OBJECTIVES: Traditional approaches for analyzing real-world data are often ill-equipped to handle the complexity of real-world clinical decision-making. Methods such as propensity score matching were designed to mimic static treatments at a single timepoint; they cannot tackle situations in which treatment decisions are made over time in response to changing patient characteristics. For example, doses may be adjusted to achieve desired biomarker levels. Treatments may be stopped if adverse events occur. Clinicians may switch or add medications to improve treatment response. Insurance coverage disruptions may cause periods of uninsurance, delayed care, and unfilled prescriptions, all of which impact real-world effectiveness but are often ignored in standard approaches. We describe a novel approach that improves upon the ubiquitous propensity-score matching paradigm and can evaluate complex questions around optimal disease management over time.

METHODS: In our approach, known as clone-censor-weighting, patients are “cloned” into one cohort per treatment protocol, allowing for comparison of specific treatment sequences observed in the real world. In each cloned cohort, patients are artificially censored upon deviation from the treatment sequence associated with that cohort. To account for the artificial censoring, patients are reweighted using inverse probability of censoring weights to reflect the original target population.

RESULTS: This approach has been successfully implemented across multiple therapeutic areas and patient populations. We recently used clone-censor-weighting to examine effectiveness of various remdesivir treatment protocols for preventing disease progression among patients hospitalized with COVID-19. We demonstrated that failure to account for complex, time-varying patient characteristics underestimated the real-world effectiveness of remdesivir.

CONCLUSIONS: As pharmaceutical therapies advance, and as access to data about real-world use of those therapies grows, we must update our analytic methods accordingly. Randomized controlled trials are too costly and time-consuming to answer every treatment question. The clone-censor-weight approach bridges this evidence gap between clinical trials and real-world practice.