Using Bayesian Sequential Logistic Regression to Model the Organ Transplant Selection Process
Abstract
Organ transplantation evaluation is a multi-stage process in which only patients who complete one stage remain eligible for the next, leading to patient attrition and reducing later-stage sample sizes. Covariate associations may differ across decision points, and identifying these stage-specific effects is necessary to target interventions and understand disparities. Conventional approaches that model each transition independently fail to distinguish stage-specific from shared associations and tend to produce imprecise estimates at later stages. We propose a Bayesian sequential logistic regression (BSLR) framework that represents progression as conditional Bernoulli distributions within a joint likelihood that preserves chronological ordering; we further show its equivalence to an adjusted continuation-ratio model. A hierarchical prior decomposes each covariate effect into an overall effect and a transition-specific deviation. This partial-pooling structure shares information across stages and stabilizes estimation where data are sparse, while a masking mechanism accommodates covariates structurally unavailable at particular stages without discarding records or imputing values. In simulations, BSLR achieved lower bias, posterior variance, and mean squared error than independent logistic regression. We demonstrated the utility of our model using a cohort of patients referred for kidney or liver transplant. Posterior estimates indicated that the associations of clinical indicators and social determinants of health with progression differed across care continuum phases. BSLR identified stage-specific heterogeneity in covariate associations, revealing factors whose effects reversed direction across evaluation stages and detecting associations at late-stage transitions that independent models failed to resolve due to insufficient precision.
Related
- SGD in Multiclass Logistic Regression: Sequential Learning and Scaling Laws
- Logistic Gaussian process density regression: a generalized Bayesian approach
- Bayesian Deck-of-cards-based Ordinal Regression with Sequential Preference Elicitation
- SelectPFN: Amortized Bayesian Variable Selection for Sparse Nonlinear Regression with Prior-Data Fitted Networks
- Data generation for the logistic regression model under controlled conditions
BibTeX
@misc{zhu2026using,
title = {{Using Bayesian Sequential Logistic Regression to Model the Organ Transplant Selection Process}},
author = {Xukun Zhu and Tyler Schappe and Ursula Rogers and Roland Matsouaka and Jacqueline B. Henson and Norine W. Chan and Goni Katz-Greenberg and Lisa M. McElroy and Samuel I. Berchuck},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.25744},
eprint = {2610.25744},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.25744},
}
RIS
TY - UNPB
TI - Using Bayesian Sequential Logistic Regression to Model the Organ Transplant Selection Process
AU - Xukun Zhu
AU - Tyler Schappe
AU - Ursula Rogers
AU - Roland Matsouaka
AU - Jacqueline B. Henson
AU - Norine W. Chan
AU - Goni Katz-Greenberg
AU - Lisa M. McElroy
AU - Samuel I. Berchuck
PY - 2026
DA - 2026/10/19
PB - arXiv
DO - 10.48550/arxiv.2610.25744
UR - https://arxiv.org/abs/2610.25744
AB - Organ transplantation evaluation is a multi-stage process in which only patients who complete one stage remain eligible for the next, leading to patient attrition and reducing later-stage sample sizes. Covariate associations may differ across decision points, and identifying these stage-specific effects is necessary to target interventions and understand disparities. Conventional approaches that model each transition independently fail to distinguish stage-specific from shared associations and tend to produce imprecise estimates at later stages. We propose a Bayesian sequential logistic regression (BSLR) framework that represents progression as conditional Bernoulli distributions within a joint likelihood that preserves chronological ordering; we further show its equivalence to an adjusted continuation-ratio model. A hierarchical prior decomposes each covariate effect into an overall effect and a transition-specific deviation. This partial-pooling structure shares information across stages and stabilizes estimation where data are sparse, while a masking mechanism accommodates covariates structurally unavailable at particular stages without discarding records or imputing values. In simulations, BSLR achieved lower bias, posterior variance, and mean squared error than independent logistic regression. We demonstrated the utility of our model using a cohort of patients referred for kidney or liver transplant. Posterior estimates indicated that the associations of clinical indicators and social determinants of health with progression differed across care continuum phases. BSLR identified stage-specific heterogeneity in covariate associations, revealing factors whose effects reversed direction across evaluation stages and detecting associations at late-stage transitions that independent models failed to resolve due to insufficient precision.
ER -