.png)
According to the World Health Organization (WHO), 537 million adults worldwide live with diabetes, a figure projected to climb to 643 million by 2030. The vast majority, approximately 90%, suffer from Type 2 Diabetes (T2D), a condition primarily driven by lifestyle factors and preceded by a "silent" physiological state known as insulin resistance (IR). In a healthy body, insulin facilitates glucose uptake into cells; however, in those with IR, tissues become less responsive, forcing the pancreas to overproduce insulin to maintain stable blood sugar levels.
The challenge for modern healthcare is that IR often remains invisible until it has already progressed to pre-diabetes or clinical T2D. Current diagnostic standards, such as the hyperinsulinaemic euglycaemic clamp, are resource-intensive and restricted to research facilities. Even more accessible metrics like HOMA-IR (Homeostatic Model Assessment of Insulin Resistance) require clinical laboratory visits for fasting insulin and glucose tests, which are not typically part of annual medical exams.
This diagnostic gap represents a significant opportunity for digital health innovation. New research, specifically the WEAR-ME study, demonstrates that we can now predict IR using the devices already on our wrists. By combining wearable sensor data with routine blood biomarkers and deep learning, we are moving toward a future where metabolic risk is detected continuously and non-invasively, shifting the paradigm from reactive treatment to proactive prevention.
Accurately measuring insulin resistance is possible, but not practical at scale. The gold-standard method, the hyperinsulinaemic euglycaemic clamp, provides precise results but is time-consuming, expensive, and limited to specialized research settings. As a result, it is rarely used in routine care.
A more accessible alternative is HOMA-IR, which estimates insulin resistance using fasting glucose and insulin levels. While easier to implement, it still requires blood draws and laboratory testing, making it unsuitable for frequent or continuous monitoring. This creates a gap between what is clinically ideal and what is realistically applied in everyday healthcare.
In practice, most metabolic screening relies on snapshot-based markers such as fasting glucose, HbA1c, or oral glucose tolerance tests (OGTT). These metrics are valuable, but they primarily capture later-stage metabolic dysfunction.
Insulin resistance, however, develops gradually and can remain undetected while these markers still appear normal. This means early physiological changes often go unnoticed, limiting the ability to intervene at a stage where lifestyle adjustments could have the greatest impact.
Together, these limitations create a clear prevention gap. Detection methods are either too complex for routine use or not sensitive enough to identify early-stage risk.
By the time metabolic issues become visible through standard screening, progression toward prediabetes or type 2 diabetes may already be underway. Closing this gap requires approaches that are both scalable and capable of capturing subtle, continuous changes in metabolic health over time.
Modern wearables have transitioned from basic step counters into sophisticated tools capable of capturing high-dimensional representations of our biology. By leveraging advanced sensors and AI, these devices provide a continuous window into our metabolic health that was previously impossible outside of a lab.
The foundation of wearable-based monitoring lies in physiological signals like heart rate. Research confirms that higher resting heart rates (RHR) and lower heart rate variability (HRV) are clinically associated with insulin resistance (IR). Unlike a single clinical reading, a smartwatch tracks these rhythms during sleep and daily activity, providing a much clearer picture of how your body manages metabolic stress.
While standard apps use simple math, the WEAR-ME study utilized a Wearable Foundation Model (WFM) pretrained on 40 million hours of sensor data. This advanced AI understands complex physiological dynamics and "physiological rhythms" that simple aggregates miss. The WFM is even "missingness-agnostic," meaning it can accurately predict IR even if you occasionally forget to wear the device.
The true breakthrough occurs when wearable data is combined with other health markers. By integrating sensor data with demographics (age and BMI) and routine blood biomarkers, such as fasting glucose and lipid panels, these models achieved a high performance of 0.88 AUROC in independent validation. This multimodal approach proves that wearables significantly enhance the predictive power of standard medical tests.
Finally, wearables offer a longitudinal view of your health that a single lab visit cannot capture. By monitoring activity and sleep "complex dynamics" 24/7, they detect metabolic dysfunction early, often while blood sugar levels still appear "healthy". This shift toward continuous monitoring allows for timely lifestyle interventions that can reverse the progression toward type 2 diabetes
Transitioning from clinical research to an actual clinical solution requires a clear framework that balances data science with user engagement. Moving beyond "raw data" to actionable health insights involves a strategic, four-step technical implementation.
Digital health platforms must first ingest high-resolution time-series data from various consumer devices, such as Fitbit or Pixel watches. This raw sensor data is preprocessed and summarized into embedded representations. To ensure the AI modeling remains accurate across diverse users, input features are standardized to have zero mean and unit variance, making the data "agnostic" to the learning model.
The platform utilizes a Wearable Foundation Model (WFM) to learn robust, high-dimensional feature representations. A critical advantage of the WFM is that it is "missingness-agnostic". Because it is trained on millions of hours of sensor data, it can still extract accurate health insights even when data is incomplete, a common occurrence when users forget to charge their devices or leave them off-body.
To make clinical data useful for consumers, platforms integrate an "IR Agent" powered by Large Language Models like Gemini 2.0 Flash. Using a "Reason and Act" (ReAct) framework, the agent translates complex HOMA-IR scores into personalized, holistic insights. It contextualizes results, offering illustrative explanations and lifestyle recommendations tailored to the individual's specific profile.
Finally, digital health providers must ensure all AI-generated responses are factually accurate and clinically safe. In the WEAR-ME study, board-certified endocrinologists rated the IR Agent’s outputs for factuality and safety, ensuring the system remains a trustworthy tool for metabolic literacy.
Wearable devices and AI models evolve at a breakneck pace. New sensors, improved algorithms, and sophisticated foundation models, such as the one used in the WEAR-ME study, emerge every year, prompting users to upgrade their hardware to track metabolic risk better. But while devices change frequently, a person’s metabolic health history must remain continuous to accurately predict risks like insulin resistance.
At Thryve, we focus on making metabolic health data truly portable by building infrastructure that harmonizes wearable data across ecosystems. Our API ensures that your health journey isn't reset every time you switch devices. With Thryve, you get:
If you are building products that depend on continuous health data to detect risks like insulin resistance, explore how our diabetes infrastructure can make metabolic monitoring seamless.
Book a demo with Thryve!
Paul Burggraf, co-founder and Chief Science Officer at Thryve, is the brain behind all health analytics at Thryve and drives our research partnerships with the German government and leading healthcare institutions. As an economical engineer turned strategy consultant, prior to Thryve, he built the foundational forecasting models for multi-billion investments of big utilities using complex system dynamics. Besides applying model analytics and analytical research to health sensors, he’s a guest lecturer at the Zurich University of Applied Sciences in the Life Science Master „Modelling of Complex Systems“
.png)