5B-04 - To Clone or Not to Clone? Baseline Incompatibility with Regimens when Cloning, Censoring, and Weighting

Background: Cloning, censoring, and weighting (CCW) is a popular analytic approach to contrast regimens that are indistinguishable at baseline. Initially developed to compare regimens of the form “never start X” with “start X and continue N years” in randomized trials, use of CCW has extended to cases where people are incompatible with some regimens at baseline (e.g., people treated at time 0 are incompatible with “never treat” regimens at time 0). There is a need to evaluate guidance on CCW in this new context with regards to which regimens to clone patients into.

Objectives: Examine differences between A) cloning into all regimens regardless of baseline compatibility and B) cloning only into compatible regimens then weighting to address baseline imbalances in confounders in simulated CCW scenarios.

Methods: We simulated 2,000 iterations of a 50,000 patient CCW study comparing the risk of an outcome Y by time 3 under three regimens: “start X by time 1”, “start X at time 0”, and “never start X.” A third variable W confounded the X-Y association. We calculated risk differences (RDs) between: A) cloning all patients into all regimens, censor patients when they become incompatible with their assigned regimen, with some patients potentially censored at time 0, and create censoring weights; and B) cloning patients only into compatible regimens at baseline, censor and weight within regimens, and account for confounding between regimens with contrast-specific IPTW. We calculated RDs between regimens using A and B with no effect of X (scenario 1), X causing a fixed increase in the risk of Y of 0.05 (scenario 2), and a heterogeneous effect involving X causing an increase of 0.20 in the risk of Y only in patients with W=1 (scenario 3). We calculated relative bias of RDs vs a true value obtained via separate Monte Carlo simulations of 50,000,000 patients for each regimen.

Results: In scenario 1 (no effect), A and B were unbiased for all contrasts. In scenario 2 (fixed effect), A and B were both unbiased when comparing “start at time 0” vs “never start” but B was very slightly biased when comparing “start by time 1” vs “never start” (bias = 1.01 for B) and “start by time 1” vs “start at time 0” (bias = 0.88 for B). Bias was larger for B in Scenario 3 (bias = 0.81 for “start by time 1” vs “never start”, bias = 1.28 for “start by time 1” vs “start at time 0”).

Conclusions: Cloning patients into all regimens with possible censoring at time 0 enables consistent estimation of contrasts between regimens for eligible participants. Cloning only into compatible regimens and subsequently addressing confounding can bias estimates when there is a treatment effect. This supports cloning patients into all regimens, regardless of baseline compatibility, in real-world applications of CCW.