Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Contrastive Representation Learning Helps Cross-institutional Knowledge Transfer: A Study in Pediatric Ventilation Management

Yuxuan Liu, Jinpei Han, Padmanabhan Ramnarayan, A. Aldo Faisal

TL;DR

Clinical ML deployment across institutions is hindered by population and practice differences. The paper shows that contrastive predictive coding (CPC) pre-training on a large source PICU, followed by targeted fine-tuning on a smaller cardiac-focused unit, enables effective cross-institution knowledge transfer, especially in few-shot settings. A systematic evaluation across Target-Only, Full Fine-Tuning (FTF), and Decoder-only Fine-Tuning (FTD) reveals that CPC with full fine-tuning bridges performance gaps, while decoder-only fine-tuning underperforms; transfer is easier for temporal progression tasks than for point-of-care decisions, highlighting practical deployment pathways. The work provides a framework for generalizable clinical decision support and demonstrates how smaller, specialized units can leverage knowledge from larger centers to improve pediatric ventilation management.

Abstract

Clinical machine learning deployment across institutions faces significant challenges when patient populations and clinical practices differ substantially. We present a systematic framework for cross-institutional knowledge transfer in clinical time series, demonstrated through pediatric ventilation management between a general pediatric intensive care unit (PICU) and a cardiac-focused unit. Using contrastive predictive coding (CPC) for representation learning, we investigate how different data regimes and fine-tuning strategies affect knowledge transfer across institutional boundaries. Our results show that while direct model transfer performs poorly, CPC with appropriate fine-tuning enables effective knowledge sharing between institutions, with benefits particularly evident in limited data scenarios. Analysis of transfer patterns reveals an important asymmetry: temporal progression patterns transfer more readily than point-of-care decisions, suggesting practical pathways for cross-institutional deployment. Through a systematic evaluation of fine-tuning approaches and transfer patterns, our work provides insights for developing more generalizable clinical decision support systems while enabling smaller specialized units to leverage knowledge from larger centers.

Contrastive Representation Learning Helps Cross-institutional Knowledge Transfer: A Study in Pediatric Ventilation Management

TL;DR

Clinical ML deployment across institutions is hindered by population and practice differences. The paper shows that contrastive predictive coding (CPC) pre-training on a large source PICU, followed by targeted fine-tuning on a smaller cardiac-focused unit, enables effective cross-institution knowledge transfer, especially in few-shot settings. A systematic evaluation across Target-Only, Full Fine-Tuning (FTF), and Decoder-only Fine-Tuning (FTD) reveals that CPC with full fine-tuning bridges performance gaps, while decoder-only fine-tuning underperforms; transfer is easier for temporal progression tasks than for point-of-care decisions, highlighting practical deployment pathways. The work provides a framework for generalizable clinical decision support and demonstrates how smaller, specialized units can leverage knowledge from larger centers to improve pediatric ventilation management.

Abstract

Clinical machine learning deployment across institutions faces significant challenges when patient populations and clinical practices differ substantially. We present a systematic framework for cross-institutional knowledge transfer in clinical time series, demonstrated through pediatric ventilation management between a general pediatric intensive care unit (PICU) and a cardiac-focused unit. Using contrastive predictive coding (CPC) for representation learning, we investigate how different data regimes and fine-tuning strategies affect knowledge transfer across institutional boundaries. Our results show that while direct model transfer performs poorly, CPC with appropriate fine-tuning enables effective knowledge sharing between institutions, with benefits particularly evident in limited data scenarios. Analysis of transfer patterns reveals an important asymmetry: temporal progression patterns transfer more readily than point-of-care decisions, suggesting practical pathways for cross-institutional deployment. Through a systematic evaluation of fine-tuning approaches and transfer patterns, our work provides insights for developing more generalizable clinical decision support systems while enabling smaller specialized units to leverage knowledge from larger centers.
Paper Structure (12 sections, 4 equations, 2 figures, 4 tables)

This paper contains 12 sections, 4 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Problem Formulation
  4. Study Design
  5. Clinical Setting and Data Preparation
  6. Prediction Tasks
  7. Pre-training with Contrastive Learning
  8. CPC Architecture
  9. Clinical Domain Adaptation
  10. Implementation and Evaluation
  11. Results & Discussion
  12. Conclusion

Figures (2)

  • Figure 1: Overview of Cross-institutional Knowledge Transfer in Healthcare: From Large Source Data to Specialized Adaptation.
  • Figure 2: (A) Ventilation Time Series Data Types: Our dataset contains successful extubation, reintubation, failed extubation and censored data with unobservable outcomes. (B) Framework Architecture: Network $f_\theta$ Pre-training on source institution data using modified CPC, then fine-tunned with classification layer $h_\psi$ in downstream tasks.