Abstract: Background: An accurate tool with good discrimination for bleeding would be useful to clinicians for improved management of all their patients. Bleeding risk models have been published but not externally validated in independent clinical dataset. We chose the NCDR PCI score to validate within a large, multi-site community datasets. The aim of the study was to determine the diagnostic utility of this bleeding risk score tool.

Methods: This is a large-scale retrospective analysis utilizing American College of Cardiology data from a 37-hospital health system. The central repository of PCI procedures between 6-1-2009 and 6-30-2012 was utilized to validate the NCDR PCI bleeding risk score (BRS) among 4693 patients. The primary endpoint was major bleeding. Discriminant analysis calculating the receiver operating characteristic curve was performed.

Results:There were 143 (3.0%) major bleeds. Mean bleeding risk score was 14.7 (range 3 – 42). Incidence of bleeding by risk category: low (0.5%), intermediate (1.7%), and high risk (7.6%). Patients given heparin had 113 (3.7%) major bleeds and those given bivalirudin had 30 (2.1%) major bleeds. Tool accuracy was poor to fair (AUC 0.78 heparin, 0.65 bivalirudin). Overall accuracy was 0.71 (CI: 0.66-0.76). Accuracy did not improve when confined to just the intermediate risk group (AUC 0.58; CI: 0.55-0.67).

Conclusion:Bleeding risk tools have low predictive value. Adjustment for anticoagulation use resulted in poor discrimination because bivalirudin differentially biases outcomes toward no bleeding. The current state of bleeding risk tools provides little support for diagnostic utility in regards to major bleeding and therefore have limited clinical applicability.

Keywords: Major bleeding, bleeding risk model, anticoagulant, percutaneous coronary intervention, cardiovascular.

Abstract: This paper discusses the use of PP-plots for survival distributions where for a pair of survival distributions, one is plotted against the other. This is another way of visualizing the nature of the relationship between the two survival distributions along with typical Kaplan-Meier plots. For three survival distributions, the PPP-plot is introduced where the survival distributions are plotted against each other in three-dimensions. At the population level, measures of divergence between distributions are introduced based on areas and lengths associated with the PP- and PPP- plots. At the sample level, two test statistics are defined, based on these areas and lengths, to test the null hypothesis of equivalent survival curves. A simulation exercise showed that, overall, the new tests are worthy competitors to the log-rank and Wilcoxon tests and also to a Levine-type test and a Kolmogorov-Smirnov type test for the case of crossing survival curves. The paper also shows how the PP-plot can be used to estimate the hazard ratio and to assess the ratio of hazard functions if proportional hazards are not appropriate. Finally, the methods introduced are illustrated on two cancer data sets.

Keywords: Crossing survival curves, Hazard ratio, Kaplan-Meier, Log-rank test, PP-plot, Wilcoxon test.