5. The impact of data and safety monitoring board decisions to drop treatments

Team: Dr Franz König, Prof Peter Bauer (Medical University Vienna), Dr Simon Day (CTCT)
Clinical Advisor: Prof Harald Herkner,
an expert in emergency medicine and intensive care medicine at the Medical University of Vienna. He is also a member of the pharmacovigilance and risk assessment committee at the European Medicines Agency.
Location: Medical University Vienna (Vienna, Austria)

In phase II studies it is common to compare several doses to a control. Besides estimating the dose-response curve, it is of interest to detect which doses have a statistically different effect. To adjust for multiplicity popular adjustments include hierarchical (pre-defined sequence) Bonferroni and Dunnett tests. Such trials are often accompanied by a data safety monitoring board (DSMB). To protect the study participants a DSMB may take decisions to terminate dose groups during the conduct of the trial because of safety concerns, or because the risks are not outweighed by the (small) benefits seen so far.

This work will explore the impact of such decisions on the originally planned analysis and the bias in effect estimates. The frequent seen argument that the properties of the design remain acceptable, as long as the termination of a dose is based only on safety data, will be scrutinized. We will thoroughly explore under which statistical conditions it suffices to only consider the selected treatments for different multiple testing strategies. E.g., if a Bonferroni split was originally planned, can the significance level be divided by the number of treatments finally selected? We will explore several multiple testing procedures such as (sequentially-rejective) weighted Bonferroni tests, Dunnett type tests extending earlier work for the hierarchical test8 . Testing strategies based on the observed correlation between safety and efficacy will be developed and compared to adaptive tests accounting for unblinded design adaptations9 . In a next step graphical methods when testing primary and secondary endpoints will be evaluated as well.

The investigation is subsequently expanded to evaluate the impact of other common decisions of the DSMB, such as reallocating the sample size of dropped treatment arms to the remaining arms or dropping a subgroup due to safety issues or lack of efficacy at interim.

Meet our Early Stage Researcher: Julia Niewczas, Medical University Vienna

Julia Niewczas_new

I am from Poland where I lived for most of my life.  I studied at Lancaster University where I first completed a BSc in Mathematics with Statistics and then an MSc in Statistics with a medical focus. I received a PSI student prize for academic excellence in my MSc. Now I am at the Medical University of Vienna and my research is focused on adaptive designs and multiple testing procedures. I enjoy travelling and discovering new places and cultures.

8 König F, Bauer P, Brannath W. (2006) Biometrical J, 48(4):663-678.
9 Bauer P, Kieser M. (1999). Combining different phases in the development of medical treatments within a single trial. Stat Med, 18:1833-48.


Project News 1 of 9 pages

IDEAS welcomes AstraZeneca to our H2020 project

IDEAS would like to welcome AstraZeneca to our innovative training network as a partner.

Meet the IDEAS Network partners

The IDEAS network brings together leading institutions in the training of medical statisticians, large and small sized pharmaceutical companies and is supplemented by clinical experts.