Composite Variable Strategies for Continuous Endpoints
Abstract
Composite endpoints are common in clinical trials for binary, time-to-event or recurrent outcomes. Recently, there has been interest in estimands for continuous endpoints that employ a composite variable strategy for handling intercurrent events, whereby a value on the continuous scale is defined as the outcome. In this paper we describe these continuous composite endpoints where participants who experience the intercurrent event are assigned a pre-specified value on the outcome scale (which we call the failure value) instead of a measurement. We describe these endpoints and the difficulty of choosing the failure value when the mean is the chosen summary statistic. The large impact of the choice of failure value on the variance of the distribution of the endpoint is quantified as a function of the probability of the event and the standardised difference of the failure value from the mean of the unaffected values. Strategies for choosing failure values are discussed, emphasising the need to balance the clinical relevance and interpretability of the estimand with the need to reliably estimate it. Two potential solutions to situations where there is no relevant and viable choice of failure value are proposed; using failure values based on distributional characteristics to generate clinical relevance and using alternative summary measures to avoid the need to select specific values.
Related
- Interpreting Hierarchical Composite Endpoints with Survival and Longitudinal Outcomes: Application to Amyotrophic Lateral Sclerosis Trials
- The problem of continuous endpoints for clinical decision making: a decision curve analysis approach
- Reasoning about Continuous-Variable Quantum Systems
- Classical methods for modern applications: synthetic tabular data generation using lightweight discrete response continuous latent variable model
- Optimal Nonparametric Estimation of Phase-Space Representations for Non-Gaussian Continuous Variable Quantum States
BibTeX
@misc{bell2026composite,
title = {{Composite Variable Strategies for Continuous Endpoints}},
author = {James Bell and Oliver N. Keene and Christian Bressen Pipper and Thomas Drury and Lorenzo Guizzaro and Marian Mitroiu and Tobias Mütze and Khadija Rerhou Rantell and Marcel Wolbers and David Wright},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.28265},
eprint = {2610.28265},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.28265},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - UNPB
TI - Composite Variable Strategies for Continuous Endpoints
AU - James Bell
AU - Oliver N. Keene
AU - Christian Bressen Pipper
AU - Thomas Drury
AU - Lorenzo Guizzaro
AU - Marian Mitroiu
AU - Tobias Mütze
AU - Khadija Rerhou Rantell
AU - Marcel Wolbers
AU - David Wright
PY - 2026
DA - 2026/10/20
PB - arXiv
DO - 10.48550/arxiv.2610.28265
UR - https://arxiv.org/abs/2610.28265
AB - Composite endpoints are common in clinical trials for binary, time-to-event or recurrent outcomes. Recently, there has been interest in estimands for continuous endpoints that employ a composite variable strategy for handling intercurrent events, whereby a value on the continuous scale is defined as the outcome. In this paper we describe these continuous composite endpoints where participants who experience the intercurrent event are assigned a pre-specified value on the outcome scale (which we call the failure value) instead of a measurement. We describe these endpoints and the difficulty of choosing the failure value when the mean is the chosen summary statistic. The large impact of the choice of failure value on the variance of the distribution of the endpoint is quantified as a function of the probability of the event and the standardised difference of the failure value from the mean of the unaffected values. Strategies for choosing failure values are discussed, emphasising the need to balance the clinical relevance and interpretability of the estimand with the need to reliably estimate it. Two potential solutions to situations where there is no relevant and viable choice of failure value are proposed; using failure values based on distributional characteristics to generate clinical relevance and using alternative summary measures to avoid the need to select specific values.
ER -