Using randomized trial data to estimate effects of complex treatment regimens in real-world patients.

Abstract

e18706 Background: Real-world patients often differ from trial participants in prognostic factors such as age, sex, and cancer substage. New methods combine covariate data from real-world patients (the “target population”) with outcome and covariate data from a trial to estimate treatment effects in the target population that take these differences into account. With some assumptions, these methods can also estimate outcomes under treatment regimens not studied in the trial such as “what if we only gave six cycles of chemotherapy?” or “what if patients all perfectly followed a protocol?” Methods: Data from the MOSAIC trial of 5-fluorouracil (5FU) vs oxaliplatin + 5FU (FOLFOX) were combined with covariate data from a target population of stage III colon cancer patients in the US Oncology Network meeting trial eligibility criteria. We used weighting and G-computation to estimate five-year mortality and treatment-related paresthesia risk in the target population for four regimens: treatment with up to 12 cycles of 5FU, if providers used their discretion on dose reductions and delays (5FU-MD); up to 12 cycles of FOLFOX with similar physician discretion (FOLFOX-MD); up to 6 cycles of FOLFOX, with providers perfectly following a strict and specific protocol of dose reductions and delays (6-cycle FOLFOX-P, “P” for “per protocol”); and up to 12 cycles of FOLFOX, following the same strict protocol (12-cycle FOLFOX-P). Results: Tablepresents five-year all-cause mortality and paresthesia risk under each regimen in the stage III target population estimated from the models built in trial participants. Paresthesia risk increased with cumulative oxaliplatin dose. Estimated 5-year mortality was lowest with 12-cycle FOLFOX-P. Conclusions: In a target population of US Oncology Network patients with stage III colon cancer, strict protocols of 12 cycles of FOLFOX were predicted to improve survival compared to strict 6-cycle FOLFOX regimens or less strict 12-cycle FOLFOX and 5FU regimens at the cost of substantial increases in side effects. While estimates of risk differences in 5-year mortality were imprecise, combining trial and real-world data and then using weights and G-computation allowed estimation of benefits and harms of multiple regimens in a clinically relevant patient population.[Table: see text]