Proportional Randomization Method to Identify Index Line of Therapy in Externally Controlled Trials

Speaker(s)

Cui Z, Khanal M, Chen Y, Li X, Winfree K
Eli Lilly and Company, Indianapolis, IN, USA

Presentation Documents

OBJECTIVES: In some externally controlled trials (ECT), patients in the treatment arm receive the investigational drug at different line of therapy (index LOT) and the index LOTs for patients in the external control arm (ECA) are to be determined. Most existing methods to determine index LOT fail to balance LOT distribution between the two arms. This study proposes and validates a new method to identify index LOT for patients in the ECA based on the known index LOT distribution from the treatment arm.

METHODS: We proposed a proportional randomization algorithm to assign patients from the ECA into a frequency distribution similar to the index LOTs of patients in the treatment arm. The index LOT for a patient from the ECA is then defined as one of the available eligible LOTs closest to the assigned LOT in the distribution. This LOT assignment was initially assessed by comparing the LOT distributions between the two arms and further validated by assessing the stability and variability of the standardized mean differences (SMD) from 1000 proportional randomization samples. An SMD≤0.1 is considered balance in LOT between the two arms. In a case study of ECT with single arm trial BRUIN-18001 and real-word ConcertAI Oncology Dataset for mantle cell lymphoma (MCL), the index LOT frequency distribution was obtained from BRUIN and applied in the proportional randomization method to identify index LOT for patients from ConcertAI data.

RESULTS: The proportional randomization method produced a frequency distribution of index LOTs for real-world patients similar to that for BRUIN patients (mean (SD)=4.25(1.44) vs 4.22(1.65), t-test p-value=0.91, SMD=0.016). The mean and 95% confidence interval of SMDs based on the 1000 proportional randomization samples were 0.035(0.001, 0.093).

CONCLUSIONS: The proportional randomization method was successfully used to identify index LOT in this ECT, and its applicability can be explored in future ECTs.

Code

MSR105

Topic

Methodological & Statistical Research

Topic Subcategory

Confounding, Selection Bias Correction, Causal Inference

Disease

Oncology