Bootstrapping-based Regularisation for Reducing Individual Prediction Instability in Clinical Risk Prediction Models

2026-02-13 · Source: stat.ML updates on arXiv.org · Field: Health & Wellbeing — Health & Medical Research, Clinical Care & Medical Practice, Medical Devices & Health Technology · Depth: Advanced, extended

Summary

A novel bootstrapping-based regularisation framework has been developed to enhance the individual prediction stability of deep learning clinical risk models. This method embeds the bootstrapping process directly into the training of deep neural networks, constraining prediction variability across resampled datasets to produce a single, stable model. Evaluated against conventional and ensemble models using simulated data and three clinical datasets (GUSTO-I, Framingham, SUPPORT), the proposed approach consistently demonstrated improved prediction stability with lower mean absolute differences (e.g., 0.019 vs. 0.059 in GUSTO-I; 0.057 vs. 0.088 in Framingham) and significantly fewer deviating predictions. Crucially, it maintained discriminative performance (AUC) and feature importance consistency (high SHAP correlations, e.g., 0.894 for GUSTO-I), addressing the interpretability issues often associated with ensemble methods.

Key takeaway

Research scientists developing clinical prediction models should consider implementing this bootstrapping-based regularisation framework. This approach offers a practical route to achieve greater robustness and reproducibility in deep learning models, particularly in data-limited healthcare settings, without compromising the interpretability essential for clinical trust and adoption. You can achieve ensemble-like stability within a single, explainable model.

Key insights

A novel regularisation method improves clinical prediction model stability without sacrificing interpretability, crucial for healthcare adoption.

Principles

• Prediction instability undermines clinical reliability.
• Ensemble methods improve stability but reduce interpretability.
• Regularisation can embed ensemble-like stability into a single model.

Method

The method introduces a penalised likelihood function that includes a regularisation term measuring the expected difference in predictions between the target model and pre-computed bootstrapped models, balancing data fit with diversity.

In practice

• Use bootstrapping-based regularisation for stable clinical models.
• Pre-compute bootstrapped models to reduce computational burden.
• Tune regularisation strength (λ) and bootstrap samples (M) for optimal balance.

Topics

• Bootstrapping Regularisation
• Clinical Prediction Models
• Prediction Instability
• Deep Neural Networks
• Model Interpretability

Best for: Research Scientist, AI Researcher, AI Scientist, AI Data Scientist

Related on AIssential

• Regularisation in Neural Networks — Regularization techniques are crucial for neural networks to prevent overfitting and improve generalization on unseen data.
• A Reproducible Log-Driven AutoML Framework for Interpretable Pipeline Optimization in Healthcare Risk Prediction — Reproducible, log-driven AutoML reveals healthcare risk prediction pipeline component importance and search space redundancy.
• The limits of interpretability in multiple linear regression — Multicollinearity fundamentally limits multiple linear regression interpretability by destabilizing feature weights.
• Observable Neural ODEs for Identifiable Causal Forecasting in Continuous Time — Observability of latent states is crucial for identifying dynamic treatment effects in continuous-time causal forecasting with hidden confounders.
• ADAPTive Input Training for Many-to-One Pre-Training on Time-Series Classification — ADAPT enables mixed-batch pre-training for time-series data by aligning physical properties across diverse datasets.
• An interpretable and trustworthy AI framework for large-scale longitudinal structure-pain association studies using data from the Osteoarthritis Initiative (OAI) — Combining uncertainty-aware deep learning with longitudinal statistical modeling improves osteoarthritis structure-pain association studies.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by stat.ML updates on arXiv.org.