Statistical methods for elimination of guarantee-time bias in cohort studies: a simulation study

Table 1 Estimated Empirical type I error of Cox regression, Time-dependent Cox regression and Landmark methods on simulated data under the null hypothesis that aspirin is not significantly related to cancer

β ₀	Cumulative dose	Cox regression	Time-dependent Cox regression	Landmark analyses
β ₀	Cumulative dose	Cox regression	Time-dependent Cox regression	τ = 5	τ = 7
log(0.015)	Low	214	58	56	57
	Moderate	1000	52	42	45
	High	1000	43	55	47
log(5^∗0.015)	Low	723	47	56	66
	Moderate	1000	50	40	44
	High	1000	47	46	46
log(10^∗0.015)	Low	957	52	62	52
	Moderate	1000	59	50	48
	High	1000	46	46	44

Each value represents the number of cases out of 1000 replications for which the null hypothesis is incorrectly rejected
Two values of landmark time (τ) are used, 5 and 7. β ₀ values represent disease incidence rate

ISSN: 1471-2288