Abstract P4: Does the Expected Reduction in Restenosis from Drug-Eluting Stents Predict Actual Use? An Assessment of Treatment-Risk Paradox Using the C-Statistic

Abstract

Background: Beneficial treatments are often least prescribed to patients at highest risk for poor outcomes. Quantitative assessments of “treatment-risk paradox” have relied on comparisons across risk strata defined by single variables or arbitrary categories of predicted risk. We examined whether DES were preferentially used in patients at highest risk for target vessel revascularization (TVR) after PCI using a novel quantitative method. Methods: We used a validated model to predict TVR at 1 year for all PCI hospitalizations in MA between 10/04 and 9/07. For each patient, we calculated predicted absolute risk reduction (ARR) in TVR assuming treatment with DES (vs. BMS). Using logistic regression, we assessed the association between predicted ARR and observed stent treatment for the entire population and each of 21 hospitals, and calculated the c-statistics for each as a measure of the extent to which predicted benefit discriminated observed DES use. Results: A total of 21933 (80.9%) PCIs with DES and 5174 (19.1%) PCIs with BMS were included. The predicted ARR in TVR associated with PCI ranged from 1.2% to 15.9%. Overall, predicted benefit from DES poorly discriminated receipt of DES (c-statistic 0.545, 95% CI 0.537 - 0.554). Of 22 hospitals, 15 had c-statistics with CIs that crossed 0.5, implying no relationship between predicted benefit of DES and stent selection. (Figure) Conclusions: We observed a minimal relationship between predicted benefit from DES and actual DES use. The c-statistic and ROC curve may be a useful metrics to assess the relationship between predicted benefit and observed receipt of therapies, and identify potential areas for quality improvement.