Long glucose data patterns improve health predictions across groups
Transferable Evidence Reconstruction for Longitudinal Glucose Representations
Machine Learning
Summary
Predicting health events from long glucose monitoring data is hard because the useful signals are rare and spread out. The authors developed a new way to train computers to spot important patterns by teaching a small reader network to find clues that work across different groups of people. This method improves the accuracy of predicting fourteen glucose-related health tasks compared to earlier approaches. Their approach also learns to organize glucose data over several days, helping the computer understand daily rhythms.
What this means in practice
- •For hospital data teams: Improve accuracy of glucose-related health forecasts by using transferable evidence from long-term monitoring data.
- •For wearable device developers: Enhance continuous glucose monitors by integrating time-aware models that better capture patterns over several days for personalized alerts.
Authors
Tian Zhou, Bingqing Peng, Linxiao Yang, Wenwei Wang, Mengni Ye, Beverly Jin, Zuyi Zhu, Jinjie Gu, Liang Sun
Abstract
Long physiological recordings contain many routine measurements, while predictive information is often concentrated in rare events, sustained burden, and recurring temporal patterns. Masked autoencoding recovers measurements; contrastive learning aligns views. We study self-supervision that explicitly prioritizes structured signal evidence. We introduce transferable evidence reconstruction (TER), which constructs evidence from unlabeled recordings, fits a fresh low-capacity reader on one recording group, and requires that reader to recover the same evidence in another group without refitting. Differentiating through this cross-group test learns representations with transferable evidence-decoding rules; the evidence guides self-supervision but is not used as a downstream feature. For continuous glucose monitoring (CGM), an observation-aware daily encoder and clock-aware multi-day memory bind glucose level and change to recorded time while organizing up to seven days of history. On the 14-task leaderboard, TER improves the strongest prior overall PR-AUC/ROC-AUC/Macro-F1 scores by 5.51/4.43/2.80 percentage points and sets a new best metric on 12/14 tasks. These leaderboard gains are 2.0-2.9 times the respective gaps between the two strongest baselines. With public pretraining data, folds, and the linear probe matched, TER outperforms our GlucoFM reproduction by 6.09/5.52/2.72 points. Target-reader ablations, same-history controls, and cross-person readouts support the combination of structured evidence, cross-group reader fitting, and learned multi-day organization.