Introduction. Sequential trial designs foresee one or more interim analyses (IA) before the full sample size has been reached. Such IA has the primary purpose to terminate the trial when futility or superiority of one of the interventions becomes clear, according to pre-specified stopping rules. Overrunning occurs when data continue to be collected also if a stopping criterion has been reached. Overrunning data collected according to the trial protocol are considered valid and should be included in the analyses but they could influence the results and change the conclusions. Over the years many proposals to deal with overrunning were proposed. Deletion method includes overrunning data, ignoring the interim analysis that has led to the stopping of the trial. The methods of combining p-values rely on the idea to make two different analyses, one on the sequential portion of the data and one on the overrunning part, and to combine them by weighting their p-values. The repeated confidence interval approach is a further alternative to adopt for the overrunning problem. Objective. Comparing different methods for overrunning under a variety of data generating mechanisms. Methods. Two real clinical trials are considered as motivating examples. The first trial was designed to test superiority assuming response rates (death) for Test and Reference treatment respectively of 9% and 15% and a power of 90%. The second trial was designed assuming response rates for Test and Reference drug respectively of 50% and 45%, a non-inferiority margin of 15% and a power of 80%. Both the trial designs considered also O’Brien and Fleming stopping criteria for three IAs and 2.5% one-sided significance levels. These motivating examples are used as base for simulation studies. Results. Preliminary results show similar behaviors for deletion and combining p-values methods. Repeated confidence interval approach show null hyphothesis refusal rates of 1-2% smaller in some of the considered simulation scenarios. Conclusion. Repeated confidence interval approach seems to be the most conservative method.

OVERRUNNING DATA METHODS: COMPARISONS BASED ON REAL DATA TRIALS

SORIANI, NICOLA;BALDI, ILEANA;GREGORI, DARIO
2013

Abstract

Introduction. Sequential trial designs foresee one or more interim analyses (IA) before the full sample size has been reached. Such IA has the primary purpose to terminate the trial when futility or superiority of one of the interventions becomes clear, according to pre-specified stopping rules. Overrunning occurs when data continue to be collected also if a stopping criterion has been reached. Overrunning data collected according to the trial protocol are considered valid and should be included in the analyses but they could influence the results and change the conclusions. Over the years many proposals to deal with overrunning were proposed. Deletion method includes overrunning data, ignoring the interim analysis that has led to the stopping of the trial. The methods of combining p-values rely on the idea to make two different analyses, one on the sequential portion of the data and one on the overrunning part, and to combine them by weighting their p-values. The repeated confidence interval approach is a further alternative to adopt for the overrunning problem. Objective. Comparing different methods for overrunning under a variety of data generating mechanisms. Methods. Two real clinical trials are considered as motivating examples. The first trial was designed to test superiority assuming response rates (death) for Test and Reference treatment respectively of 9% and 15% and a power of 90%. The second trial was designed assuming response rates for Test and Reference drug respectively of 50% and 45%, a non-inferiority margin of 15% and a power of 80%. Both the trial designs considered also O’Brien and Fleming stopping criteria for three IAs and 2.5% one-sided significance levels. These motivating examples are used as base for simulation studies. Results. Preliminary results show similar behaviors for deletion and combining p-values methods. Repeated confidence interval approach show null hyphothesis refusal rates of 1-2% smaller in some of the considered simulation scenarios. Conclusion. Repeated confidence interval approach seems to be the most conservative method.
File in questo prodotto:
Non ci sono file associati a questo prodotto.
Pubblicazioni consigliate

Caricamento pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11577/2754079
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact