Unanchored simulated treatment comparisons (STCs) are a valuable tool for manufacturers navigating the health technology assessment (HTA) landscape. When head-to-head clinical trials are unavailable, STCs allow for population-adjusted indirect comparisons between a single-arm trial and an external control arm.

Using regression modeling to predict outcomes based on patient characteristics, STCs enable comparisons in the absence of a common comparator. This is particularly valuable when evaluating novel therapies, especially in rare or specialized disease areas where randomized controlled trials may be limited.

The Challenge of Censoring in STCs

Recently, we had the opportunity to conduct an unanchored STC for a client in an oncology indication, using their single-arm trial data to bolster the evidence for their HTA submission. However, in the process, we identified a key challenge — the absence of formal guidance on how to appropriately simulate censoring for the time-to-event outcomes in the simulated comparator arm.

Censoring assumptions can significantly influence survival estimates; for example, censoring early in a trial can decrease the observed median survival time due to estimates based on fewer events, potentially leading to less reliable results. Conversely, censoring towards the end of a trial may artificially inflate the median survival time, possibly overestimating survival probabilities as individuals who might have experienced the event get censored prematurely. Recognizing the potential impact of censoring assumptions on the STC results, we decided to undertake a methodological study to explore this issue in depth. Our goal was to provide much-needed guidance to researchers and HTA bodies on best practices for handling censoring in unanchored STCs.

Impact of Censoring Approaches

In our study, we examined the influence of various censoring assumptions on the results of unanchored STCs across three oncology indications. Using digitized pseudo-IPD, we conducted STCs with extreme differences in baseline characteristics between the single-arm trial and external comparator to highlight the impact of the adjustment while ensuring these were nonsignificant treatment effect modifiers.

Our findings demonstrated that the choice of censoring approach can significantly impact the estimated hazard ratios and median overall survival.

Applying different censoring assumptions—including:

• censoring at the end of the trial,
• censoring evenly across the time,
• censoring centered around the middle of the trial,
• predominantly early or late censoring,
• and censoring based on trial-specific probabilities —
led to notable variability in the STC-adjusted results compared to the true data.

Key Learnings

Our study identified one standout approach that provided the closest approximation to the true value — with the median survival aligning most closely with the original unadjusted estimate, thereby minimizing prediction error. These results underscore the importance of carefully considering censoring assumptions when implementing unanchored STCs. While the NICE Decision Support Unit Technical Support Document 18 provides guidance on conducting STCs, it does not specifically address the simulation of censoring for the comparator arm. Our study identified the most effective approach for censoring in the simulation of survival data for STCs.

Final takeaway

As the use of unanchored comparisons continues to grow in health technology assessment and regulatory submissions, researchers must have a clear understanding of the potential pitfalls and best practices for implementing these methods. We identified a need for further research and guidance on handling censoring in unanchored STCs to ensure the validity and reliability of the results.
While simulated treatment comparisons offer a valuable solution for making population-adjusted indirect comparisons, the choice of censoring approach can significantly impact the outcomes. From our findings, we determined one approach that notably minimized the prediction error compared to the others. As the field continues to evolve, further research and consensus on best practices will be essential to support robust and reliable unanchored indirect comparisons.

We plan to present our results at ISPOR Europe 2024. Meet the Cytel team at booth 1500!

Share Via