Exploring Causal Effects of Hormone‐ and Radio‐Treatments in an Observational Study of Breast Cancer Using Copula‐Based Semi‐Competing Risks Models

Tonghui Yu, Mengjiao Peng, Yifan Cui, Elynn Chen, Chixiang Chen

TL;DR

This paper introduces a frequentist, semi-parametric framework using copula-based models to estimate the causal effects of treatments in the presence of "semi-competing risks" (e.g., cancer relapse vs. death). By accounting for the dependency between these events and offering a robust sensitivity analysis for unmeasured factors, the authors provide a more reliable way to evaluate how hormone and radiotherapies impact patient outcomes over time.

What Problem Does This Paper Address?

In clinical studies for breast cancer, researchers typically track two major milestones: Relapse-Free Survival (RFS) and Overall Survival (OS). However, analyzing these together is tricky because they are "semi-competing risks."

While a relapse doesn't prevent death, death does prevent us from ever observing a relapse. This creates a "dependent censoring" problem. If we ignore the fact that the patients who die might have been the ones most likely to relapse, our estimates of treatment effectiveness become biased.

Existing methods often struggle to capture the complex "net" effect of a treatment—that is, the effect on relapse if death were theoretically removed from the equation. Furthermore, applying rigorous causal inference (determining if the treatment caused the outcome rather than just being correlated with it) to these types of survival data is still a developing field.

Methods

The authors propose a framework built on three pillars:

  1. Copula-Based Joint Modeling: Instead of looking at RFS and OS in isolation, they use Archimedean copulas (specifically the Frank copula) to model the joint distribution of the two events. This allows them to quantify the "net" treatment effects while accounting for the inherent association between relapsing and dying.
  2. Principal Stratification: To ensure "apples-to-apples" comparisons, they use the concept of Survivor Average Causal Effect (SACE). They focus on specific strata of patients, such as the "Always-Diseased" (those who would relapse regardless of treatment) and "Never-Diseased," to provide clinically meaningful interpretations.
  3. Advanced Estimation Techniques: They enhance standard Nonparametric Maximum Likelihood Estimation (NPMLE) by using better initial values to avoid local optima. For scenarios involving unmeasured confounding, they implement a modified Monte Carlo Expectation-Maximization (MCEM) algorithm to handle "frailty" variables (latent factors that affect both events).

Datasets

The researchers validated their model using:

  • METABRIC Cohort: A well-known dataset from the Molecular Taxonomy of Breast Cancer International Consortium. They analyzed 1,114 patients who underwent mastectomies, focusing on subsets comparing hormone therapy vs. no treatment, and combined hormone+radio therapy vs. hormone therapy alone.
  • Extensive Simulations: Four distinct scenarios (Ex1-Ex4) were used to test the model's robustness against different censoring levels, varying numbers of covariates, and the presence of unmeasured confounding factors.

Key Insights and Ideas

  • Net Quantities Matter: By focusing on "net" effects rather than "crude" rates, the model provides a "purified" view of how a drug slows down cancer progression, independent of other causes of death.
  • The "Always-Diseased" Perspective: Using principal stratification helps clinicians understand treatment benefits for specific risk groups. For instance, they found that combined therapy might be more beneficial specifically for patients in the "Always-Diseased" stratum (those at higher risk of relapse).
  • Sensitivity to the Unseen: The inclusion of a frailty-based sensitivity analysis is a major contribution. It allows researchers to ask: "How much would an unmeasured factor have to change my data to flip my conclusion?" This adds a layer of honesty and rigor to observational studies.
  • Copula Flexibility: Unlike traditional frailty models that often struggle to describe negative correlations between event times, copulas can simultaneously model marginal distributions and complex associations.

Key Results

  • Clinical Findings: In the METABRIC study, hormone therapy alone showed a significant positive impact on survival for patients with a low risk of relapse. For those at higher risk, the combination of hormone and radiotherapy showed a detectable (though weaker) benefit for overall survival compared to hormone therapy alone.
  • Statistical Reliability: Simulations showed that the proposed estimators have minimal bias and that their 95% confidence intervals are highly accurate (coverage probabilities near 0.95), even in high-censoring environments.
  • Comparison to Standard Methods: The authors demonstrated that traditional Kaplan-Meier analysis—which ignores the dependence between relapse and death—can lead to misleading survival curves that contradict clinical expectations. Their copula-based approach yielded more conservative and reliable results.

Limitations and Future Work

  • Copula Selection: The current study uses a relatively simple copula with a homogeneous association. The authors suggest that future versions could incorporate time-dependent or covariate-dependent associations to better reflect biological realities.
  • Multiple Treatments: The framework is currently designed for binary comparisons (Treatment A vs. B). Extending this to handle multiple simultaneous interventions or continuous dosages is a logical next step.
  • Identifying Latent Strata: In practice, it remains difficult to identify exactly which stratum (e.g., "Always-Diseased") a patient belongs to. The authors highlight the need for better "discrimination measures" to help clinicians apply these insights to individual patients.

Summary updated

Abstract

Breast cancer patients may experience relapse or death after surgery during the follow-up period, leading to dependent censoring of relapse. This phenomenon, known as semi-competing risk, imposes challenges in analyzing treatment effects on breast cancer and necessitates advanced statistical tools for unbiased analysis. Despite progress in estimation and inference within semi-competing risks regression, its application to causal inference is still in its early stages. This article aims to propose a frequentist and semi-parametric framework based on copula models that can facilitate valid causal inference, net quantity estimation and interpretation, and sensitivity analysis for unmeasured factors under right-censored semi-competing risks data. We also propose novel procedures to enhance parameter estimation and its applicability in practice. After that, we apply the proposed framework to a breast cancer study and detect the time-varying causal effects of hormone- and radio-treatments on patients' relapse and overall survival. Moreover, extensive numerical evaluations demonstrate the method's feasibility, highlighting minimal estimation bias and reliable statistical inference.
BibTeX
@inproceedings{yu2025exploring,
  author = {Tonghui Yu and Mengjiao Peng and Yifan Cui and Elynn Chen and Chixiang Chen},
  title = {{Exploring Causal Effects of Hormone‐ and Radio‐Treatments in an Observational Study of Breast Cancer Using Copula‐Based Semi‐Competing Risks Models}},
  booktitle = {Statistics},
  year = {2025},
  doi = {10.1002/sim.70131},
}

Loading related content...

Figure