SMART Design, Conduct, and Analysis in Oncology

This talk is based on my experiences as a biostatistician working with oncologists and statisticians in clinical trial design and analysis. It is about dynamic treatment regimes, clinical trial designs that were both SMART and stupid, and how clinical evaluation of therapies for cancer actually plays out in practice. I will discuss three trials. The first was a trial of four combination chemotherapies for advanced prostate cancer, with a design that included re-randomization. I will tell the saga of this trial, from its inception in 1999 to the most recent data analyses in 2012. The second is an ongoing trial of three targeted agents for metastatic renal cancer. Both of these trial designs turned out to be SMART, although we constructed them without knowledge of this useful acronym. The third was a randomized trial of four chemotherapy combinations for acute leukemia that was not as smart as it might have been.