Standardized survival probabilities and contrasts between hierarchical units in multilevel survival models

多层生存模型中各层级单元的标准化生存概率和对比

阅读:2

Abstract

BACKGROUND: In the medical literature, in which time-to-event (such as time to death or disease recurrence) outcomes are commonly studied, hierarchical data is frequently encountered with patients nested within hospitals or regions. Multilevel hierarchical mixed-effects survival models are routinely used in these settings to accommodate the correlation between study subjects belonging to the same cluster and any potential unobserved heterogeneity. However, these analyses usually focus on fixed effects while marginalizing over the random effects, with fully conditional or marginal (on the random effects) post-estimation predictions. METHODS: In this work, we combine regression standardization over the observed covariates with posterior predictions of the random effects to obtain standardized survival probabilities. These predictions quantify how the entire study population would have fared under the performance of each cluster and can be used to obtain fair comparisons between hierarchical units. Compared to other common approaches, such as the median hazard ratio, this proposal yields quantities that are easier to interpret and that, under certain assumptions, can have a causal interpretation. RESULTS: We illustrate the new methodology in practice using data on bladder cancer patients with a three-level hierarchical structure composed of patients nested within surgeons and centers. Then, we illustrate a variety of standardized survival predictions benchmarking, e.g., best/average/worst surgeons and centers, surgeons within a center, or centers directly — all from a single and unified modeling framework. CONCLUSIONS: We introduced an analytical approach that can be used to quantify and fairly compare the differences between hierarchical units using easily interpretable measures such as (standardized) survival probabilities and contrasts thereof. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12874-026-02782-8.

特别声明

1、本页面内容包含部分的内容是基于公开信息的合理引用;引用内容仅为补充信息,不代表本站立场。

2、若认为本页面引用内容涉及侵权,请及时与本站联系,我们将第一时间处理。

3、其他媒体/个人如需使用本页面原创内容,需注明“来源:[生知库]”并获得授权;使用引用内容的,需自行联系原作者获得许可。

4、投稿及合作请联系:info@biocloudy.com。