arXiv:2512.05216v1 Announce Type: new 
Abstract: Masked autoencoders (MAEs) are increasingly applied to electronic health records (EHR) for learning general-purpose representations that support diverse clinical tasks. However, existing approaches typically rely on uniform random masking, implicitly assuming all features are equally predictable. In reality, laboratory tests exhibit substantial heterogeneity in volatility: some biomarkers (e.g., sodium) remain stable, while others (e.g., lactate) fluctuate considerably and are more difficult to model. Clinically, volatile biomarkers often signal acute pathophysiology and require more sophisticated modeling to capture their complex temporal patterns. We propose a volatility-aware pretraining strategy, Coefficient of Variation Masking (CV-Masking), that adaptively adjusts masking probabilities according to the intrinsic variability of each feature. Combined with a value-only masking objective aligned with clinical workflows, CV-Masking yields systematic improvements over random and variance-based strategies. Experiments on a large panel of laboratory tests show that CV-Masking enhances reconstruction, improves downstream predictive performance, and accelerates convergence, producing more robust and clinically meaningful EHR representations.

تم اقتراح استراتيجية جديدة تُسمى Coefficient of Variation Masking (CV-Masking) لتعزيز تطبيقات المحولات الماسكة في السجلات الصحية الإلكترونية (EHR). تتكيف هذه الطريقة مع احتمالات التعتيم بناءً على تقلبات المؤشرات الحيوية، مما يعالج تحدي نمذجة الميزات ذات التنبؤ المتغير، مثل مستويات الصوديوم المستقرة مقابل مستويات اللاكتات المتقلبة.

Se ha propuesto una nueva estrategia llamada Coefficient of Variation Masking (CV-Masking) para mejorar la aplicación de autoencoders enmascarados en los registros electrónicos de salud (EHR). Este enfoque adapta las probabilidades de enmascaramiento según la volatilidad de los biomarcadores, abordando el desafío de modelar características con variabilidad en la predictibilidad, como los niveles estables de sodio frente a los fluctuantes de lactato.

Une nouvelle stratégie appelée Coefficient de Variation Masking (CV-Masking) a été proposée pour améliorer l'application des autoencodeurs masqués dans les dossiers de santé électroniques (EHR). Cette approche adapte les probabilités de masquage en fonction de la volatilité des biomarqueurs, abordant le défi de modéliser des caractéristiques avec une prévisibilité variable, telles que les niveaux de sodium stables par rapport aux niveaux de lactate fluctuants.

A new strategy called Coefficient of Variation Masking (CV-Masking) has been proposed to enhance the application of masked autoencoders in electronic health records (EHR). This approach adapts masking probabilities based on the volatility of biomarkers, addressing the challenge of modeling features with varying predictability, such as stable sodium levels versus fluctuating lactate levels.

Coefficient of Variation Masking: A Volatility-Aware Strategy for EHR Foundation Models

arXiv:2512.06652v1 Announce Type: new 
Abstract: Accurate prediction of the need for invasive mechanical ventilation (IMV) in intensive care units (ICUs) patients is crucial for timely interventions and resource allocation. However, variability in patient populations, clinical practices, and electronic health record (EHR) systems across institutions introduces domain shifts that degrade the generalization performance of predictive models during deployment. Test-Time Training (TTT) has emerged as a promising approach to mitigate such shifts by adapting models dynamically during inference without requiring labeled target-domain data. In this work, we introduce Adaptive Test-Time Training (AdaTTT), an enhanced TTT framework tailored for EHR-based IMV prediction in ICU settings. We begin by deriving information-theoretic bounds on the test-time prediction error and demonstrate that it is constrained by the uncertainty between the main and auxiliary tasks. To enhance their alignment, we introduce a self-supervised learning framework with pretext tasks: reconstruction and masked feature modeling optimized through a dynamic masking strategy that emphasizes features critical to the main task. Additionally, to improve robustness against domain shifts, we incorporate prototype learning and employ Partial Optimal Transport (POT) for flexible, partial feature alignment while maintaining clinically meaningful patient representations. Experiments across multi-center ICU cohorts demonstrate competitive classification performance on different test-time adaptation benchmarks.

تم تقديم إطار عمل جديد يسمى Adaptive Test-Time Training (AdaTTT) لتحسين التنبؤ بالحاجة إلى التهوية الميكانيكية الغازية (IMV) لدى مرضى وحدات العناية المركزة. تتناول هذه الطريقة التحديات التي تطرحها التباينات في مجموعات المرضى والممارسات السريرية عبر مؤسسات مختلفة، مما قد يعيق فعالية النماذج التنبؤية أثناء نشرها.

Se ha introducido un nuevo marco llamado Adaptive Test-Time Training (AdaTTT) para mejorar la predicción de la necesidad de ventilación mecánica invasiva (IMV) en pacientes de UCI. Este enfoque aborda los desafíos que plantea la variabilidad en las poblaciones de pacientes y las prácticas clínicas en diferentes instituciones, lo que puede obstaculizar la efectividad de los modelos predictivos durante su implementación.

Un nouveau cadre appelé Adaptive Test-Time Training (AdaTTT) a été introduit pour améliorer la prédiction des besoins en ventilation mécanique invasive (IMV) chez les patients en soins intensifs. Cette approche traite les défis posés par la variabilité des populations de patients et des pratiques cliniques entre différentes institutions, ce qui peut entraver l'efficacité des modèles prédictifs lors de leur déploiement.

A new framework called Adaptive Test-Time Training (AdaTTT) has been introduced to improve the prediction of invasive mechanical ventilation (IMV) needs in ICU patients. This approach addresses the challenges posed by variability in patient populations and clinical practices across different institutions, which can hinder the effectiveness of predictive models during deployment.

Coefficient of Variation Masking: A Volatility-Aware Strategy for EHR Foundation Models

Was this article worth reading? Share it

LucidQuery AI

Twofold Health

Supanote