Scholarlink

<p>The average treatment effect (ATE) is a canonical estimand in causal inference analysis. A key assumption for ATE estimation is the positivity requirement, ensuring that every study subject has a nonzero probability of receiving either treatment modality. The incremental propensity score intervention (IPSI) offers an alternative approach that bypasses this requirement. By shifting the observational propensity score (PS) by a fixed amount, IPSI assesses how this shift influences the mean outcome, allowing for the construction of an incremental effect curve over a range of shifts. However, existing IPSI methods are not applicable to survival outcomes subject to right-censoring. We apply the IPSI framework to survival data to evaluate causal effects of stochastic PS shifts on survival functions in a classical setting with time-fixed treatment. We introduce the incremental survival function surface (ISFS), which captures how survival probabilities evolve jointly over the shifting factor and time. Uniform confidence bands for the ISFS are constructed via bootstrap procedures. We further consider incremental derivative effects (IDEs), which quantify the instantaneous change in the survival function under infinitesimal shifts in treatment distribution, offering a more interpretable causal measure. We employ parametric models for estimation and uniform inference of both the ISFS and IDEs. We demonstrate the practical utility of the proposed method by comparing survival functions between peritoneal dialysis and hemodialysis groups, where elderly hemodialysis patients with comorbidities are unlikely to receive peritoneal dialysis. The uniform inference for the ISFS reveals non-uniform incremental effects, with the survival benefits of peritoneal dialysis diminishing at later follow-up times, as clinically expected. The proposed IPSI framework offers a valuable approach for assessing how shifts in the observational PS influence survival functions over time, providing new insights into treatment effects in observational studies with right-censored outcomes.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Estimating incremental propensity score effects on survival functions

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月17日

查看全文
  • Byeong Yeob Choi

摘要

The average treatment effect (ATE) is a canonical estimand in causal inference analysis. A key assumption for ATE estimation is the positivity requirement, ensuring that every study subject has a nonzero probability of receiving either treatment modality. The incremental propensity score intervention (IPSI) offers an alternative approach that bypasses this requirement. By shifting the observational propensity score (PS) by a fixed amount, IPSI assesses how this shift influences the mean outcome, allowing for the construction of an incremental effect curve over a range of shifts. However, existing IPSI methods are not applicable to survival outcomes subject to right-censoring. We apply the IPSI framework to survival data to evaluate causal effects of stochastic PS shifts on survival functions in a classical setting with time-fixed treatment. We introduce the incremental survival function surface (ISFS), which captures how survival probabilities evolve jointly over the shifting factor and time. Uniform confidence bands for the ISFS are constructed via bootstrap procedures. We further consider incremental derivative effects (IDEs), which quantify the instantaneous change in the survival function under infinitesimal shifts in treatment distribution, offering a more interpretable causal measure. We employ parametric models for estimation and uniform inference of both the ISFS and IDEs. We demonstrate the practical utility of the proposed method by comparing survival functions between peritoneal dialysis and hemodialysis groups, where elderly hemodialysis patients with comorbidities are unlikely to receive peritoneal dialysis. The uniform inference for the ISFS reveals non-uniform incremental effects, with the survival benefits of peritoneal dialysis diminishing at later follow-up times, as clinically expected. The proposed IPSI framework offers a valuable approach for assessing how shifts in the observational PS influence survival functions over time, providing new insights into treatment effects in observational studies with right-censored outcomes.