Regression models for interval censored data using parametric pseudo-observations

Table 1 Results of the simulations in the general set-up based on estimation of cumulative incidence, risk difference and the logarithm of relative risk

Method	Bias	empSE	Relative empSE	modSE	RMSE	Coverage (95% CI)	Valid est.
Cumulative incidence (true value: 0.1236)
Exact	-0.002	0.019	1 (ref.)	0.019	0.019	95.5 (94.0 to 96.6)	999
IC	-0.005	0.026	1.32	0.025	0.026	94.7 (93.1 to 96.0)	942
Sabathé et al.	-0.005	0.023	1.17	0.023	0.023	95.3 (93.8 to 96.5)	1000
Risk difference (true value: 0.0673)
Exact	0.000	0.038	1 (ref.)	0.039	0.038	95.5 (94.0 to 96.6)	999
IC	-0.005	0.059	1.52	0.052	0.059	94.2 (92.6 to 95.5)	953
Sabathé et al.	-0.003	0.044	1.14	0.045	0.044	95.3 (93.8 to 96.5)	1000
Logarithm of relative risk (true value: 0.5590)
Exact	0.010	0.348	1 (ref.)	0.341	0.349	96.2 (94.8 to 97.2)	999
IC	-0.025	0.456	1.31	0.454	0.456	95.8 (94.3 to 96.9)	944
Sabathé et al.	-0.014	0.414	1.19	0.409	0.415	96.6 (95.3 to 97.6)	1000

IC: The proposed method for calculating pseudo-observations for interval censored data
empSE: Empirical standard error, defined as standard deviation of parameter estimates
Relative empSE: Defined as empSE divided by empSE of parameter estimates using the exact method
modSE: Average model standard error
RMSE: Root mean squared error
CI: Confidence interval

ISSN: 1471-2288