Mann-Whitney-Wilcoxon results testing the robustness of the 7-gene signature to assignments
| Treatment (data set) . | Comparison . | Median HR* observed . | Median HR* null . | Mann-Whitney U . | P . |
|---|---|---|---|---|---|
| Bortezomib based (C) | Bortezomib standard vs bortezomib good (PFS) | 0.44 | 0.94 | 348 | 5.67 × 10−30 |
| RD (C) | Bortezomib standard vs bortezomib good (PFS) | 2.01 | 0.93 | 1526 | 1.91 × 10−17 |
| Bortezomib based or RD (C) | Incorrectly treated vs correctly treated (PFS) | 0.48 | 0.92 | 185 | 5.44 × 10−32 |
| Bortezomib based or RD (C) | Incorrectly treated vs correctly treated (OS) | 0.46 | 0.85 | 3 | 2.52 × 10−34 |
| Bortezomib alone (M) | Lenalidomide best vs bortezomib best (PFS) | 0.68 | 0.97 | 489 | 2.75 × 10−28 |
| Bortezomib alone (M) | Lenalidomide best vs bortezomib best (OS) | 0.53 | 1.00 | 108 | 5.67 × 10−33 |
| RD (PCL) | Bortezomib best vs lenalidomide best (PFS) | 0.24 | 0.83 | 1334 | 2.84 × 10−19 |
| RD (PCL) | Bortezomib best vs lenalidomide best (OS) | 0.20 | 0.96 | 1117 | 1.98 × 10−21 |
| Treatment (data set) . | Comparison . | Median HR* observed . | Median HR* null . | Mann-Whitney U . | P . |
|---|---|---|---|---|---|
| Bortezomib based (C) | Bortezomib standard vs bortezomib good (PFS) | 0.44 | 0.94 | 348 | 5.67 × 10−30 |
| RD (C) | Bortezomib standard vs bortezomib good (PFS) | 2.01 | 0.93 | 1526 | 1.91 × 10−17 |
| Bortezomib based or RD (C) | Incorrectly treated vs correctly treated (PFS) | 0.48 | 0.92 | 185 | 5.44 × 10−32 |
| Bortezomib based or RD (C) | Incorrectly treated vs correctly treated (OS) | 0.46 | 0.85 | 3 | 2.52 × 10−34 |
| Bortezomib alone (M) | Lenalidomide best vs bortezomib best (PFS) | 0.68 | 0.97 | 489 | 2.75 × 10−28 |
| Bortezomib alone (M) | Lenalidomide best vs bortezomib best (OS) | 0.53 | 1.00 | 108 | 5.67 × 10−33 |
| RD (PCL) | Bortezomib best vs lenalidomide best (PFS) | 0.24 | 0.83 | 1334 | 2.84 × 10−19 |
| RD (PCL) | Bortezomib best vs lenalidomide best (OS) | 0.20 | 0.96 | 1117 | 1.98 × 10−21 |
Mann-Whitney-Wilcoxon results testing the robustness of the 7-gene signature to assignments across multiple (n = 100) training/validation data splits; 100 assignments were made in each data set by the 7-gene signature after random training/validation splits (observed). Each of these assignments was then permuted to maintain assignment ratios (null). The performance of the observed and null assignments for predicting PFS and OS was compared in terms of HRs by the Mann-Whitney-Wilcoxon test.
C, CoMMpass data set; H, HOVON/GMMG data set; M, Millennium data set.
HR: second group vs first group.