[论文解读] Heterogeneous readmission prediction with hierarchical effect decomposition and regularization

Ziren Jiang, Lingfeng Huo+4 more|arXiv (Cornell University)|Mar 20, 2026

Machine Learning in Healthcare被引用 0

一句话总结

论文提出了 hierNest，一个分层重新参数化与正则化框架，用于在 MDC 内的 DRG 子组之间预测再入院风险，借力共享信息同时保持可解释性。

ABSTRACT

Accurately predicting hospital readmission risks using electronic health records (EHRs) is critical for effective patient management and healthcare resource allocation. Patient populations in health systems are highly heterogeneous across different primary diagnoses, necessitating tailored yet interpretable prediction models. We propose a hierarchical modeling framework incorporating hierarchical nested re-parameterization and structured regularization methods, which we call hierNest. Specifically, our approach leverages the inherent hierarchical structure present in primary diagnoses and groupings of these diagnoses into major diagnostic categories. Our methodology facilitates information borrowing across related patient subgroups and preserves interpretability at different hierarchical levels. Simulation studies demonstrate superior predictive accuracy of the proposed method, particularly with small subgroup sample sizes and varying degrees of hierarchical effects. We apply our methods to a large EHR dataset comprising Medicare patients.

研究动机与目标

• Motivate accurate readmission risk prediction within heterogeneous patient populations.
• Leverage the MDC-DRG hierarchy to borrow strength across related subgroups.
• Develop interpretable, subgroup-aware risk models for large, high-dimensional EHR covariates.
• Provide a computationally efficient framework suitable for large-scale clinical data.

提出的方法

• Re-parameterize DRG-specific effects as a sum of a common effect, an MDC-specific effect, and a DRG-specific effect.
• Regularize these hierarchical effects with lasso or overlapping group lasso penalties to encourage similarity within MDCs and DRGs.
• Use a modified design matrix X_H to encode hierNest decomposition for standard lasso software.
• Optionally employ an overlapping group lasso with a majorization-minimization algorithm for efficient computation.
• Tune penalties via cross-validation and ensure interpretability across hierarchical levels.

Figure 1: Illustration of the hierarchical nested structure of the DRGs and MDCs and how our parameter decomposition aligns with this structure.

实验结果

研究问题

• RQ1 Can hierarchical re-parameterization improve readmission prediction by borrowing strength across related DRGs within MDCs?
• RQ2 Do regularization schemes (lasso and overlapping group lasso) provide accurate, interpretable subgroup-specific effects?
• RQ3 How does hierNest perform in high-dimensional EHR settings with small subgroup sample sizes?
• RQ4 Is the approach computationally feasible for large-scale clinical datasets?

主要发现

• Simulation studies show superior predictive accuracy of hierNest, especially for small subgroup sizes and varying hierarchical effects.
• Applied to a Medicare dataset with over 60,000 observations and over 1 million parameters, the method runs within a few hours using 10-fold cross-validation.
• The framework maintains interpretability from global to MDC- and DRG-level covariate effects.
• Penalties enable adaptive borrowing of strength across DRGs within MDCs, collapsing effects when appropriate.

Figure 3: Average AUROC across all DRG subgroups for different methods. The sample size $n$ here represents the sample size for each DRG.