Correlation Matters! Streamlining the Sample Size Procedure with Hierarchical Time-to-event Endpoints
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Statistics. The day in statistics: methods, theory and the inference under them.
Tom: We're looking at the paper: "Correlation Matters! Streamlining the Sample Size Procedure with Hierarchical Time-to-event Endpoints".
Jane: Here's the gist of the paper. Composite endpoints are widely used in cardiovascular clinical trials to improve statistical efficiency while preserving clinical relevance The Win Ratio (WR) measure and more general frameworks of Win Statistics (WS) have emerged as increasingly popular alternatives to traditional time-to-first-event analyses To address this challenge, the authors derive refined formulas to calculate the proportions of wins, losses, and ties for multiple prioritized time-to-event endpoints These formulas rely on familiar design inputs and become directly applicable when integrated with existing sample size methods The paper conducts a comprehensive assessment of how correlation among endpoints affects sample size requirements across varying design features Pruning redundant reasoning tokens The primary motivation for the research is to provide practical tools that make the impact of correlation transparent at the design stage The authors address a fundamental tension in trial design where prioritized composite time-to-event endpoints have clear interpretability advantages, but translating clinical intuition into formal sample size calculations remains difficult They propose a streamlined, formula-based procedure for sample size and power calculation based on the YG formula This procedure begins by computing the probabilities of wins, losses, and ties under common clinical design settings using integral-based expressions that accommodate potential correlation through a copula framework How the training signal is built
Title and authors: Tom: So we're talking about this paper titled "Correlation Matters! Streamlining the Sample Size Procedure with Hierarchical Time-to-event Endpoints." It sounds really technical, but at its core, it's about making the math for clinical trials with multiple outcomes a lot more practical.
Jane: Right, so it addresses how we figure out how many patients we need in a trial when you have several different time-to-event endpoints that are all related. It’s about simplifying that process for people who aren't deep into the math.
Lu: Exactly, it tackles the difficulty of translating clinical intuition about correlated endpoints into actual sample size calculations. It suggests using a streamlined, formula-based procedure instead of relying on older methods that might be too complex to use in practice.
Meng: What does that streamline actually mean for us when we're thinking about running a trial? Does it just make the calculation faster, or does it actually give us better results because of how correlation is handled?
Lalam: I think the main point is they propose using something called the YG formula to calculate the probabilities of winning, losing, and tying for multiple endpoints. This lets them build in a way to account for that correlation directly from the start.
The paper's summary: Tom: So what’s this new approach? Basically, they are using these integral-based expressions within a copula framework to figure out the probabilities of wins, losses, and ties under various clinical settings like specific hazard ratios or dropout patterns.
Jane: That sounds complicated. Can you put that in terms a listener who is just trying to understand the concept? It’s about modeling how these different outcomes interact when they are correlated.
Lu: Yeah, it’s about taking those probabilities—the wins, the losses, and the ties—and plugging them into this YG formula to get a sample size and power estimate. They aren't just calculating one thing; they are calculating it for several endpoints simultaneously in a structured way.
Meng: So instead of running separate, complicated simulations for every possible combination of correlation structures, they have this single procedure that takes the correlation information and feeds it into the formula to get a result. That sounds like something you could actually use in planning.
Lalam: It’s about building these probabilities first using the copula framework, which lets them handle that correlation structure, and then feeding those into the YG formula for sample size calculation based on common clinical design settings like specified hazard ratios or dropout patterns.
The paper's improvements: Tom: So what are they actually improving? The paper claims this streamlined procedure lets them conduct a comprehensive investigation into how correlation among endpoints influences needed sample sizes when you change things like the magnitude of treatment effects or the follow-up duration.
Jane: That means they aren't just giving you one number; they’re showing you *when* correlation matters most. They want to tell researchers precisely which design features, like how long patients stay in the study, cause the biggest swings in sample size requirements because of that correlation.
Lu: They are essentially creating a way to quantify the impact of correlation on trial efficiency at the design stage, which is huge because it lets you see how sensitive your sample size calculation is to changes in those factors.
Meng: I’m interested in that part about evaluating the trend under varying magnitudes of treatment effects and follow-up durations. It tells us if we should be more conservative when our treatment effect is small or when patients are followed for a long time, which directly affects how many patients we need to enroll.
Lalam: They also propose this as a way to adapt other Win Statistics measures, like net benefit or win odds, by first calculating the probabilities of winning, losing, and tying in one stage and then applying sample size formulas derived as transformations of that YG formula.
Conclusion: Tom: So to wrap up on "Correlation Matters! Streamlining the Sample Size Procedure with Hierarchical Time-to-event Endpoints," the core idea is providing a practical, formula-based procedure using the YG formula to handle correlated endpoints in clinical trials.
Jane: They’ve shown that by using this method, you can investigate exactly how correlation affects sample size needs across different treatment effects and follow-up durations, which gives trial designers much more concrete guidance.
Lu: It shifts the focus from just calculating a single number to understanding the sensitivity of that calculation to various design choices and treatment effects when correlation is present.
Meng: For us at the startup side, this means we can build tools that give us a better estimate of what’s needed without having to run massive, slow simulations for every possible correlation scenario.
Lalam: I think the implication is that we can move from abstract ideas about composite endpoints to a concrete, scalable method for planning trials that account for correlation upfront.
Yunhan Mou, Fan Li, Denise Esserman, Yuan Huang
Yale School of Public Health
stat.ME
Submitted: 2025-11-20
Updated: 2026-10-31
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 83/100
The gist: Composite endpoints are widely used in cardiovascular clinical trials to improve statistical efficiency while preserving clinical relevance The Win Ratio (WR) measure and more general frameworks of…
Key concepts
- Composite Endpoints
- These are multiple time-to-event measurements used together in a trial, often for cardiovascular studies. They help improve statistical efficiency by combining information from several related outcomes into one analysis.
- Win Ratio (WR) / Win Statistics (WS)
- These are modern ways to measure success in trials instead of just looking at the time until an event occurs. The paper uses these frameworks to calculate proportions of wins, losses, and ties across different endpoints.
- Copula Framework
- This mathematical tool is used within the formulas to model the relationship or correlation between different time-to-event endpoints. It allows researchers to incorporate how outcomes are linked without needing complex assumptions about their individual distributions.
Terminology
Summary
Composite endpoints are widely used in cardiovascular clinical trials to improve statistical efficiency while preserving clinical relevance The Win Ratio (WR) measure and more general frameworks of Win Statistics have emerged as increasingly popular alternatives to traditional time-to-first-event analyses To address this challenge, the authors derive refined formulas to calculate the proportions of wins, losses, and ties for multiple prioritized time-to-event endpoints They conduct a comprehensive assessment of how correlation among endpoints affects sample size requirements across varying design features
Streamlining the Procedure
The authors propose a streamlined, formula-based procedure for sample size and power calculation for clinical trials with multiple prioritized time-to-event endpoints based on the YG formula. This procedure begins by computing the probabilities of wins, losses, and ties under common clinical design settings using integral-based expressions that accommodate potential correlation among multiple prioritized time-to-event endpoints through a copula framework These quantities are then incorporated into the YG formula for sample size and power estimation. p. 2
Estimation of Win, Loss and Tie Probabilities
The authors define pairwise comparisons hierarchically across prioritized endpoints based on event times under common clinical design settings The win probability at the first endpoint can be expressed using an integral expression involving the shared underlying censoring-free probability function and hazard functions. The overall WR and probability of ties are obtained as specific combinations of these calculated probabilities. p. 5
Influence of Correlation for Study Planning
The numerical studies investigate how correlation among endpoints influences needed sample sizes across varying magnitudes of treatment effects, follow-up durations, and other practical design factors They observe that when the treatment effect is stronger on the first endpoint, increasing correlation tends to require a larger sample size. The relative change rate (RCR) measure is employed to quantify this rate of change in required sample sizes as correlation varies
Case Studies and Practical Implications
The authors examine two landmark cardiovascular studies, SPRINT and STICH, to illustrate the application of the proposed streamlined procedure in real-world contexts In the SPRINT trial analysis, required sample sizes show a monotonic increase as correlation between endpoints strengthens For STICH, required sample sizes exhibit a modest non-monotonic pattern with respect to the correlation level The discussion suggests that performing a supplementary simulation-based power assessment using the estimated sample size as a validation step is advisable in practical trial design
Discussion and Future Directions
It is noted that the YG formula generally yields power estimates comparable to those obtained from simulation-based methods, though it may be moderately conservative in certain scenarios It is recommended to construct a sample size table across a range of correlation levels, such as from 0 to 0.9, to obtain a broader understanding of how varying degrees of dependence may affect sample size requirements The proposed procedure is readily applicable to other commonly used WS measures like the net benefit (NB) and win odds (WO) by transforming the YG formula.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Improved Sample Size Determination for Composite Endpoints: The system can now calculate required sample sizes for cardiovascular clinical trials using a streamlined procedure based on the YG formula, which incorporates probabilities of wins, losses, and ties derived from integral-based expressions that accommodate potential correlation through a copula framework.
-
Correlation Impact Assessment: The AI system can quantify how correlation among time-to-event endpoints affects sample size requirements by calculating relative change rates (RCR) across varying magnitudes of treatment effects and follow-up durations, providing specific guidance on when correlation exerts the greatest influence on trial efficiency.
-
Real-World Data Application: The system can leverage case studies from landmark trials like SPRINT and STICH to estimate necessary design parameters from observed data, allowing it to generate sample size tables that account for empirically estimated correlations, thus making planning more robust for future clinical trials.
-
Validation of Analytic Formulas: The AI system can perform a cross-checking step by comparing the estimates obtained from the proposed formula-based approach with results from conventional simulation-based methods to validate the accuracy of its analytic calculations, especially in scenarios with high overall probability of ties where conservatism is noted.
-
Flexible Win Measure Adaptation: The system can be extended to incorporate other Win Statistics measures, such as net benefit or win odds, by computing the necessary probabilities of winning, losing, and tying in a first stage and then applying corresponding sample size formulas derived as transformations of the YG formula.
Sources
Related papers
- Learning Robust Treatment Rules for Censored Data
- Neural Estimation of Summary Statistics for Simulation-Based Inference with Non-Asymptotic Guarantees
- Spectral divide-and-conquer MCMC for long stationary time series
- Improving the Efficiency of Subgroup Analysis in Randomized Controlled Trials with TMLE
- Personalized Federated Vector Autoregression with Personalization Diversity
- Doubly Robust and Efficient Calibration of Prediction Sets for Right-Censored Time-to-Event Outcomes