Gatekeeping strategies for clinical trials that do not require all primary effects to be significant

Abstract

SummaryIn this paper we describe methods for addressing multiplicity issues arising in the analysis of clinical trials with multiple endpoints and/or multiple dose levels. Efficient "gatekeeping strategies" for multiplicity problems of this kind are developed. One family of hypotheses (comprised of the primary objectives) is treated as a "gatekeeper," and the other family or families (comprised of secondary and tertiary objectives) are tested only if one or more gatekeeper hypotheses have been rejected. We discuss methods for constructing gatekeeping testing procedures using weighted Bonferroni tests, weighted Simes tests, and weighted resampling-based tests, all within a closed testing framework. The new strategies are illustrated using an example from a clinical trial with co-primary endpoints, and using an example from a dose-finding study with multiple endpoints. Power comparisons with competing methods show the gatekeeping methods are more powerful when the primary objective of the trial must be met.

Ferric Citrate Reduces Intravenous Iron and Erythropoiesis-Stimulating Agent Use in ESRD

“…The study's analysis plan used a sequential gatekeeping strategy 36 with two-sided a=0.05 at each step to protect the study-wise a-level for the treatment comparisons of the primary and four prespecified main secondary outcomes in the following sequence: (1) change in serum phosphorus during the final 4-week placebo-control period, changes in (2) ferritin and (3) TSAT from baseline to week 52 in the 52-week period, and finally, cumulative use of (4) iv iron and (5) ESA over the 52-week period. Because each of these five comparisons attained the designated a-level of 0.05, we report comparisons of ferritin, TSAT, iv iron, and ESA between the treatment groups using a two-sided a=0.05 in accordance with the prespecified gatekeeping rule.…”

Tissue Type-Specific Bioenergetic Abnormalities in Adults with Major Depression

“…We used an analogous approach to our test of the primary hypothesis, except, in place of condition, we inserted symptom intensity scores. Therefore, separate linear mixed effects models were constructed for each depressive symptom domain of the Symptom Questionnaire (depression, anxiety, somatic, and anger-hostility) and the HAM-D-17 using metabolites where we had previously observed a significant effect of condition (Dmitrienko et al, 2003). We tested for the following: (1) associations of the differences between gray matter and white matter metabolite concentrations with SQ domain or HAM-D-17 total score (significance of the gray-white matter difference by symptom test interaction), and (2) when these higher order interactions were non-significant, we tested for associations in total tissue as defined above.…”

“…We tested for the following: (1) associations of the differences between gray matter and white matter metabolite concentrations with SQ domain or HAM-D-17 total score (significance of the gray-white matter difference by symptom test interaction), and (2) when these higher order interactions were non-significant, we tested for associations in total tissue as defined above. Bonferroni corrections were applied for the total number of symptom dimensions (5) multiplied by the number of metabolites passing the gatekeeper test (2) (Dmitrienko et al, 2003). Therefore, the p-value was multiplied by 10.…”

A Dunnett-Type Procedure for Multiple Endpoints

“…Simpler procedures (Holm, 1979, Hochberg, 1988, Hommel, 1988 yield conservative and even biased test decisions, because the information about correlations of the endpoints is not exploited, or the correlations are assumed non-negative, respectively. Gatekeeping procedures (Bauer, 1991, Dmitrienko, Offen, andWestfall, 2003) suffer from similar drawbacks. The T 2 test of Hotelling (1951) takes correlations into account, but because of a square sum test statistic it is non-directional and hence not meaningful in many application areas.…”