomy Prostate Cancer Patients: Simulated Annealing versus Co-Morbidity Indexes

Abstract

Background: The Deyo/Charlson co-morbidity index (CCI) and Klabunde co-morbidity index (KCI) co-morbidity indexes represent outdated indexes when the endpoint of complications after radical prostatectomy (RP) is considered. A novel group of co-morbidities derived from International Classification of Diseases-9 diagnostic codes in a contemporary RP database could provide better accuracy. Research Design, Subjects and Measures: We relied on 20,484 patients with clinically localized non-metastatic prostate cancer treated with RP between 2000 and 2009 in the Surveillance, Epidemiology, and End Results-Medicare linked database. We examined 2 endpoints, namely, 90-day medical complication rate and 90-day surgical complication rate after RP. Simulated annealing (SA) was used to develop a novel co-morbidity index. Finally, the newly identified groups of co-morbid conditions were compared with the CCI and Klabunde indexes. Results: Our SA identified 10 and 7 individual co-morbid conditions able to predict 90-day medical and surgical complications respectively. This novel model showed improved predictive accuracy over CCI and KCI for the 2 endpoints considered (respectively: 59.4 vs. 58.1 and 58.0% for medical complications, 58.0 vs. 56.8 and 56.7% for surgical complications). Conclusions: The newly defined groupings of co-morbid conditions resulted in better ability to predict the 2 endpoints of interest compared to CCI and KCI. However, the gain was marginal. This implies that better tools should be defined to more accurately predict these outcomes.