(209) What Exactly Does Clone-Censor-Weighting Estimate When Studying Treatment Initiation Windows, Anyway?

Background: Clone-censor-weighting (CCW) can compare treatment regimens that are indistinguishable from one another at baseline without relying on landmark dates or creating immortal person time. While this is useful in observational studies comparing starting treatment inside time windows (e.g., starting inhaled corticosteroids within 30 days vs 30 to 90 days after hospitalization for chronic obstructive pulmonary disease exacerbation), the intervention that CCW corresponds to in this context has not been discussed explicitly.

Objectives: Clarify the population-specific hypothetical intervention corresponding to CCW when studying regimens of “start treatment within a time window” (e.g., starting treatment by day 30).

Methods: Using CCW to study the effect of “start treatment by day 30” means estimating potential outcomes under a hypothetical intervention where A) before day 30, everyone follows the covariate-conditional start time distribution of the study population and B) those who have not initiated by day 30 initiate on day 30. We simulated 5,000,000 individuals to either start treatment immediately (time 0), start treatment at time 1 (i.e., day 30), or never start treatment. The probability of starting treatment depended on a confounder. We conducted two CCW analyses to estimate outcomes if individuals started treatment by time 1. The first (“full CCW”) included all survivors to time 1 in inverse probability of censoring weights (IPCW). The second (“limited CCW”) excluded surviving time 0 initiators when calculating IPCW and gave them weights of 1, meaning only those who started treatment at time 1 stood in for those censored at time 1. For comparison purposes, we estimated “true” risks under an intervention where the population followed the covariate-conditional trajectory of initiation until time 1 when all remaining individuals initiated. We compared results in a base simulation with a complex treatment effect to scenarios with simpler effects.

Results: In the base case, full CCW yielded a time 3 (i.e., 90 day) risk of 59.1% while limited CCW yielded a time 3 risk of 56.6%. The approximated true risk was 56.6%. Full CCW only estimated the true time 3 risk in scenarios in which exposure effects were constant over time (e.g., null). In other scenarios with a time-varying exposure effect (e.g., delayed or cumulative effects), full CCW yielded the estimate of an intervention involving impossible exposure histories.

Conclusions: While CCW can examine regimens like “start by 30 days,” providing information on the timing of treatment initiation adds key context to the hypothetical intervention under study. Researchers should be wary about including all individuals who initiate prior to the end of the time interval in IPCW calculations if treatment effects vary over time.