Abstract: The DiaTeleMed Study (NCT05046886) is a fully remote randomized clinical trial evaluating whether machine-learning-driven "precision nutrition" outperforms standard dietary advice for managing Type 2 Diabetes (T2D).
Introduction: The Challenge of Inter-Individual Variability
Standard dietary recommendations for Type 2 Diabetes (T2D), such as the Mediterranean diet or low-glycemic-index approaches, often struggle with inconsistent results across different patients. This is largely due to inter-individual variability: two people consuming the same meal can exhibit vastly different postprandial glycemic responses (PPGR). These differences are driven by a complex interplay of genetics, lifestyle, metabolism, and the gut microbiome. The DiaTeleMed study aims to address this gap by testing a "precision nutrition" approach that leverages a machine learning algorithm to predict individual responses to food.
Methods and Study Design
The DiaTeleMed study is a three-arm, fully remote randomized clinical trial (RCT) involving 255 participants with moderately controlled T2D (HbA1c 6.0–8.0%). Participants are randomized into one of three groups:
- Personalized Intervention: Participants receive behavioral counseling and use the DayTwo app, which provides predicted PPGR scores (1–10) for meals based on the Personal Nutrition Project (PNP) algorithm.
- Standardized Intervention: Participants receive the same behavioral counseling and app for tracking, but follow standard Mediterranean diet targets without personalized predictions.
- Usual Care Control (UCC): Participants receive basic diabetes education and use the app for self-monitoring only.
Technological Implementation
The study leverages a sophisticated "tech stack" to deliver and monitor the intervention entirely remotely:
- Continuous Glucose Monitors (CGM): Participants wear CGMs for 12-day periods at baseline, 3 months, and 6 months to track real-time glucose fluctuations.
- Gut Microbiome Profiling: Those in the personalized arm provide stool samples for metagenomic sequencing, which informs the PNP algorithm originally trained on large-scale cohort data.
- Telemedicine Delivery: All counseling sessions and clinical assessments are conducted via synchronous videoconferencing (WebEx), ensuring the trial is accessible regardless of geographic location.
Outcome Measures
The primary outcome of the study is the change in Mean Amplitude of Glycemic Excursions (MAGE). While HbA1c is a standard measure of long-term control, MAGE specifically captures glycemic variability—the "rollercoaster" of blood sugar spikes and dips that contributes to diabetic complications. Secondary and exploratory outcomes include changes in HbA1c, $\beta$-cell function, and behavioral metrics such as self-efficacy and dietary adherence.
Discussion and Scalability
By conducting the entire trial remotely—from stool kit mailing to virtual counseling—the researchers are testing a scalable model for precision medicine. The study is specifically powered to determine if the personalization provided by the ML algorithm adds significant value above and beyond high-quality behavioral counseling. If successful, this model could be integrated into standard primary care to provide scalable, precision-based chronic disease management for diverse populations.
Summary updated