Psst… We received a tip…

Revolutionary metabolic health monitoring that runs on passive data from a device already on millions of wrists. Blood tests and clinic visits move to follow-up, rather than first step. The market opportunity runs into the tens of billions. The technology sounds too good to be true.

You've seen this movie before.

Elizabeth Holmes stood in front of cameras claiming Theranos could run hundreds of tests from a single small blood sample. Investors poured in $9 billion. The technology didn't work. When the house of cards collapsed, it took down patient trust, investor capital, and an entire category's credibility… it also put her in prison.

Now Google drops a patent with a familiar vibe. This time the input signals come from heart rate, sleep patterns, and step counts. The promise is a wearable that predicts insulin resistance without a blood draw.

Google frames this as an early warning signal to discuss with your clinician, not a final diagnosis. They avoid claims of diagnostic certainty. And they have something Theranos lacked: Fitbit users already generating the data needed to train the model along with their blood sugar levels data.

And there is another question we need to ask. Wearable buyers are unlikely to be those carrying the highest diabetes burden.

This week we're uncovering a patent that turns your wrist into a metabolic early warning system, and asking whether Google's data aligns with the people who need this technology most.

Google LLC filed this patent in May 2025, but the real story starts four years earlier. When Google acquired Fitbit for $2.1 billion in 2021, regulators focused on controlling Google’s use of Fitbit data for advertising purposes. But a bigger asset sat in plain sight. Tens of millions of people generating continuous biometric data that can train metabolic AI.

Insulin resistance is the body's early warning system for Type 2 diabetes. Your cells respond less efficiently to insulin, forcing your pancreas to produce more insulin to keep blood sugar stable. Over time, this can progress to prediabetes and eventually full diabetes. 

The gold standard of diabetes detection relies on blood draws, and many people only get tested after symptoms appear, which shrinks the window for early intervention.

Google’s proposed workaround is prediction instead of measurement. The patent aggregates 1 to 4 weeks of data (resting heart rate, sleep duration, movement patterns) and feeds it through a machine learning model trained to recognize signals associated with insulin resistance.

The model classifies users into three buckets: insulin sensitive, insulin impaired, or insulin resistant. The minimal feature set that delivers useful predictions is surprisingly lean, monitoring resting heart rate during sleep, sleep duration, and BMI. Three inputs most fitness trackers already capture.

Continuous glucose monitors from Dexcom and Abbott use sensors placed under the skin, replaced every 10 to 14 days, and often cost hundreds per month without insurance (Dexcom, Abbott). Apple has reportedly pursued non-invasive glucose monitoring for years. The direct optical measurement route remains a hard engineering problem, and the industry still treats FDA clearance as a major hurdle for truly non-invasive devices.

Google takes a different route. The system estimates metabolic dysfunction associated with glucose problems, using sensors that shipped years ago.

Training this AI requires two things that are hard to obtain at scale

Google's WEAR-ME study with Quest Diagnostics, launched January 2024, paired Fitbit data with clinical metabolic testing (Google Blog). That creates exactly the kind of labeled dataset this patent needs.

The composition of that dataset matters. Fitbit's user base skews toward early adopters and fitness enthusiasts. Insulin resistance prevalence varies across ethnic groups, and biometric patterns may differ with genetics and environment. The patent notes that "different populations may have different baseline characteristics" but provides no detail on diversity targets, sampling strategy, or performance reporting across subgroups.

Research is already moving in this direction. A 2025 Journal of Medical Internet Research systematic review on wearables for early Parkinson’s disease diagnosis summarizes studies with high diagnostic performance, including an included study reporting 92% accuracy for distinguishing Parkinson’s disease from essential tremor, and it also summarizes a longitudinal study reporting 92.3% accuracy. (JMIR)

A Nature Reviews Cardiology review argues that evidence supports wearable devices in cardiovascular risk assessment and in prevention, diagnosis, and management across cardiovascular conditions. (Nature)

Researchers are also exploring wearable and smartphone-derived features for predicting depression and anxiety, including 2025 work in the Journal of Affective Disorders analyzing wearable-derived indicators in population data. (sciencedirect.com)

The same Fitbit users generating training data for one model could generate training data for dozens. Every condition with a biometric fingerprint becomes a potential product feature.

The uncomfortable math is that diabetes affects 13% of U.S. adults, but the burden falls unevenly. American Indians and Alaska Natives face the highest rates at 14.7%, followed by the Hispanic population at 12.5% and African Americans at 11.7%, compared to 7.5% among White Americans. CDC data shows that individuals with family income below the federal poverty level have the highest diabetes prevalence.

Now consider who buys Fitbits. Apple captures 58% of the U.S. smartwatch market, with Fitbit at 25%. Fitness enthusiasts are the largest buyer segment. The highest-burden groups sit elsewhere.

Google's teams will be aware of this distribution mismatch. The patent acknowledges baseline differences across populations. Still, the document offers no concrete diversity requirements for training data, and public readers cannot see subgroup performance. The core risk is performance drift between the data-generating population and the populations with the highest need, even under careful adjustment.

The FDA recently clarified a path for more wearables to sit in a general wellness category, with lighter oversight, as long as intended use stays strictly wellness-focused. Non-invasive products estimating blood pressure, blood glucose, or other physiologic parameters for wellness uses can qualify when the claims stay on the wellness side rather than diagnostics.

Google can position metabolic prediction as a wellness signal that prompts a conversation with a clinician, and operates within that pathway as long as the claims and labeling stays disciplined. That same door is open to founders building similar predictive features.

Google isn't alone. Bloomberg has described Apple's noninvasive glucose monitoring effort as "many years away" after more than a decade of work. Apple has invested heavily which signals how strategically important the category is.

Apple and Google are placing fundamentally different bets. Apple is pursuing direct glucose measurement via optical spectroscopy, an engineering moonshot. Google is pursuing metabolic status prediction from heart rate and sleep using the same technology, an algorithmic shortcut built on existing hardware and a large installed base. If Apple achieves reliable direct measurement, Google's approach becomes supporting context. A long timeline also leaves plenty of room for predictive approaches to win distribution and mindshare first.

For founders watching this space, the opportunities cluster around Google's patent rather than head-on competition.

The access gap. Consumer wearables concentrate in higher-income groups. Community health centers serve higher-risk populations. A bridge can form through employer programs, Medicaid/public healthcare partnerships, and hardware subsidies that put metabolic screening in the hands that need it.

The intervention layer. A risk signal is only the first step. The product gap sits between "your patterns suggest elevated risk" and "here is a personalized plan and support to change behavior." Companies like Noom and Virta Health pair metabolic programs with coaching and support. That middle layer is where behavior-change companies can build.

The acqui-hire path. Oura closed a $200 million Series D at a $5.2 billion valuation, with a $75 million strategic investment from Dexcom after acquiring Veri. The giants are buying what they cannot build fast enough. Startups with validated metabolic algorithms or unique training data become acquisition targets.

Capital is flowing into wearables, and into everything that connects sensors to intervention.

Function Health raised $298 million in a Series B round in November 2025 at a $2.5 billion valuation. The Austin-based company offers nutrition, supplement, and lifestyle guidance based on lab tests and body scans . Since launching in 2023, Function has completed more than 50 million lab tests and recently lowered its annual membership from $499 to $365, positioning preventive screening as a subscription product (Fierce Healthcare).

Neko Health, the Swedish body-scanning startup, raised $260 million in January 2025 at a $1.8 billion valuation. The company delivered six times more scans in 2025 than 2024, with global signups exceeding 300,000 (nekohealth.com). Of the thousands of scans completed in Stockholm during 2024, 1.2% revealed life-threatening conditions and 6.4% found medically significant findings requiring clinical attention.

Prenuvo raised $120 million in 2024 for its whole-body MRI screening service, surpassing 110,000 members (prenuvo.com). The pattern across all three: investors are betting consumers will pay out-of-pocket for early detection, bypassing the traditional insurance-first healthcare model.

Oura closed a $200 million Series D in December 2024, valuing the smart ring maker at $5.2 billion (mobihealthnews.com). Oura CEO Tom Hale said the company is on track for $1 billion in 2025 sales and could reach "close to $2 billion" in 2026 (Sacra).

In September 2024, Oura acquired Veri, a personalized metabolic health company that uses CGM (continuous glucose monitoring) data to guide food choices. That acquisition was a factor in Dexcom's $75 million strategic investment in Oura's round (mobihealthnews.com). Metabolic health is a stated priority. 97% of Oura members expressed interest in understanding how food impacts their health (Latka).

WHOOP hit $260 million in revenue in 2025 with a subscription-first model. Users subscribe for $199 to $359 annually. In September 2025, WHOOP opened its Advanced Labs blood-testing service to a 350,000-person waitlist, offering Quest Diagnostics panels integrated with continuous monitoring (Business of Apps).

Health and fitness apps generated over $5 billion in revenue in 2024, up 14% year-over-year, with about three-quarters coming from subscriptions (Business of Apps).

MyFitnessPal led fitness apps with $12 million in monthly in-app revenue in January 2024, followed by Strava at $5.68 million and Peloton at $5 million. WeightWatchers generated $452 million in 2024, followed by Noom. Noom has raised $632 million over 16 rounds and hit $500 million in revenue in 2023 (Latka) with its psychology-based weight loss approach.

In October 2025, Noom added AI-powered Face Scan to its free tier, using remote photoplethysmography (light measured through your phone camera) to generate a 30-second health screening with biological age, vital signs, and cardio-metabolic indicators. In November 2025, Noom introduced Glucose Forecasting, which predicts the glucose impact of meals without requiring CGM hardware (noom.com). Same sensor signals, predictive algorithms, metabolic outcomes. The patent Google just filed fits a playbook these companies have already validated.

In 2022, diabetes cost the U.S. $413 billion annually, with $307 billion in direct medical expenses (PubMed). Prediabetes alone costs the U.S. economy $43.4 billion a year (American Medical Association). Every company betting on metabolic health is betting that catching 115 million prediabetic people earlier shifts the cost curve.

Google's patent turns your smartwatch into a metabolic early warning system, predicting insulin resistance from heart rate and sleep patterns you're already generating. If performance holds up, people at elevated risk can get flagged earlier, then confirm with clinical testing when needed. 

Your wrist is already collecting the data. The question is whether you trust the algorithm interpreting it.

Got thoughts on where this is heading? Hit reply, we read every response.

Read the source: WO2025231209A1, Non-Invasive Prediction of Insulin Resistance Using Wearable Sensor Data · Published January 2025 · Assignee: Google LLC · Status: Pending