A simulation study on estimating biomarker–treatment interaction effects in randomized trials with prognostic variables

Table 1 Proportions of rejected null hypotheses and numbers of included covariates stratified for number of potential prognostic variables (K), strength of interaction and proportion of censored observations

K	Interaction	Censoring	Main	True	AIC_A	AIC_B	Significance	Full
12	No	Low	0.058	0.060	0.062	0.056	0.059	0.060
12	No	High	0.054	0.051	0.056	0.048	0.053	0.054
12	Quantitative	Low	0.118	0.133	0.139	0.131	0.130	0.134
12	Quantitative	High	0.095	0.100	0.108	0.099	0.096	0.099
12	Qualitative	Low	0.579	0.663	0.663	0.654	0.653	0.661
12	Qualitative	High	0.393	0.437	0.441	0.424	0.432	0.436
24	No	Low	0.065	0.058	0.067	0.056	0.057	0.058
24	No	High	0.057	0.062	0.073	0.054	0.059	0.063
24	Quantitative	Low	0.115	0.135	0.150	0.131	0.131	0.136
24	Quantitative	High	0.094	0.100	0.114	0.092	0.100	0.103
24	Qualitative	Low	0.465	0.634	0.645	0.616	0.610	0.633
24	Qualitative	High	0.349	0.411	0.426	0.383	0.399	0.415
36	No	Low	0.065	0.059	0.078	0.056	0.061	0.063
36	No	High	0.067	0.068	0.085	0.057	0.066	0.071
36	Quantitative	Low	0.114	0.132	0.153	0.118	0.130	0.134
36	Quantitative	High	0.093	0.102	0.127	0.093	0.101	0.108
36	Qualitative	Low	0.412	0.618	0.629	0.578	0.576	0.610
36	Qualitative	High	0.302	0.406	0.431	0.367	0.382	0.402

Results are aggregated over different values of β_k and Σ. For the scenarios with no true biomarker–treatment interaction, results for methods/strategies with an observed type I error probability above 7% are in italics. For scenarios with a true biomarker–treatment interaction, the observed power is in bold if the type I error probability did not exceed 7%

ISSN: 1745-6215