Optimal Threshold Time Selection in RMST Analysis for Comparative Clinical Studies

Overview

This study introduces a novel statistical method to determine the optimal threshold time (τ) in restricted mean survival time (RMST) analysis, enhancing the comparison of two groups in clinical and epidemiological research. Simulation studies demonstrate that this method achieves higher statistical power and controlled type I error rates compared to existing approaches. The method is exemplified in oncology and gerontology applications, illustrating its practical utility.

Background

Restricted mean survival time (RMST) analysis is a robust alternative to traditional survival analysis methods like the log-rank test and Cox proportional hazards regression, especially when the proportional hazards assumption is violated. RMST calculates the expected survival time up to a pre-specified threshold time (τ), allowing meaningful group comparisons over a fixed period. However, selecting an appropriate τ is challenging, as it significantly influences the analysis outcome, statistical power, and type I error rates. Existing approaches often rely on clinical knowledge or arbitrary choices, lacking a systematic method to optimize τ based on pilot data.

Data Highlights

Simulation studies showed that the proposed method for selecting τ led to higher statistical power and maintained appropriate type I error rates compared to traditional methods. The method was applied to two real-world datasets: (1) a non–small-cell lung cancer clinical trial comparing treatments based on PD-L1 biomarker status, and (2) a gerontology study assessing instrumental activities of daily living in dementia care recipients. In both cases, the method facilitated more informative and statistically robust comparisons.

Key Findings

• The choice of threshold time τ critically affects RMST analysis results, influencing power and error rates.
• The proposed method estimates an optimal τ by modeling survival distributions and identifying crossing points of survival curves when proportional hazards do not hold.
• Simulation results indicate that this method outperforms existing approaches by increasing power and controlling type I error.
• Application to oncology and gerontology datasets demonstrates practical utility in diverse clinical contexts.
• Using pilot or early-phase data to select τ enables better planning and analysis of future comparative studies.

Clinical Implications

Clinicians and researchers should consider employing the proposed method to select the RMST threshold time based on pilot data, especially when proportional hazards assumptions are questionable. This approach can improve the interpretability and statistical robustness of survival comparisons, aiding decision-making in clinical trials and observational studies. Incorporating this method may enhance the design and analysis of phase IIb/III trials by optimizing power and controlling error rates.

Conclusion

The novel method for determining the optimal threshold time in RMST analysis provides a statistically sound and practical tool for comparative survival studies. Its application can improve the accuracy and interpretability of treatment effect assessments across various clinical and epidemiological settings.

References

1. Huang and Kuan 2021 -- Impact of Threshold Time on RMST Analysis
2. Zhang et al. 2020 -- Inference Challenges in RMST with Inappropriate τ
3. Irwin 1949 -- Early Development of Restricted Mean Survival Time