Choosing an Index Date for Untreated Patients in External Comparator Studies

Selecting an index date (also called time zero or baseline) can be challenging for External Comparator (EC) studies when comparing against untreated patients. Existing literature addresses methods for defining an index date for untreated patients in observational studies generally, but not for EC studies specifically, which are likely to benefit from customized approaches.

A simulation study was performed to assess different index date assignments and analytical approaches in terms of bias and other performance characteristics: The first approach took the time from a major clinical event (say, diagnosis date) to treatment start as observed in the treated cohort and randomly assigned these times to the untreated cohort to derive the index dates. This approach was applied without and with the condition that the emulated index dates in the untreated cohort needed to be before the observed event times (index date emulation [IDE] and modified index date emulation approach [mIDE]). The second approach was to start the follow-up period at the diagnosis date (early index date approach [EID]) and to perform an analysis according to a time-dependent Cox model (or its generalization, e.g., a Marginal Structural Cox Model). This model was applied both in a traditional but also in a modified manner (modified early index date approach, mEID), where the modified model coded the treatment cohorts before the true (treated patients) and emulated (untreated patients, using IDE) treatment start dates to belong to a third treatment category. This allowed the treatment comparison of interest to be restricted to the time after the true and emulated treatment start dates.

The IDE and mEID approaches were shown to be unbiased with identical performance, while mIDE and EID exhibited significant bias.

We showed that our EC analysis approach based on emulated index dates for untreated patients constitutes a valid concept, which may be advantageous for many external comparator studies.