V. S. Purohit,1 M. M. Hutmacher2; 1Pfizer, Groton, CT, 2Ann Arbor Pharmacometrics Group, Ann Arbor, MI
BACKGROUND: In dose ranging studies teams often debate between choosing conventional pairwise comparisons (PC) or regression-based dose response (DR) analyses as the primary method for making clinical development decisions. This work evaluated the operating characteristics for decision making and dose selection of joint decision rules based on a PC and DR analysis using simulation of a case study.
METHODS: ADR relationship based on existing data was used for data generation. Landmark data for placebo and three active dose levels with various sample sizes (40, 50, 60, 70 and 80) per arm were simulated. Each of the 1,000 replicates was analyzed by a family (Linear, Emax and Step-function) of DR models with final model selection based on Akaike information criterion and PC using Dunnetts method for multiplicity correction. Decision rules were applied to the results to determine their operating characteristics. The decision rules were constructed using meaningful target values of difference from placebo (0% and 15%) with different levels of confidence. A joint rule for a considered dose was defined by requiring a PC confidence interval and DR estimated mean be >15%.
RESULTS: The DR analysis had higher power, yet with a slightly higher type 1 error relative to PC to detect difference from placebo. Typically, you need more subjects with PC to achieve the same operating characteristics as with DR. However, for this example, the advantage of DR was limited. The probability of a correct Go/No-Go decision (PCD) for the joint rule was comparable to those for PC or DR only, yet was 5-10% higher for smaller sample sizes and treatment effects.
CONCLUSION: Joint decision rules are viable options that can improve PCD, and focus the teams on integration of methods, viewpoints, and making correct development decisions.