Research Article
A Simulation Study Comparing Different Statistical Approaches for the Identification of Predictive Biomarkers
Table 3
Estimated power with exact
confidence intervals (in brackets) for Scenarios 5 and 6 for all investigated methods.
| | | n = 250 | n = 500 | n = 1000 | | Low cens. | High cens. | Low cens. | High cens. | Low cens. | High cens. |
| | Scenario 5 | Median split | 16.1% | 10.4% | 26.7% | 17.7% | 46.6% | 29.3% | | (13.9–18.5%) | (8.6–12.5%) | (24.0–29.6%) | (15.4–20.2%) | (43.5–49.7%) | (26.5–32.2%) | | Quartile split | 23.0% | 10.5% | 41.3% | 21.2% | 70.9% | 40.1% | | (20.4–25.7%) | (8.7–12.6%) | (38.2–44.4%) | (18.7–23.9%) | (68.0–73.7%) | (37.0–43.2%) | | Optimal split | 79.7% | 60.2% | 93.8% | 78.3% | 99.4% | 91.9% | | (77.1–82.2%) | (57.1–63.2%) | (92.1–95.2%) | (75.6–80.8%) | (98.7–99.8%) | (90.0–93.5%) | | STEPP | 15.9% | 7.4% | 35.0% | 14.8% | 68.7% | 31.0% | | (13.7–18.3%) | (5.9–9.2%) | (32.0–38.0%) | (12.7–17.2%) | (65.7–71.6%) | (28.1–34.0%) | | Cox (linear int.) | 25.6% | 16.3% | 47.0% | 30.6% | 76.2% | 51.1% | | (22.9–28.4%) | (14.1–18.7%) | (43.9–50.1%) | (27.8–33.6%) | (73.4–78.8%) | (48.0–54.2%) | | MFPI (FP1-flex3) | 39.7% | 22.6% | 67.3% | 39.7% | 93.8% | 66.5% | | (36.7–42.8%) | (20.0–25.3%) | (64.3–70.2%) | (36.7–42.8%) | (92.1–95.2%) | (63.5–69.4%) | | MFPI (FP2-flex1) | 26.8% | 13.2% | 50.6% | 27.1% | 87.5% | 51.8% | | (24.1–29.7%) | (11.2–15.5%) | (47.5–53.7%) | (24.4–30.0%) | (85.3–89.5%) | (48.7–54.9%) | | LPLB | 22.5% | 9.3% | 46.9% | 21.9% | 83.4% | 45.3% | | (19.9–25.2%) | (7.6–11.3%) | (43.8–50.0%) | (19.4–24.6%) | (80.9–85.7%) | (42.2–48.4%) |
| | Scenario 6 | Median split | 5.3% | 3.9% | 5.1% | 5.1% | 6.4% | 4.9% | | (4.0–6.9%) | (2.8–5.3%) | (3.8–6.7%) | (3.8–6.7%) | (5.0–8.1%) | (3.6–6.4%) | | Quartile split | 24.3% | 13.0% | 42.0% | 16.0% | 73.8% | 36.5% | | (21.7–27.1%) | (11.0–15.2%) | (38.9–45.1%) | (13.8–18.4%) | (71.0–76.5%) | (33.5–39.6%) | | Optimal split | 73.8% | 56.5% | 88.1% | 67.1% | 97.6% | 86.3% | | (71.0–76.5%) | (53.4–59.6%) | (85.9–90.0%) | (64.1–70.0%) | (96.4–98.5%) | (84.0–88.4%) | | STEPP | 14.8% | 7.9% | 31.4% | 12.7% | 61.6% | 25.9% | | (12.7–17.2%) | (6.3–9.7%) | (28.5–34.4%) | (10.7–14.9%) | (58.5–64.6%) | (23.2–28.7%) | | Cox (linear int.) | 6.1% | 5.9% | 4.8% | 5.1% | 6.2% | 4.6% | | (4.7–7.8%) | (4.5–7.5%) | (3.6–6.3%) | (3.8–6.7%) | (4.8–7.9%) | (3.4–6.1%) | | MFPI (FP1-flex3) | 23.0% | 16.5% | 30.6% | 18.5% | 50.9% | 27.3% | | (20.4–25.7%) | (14.3–18.9%) | (27.8–33.6%) | (16.1–21.0%) | (47.8–54.0%) | (24.6–30.2%) | | MFPI (FP2-flex1) | 20.0% | 11.2% | 28.6% | 12.8% | 45.9% | 26.9% | | (17.6–22.6%) | (9.3–13.3%) | (25.8–31.5%) | (10.8–15.0%) | (42.8–49.0%) | (24.2–29.8%) | | LPLB | 22.2% | 10.4% | 41.5% | 17.9% | 74.6% | 36.7% | | (19.7–24.9%) | (8.6–12.5%) | (38.4–44.6%) | (15.6–20.4%) | (71.8–77.3%) | (33.7–39.8%) |
|
|
Due to increased type I error probabilities, results for Optimal split and MFPI (FP1-flex3) are presented in italics.
|
