Calibration of combination models. Calibration was calculated after 12 and 24 months using the calibrate function in the rms package of R for combination model 1 (A) and combination model 2 (B). Bootstrapping (1000 repeats) was performed to obtain bias-corrected estimates of predicted vs. observed values. Non-convergence reduced the number of included bootstrapping steps for combination model 2 to 981 or 990 after 12 or 24 months, respectively. "Predicted" survival probabilities at 12 or 24 months are those predicted by the Cox model, and "observed" refers to the corresponding Kaplan-Meier survival estimate at the given time-point. Mean absolute error in predictions, the mean squared error, and the 0.9 quantile of the absolute error is reported. "Error" refers to the difference between the predicted values and the corresponding bias-corrected calibrated values. Mean error was <3% for both combination models and both time-points. The calibration according to Kolmogorov Smirnov was excellent for combination model 1 and satisfactory for model 2 (p=0.657 and p=0.021, respectively).
ARTICLE ABSTRACT
Purpose: To identify baseline peripheral blood biomarkers associated with clinical outcome following ipilimumab treatment in advanced melanoma patients.Experimental Design: Frequencies of myeloid-derived suppressor cells (MDSC) and regulatory T cells (Treg), serum lactate dehydrogenase (LDH), routine blood counts, and clinical characteristics were assessed in 209 patients. Endpoints were overall survival (OS) and best overall response. Statistical calculations were done by Kaplan–Meier and Cox regression analysis, including calibration and discrimination by C-statistics.Results: Low baseline LDH, absolute monocyte counts (AMC), Lin−CD14+HLA-DR−/low-MDSC frequencies, and high absolute eosinophil counts (AEC), relative lymphocyte counts (RLC), and CD4+CD25+FoxP3+-Treg frequencies were significantly associated with better survival, and were considered in a combination model. Patients (43.5%) presenting with the best biomarker signature had a 30% response rate and median survival of 16 months. In contrast, patients with the worst biomarkers (27.5%) had only a 3% response rate and median survival of 4 months. The occurrence of adverse events correlated with neither baseline biomarker signatures nor the clinical benefit of ipilimumab. In another model, limited to the routine parameters LDH, AMC, AEC, and RLC, the number of favorable factors (4 vs. 3 vs. 2–0) was also associated with OS (P < 0.001 for all pairwise comparisons) in the main study and additionally in an independent validation cohort.Conclusions: A baseline signature of low LDH, AMC, and MDSCs as well as high AEC, Tregs, and RLC is associated with favorable outcome following ipilimumab. Prospective investigation of the predictive impact of these markers following ipilimumab and other treatments, e.g., PD-1 antibodies, is warranted. Clin Cancer Res; 22(12); 2908–18. ©2016 AACR.
