Robust estimation of the expected survival probabilities from high-dimensional Cox models with biomarker-by-treatment interactions in randomized clinical trials

Table 3 Accuracy and precision of the survival probabilities, and coverage probability of their 95% confidence intervals of the selected models by the adaptive lasso penalty

Scenarios	Point estimate of the 5-year survival probability				95% CI of the expected survival
	Mean bias		Standard error		Coverage probability
	Pointwise	Spline	Pointwise	Spline	Pointwise		Spline
					Anly	Boot	Anly	Boot
(1) Complete null	−0.002	−0.001	0.05	0.05	0.93	0.97	0.94	1.00
(2) Treatment effect only	−0.001	−0.001	0.06	0.05	0.93	0.96	0.93	1.00
(3) 20 prognostic biomarkers	−0.002	0.001	0.09	0.09	0.91	0.97	0.89	0.98
(4) 15 treatment-effect modifiers	−0.003	−0.001	0.11	0.10	0.88	0.96	0.89	0.98
(5) Treatment effect + (4)	−0.002	0.000	0.11	0.10	0.88	0.96	0.89	0.98
(6) 20 prognostic biomarkers + (5)	−0.005	0.000	0.13	0.13	0.88	0.96	0.87	0.96

Anly: analytical approach, Boot: non-parametric bootstrap approach, CI: confidence interval. Average quantities across 250 replications

ISSN: 1471-2288