Research Article

A Simulation Study Comparing Different Statistical Approaches for the Identification of Predictive Biomarkers

Table 1

Estimated type I error probabilities with exact confidence intervals (in brackets) for Scenarios 1 and 2 for all investigated methods.

n = 250n = 500n = 1000
Low cens.High cens.Low cens.High cens.Low cens.High cens.

Scenario 1Median split4.7%5.3%4.9%4.4%5.0%6.4%
(3.5–6.2%)(4.0–6.9%)(3.6–6.4%)(3.2–5.9%)(3.7–6.5%)(5.0–8.1%)
Quartile split6.3%5.3%4.2%5.1%3.5%5.7%
(4.9–8.0%)(4.0–6.9%)(3.0–5.6%)(3.8–6.7%)(2.4–4.8%)(4.3–7.3%)
Optimal split43.6%39.9%45.8%40.9%45.5%46.6%
(40.5–46.7%)(36.8–43.0%)(42.7–48.9%)(37.8–44.0%)(42.4–48.6%)(43.5–49.7%)
STEPP4.8%5.9%5.1%4.2%4.0%6.3%
(3.6–6.3%)(4.5–7.5%)(3.8–6.7%)(3.0–5.6%)(2.9–5.4%)(4.9–8.0%)
Cox (linear int.)4.9%5.2%4.8%4.4%4.8%5.8%
(3.6–6.4%)(3.9–6.8%)(3.6–6.3%)(3.2–5.9%)(3.6–6.3%)(4.4–7.4%)
MFPI (FP1-flex3)10.1%10.6%10.4%12.0%10.3%13.5%
(8.3–12.1%)(8.8–12.7%)(8.6–12.5%)(10.1–14.2%)(8.5–12.4%)(11.4–15.8%)
MFPI (FP2-flex1)4.9%5.6%4.9%4.7%6.0%7.0%
(3.6–6.4%)(4.3–7.2%)(3.6–6.4%)(3.5–6.2%)(4.6–7.7%)(5.5–8.8%)
LPLB4.6%4.5%4.2%4.4%3.9%5.8%
(3.4–6.1%)(3.3–6.0%)(3.0–5.6%)(3.2–5.9%)(2.8–5.3%)(4.4–7.4%)

Scenario 2Median split4.2%6.0%5.7%4.2%5.7%4.3%
(3.0–5.6%)(4.6–7.7%)(4.3–7.3%)(3.0–5.6%)(4.3–7.3%)(3.1–5.7%)
Quartile split5.3%6.7%5.1%4.5%5.0%5.4%
(4.0–6.9%)(5.2–8.4%)(3.8–6.7%)(3.3–6.0%)(3.7–6.5%)(4.1–7.0%)
Optimal split50.8%42.8%53.9%45.9%52.0%47.4%
(47.7–53.9%)(39.7–45.9%)(50.8–57.0%)(42.8–49.0%)(48.9–55.1%)(44.3–50.5%)
STEPP5.4%8.2%6.8%6.8%7.8%6.9%
(4.1–7.0%)(6.6–10.1%)(5.3–8.5%)(5.3–8.5%)(6.2–9.6%)(5.4–8.7%)
Cox (linear int.)4.5%8.2%4.8%6.4%5.0%5.2%
(3.3–6.0%)(6.6–10.1%)(3.6–6.3%)(5.0–8.1%)(3.7–6.5%)(3.9–6.8%)
MFPI (FP1-flex3)8.1%12.6%9.1%10.3%8.1%7.8%
(6.5–10.0%)(10.6–14.8%)(7.4–11.1%)(8.5–12.4%)(6.5–10.0%)(6.2–9.6%)
MFPI (FP2-flex1)5.5%6.5%6.5%5.6%4.1%5.9%
(4.2–7.1%)(5.1–8.2%)(5.1–8.2%)(4.3–7.2%)(3.0–5.5%)(4.5–7.5%)
LPLB6.2%5.8%7.3%4.8%7.7%6.1%
(4.8–7.9%)(4.4–7.4%)(5.8–9.1%)(3.6–6.3%)(6.1–9.5%)(4.7–7.8%)