Research Article
A Simulation Study Comparing Different Statistical Approaches for the Identification of Predictive Biomarkers
Table 2
Estimated power with exact
confidence intervals (in brackets) for Scenarios 3 and 4 for all investigated methods.
| | n = 250 | n = 500 | n = 1000 | Low cens. | High cens. | Low cens. | High cens. | Low cens. | High cens. |
| Scenario 3 | Median split | 23.5% | 14.1% | 40.8% | 25.1% | 70.2% | 40.9% | (20.9–26.3%) | (12.0–16.4%) | (37.7–43.9%) | (22.4–27.9%) | (67.3–73.0%) | (37.8–44.0%) | Quartile split | 19.2% | 12.4% | 36.4% | 17.7% | 66.0% | 31.2% | (16.8–21.8%) | (10.4–14.6%) | (33.4–39.5%) | (15.4–20.2%) | (63.0–68.9%) | (28.3–34.2%) | Optimal split | 71.4% | 57.6% | 86.4% | 71.7% | 97.1% | 84.6% | (68.5–74.2%) | (54.5–60.7%) | (84.1–88.5%) | (68.8–74.5%) | (95.9–98.0%) | (82.2–86.8%) | STEPP | 13.4% | 10.5% | 29.2% | 15.5% | 55.1% | 26.4% | (11.3–15.7%) | (8.7–12.6%) | (26.4–32.1%) | (13.3–17.9%) | (52.0–58.2%) | (23.7–29.2%) | Cox (linear int.) | 29.9% | 15.2% | 54.2% | 31.6% | 83.8% | 51.9% | (27.1–32.8%) | (13–17.6%) | (51.1–57.3%) | (28.7–34.6%) | (81.4–86.0%) | (48.8–55.0%) | MFPI (FP1-flex3) | 30.2% | 18.2% | 54.2% | 32.5% | 82.8% | 51.1% | (27.4–33.2%) | (15.9–20.7%) | (51.1–57.3%) | (29.6–35.5%) | (80.3–85.1%) | (48–54.2%) | MFPI (FP2-flex1) | 20.4% | 12.7% | 42.9% | 21.0% | 74.7% | 41.2% | (17.9–23.0%) | (10.7–14.9%) | (39.8–46.0%) | (18.5–23.7%) | (71.9–77.4%) | (38.1–44.3%) | LPLB | 15.2% | 9.1% | 32.7% | 16.2% | 61.2% | 30.9% | (13.0–17.6%) | (7.4–11.1%) | (29.8–35.7%) | (14.0–18.6%) | (58.1–64.2%) | (28.0–33.9%) |
| Scenario 4 | Median split | 26.8% | 14.1% | 52.3% | 24.4% | 78.7% | 38.7% | (24.1–29.7%) | (12.0–16.4%) | (49.2–55.4%) | (21.8–27.2%) | (76–81.2%) | (35.7–41.8%) | Quartile split | 22.8% | 14.2% | 48.6% | 24.2% | 79.4% | 37.6% | (20.2–25.5%) | (12.1–16.5%) | (45.5–51.7%) | (21.6–27.0%) | (76.8–81.9%) | (34.6–40.7%) | Optimal split | 77.6% | 57.8% | 92.3% | 72.6% | 99.1% | 88.8% | (74.9–80.1%) | (54.7–60.9%) | (90.5–93.9%) | (69.7–75.3%) | (98.3–99.6%) | (86.7–90.7%) | STEPP | 14.8% | 8.3% | 36.8% | 17.6% | 68.9% | 29.9% | (12.7–17.2%) | (6.7–10.2%) | (33.8–39.9%) | (15.3–20.1%) | (65.9–71.8%) | (27.1–32.8%) | Cox (linear int.) | 33.9% | 17.7% | 64.4% | 32.4% | 89.8% | 49.2% | (31.0–36.9%) | (15.4–20.2%) | (61.3–67.4%) | (29.5–35.4%) | (87.8–91.6%) | (46.1–52.3%) | MFPI (FP1-flex3) | 40.4% | 25.8% | 72.0% | 38.8% | 92.4% | 56.3% | (37.3–43.5%) | (23.1–28.6%) | (69.1–74.8%) | (35.8–41.9%) | (90.6–94.0%) | (53.2–59.4%) | MFPI (FP2-flex1) | 22.3% | 14.5% | 51.2% | 25.6% | 84.7% | 41.3% | (19.8–25.0%) | (12.4–16.8%) | (48.1–54.3%) | (22.9–28.4%) | (82.3–86.9%) | (38.2–44.4%) | LPLB | 19.4% | 8.4% | 44.1% | 20.1% | 75.2% | 33.8% | (17.0–22.0%) | (6.8–10.3%) | (41.0–47.2%) | (17.7–22.7%) | (72.4–77.8%) | (30.9–36.8%) |
|
|
Due to increased type I error probabilities, results for Optimal split and MFPI (FP1-flex3) are presented in italics.
|