Clinical Scorecard: Establishing Optimal Threshold Time in Restricted Mean Survival Time Analysis for Comparative Studies in Clinical and Epidemiological Research

At a Glance

Key Highlights

• RMST analysis is a robust alternative to logrank test and Cox regression, especially when proportional hazards assumption is violated.
• Choice of threshold time (τ) critically impacts RMST comparison results, affecting statistical power and type I error rates.
• A novel statistical method is proposed to determine an optimal threshold time based on pilot data, improving treatment group comparisons.

Guideline-Based Recommendations

Diagnosis

• Use RMST analysis for time-to-event data, particularly when proportional hazards assumption may not hold.
• Estimate survival functions for each group to identify potential crossing points of survival curves.

Management

• Select threshold time τ for RMST as the crossing time of survival curves when curves cross, to maximize statistical power.
• When survival curves diverge, choose the largest possible τ to capture maximal difference in restricted mean survival.
• Incorporate pilot or early phase data to inform threshold time selection rather than relying solely on clinical knowledge or maximum observed event times.

Monitoring & Follow-up

• Monitor statistical power and type I error rates when applying RMST analysis with chosen threshold time.
• Validate threshold time choice through simulation studies or pilot data to ensure robust inference.

Risks

• Inappropriate selection of threshold time can lead to invalid inference, reduced power, or inflated type I error.
• Changing threshold time after study commencement is discouraged to avoid bias.

Patient & Prescribing Data

Patients enrolled in clinical trials or observational studies with time-to-event endpoints, including oncology and gerontology populations.

Optimal threshold time selection in RMST analysis enhances detection of treatment effects and improves comparative study design.

Clinical Best Practices

• Do not rely solely on maximum observed event times or clinical knowledge to select threshold time; use pilot data when available.
• Model survival distributions to identify crossing points for threshold time selection in non-proportional hazards scenarios.
• Maintain fixed threshold time throughout study to preserve validity of statistical inference.
• Use RMST analysis as a complementary or alternative method to logrank and Cox regression tests.

References

• Restricted Mean Survival Time in Clinical Trials
• Huang and Kuan on Threshold Time Impact
• Zhang et al. on RMST Threshold Selection