Re-evaluating Heart Rate Variability Biomarkers for Glucose Sensing: The Impact of Age Normalisation and Subject-Independent Validation
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Electrical Engineering. The day in electrical engineering: signals, systems and the hardware that carries them.
Rosa: We're looking at the paper: "Re-evaluating Heart Rate Variability Biomarkers for Glucose Sensing".
Dev: Here's the gist of the paper. Non-invasive glucose monitoring remains a critical challenge in the management of diabetes HRV during sleep shows promise for glucose prediction however, age-related autonomic changes significantly confound traditional HRV analyses Age-normalized HRV features achieved R2 = 0.161 (MAE = 0.182) for log-glucose prediction, representing a 25.6% improvement over non-normalized features (R2 = 0.132) Study Objective This study introduced a novel age-normalization technique for HRV features derived from sleep stage specific ECG analysis and evaluated its impact on non-invasive glucose prediction accuracy The hypothesis was that age-normalized, sleep-aware HRV features will significantly outperform conventional approaches by addressing fundamental limitations in autonomic function assessment across diverse populations Model Performance with Age-Normalized HRV The age-normalized model achieved a 5-fold cross-validated coefficient of determination of R2 = 0.161 ± 0.010, with a mean absolute error (MAE) of 0.182 ± 0.006 and a Pearson correlation of r = 0.443 (p < 0.01) Compared to the non-normalized baseline (R2 = 0.132, MAE = 0.185), this represents a significant improvement of +25.6% in R2 (+0.034 absolute), as confirmed by the paired t-test (p < 0.01) Feature Importance and Ablation Analysis The top three predictive features were age-normalized HRV metrics from sleep stages, as described in Table 1: hrv rem mean rr age normalized (r = 0.443, p = 0.004), hrv ds mean rr age normalized (r = 0.438, p = 0.005), and diastolic blood pressure (DBP, r = 0.437) The dominance of age-normalized features provides direct evidence for the efficacy of our normalization strategy in mitigating age-related confounding factors Systematic ablation revealed critical architectural contributions as described in Table 2: Age normalization provides +25.6% performance improvement Sleep-stage HRV features are essential for positive prediction performance Multi-modal integration significantly outperforms single-modality approaches Systematic Ablation Results The systematic ablation study results are summarized in Fig 4, demonstrating that multi-modal integration significantly outperforms single-modality approaches, with the 15-feature configuration proving optimal for glucose prediction The ablation study revealed that clinical features alone show negative predictive capability This confirms the necessity of both age normalization and multi-modal integration—particularly sleep-resolved HRV—for robust metabolic prediction This pilot study presents a preliminary age-normalization technique for heart rate variability features that shows initial promise for enhancing non-invasive glucose prediction accuracy from sleep-stage ECG analysis The methodology achieved a 25.6% improvement in prediction performance (R2 = 0.161 vs.
Rosa: 0.132 for non-normalized features, p < 0.01) Critical limitations include the small sample size (n=43), lack of demographic diversity, and the preliminary nature of the validation Future work must focus on large-scale multi-site validation (n>200), diverse demographic testing, and prospective clinical trials before any clinical utility can be established Beyond glucose prediction, the age-normalization framework addresses fundamental challenges in autonomic function assessment across heterogeneous populations The study included 43 adult subjects with complete multi-modal data comprising overnight ECG recordings, extracted RR-intervals, clinical glucose measurements, and validated sleep quality assessments Single-lead ECG signals were acquired at a sampling frequency of 250 Hz using a standard monitoring equipment Sleep stages were identified using criteria established by the American Academy of Sleep Medicine A novel mathematical age-normalization approach was developed to account for agerelated autonomic decline as described in Eq 1 Target Variable Engineering Primary Target Clinical glucose measurements were log-transformed to address distributional skewness and improve regression stability as described in Eq 2 This transformation normalizes the glucose distribution and stabilizes variance across the physiological range, thereby enhancing the machine learning performance Feature Selection Strategy A systematic correlation-based feature selection approach was employed: 1.
Dev: Correlation Analysis Pearson correlation coefficients were calculated for all features and the log-transformed glucose target 2.
Rosa: Statistical Filtering Features with p-values < 0.2 were retained to balance inclusivity with statistical relevance 3.
Title and authors: Rosa: We’re diving into the details now on "Re-evaluating Heart Rate Variability Biomarkers for Glucose Sensing: The Impact of Age Normalisation and Subject-Independent Validation". So, what exactly is this paper proposing beyond just the title?
Dev: They are essentially arguing that conventional heart rate variability analysis doesn't work well because it ignores how much autonomic function changes with age, and they are testing if a systematic way to normalize those features based on age actually improves glucose prediction.
Taro: So the core idea is that by accounting for the fact that your body’s autonomic system shifts as you age, you can get a more consistent and accurate reading for glucose prediction across different ages.
Rosa: Right, so it's about moving away from just looking at raw heart rate variability metrics and instead applying this novel age normalization technique to the features derived from sleep stage specific ECG analysis.
Dev: They used Bayesian Ridge regression with five-fold cross-validation to test their model, and they got an R squared score of zero point one six one with a mean absolute error of zero point one eight two for log glucose prediction, which is a definite improvement over the non-normalized baseline of zero point one three two <ref:2508.11682#pg1,0.182 for log glucose prediction>.
Taro: That improvement is significant because it's not just noise reduction; it’s showing that this specific approach to normalization yields better performance across the board, and they even found the top three predictive features were those age-normalized heart rate variability metrics from sleep stages, specifically hrv rem mean rr age normalized.
Rosa: So they are pointing directly to these normalized sleep stage metrics as the most important predictors, which strongly supports their method's effectiveness in capturing that age-related autonomic pattern.
Dev: That leads us into how they structured their experiment; they systematically removed features and found that multi-modal integration, combining those sleep stage HRV metrics with clinical measurements like diastolic blood pressure, actually outperformed using just one type of data alone.
The paper's summary: Rosa: So to summarize what they found in this paper, they introduced a novel age-normalization technique for heart rate variability features from sleep stage specific ECG analysis and showed that this method significantly improves non-invasive glucose prediction accuracy.
Dev: They hypothesized that these age-normalized, sleep-aware HRV features would outperform conventional approaches because they directly address the issue of age-related autonomic changes that confound traditional heart rate variability analyses.
Taro: The key finding is the twenty-five point six percent improvement in R squared score, moving from a baseline of zero point one three two to a model performance of zero point one six one when using their age-normalized features for log glucose prediction, which they confirmed with a paired t-test showing significance at p less than zero point zero one <ref:2508.11682#pg1>.
Rosa: It means that if you're trying to build a non-invasive glucose predictor based on sleep data, ignoring the age factor is introducing a noticeable error, and this normalization fixes it substantially.
Dev: They also looked at feature importance and ablation studies which showed that the top three predictive features were hrv rem mean rr age normalized, hrv ds mean rr age normalized, and diastolic blood pressure.
Taro: That confirms that these specific sleep stage metrics scaled by age are what provide the most direct evidence for how their normalization strategy mitigates those confounding factors in the autonomic data.
The paper's improvements: Rosa: The paper suggests several key improvements to existing models, and one big one is this novel mathematical age-normalization approach described in equation one, which scales the raw heart rate variability features by dividing them by an age-scaled factor.
Dev: Specifically, the formula they use is hrv age normalized equals hrv raw divided by (age over sixty five point zero plus zero point one), and they say this is how you account for the agerelated autonomic decline.
Taro: That scaling method is crucial because it directly addresses what they identified as a major limitation in previous research, which was treating age as just a coarse demographic group instead of applying a systematic normalization across diverse populations.
Rosa: They also improved their approach by incorporating multi-modal deep learning systems that integrate diverse sensor data, and they found that this multi-modal integration significantly outperforms single-modality approaches when trying to achieve clinically acceptable glucose prediction accuracy.
Dev: In terms of practical application, they are moving beyond just providing discrete categories and instead focusing on generating continuous glucose predictions, aiming for something like achieving eighty four point one percent of predictions within plus or minus one point five millimol per deciliter.
Taro: That continuous prediction capability is much more useful than just giving a yes or no answer; it gives you a better sense of the actual trend in glucose levels over time, which is what's needed for real-time management.
Conclusion: Rosa: So wrapping up this look at "Re-evaluating Heart Rate Variability Biomarkers for Glucose Sensing: The Impact of Age Normalisation and Subject-Independent Validation", the main takeaway is that age normalization applied to sleep stage specific heart rate variability features provides a substantial, statistically significant improvement in non-invasive glucose prediction accuracy.
Dev: They showed that by applying this technique, they moved from an R squared of zero point one three two for non-normalized features up to zero point one six one when using age-normalized ones, which is a twenty five point six percent gain in performance over the baseline <ref:2508.11682#pg1>.
Taro: The implication here is that for anyone developing these kinds of models, you can't just rely on raw HRV metrics; you have to systematically account for how age shifts the autonomic system because that’s where the real signal lies.
Rosa: Before we wrap up, Taro, what's your final thought on this work?
Taro: I think this research really validates that when you combine sleep-stage specific data with an age-aware normalization method, you can build a much more robust system for metabolic prediction that works better outside the lab.
Dev: And from an engineering standpoint, the fact that they identified the top three features—the age normalized REM mean rr, the age normalized deep sleep mean rr, and diastolic blood pressure—gives us a clear roadmap for what data inputs we need to prioritize.
Rosa: So we're looking at this paper as preliminary evidence that incorporating systematic age normalization into HRV features is a necessary step toward making non-invasive glucose monitoring more reliable for people outside of controlled lab settings.
Department of Computer Science & Engineering, Tezpur University
eess.SP, cs.AI, cs.LG
Submitted: 2025-08-09
Updated: 2026-11-10
Journal ref: BMC Med Inform Decis Mak 26, 166 (2026)
DOI: 10.1186/s12911-026-03455-8
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 73/100
The gist: Non-invasive glucose monitoring remains a critical challenge in the management of diabetes HRV during sleep shows promise for glucose prediction however…
Key concepts
- Age Normalization
- A mathematical method developed to adjust raw HRV features based on the subject's age. This addresses how autonomic nervous system function changes with age, ensuring that the features are comparable across different individuals and mitigating confounding factors related to aging.
- HRV Features from Sleep Stages
- Heart Rate Variability (HRV) metrics extracted specifically during different sleep stages (like REM or deep sleep) from ECG recordings. The study found that features derived from these specific sleep periods are particularly important for accurate glucose prediction due to complex metabolic regulation occurring during sleep.
- Systematic Ablation
- A rigorous testing method used to determine which parts of the model are most important. By systematically removing different feature sets, researchers identified that age normalization and multi-modal integration (combining different data types) were critical for achieving the best prediction performance.
- Continuous Prediction Capability
- The goal of the improved system is to provide a continuous estimate of glucose levels rather than just a single category. This allows for more precise clinical monitoring and tolerance metrics, such as predicting if glucose will fall within a specific safe range (e.g., ±1.5 mmol/L).
Terminology
Summary
Non-invasive glucose monitoring remains a critical challenge in the management of diabetes HRV during sleep shows promise for glucose prediction however, age-related autonomic changes significantly confound traditional HRV analyses Age-normalized HRV features achieved R2 = 0. p. 0 in;
Abstract
Non-invasive glucose monitoring remains a critical challenge in the management
of diabetes. HRV during sleep shows promise for glucose prediction however,
age-related autonomic changes significantly confound traditional HRV analyses. Age-normalized HRV features achieved
R2 = 0. 161 (MAE = 0. 161 (MAE = 0. 182) for log-glucose prediction, representing a 25. 182) for log-glucose prediction, representing a 25. 6%
improvement over non-normalized features (R2 = 0.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Study Objective
This study introduced a novel age-normalization technique for HRV features derived from sleep stage specific ECG analysis and evaluated its impact on non-invasive glucose prediction accuracy The hypothesis was that age-normalized, sleep-aware HRV features will significantly outperform conventional approaches by addressing fundamental limitations in autonomic function assessment across diverse populations
Model Performance with Age-Normalized HRV
The age-normalized model achieved a 5-fold cross-validated coefficient of determination of R2 = 0.161 ± 0.010, with a mean absolute error (MAE) of 0. p. 2 1 Model Performance with Age-Normalized HRV
The age-normalized model achieved a 5-fold cross-validated coefficient of determination of R2 = 0. 010, with a mean absolute error (MAE) of 0. 010, with a mean absolute error (MAE) of 0. 006 and
a Pearson correlation of r = 0. The top predictive features were hrv rem mean rr age normalized (r = 0. 443, p = 0. p. 0 The top predictive features were hrv rem mean rr age normalized (r = 0. 443, p = 0. 409 (p < 0. Compared to the non-normalized
baseline (R2 = 0. Compared to the non-normalized
baseline (R2 = 0. 132, MAE = 0. 132, MAE = 0. 185), this represents a significant improvement of
+25. 6% in R2
(+0. 034 absolute), as confirmed by the paired t-test (p < 0.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Feature Importance and Ablation Analysis
The top three predictive features were age-normalized HRV metrics from sleep stages, as described in Table 1: hrv rem mean rr age normalized (r = 0.443, p = 0.004), hrv ds mean rr age normalized (r = 0. p. 0 443, p = 0. 004),
hrv ds mean rr age normalized (r = 0. 004),
hrv ds mean rr age normalized (r = 0. 438, p = 0. 438, p = 0. 005), and diastolic blood pressure (DBP, r = 0. 005), and diastolic blood pressure (DBP, r = 0. The dominance of age-normalized features provides direct
evidence for the efficacy of our normalization strategy in mitigating age-related confounding factors.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Systematic Ablation Results
The systematic ablation study results are summarized in Fig 4, demonstrating that multi-modal integration significantly outperforms single-modality approaches, with the 15-feature configuration proving optimal for glucose prediction The ablation study revealed that clinical features alone show negative predictive capability This confirms the necessity of both age normalization and multi-modal integration—particularly sleep-resolved HRV—for robust metabolic prediction
Conclusion
This pilot study presents a preliminary age-normalization technique for heart rate variability features that shows initial promise for enhancing non-invasive glucose prediction accuracy from sleep-stage ECG analysis The methodology achieved a 25.6% improvement in prediction performance (R2 = 0. p. 5 6% performance improvement. • Sleep-stage HRV features are essential for positive prediction performance. 6%
improvement in prediction performance (R2 = 0. 132 for non-normalized features, p < 0.Where the paper says it
Where the paper says it
Methods
The study included 43 adult subjects with complete multi-modal data comprising overnight ECG recordings, extracted RR-intervals, clinical glucose measurements, and validated sleep quality assessments Single-lead ECG signals were acquired at a sampling frequency of 250 Hz using a standard monitoring equipment Sleep stages were identified using criteria established by the American Academy of Sleep Medicine A novel mathematical age-normalization approach was developed to account for agerelated autonomic decline as described in Eq 1
Target Variable Engineering
Primary Target Clinical glucose measurements were log-transformed to address distributional skewness and improve regression stability as described in Eq 2 This transformation normalizes the glucose distribution and stabilizes variance across the physiological range, thereby enhancing the machine learning performance
Feature Selection Strategy
A systematic correlation-based feature selection approach was employed: 1. Correlation Analysis Pearson correlation coefficients were calculated for all features and the log-transformed glucose target 2. p. 9 5 Feature Selection Strategy
A systematic correlation-based feature selection approach was employed:
1. Correlation Analysis: Pearson correlation coefficients were calculated for all features
and the log-transformed glucose target. Correlation Analysis: Pearson correlation coefficients were calculated for all features
and the log-transformed glucose target. Statistical Filtering: Features with p-values < 0. Statistical Filtering: Features with p-values < 0. 2 were retained to balance
inclusivity with statistical relevance.Where the paper says it
Where the paper says it
Where the paper says it
Cross-Validation
5-fold cross-validation with stratified sampling ensured robust performance estimation and prevents overfitting The performance metrics are as follows: Coefficient of determination (R2), Mean absolute error (MAE), Pearson correlation coefficient with significance testing Cross-validation stability was high, with a mean R2 = 0.161 ± 0.010 and coefficient of variation of 5. p. 6 Cross-validation stability was high, with a mean R2 = 0. 010 and coefficient
of variation of 5. 010 and coefficient
of variation of 5. 9% across folds—all folds achieving R2 > 0.Where the paper says it
Where the paper says it
Sleep-Stage Specific Analysis
The superior performance of REM sleep features aligns with the known physiology REM sleep exhibits variable autonomic activity reflecting complex metabolic regulation, potentially providing enhanced sensitivity to glucose-related autonomic modulation as described in Table 3 Parasympathetic dominance of deep sleep offers complementary metabolic insights The preliminary R2 = 0.161 shows initial promise compared to recent ECG-based glucose detection studies achieving 75-78% binary classification accuracy, while providing continuous prediction rather than discrete categorization This work should be considered exploratory research providing preliminary evidence for the potential of age-normalized HRV features in glucose prediction
Data Availability
The dataset used and/or analyzed during the current study is publicly available in the Mendeley Data repository, accessible at https://data.mendeley.com/datasets/9c47vwvtss/4 The processed data required to reproduce these findings and the code used for the analysis in this study are available at https://github. p. 2 Further validation in larger cohorts is required to confirm these findings. Cross-validation analysis revealed robust model stability across all folds, as shown in
Fig 2.Where the paper says it
References
[1] Magliano DJ, Boyko EJ, IDF Diabetes Atlas 10th edition scientific committee. IDF DIABETES ATLAS. p. 0 in;
Contributing authors: sis@tezu. in;
Abstract
Non-invasive glucose monitoring remains a critical challenge in the management
of diabetes. in;
Contributing authors: sis@tezu. in;
Abstract
Non-invasive glucose monitoring remains a critical challenge in the management
of diabetes.Where the paper says it
Where the paper says it
[2] Alghlayini S, Al-Betar MA, Atef M. Enhancing Non-Invasive Blood Glucose Prediction from Photoplethysmography Signals via Heart Rate VariabilityBased Features Selection Using Metaheuristic Algorithms. p. 8 Raw signals underwent scaling validation to ensure
physiological amplitude range (±5 mV) [17], addressing potential instrumentation
variations that could affect the subsequent analysis. RR-Interval Extraction
R-peak detection was performed using adaptive threshold algorithms [18] with manual
verification for quality control. Parasympathetic dominance of deep sleep offers complementary metabolic insights. 3 Conclusions
This pilot study presents a preliminary age-normalization technique for heart rate
variability features that shows initial promise for enhancing non-invasive glucose prediction accuracy from sleep-stage ECG analysis. 11682v2 [eess. p. 0 Keywords: Heart rate variability (HRV), Glucose Prediction, Age normalization,
Sleep stages, Diabetes monitoring
1
arXiv:2508. 11682v2 [eess.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
[3] Gusev M, Poposka L, Spasevski G, Kostoska M, Koteska B, Simjanoska M, et al. Noninvasive Glucose Measurement Using Machine Learning and Neural Network Methods and Correlation with Heart Rate Variability. p. 0 However, these results require validation in larger
cohorts before clinical consideration. Keywords: Heart rate variability (HRV), Glucose Prediction, Age normalization,
Sleep stages, Diabetes monitoring
1
arXiv:2508. The performance comparison across different machine learning models is illustrated
in Fig 1, which demonstrates that BayesianRidge regression achieved optimal glucose prediction performance. These preliminary results suggest that age normalization
may enhance the predictive utility of heart rate variability (HRV) features for glucose
estimation.Where the paper says it
Where the paper says it
[4] Heart rate variability: standards of measurement, physiological interpretation and clinical use. Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology. p. 1 A critical limitation of current HRV-based glucose prediction approaches is the
inadequate consideration of age-related autonomic changes. While Task Force guidelines recognize age-related decline in HRV, they do not prescribe formal normalization
methods [9].Where the paper says it
[5] Im SI, Kim SJ, Bae SH, Kim BJ, Heo JH, Kwon Sk, et al. Real-time heart rate variability according to ambulatory glucose profile in patients with diabetes mellitus. p. 0 However, these results require validation in larger
cohorts before clinical consideration. Keywords: Heart rate variability (HRV), Glucose Prediction, Age normalization,
Sleep stages, Diabetes monitoring
1
arXiv:2508. For instance, Im et al. (2023) reported that poor
glycemic control is associated with significantly reduced HRV, whereas well-regulated
glucose profiles correlate with enhanced HRV in patients with diabetes [5].Where the paper says it
Where the paper says it
[6] Kajisa T, Kuroi T, Hara H, Sakai T. Correlation analysis of heart rate variations and glucose fluctuations during sleep. p. 7 Parasympathetic dominance of deep sleep offers complementary metabolic insights. 3 Conclusions
This pilot study presents a preliminary age-normalization technique for heart rate
variability features that shows initial promise for enhancing non-invasive glucose prediction accuracy from sleep-stage ECG analysis. RR-intervals were extracted with systematic artifact
removal using statistical outlier detection (intervals beyond three standard deviations
from the local mean were excluded) [17]. Sleep-Stage Classification: Sleep stages were identified using criteria established by
the American Academy of Sleep Medicine [19]:
1.Where the paper says it
Where the paper says it
[7] Rothberg LJ, Lees T, Clifton-Bligh R, Lal S. Association Between Heart Rate Variability Measures and Blood Glucose Levels: Implications for Noninvasive Glucose Monitoring for Diabetes. p. 0 However, these results require validation in larger
cohorts before clinical consideration. Keywords: Heart rate variability (HRV), Glucose Prediction, Age normalization,
Sleep stages, Diabetes monitoring
1
arXiv:2508. in;
Contributing authors: sis@tezu. in;
Abstract
Non-invasive glucose monitoring remains a critical challenge in the management
of diabetes.Where the paper says it
Where the paper says it
[8] Vv K, Ne M, Nv T. Heart rate variability is associated with interstitial glucose fluctuations in type 2 diabetic women treated with insulin. p. 1 (2016) demonstrated significant associations between specific HRV frequency-domain
parameters and blood glucose levels in both diabetic and non-diabetic populations [7],
whereas Klimontov et al. (2016) found that HRV is associated with interstitial glucose
fluctuations in women with type 2 diabetes treated with insulin [8].Where the paper says it
[9] Electrophysiology TFotESoCtNASoP. Heart Rate Variability. Circulation. 1996 Mar;93(5):1043–1065 Publisher: American Heart Association
[10] Stojmenski A, Gusev M, Chorbev I, Tudjarski S, Poposka L, Vavlukis M. Age and Gender Impact on Heart Rate Variability towards Noninvasive Glucose Measurement. p. 0 However, these results require validation in larger
cohorts before clinical consideration. Keywords: Heart rate variability (HRV), Glucose Prediction, Age normalization,
Sleep stages, Diabetes monitoring
1
arXiv:2508.Where the paper says it
[11] Martyn-Nemeth P, Phillips SA, Mihailescu D, Farabi SS, Park C, Lipton R, et al. Poor sleep quality is associated with nocturnal glycaemic variability and fear of hypoglycaemia in adults with type 1 diabetes. p. 1 Martyn-Nemeth et al. reported that individuals with type 1 diabetes and poor
sleep quality exhibited greater nocturnal glycemic variability and heightened fear of
hypoglycemia [11].Where the paper says it
[12] Cheng W, Chen H, Tian L, Ma Z, Cui X. Heart rate variability in different sleep stages is associated with metabolic function and glycemic control in type 2 diabetes mellitus. p. 0 However, these results require validation in larger
cohorts before clinical consideration. Keywords: Heart rate variability (HRV), Glucose Prediction, Age normalization,
Sleep stages, Diabetes monitoring
1
arXiv:2508. For instance, Im et al. (2023) reported that poor
glycemic control is associated with significantly reduced HRV, whereas well-regulated
glucose profiles correlate with enhanced HRV in patients with diabetes [5].Where the paper says it
Where the paper says it
[13] Radha M, Fonseca P, Moreau A, Ross M, Cerny A, Anderer P, et al. Sleep stage classification from heart-rate variability using long short-term memory neural networks. p. 7 Parasympathetic dominance of deep sleep offers complementary metabolic insights. 3 Conclusions
This pilot study presents a preliminary age-normalization technique for heart rate
variability features that shows initial promise for enhancing non-invasive glucose prediction accuracy from sleep-stage ECG analysis.Where the paper says it
[14]: Development of Non-Invasive Continuous Glucose Prediction Models Using Multi-Modal Wearable Sensors in Free-Living Conditions. Available from: //www.mdpi.com/1424-8220/25/10/3207
[15] Chowdhury MH, Chowdhury MEH, Alqahtani A. MMG-net: Multi modal approach to estimate blood glucose using multi-stream and cross modality attention. p. 4 3 Age-normalized HRV features dominate top predictive rankings with strong statistical significance (12/15 features p<0. 3 Systematic Ablation Results
The systematic ablation study results are summarized in Fig 4, demonstrating that
multi-modal integration significantly outperforms single-modality approaches, with
the 15-feature configuration proving optimal for glucose prediction. However, conventional invasive monitoring methods pose barriers to compliance and hinder real-time
metabolic assessment, highlighting the need for reliable non-invasive alternatives. Recent advances in biomedical signal processing have highlighted the potential
of physiological signals, particularly HRV, for non-invasive glucose prediction [2, 3].Where the paper says it
Where the paper says it
[16] Cheng W, Chen H, Tian L, Ma Z, Cui X. A dataset on 24-h electrocardiograph, sleep and metabolic function of male type 2 diabetes mellitus. p. 1 1 Introduction
Diabetes mellitus currently affects over 537 million adults worldwide and remains a
leading contributor to global morbidity and mortality [1]. Effective glucose monitoring
is essential for managing diabetes and preventing its complications. HRV during sleep shows promise for glucose prediction however,
age-related autonomic changes significantly confound traditional HRV analyses. We analyzed 43 subjects with multi-modal data including sleep-stage specific
ECG, HRV features, and clinical measurements.Where the paper says it
Where the paper says it
[17] Kligfield P, Gettes LS, Bailey JJ, Childers R, Deal BJ, Hancock EW, et al. Recommendations for the Standardization and Interpretation of the Electrocardiogram. p. 7 This study establishes
a preliminary foundation for sleep-aware, age-adjusted physiological monitoring in
diabetes management. Several critical limitations restrict the interpretation and generalizability of this
pilot study.Where the paper says it
[18] Pan J, Tompkins WJ. A Real-Time QRS Detection Algorithm. p. 1 Effective glucose monitoring
is essential for managing diabetes and preventing its complications. However, conventional invasive monitoring methods pose barriers to compliance and hinder real-time
metabolic assessment, highlighting the need for reliable non-invasive alternatives. Age-Normalized HRV Features for Non-Invasive
Glucose Prediction: A Pilot Sleep-Aware Machine
Learning Study
Md Basit Azam1* and Sarangthem Ibotombi Singh1
1*Department of Computer Science & Engineering,, Tezpur University,
Napaam, Tezpur, 784028, Assam, INDIA. *Corresponding author(s).Where the paper says it
Where the paper says it
[19] Berry RB, Brooks R, Gamaldo C, Harding SM, Lloyd RM, Quan SF, et al. AASM Scoring Manual Updates for 2017 (Version 2. p. 8 Raw signals underwent scaling validation to ensure
physiological amplitude range (±5 mV) [17], addressing potential instrumentation
variations that could affect the subsequent analysis. RR-Interval Extraction
R-peak detection was performed using adaptive threshold algorithms [18] with manual
verification for quality control.Where the paper says it
[20] Umetani K, Singer DH, McCraty R, Atkinson M. Twenty-Four Hour Time Domain Heart Rate Variability and Heart Rate: Relations to Age and Gender Over Nine Decades. p. 0 However, these results require validation in larger
cohorts before clinical consideration. Keywords: Heart rate variability (HRV), Glucose Prediction, Age normalization,
Sleep stages, Diabetes monitoring
1
arXiv:2508. RR-intervals were extracted with systematic artifact
removal using statistical outlier detection (intervals beyond three standard deviations
from the local mean were excluded) [17]. Sleep-Stage Classification: Sleep stages were identified using criteria established by
the American Academy of Sleep Medicine [19]:
1.Where the paper says it
Where the paper says it
[21] Zhang J. Effect of Age and Sex on Heart Rate Variability in Healthy Subjects. p. 0 However, these results require validation in larger
cohorts before clinical consideration. Keywords: Heart rate variability (HRV), Glucose Prediction, Age normalization,
Sleep stages, Diabetes monitoring
1
arXiv:2508. However, conventional invasive monitoring methods pose barriers to compliance and hinder real-time
metabolic assessment, highlighting the need for reliable non-invasive alternatives. Recent advances in biomedical signal processing have highlighted the potential
of physiological signals, particularly HRV, for non-invasive glucose prediction [2, 3].Where the paper says it
Where the paper says it
[22] Keene ON. The log transformation is special. p. 9 • hrv ds mean rr age normalized
• hrv rem mean rr age normalized
• hrv rs mean rr age normalized
4. 4 Target Variable Engineering
Primary Target
Clinical glucose measurements were log-transformed to address distributional skewness
and improve regression stability as described in Eq 2
ytarget = ln(glucose mmol L) (2)
This transformation normalizes the glucose distribution and stabilizes variance
across the physiological range, thereby enhancing the machine learning performance
[22]. RR-intervals were extracted with systematic artifact
removal using statistical outlier detection (intervals beyond three standard deviations
from the local mean were excluded) [17]. Sleep-Stage Classification: Sleep stages were identified using criteria established by
the American Academy of Sleep Medicine [19]:
1.Where the paper says it
Where the paper says it
[23] Saeys Y, Inza I, Larra˜naga P. A review of feature selection techniques in bioinformatics. p. 9 4 Target Variable Engineering
Primary Target
Clinical glucose measurements were log-transformed to address distributional skewness
and improve regression stability as described in Eq 2
ytarget = ln(glucose mmol L) (2)
This transformation normalizes the glucose distribution and stabilizes variance
across the physiological range, thereby enhancing the machine learning performance
[22]. 5 Feature Selection Strategy
A systematic correlation-based feature selection approach was employed:
1.Where the paper says it
[24] Tipping ME. Sparse Bayesian learning and the relevance vector machine. p. 0 Age-Normalized HRV Features for Non-Invasive
Glucose Prediction: A Pilot Sleep-Aware Machine
Learning Study
Md Basit Azam1* and Sarangthem Ibotombi Singh1
1*Department of Computer Science & Engineering,, Tezpur University,
Napaam, Tezpur, 784028, Assam, INDIA. *Corresponding author(s). Age-Normalized HRV Features for Non-Invasive
Glucose Prediction: A Pilot Sleep-Aware Machine
Learning Study
Md Basit Azam1* and Sarangthem Ibotombi Singh1
1*Department of Computer Science & Engineering,, Tezpur University,
Napaam, Tezpur, 784028, Assam, INDIA. *Corresponding author(s).Where the paper says it
Where the paper says it
[25] Kohavi R. A study of cross-validation and bootstrap for accuracy estimation and model selection. p. 2 Further validation in larger cohorts is required to confirm these findings. Cross-validation analysis revealed robust model stability across all folds, as shown in
Fig 2.Where the paper says it
[26] Alvarez Melis D, Jaakkola T. Towards Robust Interpretability with SelfExplaining Neural Networks. p. 2 Further validation in larger cohorts is required to confirm these findings. Cross-validation analysis revealed robust model stability across all folds, as shown in
Fig 2. However, conventional invasive monitoring methods pose barriers to compliance and hinder real-time
metabolic assessment, highlighting the need for reliable non-invasive alternatives. Recent advances in biomedical signal processing have highlighted the potential
of physiological signals, particularly HRV, for non-invasive glucose prediction [2, 3].Where the paper says it
Where the paper says it
[2508.11682] arXiv:2508. p. 0 However, these results require validation in larger
cohorts before clinical consideration. Keywords: Heart rate variability (HRV), Glucose Prediction, Age normalization,
Sleep stages, Diabetes monitoring
1
arXiv:2508.Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Improved AI system can perform non-invasive glucose prediction by integrating age-normalized HRV features derived from sleep-stage specific ECG analysis with BayesianRidge regression, achieving a 25.6% improvement in prediction performance over non-normalized features (R2 = 0.161 vs R2 = 0.132).
-
The improved system will utilize the top three predictive features identified by systematic ablation, specifically hrv rem mean rr age normalized, hrv ds mean rr age normalized, and diastolic blood pressure, to generate continuous glucose predictions rather than discrete categorization.
-
The model can be architecturally optimized by ensuring multi-modal integration across sleep-stage specific HRV metrics and clinical measurements to significantly outperform single-modality approaches in accuracy.
-
The system will incorporate a novel mathematical age normalization approach, scaling raw HRV features using the formula HRV age normalized = HRV raw / (age / 65.0 + 0.1), addressing
age-related autonomic changes
which are identified as a critical confounding factor in traditional analyses. -
The improved system can provide clinical tolerance metrics, such as achieving 84.1% of predictions within ±1.5 mmol/L, offering a continuous prediction capability that is more useful than discrete categorization methods mentioned in other studies.
Abstract
Heart rate variability (HRV) from electrocardiogram (ECG) signals may indicate glycemic status, but prior studies often combine distinct glucose targets and use validation strategies vulnerable to data leakage. Since HRV declines with age, aging may be confounded with diabetes-related autonomic dysfunction. We tested whether age-dependent HRV normalisation improves glycemic estimation. We analysed ECG features from 43 male patients with type 2 diabetes. Outcomes were glycated hemoglobin (HbA1c; n=29) and fasting blood glucose (FBG; n=38). We used leave-one-subject-out cross-validation, with feature selection and standardisation within each training fold. We compared 20 machine-learning algorithms, six age-adjustment methods, and 20 normalisation parameter combinations. Validation included 500 permutations and bootstrap confidence intervals. Extra Trees regression performed best: R2=0.222 (r=0.476, p=0.009) for HbA1c and R2=0.086 (r=0.344, p=0.034) for FBG. Mean absolute errors were 1.18 percentage points and 2.27 mmol/L, respectively. Permutation testing supported both associations (p=0.002). Contrary to our hypothesis, no age-adjustment method or parameter combination improved performance. Tree ensembles retained positive performance under strict validation, whereas linear models had negative R2 values; neural networks performed poorly. Strict within-fold preprocessing materially changed model performance and revealed potential inflation from conventional validation. Bootstrap 95% confidence intervals were [0.13, 0.82] for HbA1c R2 and [0.10, 0.72] for FBG R2. Although findings support HRV-glycemic associations, performance remains insufficient for standalone clinical use. The study highlights the importance of separating glycemic targets, subject-independent validation, within-fold preprocessing, and comprehensive baselines in non-invasive glycemic monitoring research.
Related papers
- Open-H-RF: A Large-Scale Ultrasound Raw Channel Dataset Initiative
- Scalable Attention for 5G NR Channel Estimation
- Low-Altitude Wireless Networks: A Comprehensive Survey
- Sensing-Assisted OFDM with Index Modulation: A Waveform for Optical Wireless Integrated Communication and Localisation
- Iteration-Free Candidate Pruning and an Index-Domain VLSI Architecture for Sort-Free MIMO Box Decoding up to 1024-QAM
- A Foundation Model for Large-Scale Wireless Network Planning, Operation and Optimization