GlucoGrapher: Sensor-Aware CGM Fusion with Mask-Aware Meta-Ensembles for Predicting Carbohydrate Caloric Ratio from Postprandial Glucose

Siddhant Ujjain, Pooja Singh, Ekta Srivastava, Ahmad Siraj Hashmi, Sandeep Kumar, Tapan Kumar Gandhi
Proceedings of The Second AAAI Bridge Program on AI for Medicine and Healthcare, PMLR 317:255-264, 2026.

Abstract

We study meal-level inference of the carbohydrate calorie ratio (CCR), the fraction of total calories attributable to net carbohydrates directly from early postprandial glucose responses (PPGR) recorded by continuous glucose monitors (CGMs). Using the date-shifted CGMACROS v1.0.0, we introduce GLUCOGRAPHER, a deployment-minded pipeline that (i) aligns PPGR around meal onset, (ii) performs sensor-aware fusion of Dexcom/Libre traces with explicit discordance detection and gating, (iii) leverages a dual preprocessing view of the signal (absolute $\Delta$mg/dL and percent change relative to baseline) with per-subject PPGR standardization to mitigate inter-individual scale shifts, and (iv) applies fold-wise, mask-aware nonnegative meta-learning with isotonic calibration to combine heterogeneous and sometimes missing out-of-fold (OOF) predictions without leakage. We augment PPGR shape descriptors (peak/time-to-peak, IAUC windows, slopes, late-ratio features) with lightweight behavior (steps, heart rate) and compact subject context (BMI, HbA1c buckets, selected fasting labs, and up to eight microbiome principal components). Evaluated with 5-fold Group-KFold by participant over n=663 meals, GLUCOGRAPHER attains RMSE 0.0929, NRMSErange 0.1608, NRMSEstd 0.7229, and Pearson r 0.6910. Performance is consistent across HbA1c-defined strata (Healthy/PreDM/T2D), indicating robustness to baseline glycemic status. Ablations show that the mask-aware meta-ensemble delivers a substantive lift over calibrated tree baselines, highlighting the value of reliability-aware sensor fusion, dual preprocessing, and leak-free calibration. Framing CCR as a bounded, interpretable target in [0, 1] enables actionable CGM-only feedback without perfect food logging, supporting retrospective coaching and prospective planning.

Cite this Paper

BibTeX
@InProceedings{pmlr-v317-ujjain26a, title = {GlucoGrapher: Sensor-Aware CGM Fusion with Mask-Aware Meta-Ensembles for Predicting Carbohydrate Caloric Ratio from Postprandial Glucose}, author = {Ujjain, Siddhant and Singh, Pooja and Srivastava, Ekta and Hashmi, Ahmad Siraj and Kumar, Sandeep and Gandhi, Tapan Kumar}, booktitle = {Proceedings of The Second AAAI Bridge Program on AI for Medicine and Healthcare}, pages = {255--264}, year = {2026}, editor = {Wu, Junde and Pan, Jiazhen and Zhu, Jiayuan and Luo, Luyang and Li, Yitong and Xu, Min and Jin, Yueming and Rueckert, Daniel}, volume = {317}, series = {Proceedings of Machine Learning Research}, month = {20--21 Jan}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v317/main/assets/ujjain26a/ujjain26a.pdf}, url = {https://proceedings.mlr.press/v317/ujjain26a.html}, abstract = {We study meal-level inference of the carbohydrate calorie ratio (CCR), the fraction of total calories attributable to net carbohydrates directly from early postprandial glucose responses (PPGR) recorded by continuous glucose monitors (CGMs). Using the date-shifted CGMACROS v1.0.0, we introduce GLUCOGRAPHER, a deployment-minded pipeline that (i) aligns PPGR around meal onset, (ii) performs sensor-aware fusion of Dexcom/Libre traces with explicit discordance detection and gating, (iii) leverages a dual preprocessing view of the signal (absolute $\Delta$mg/dL and percent change relative to baseline) with per-subject PPGR standardization to mitigate inter-individual scale shifts, and (iv) applies fold-wise, mask-aware nonnegative meta-learning with isotonic calibration to combine heterogeneous and sometimes missing out-of-fold (OOF) predictions without leakage. We augment PPGR shape descriptors (peak/time-to-peak, IAUC windows, slopes, late-ratio features) with lightweight behavior (steps, heart rate) and compact subject context (BMI, HbA1c buckets, selected fasting labs, and up to eight microbiome principal components). Evaluated with 5-fold Group-KFold by participant over n=663 meals, GLUCOGRAPHER attains RMSE 0.0929, NRMSErange 0.1608, NRMSEstd 0.7229, and Pearson r 0.6910. Performance is consistent across HbA1c-defined strata (Healthy/PreDM/T2D), indicating robustness to baseline glycemic status. Ablations show that the mask-aware meta-ensemble delivers a substantive lift over calibrated tree baselines, highlighting the value of reliability-aware sensor fusion, dual preprocessing, and leak-free calibration. Framing CCR as a bounded, interpretable target in [0, 1] enables actionable CGM-only feedback without perfect food logging, supporting retrospective coaching and prospective planning.} }
Endnote
%0 Conference Paper %T GlucoGrapher: Sensor-Aware CGM Fusion with Mask-Aware Meta-Ensembles for Predicting Carbohydrate Caloric Ratio from Postprandial Glucose %A Siddhant Ujjain %A Pooja Singh %A Ekta Srivastava %A Ahmad Siraj Hashmi %A Sandeep Kumar %A Tapan Kumar Gandhi %B Proceedings of The Second AAAI Bridge Program on AI for Medicine and Healthcare %C Proceedings of Machine Learning Research %D 2026 %E Junde Wu %E Jiazhen Pan %E Jiayuan Zhu %E Luyang Luo %E Yitong Li %E Min Xu %E Yueming Jin %E Daniel Rueckert %F pmlr-v317-ujjain26a %I PMLR %P 255--264 %U https://proceedings.mlr.press/v317/ujjain26a.html %V 317 %X We study meal-level inference of the carbohydrate calorie ratio (CCR), the fraction of total calories attributable to net carbohydrates directly from early postprandial glucose responses (PPGR) recorded by continuous glucose monitors (CGMs). Using the date-shifted CGMACROS v1.0.0, we introduce GLUCOGRAPHER, a deployment-minded pipeline that (i) aligns PPGR around meal onset, (ii) performs sensor-aware fusion of Dexcom/Libre traces with explicit discordance detection and gating, (iii) leverages a dual preprocessing view of the signal (absolute $\Delta$mg/dL and percent change relative to baseline) with per-subject PPGR standardization to mitigate inter-individual scale shifts, and (iv) applies fold-wise, mask-aware nonnegative meta-learning with isotonic calibration to combine heterogeneous and sometimes missing out-of-fold (OOF) predictions without leakage. We augment PPGR shape descriptors (peak/time-to-peak, IAUC windows, slopes, late-ratio features) with lightweight behavior (steps, heart rate) and compact subject context (BMI, HbA1c buckets, selected fasting labs, and up to eight microbiome principal components). Evaluated with 5-fold Group-KFold by participant over n=663 meals, GLUCOGRAPHER attains RMSE 0.0929, NRMSErange 0.1608, NRMSEstd 0.7229, and Pearson r 0.6910. Performance is consistent across HbA1c-defined strata (Healthy/PreDM/T2D), indicating robustness to baseline glycemic status. Ablations show that the mask-aware meta-ensemble delivers a substantive lift over calibrated tree baselines, highlighting the value of reliability-aware sensor fusion, dual preprocessing, and leak-free calibration. Framing CCR as a bounded, interpretable target in [0, 1] enables actionable CGM-only feedback without perfect food logging, supporting retrospective coaching and prospective planning.
APA
Ujjain, S., Singh, P., Srivastava, E., Hashmi, A.S., Kumar, S. & Gandhi, T.K.. (2026). GlucoGrapher: Sensor-Aware CGM Fusion with Mask-Aware Meta-Ensembles for Predicting Carbohydrate Caloric Ratio from Postprandial Glucose. Proceedings of The Second AAAI Bridge Program on AI for Medicine and Healthcare, in Proceedings of Machine Learning Research 317:255-264 Available from https://proceedings.mlr.press/v317/ujjain26a.html.

Related Material