Using Bias Correction Methods with Hybrid Randomized Controlled Trial (RCT) Plus Real World (RW) Controls to Replicate Results of a Phase Iib RCT of Baricitinib for Rheumatoid Arthritis (RA)

Friday, August 26, 2022

1:30 PM – 1:45 PM CEST

Location: Congress Hall D4

Publication Number: 55

Background: Studies augmenting RCT controls with data from external RW patients face potential concerns of selection bias and unmeasured confounding.

Objectives: To assess 2 methods for hybridizing RCT controls with RW controls and to compare the proportion of RA patients achieving 20% improvement in American College of Rheumatology measure (ACR20) between RCT baricitinib treated patients and hybrid controls for each method.

Methods: A random 50% sample of RCT controls was selected for this study; the remaining control patients were obtained from trial-eligible RW patients from the CorEvitas RA registry. The Match-Test-and-Pool Method (MTP) used propensity score (PS) and optimal matching to match the 50% RCT control sample to RW patients from the registry. The ACR20 was compared among RCT and matched RW controls to determine if populations could be pooled for comparison to the RCT treated patients. The Matching and Bias Adjustment Method (MBA), proposed by Stuart and Rubin in 2008, involved 3 optimal PS matches: (1) RCT treated to the 50% RCT control sample, (2) remaining unmatched RCT treated to RW patients, and (3) the 50% RCT control sample to RW patients. A bias adjustment parameter was estimated using a Bayesian logistic regression model that adjusted for covariates with remaining imbalance in match #3. 500 samples of the bias adjustment parameter were drawn from its posterior distribution and added as an offset for RW controls in logistic regressions comparing the RCT treated to the hybrid controls. Rubin's rules were used to pool treatment effect estimates.

Results: 34.7% of RCT controls vs 44.9% of matched RW controls achieved ACR20 (p=0.41); thus, the controls were pooled. In MTP method, comparing RCT treated vs hybrid controls, the marginal probabilities of ACR20 were 0.72 (95%CI: 0.63, 0.80) vs 0.40 (95% CI: 0.30, 0.49), respectively, similar to that calculated in the trial populations. Comparing RCT treated to hybrid controls, using the MBA method with bias correction, the estimated ACR20 marginal probabilities were 0.72 (95% CI: 0.63, 0.80) in RCT treated vs 0.35 (95% CI: 0.19, 0.51) in hybrid controls.

Conclusions: Using data from RA registry patients, we obtained well-matched controls and replicated the inference of the baricitinib Phase IIb trial using both hybrid control methods. MBA adjusts for additional outcome bias that may exist between trial and RW control arms and is a robust method for constructing hybrid controls in situations where PS matching is not sufficient to adjust for heterogeneity between populations.

Presenting Author(s)

Anita Loughlin

Senior Director of Epidemiology
CorEvitas
Wakefield, Massachusetts, United States

Slides

Co-Author(s)

KS

Kristin Sheffield, PhD

Research Advisor - Value Evidence and Outcomes
Eli Lilly and Company, Indiana, United States
OP

Oksana Pugach, PhD

Associate Director Query Coordination in Biostatistics
CorEvitas, LLC, United States
MS

Mingyang Shan, PhD

Senior Research Scientist
Eli Lilly and Company
Indianapolis, Indiana, United States
DF

Douglas Faries, PhD

Senior Research Fellow
Eli Lilly and Company, United States
PL

Paul Lakin, MS

Biostatistician II
CorEvitas, LLC, United States
SK

Samantha Kerti, MS

Biostatistician
University of Pittsburgh Medical Center, United States
ZC

Zhanglin Cui, PhD

Research Advisor - Real World Analytics - Oncology
Eli Lilly and Company, Indiana, United States
SC

Sarah Curtis, MPH

Researcher
Eli Lilly and Company, United States
HL

Heather Litman, PhD

Vice President of Biostatistics
CorEvitas, LLC, United States
LH

Leslie Harrold, MD, MPH

Chief Scientific Officer
CorEvitas, LLC, United States
JJ

Joseph Johnston, MD, MSc

Medical Fellow, Global Patient Outcomes and Real World Evidence
Eli Lilly and Company, Indiana, United States