Enhancing Inference for Small Cohorts via Transfer Learning and Weighted Integration of Multiple Datasets

2025-05-11

AI Summary

Clinical variable effect inference (e.g., FiO₂, creatinine, platelets, lactate) for pulmonary sepsis patients in Northeast China is severely hampered by small sample sizes, poor representativeness of regional data in the eICU database, and gender heterogeneity. Method: We propose TRANSLATE, a weighted transfer learning framework that jointly models domain-specific weights and effective sample size to enable adaptive alignment and differential down-weighting of external data—guaranteeing theoretical estimation accuracy. It integrates covariate shift correction, distribution function estimation, and weighted ensemble learning to support mean, variance, and full-distribution inference. Contribution/Results: Evaluated on synthetic and real-world Northeast Chinese cohorts, TRANSLATE significantly improves inference accuracy for key biomarkers while robustly mitigating regional and gender heterogeneity—marking the first method to formally couple domain weighting with effective sample size in clinical transfer learning.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Search and Optimization: Sampling/Simulation-based Search
• Reasoning under Uncertainty: Probabilistic Inference

Application Category

• User Modeling, Personalization and Recommendation: Accountability, Transparency, and Ethics for personalization
• Security and Privacy: Data transparency and provenance
• Responsible Web: Algorithmic accountability and transparency on the web

Abstract

Lung sepsis remains a significant concern in the Northeastern U.S., yet the national eICU Collaborative Database includes only a small number of patients from this region, highlighting underrepresentation. Understanding clinical variables such as FiO2, creatinine, platelets, and lactate, which reflect oxygenation, kidney function, coagulation, and metabolism, is crucial because these markers influence sepsis outcomes and may vary by sex. Transfer learning helps address small sample sizes by borrowing information from larger datasets, although differences in covariates and outcome-generating mechanisms between the target and external cohorts can complicate the process. We propose a novel weighting method, TRANSfer LeArning wiTh wEights (TRANSLATE), to integrate data from various sources by incorporating domain-specific characteristics through learned weights that align external data with the target cohort. These weights adjust for cohort differences, are proportional to each cohort's effective sample size, and downweight dissimilar cohorts. TRANSLATE offers theoretical guarantees for improved precision and applies to a wide range of estimands, including means, variances, and distribution functions. Simulations and a real-data application to sepsis outcomes in the Northeast cohort, using a much larger sample from other U.S. regions, show that the method enhances inference while accounting for regional heterogeneity.

Problem

• Addressing underrepresentation of Northeastern U.S. lung sepsis patients in national databases
• Improving inference for small cohorts via transfer learning and weighted data integration
• Accounting for regional and sex-based variations in sepsis clinical markers

Innovation

• Transfer learning integrates multiple datasets effectively
• Novel TRANSLATE method weights data by similarity
• Adjusts for regional heterogeneity in sepsis outcomes