Presentation (CTI)

OBJECTIVES: Real-world evidence shows that immunocompromised and other high-risk patients were disproportionately impacted by the COVID-19 pandemic (2020-2022). Thus, it has been hypothesized that trial outcomes may be affected by increased mortality during this period. Anti-cancer therapies and novel treatments for rare diseases are commonly evaluated in single-arm trials, where excess mortality would lower the treatment effect of an intervention when compared to trials unaffected by the pandemic. Methods such as naïve censoring of deaths due to COVID-19 may lead to informative censoring, thus biasing outcomes. Treatment switching methods exist to estimate true survival for patients who switch treatments upon disease progression. This research aims to explore whether these methods could also be used to estimate treatment effects in the absence of COVID-19.
METHODS: A review of existing statistical methods for treatment switching listed in NICE TSD DSU 16 was carried out to determine if they could be used to estimate patient survival in the absence of COVID-19, based on current understanding of COVID-19 and expected trial data availability, potentially providing a more unbiased estimate of treatment effect.
RESULTS: Most treatment switching methods, such as inverse probability of censoring weighting and rank preserving structural failure time models, assume that all confounders are measured at time of infection (treatment switch). Moreover, two-stage methods assume a link to disease-state. Neither assumption applies to COVID-19. However, expectation-maximization (EM) methods, which iteratively estimate expected patient survival, appear well-suited for estimating expected patient survival in the absence of COVID-19, as it relies on baseline characteristics to determine patient survival. This method was previously recommended for COVID-19 by De Felice (2023).
CONCLUSIONS: Most existing treatment switching methods rely on assumptions that are not applicable to COVID-19. However, EM methods may help address this issue, allowing for unbiased comparisons against trials unaffected by COVID-19.