Effect of designations of index date in externally controlled trials: an empirical example

Abstract

Abstract Objectives To create an external control arm (ECA) for a single arm trial, the choice of index date – when a patient becomes eligible for a study, is a complex issue. In real world data (RWD), patients commonly have multiple qualifying lines of therapy (LOT) which can be used to determine the index date. This study assessed the impact of different methods to assign the index date on the effectiveness estimates of the target drug versus conventional therapies and explored the impact of seven methods to assign the index date on the effectiveness estimates of the target drug versus conventional therapies. Methods A study using RWD was conducted in which patients received varied number of LOTs before qualifying for entry into the ECA. Two novel and five established indexing methods were examined for the ECA in this comparative effectiveness research. Baseline characteristics were adjusted by using stabilized inverse probability of treatment weighting (sIPTW). Cox proportional hazards (PH) model was used for time-to-event endpoints and risk ratio (RR) was estimated from a binomial regression for response-based end points. Results Five methods (first eligible line [FEL], restricted-line, all eligible lines, random line, and stratified random line) demonstrated close clinical outcome estimates after adjustment of baseline differences via sIPTW. The FEL resulted in an inability to adjust for number of prior LOTs due to poor overlap of line distribution in this study. The last and second last eligible line cannot be recommended due to their inability to adjust for immortal time bias. Conclusions Multiple methods are available for selecting the most appropriate index date for an ECA, and this empirical study has indicated that certain methods yield comparable outcomes when the treatment effect and sample size are large. It is important for researchers to carefully assess the specifics of their studies and justify their selection of the most appropriate indexing method. Future research including simulations to evaluate the two novel stratified random line and SLEL methods is necessary.