中国P站

Digital biomarkers; Continuous glucose monitoring; Wearable devices; Glycemic variability; Type 2 diabetes; Non-invasive monitoring; Machine learning; Electrodermal activity; Skin temperature; Heart rate variability; Accelerometry; Predictive modeling; Personalized health.

Introduction

Glycemic variability (GV) is a critical factor in the management of type 2 diabetes (T2D), influencing both short-term and long-term health outcomes. Traditional methods of monitoring blood glucose levels, such as intermittent fingerstick measurements, provide limited insight into the dynamic fluctuations of glucose throughout the day. Continuous glucose monitoring (CGM) systems have addressed this limitation by offering real-time, continuous data on interstitial glucose levels [1-5]. However, CGMs primarily capture glucose data and do not account for other physiological and behavioral factors that influence GV. Recent advancements in wearable technologies have enabled the collection of a broader range of data, including skin temperature, electrodermal activity, heart rate variability, and physical activity levels. Integrating these digital biomarkers with CGM data through machine learning models holds promise for more accurate prediction and management of GV in individuals with T2D [6-10].

Discussion

Several studies have explored the integration of digital biomarkers from CGM and wearable devices to predict GV in T2D. For instance, a proof-of-concept study recorded over 25,000 measurements from CGM and simultaneous data from wrist-worn wearables over 8–10 days in 16 participants. Machine learning models developed from this data achieved high accuracy in predicting GV metrics and HbA1c levels, with mean average percent error (MAPE) less than 10% for 11 of the 27 glucose variability models. The most influential wearable sensor features in predicting HbA1c were skin temperature, electrodermal activity, accelerometry, and heart rate, with skin temperature contributing 33% to the model's accuracy. Another approach involved using photoplethysmography (PPG) data from wearables to assess heart rate variability (HRV), which has been associated with GV. Changes in HRV have been observed up to 90 minutes prior to hypoglycemic events, indicating the potential of wearables in early detection of glucose fluctuations. Additionally, integrating contextual data such as sleep patterns, physical activity, and dietary intake with CGM data has enhanced the personalization of glucose predictions, allowing for more tailored diabetes management strategies.

The application of machine learning algorithms, including regression models and deep learning techniques, has further improved the predictive accuracy of GV models. For example, models trained on personalized data from individual participants have demonstrated better performance compared to population-based models, highlighting the importance of individualized approaches in diabetes care. These advancements suggest that integrating digital biomarkers from CGM and wearable devices can provide a comprehensive understanding of factors influencing GV, leading to more effective and personalized diabetes management.

Conclusion

The integration of digital biomarkers from CGM and wearable devices represents a significant advancement in the management of glycemic variability in type 2 diabetes. By combining continuous glucose data with physiological and behavioral information, healthcare providers can gain a more holistic view of factors affecting GV. Machine learning models utilizing these integrated data sources have shown promise in accurately predicting GV and HbA1c levels, facilitating more personalized and proactive diabetes care. However, challenges remain, including the need for larger and more diverse datasets to validate these models, as well as considerations regarding data privacy and the clinical implementation of such technologies. Future research should focus on refining these predictive models, exploring their real-world applicability, and addressing regulatory and ethical concerns to fully realize the potential of digital biomarkers in diabetes management.

References

1. Jomezadeh N, Babamoradi S, Kalantar E, Javaherizadeh H (2014) . Gastroenterol Hepatol Bed Bench 7: 218.

   ,

2. Sangeetha A, Parija SC, Mandal J, Krishnamurthy S (2014) . J Health Popul Nutr 32: 580.

   ,

3. Ranjbar R, Dallal MMS, Talebi M, Pourshafie MR (2008) . J Health Popul Nutr 26: 426.

   Indexed at, ,

4. Zhang J, Jin H, Hu J, Yuan Z, Shi W, et al. (2014) . Diagn Microbiol Infect Dis 78: 282–286.

   , ,

5. Pourakbari B, Mamishi S, Mashoori N, Mahboobi N, Ashtiani MH, et al. (2010) . Braz J Infect Dis 14: 153–157.

   , ,

6. Von-Seidlein L, Kim DR, Ali M, Lee HH, Wang X, et al. (2006) . PLoS Med 3: e353.

   , ,

7. Germani Y, Sansonetti PJ (2006) . The prokaryotes In: Proteobacteria: Gamma Subclass Berlin: Springer 6: 99-122.

8. Aggarwal P, Uppal B, Ghosh R, Krishna Prakash S, Chakravarti A, et al. (2016) . Travel Med Infect Dis 14: 407–413.

   , ,

9. Taneja N, Mewara A (2016) . Indian J Med Res 143: 565-576.

   , ,

10. Farshad S, Sheikhi R, Japoni A, Basiri E, Alborzi A (2006) . J Clin Microbiol 44: 2879–2883.

    , ,