A model assessing how variable timing of outcome assessments impacts progression-free survival (PFS)

Abstract

5130 Background: PFS has been proposed as an endpoint in prostate cancer because tumor regression cannot be assessed easily and the significance of post-therapy changes in PSA is uncertain. There is significant variability in the frequency by which outcomes are assessed across clinical trials. We sought to create a model that would define the degree of error in estimating PFS from this variability. Methods: A simulation experiment was performed. An exponential distribution was used to generate 100 progression times for 3 hypothetical risk cohorts: rapid (median PFS of 18 wks), intermediate (36 wks) and slow progressors (72 wks). We examined how reported PFS would change depending on 3 assessment schedules: every (q) 6, 8 and 12 wks for 48 wks each, then q6 months for 2 years. The logrank statistic was used to compare PFS between schedules. If different schedules have no impact the expected type 1 error rate should be 5% (where a difference in PFS time is detected when none existed). Each simulation was repeated 1000 times. Results: Nine pairwise comparisons were performed and Kaplan-Meier PFS estimates created for the 3 assessment schedules and 3 risk cohorts (see table 1 ). In the highest risk cohort, PFS of 18 wks, 38% of the time the logrank test showed a falsely prolonged PFS for pts assessed on the q12 vs 6 wks schedule. This type 1 error rate (by simulation) was reduced to 20% and 11%, when the schedules were q8 vs 12 wks and q6 vs 8 wks, respectively, but remained above the 5% expected rate for type 1 error. For lower risk pts, PFS of 72 wks, the disparity in PFS times was diminished. Conclusions: Progression free survival is significantly skewed by the schedule of assessing treatment effects in clinical trials. This argues for uniformity in the timing of outcome assessments across trials and between arms in randomized trials. Grant support: 5T32CA09207 [Table: see text] No significant financial relationships to disclose.