Using the Facts Recommender in Oncology Dose Escalation Trials: A Comparison of Dose Escalation Decisions Suggested by the Recommender with Actual Decisions Made by Project Teams

• Activity Number: 31
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Sharon Murray*+ and Tom Parke and Allison Florance and Alun Bedding
• Companies: GlaxoSmithKline and Tessella plc and GlaxoSmithKline and GlaxoSmithKline
• Keywords: CRM ; Oncology trials ; FACTS ; Dose escalation

Abstract

Many variations of Continual Reassessment Methods (CRM) have been proposed for oncology dose escalation trials. After each dosing cohort, the statistician updates the modeled toxicity curve with accruing data. A cross-functional team then meets to select the dose for the next cohort using the observed data and updated toxicity model. A challenge of implementing these methods has been the quick turnaround time required for decision-making. To address this, the FACTS software has added a Recommender that does toxicity modeling based on the N-CRM methodology (Neuenschwander et al, Stat Med, 2008). Using historical data from several oncology dose escalation trials conducted at GlaxoSmithKline, actual decisions made by project teams under modified 3+3 designs are compared to those suggested by the FACTS Recommender. In many cases, actual decisions aligned with the FACTS recommendation. Differences, which tended to be dependent on the selection of the prior, are discussed. Use of the FACTS software to understand the impact of the specification of the prior, illustrate the underlying model fit, and justify the recommended decisions should aid uptake of the N-CRM for clinical teams.