ABSTRACT
Holmgren has introduced a clever procedure to determine the cutoff (threshold) for a continuous biomarker using the group sequential methodology (GSM). The precise marker threshold for dichotomizing is unknown or the inflection point for treatment effect along the biomarker range of expression values is unknown. Holmgren calculates the number of events at each ‘stage’ or biomarker subgroup fraction, alpha spent, and power at each stage. Holmgren points out there are really four subgroups being considered (treatment vs placebo, marker vs marker-complement). This is a more complex problem to solve than simply testing for a treatment effect in subgroups. But, Holmgren’s novel approach enables the marker subgroup and its complement to be tested within the solid framework of GSM methodology of type-I error control while testing several biomarker x treatment hypotheses along the range of biomarker expression values.
