Insulin Testing Could Reveal Metabolic Risk Before Blood Sugar Rises

Mark Hyman describes Function as an attempt to move healthcare away from episodic treatment and toward continuous, patient-held information. The premise is not that people should replace clinicians. It is that they should not have to remain ignorant of their own biology until they are ill enough to generate a diagnosis, billing code, or prescription.

Hyman says the conventional sequence has been rigid: a patient goes to a doctor, knows—or does not know—what to ask for, hopes the clinician agrees to order it, then hopes insurance covers it. Patients rarely receive an integrated view of what is happening “under the hood.” Function’s model is to bring medical history, laboratory results, connected wearable data, available electronic records, and imaging into a personal health platform, then use AI-assisted explanations and action plans to make the information usable.

He distinguishes Function Health from functional medicine, though he says the company incorporates that framework. Functional medicine, as Hyman defines it, treats the body as a network rather than a collection of separate organs and diagnoses. It puts diet, exercise, sleep, stress, nutritional status, hormones, inflammation, toxins, mitochondria, and the microbiome in the same frame. Its orientation, he says, is toward root causes and the creation of health rather than treatment after disease has become manifest.

Function is broader than functional medicine in his account: conventional medical science combined with emerging forms of measurement. The intended result is a longitudinal view that lets a person identify a current issue or recognize a trajectory toward diabetes, Alzheimer’s disease, or another serious condition early enough to “get on the off ramp.”

The commercial proposition is built around the cost and fragmentation of care. Hyman says Function’s $365 annual membership includes two rounds of testing covering more than 160 biomarkers, with additional testing available for hormones, toxins, metabolic health, brain health, nutrition, and cancer screening. The platform partners with Quest rather than operating its own laboratory; Hyman says users can use more than 2,000 draw stations, with mobile phlebotomy available in some cases.

Hyman says Function has processed more than 100 million biomarkers across “hundreds and hundreds of thousands” of users. He characterizes that population as health-forward but not exclusively affluent, saying its average income is below $100,000. The offer is meant to supply part of what Jason Calacanis calls a concierge experience without concierge pricing.

Hyman’s example of the healthcare system’s pricing problem is vitamin D testing. At Cleveland Clinic, where he formerly worked, he says a vitamin D test could cost $300; at Quest, he says, it might cost $10. He compares the arrangement to buying the same Toyota for $10,000 at one lot and $100,000 at another. The point is not simply that testing should be cheaper. It is that people should be able to purchase testing, keep the results, and decide what additional questions are worth pursuing.

Calacanis frames the demand partly as an access issue, saying one in three Americans skip speaking with a doctor about a health issue because they cannot afford it. Hyman’s broader criticism is that the country has “disease insurance,” not health insurance: it is useful when something is already wrong, but offers no clear answer when someone asks how to avoid becoming sick.

Mark Hyman sees an opportunity in the expanding volume of health data, but the harder problem is operational: deciding what matters, putting information in one place, and allowing patients and clinicians to work from the same record.

Hyman contrasts conventional diagnostic practice—which he characterizes as gathering the minimum evidence needed to confirm a suspected condition—with a model that collects information before a person has a defined disease. He points to metabolomics, which can measure thousands of metabolites; proteomics, involving tens of thousands of proteins; genetics, including 20,000 genes and millions of SNPs; microbiome data; and imaging datasets that can reach terabytes per person.

The scale is central to his case for AI assistance. A clinician may take a history, order a limited set of tests, assess the results mentally, and document the encounter in an electronic medical record. Hyman calls those records a computerized version of two-dimensional paper rather than a genuinely intelligent system. In his view, AI can assemble disparate data, retrieve literature, identify patterns that warrant attention, and help clinicians ask better questions.

Hyman invokes Lee Hood’s concept of P4 medicine: preventive, predictive, personalized, and participatory. The final term matters to his argument. Testing may make a risk visible, but it does not itself change a person’s diet, activity, sleep, treatment decisions, or follow-up. The patient must participate.

Whole-body MRI illustrates both the promise and the practical complications of expansive testing. Hyman says Function’s MRI technology is FDA approved, takes 22 minutes, and does not use radiation. He considers a baseline scan valuable, especially when repeat scans establish a longitudinal record. A single scan can find something meaningful, he says, but the ability to see changes over time is where imaging becomes more valuable.

Scans can also produce incidental findings: a kidney or liver cyst, for example, that is not itself a problem. Hyman says radiologists can often identify those findings, while some require follow-up. He says Function has found aneurysms, pituitary tumors, and other serious conditions through imaging. His case for the scan is therefore not that every finding is consequential, but that earlier identification of serious disease can matter and repeated measurement can show where a person is headed.

The company also offers Galleri, GRAIL’s multi-cancer early-detection blood test.^† Hyman says it looks for more than 50 cancers, is “about 75 percent good” at detecting them, has a low false-positive rate, and does not find every cancer. He places it alongside familiar screenings such as colonoscopy, mammography, Pap tests, and PSA testing, which address particular cancers and have their own limitations and controversies.

Function’s effort to centralize records addresses a separate problem. Hyman says patients often arrive with laboratory results from one provider, imaging from another, and records scattered through hospitals and specialist practices. He calls the resulting stack of documents “chartomegaly”: information that technically exists but is difficult for a clinician to reconcile in the time available.

The platform can connect wearables, ingest outside laboratory and imaging results, and connect electronic medical records, Hyman says. Its current sharing model is still limited. A user can give a clinician login information or download reports to send them; deeper integration remains a future objective. Jason Calacanis calls the desired model “multiplayer mode”—a patient-controlled record that can be shared with the right clinicians without paper charts, CD-ROMs, passwords, and manual transfer.

AI is meant to assist that model, not operate without review. Hyman cites OpenEvidence as a tool through which a clinician can enter a case, retrieve relevant literature, and work through a question outside the clinician’s most frequently encountered conditions.^† Calacanis refers to a Harvard study which he says found large language models as accurate as or more accurate than doctors in comparable situations, with patients preferring the models’ bedside manner. Hyman’s own assessment is more guarded: medical AI has gone from “terrible” to “actually pretty good” over the last three or four years, but it is not ready to operate without safeguards.

Function uses doctors in the loop, Hyman says, with physicians reviewing language-model outputs for clinical appropriateness. He accepts Calacanis’s shorthand: “trust but verify.” AI’s role, in this account, is to make a vast and fragmented record more legible—not to remove medical judgment where an incorrect recommendation could matter.

Wearables belong in the same category of potentially useful but not universally beneficial measurement. Hyman says they can surface subtle changes in sleep, heart rate, blood pressure, and illness, while helping people see the behavioral costs of late nights or alcohol. Calacanis describes an elevated skin-temperature alert that prompted him to test for COVID. Hyman’s caveat is straightforward: if monitoring recovery, heart-rate variability, sleep, and stress makes someone more stressed, it may not be useful for that person.

Mark Hyman argues that clinicians routinely miss insulin resistance because they test blood sugar but not insulin. He calls insulin “probably one of the most important biomarkers for your health” and says the test costs only a few dollars. Yet he says Quest told him that fewer than 1% of submitted lab panels include it.

Function’s internal data, according to Hyman, suggests that 65% of users have high insulin levels. He also says nearly 70% have a nutritional deficiency at a level the laboratory defines as low, and 54% have high ApoB. Hyman calls ApoB a more important lipid measure than LDL or other cholesterol markers, citing American College of Cardiology literature.

The distinction between blood sugar and insulin does most of the work in Hyman’s argument. He describes a patient whose blood sugar was normal while insulin was around 100, which he characterizes as extremely high. In his explanation, elevated insulin can keep blood sugar within the normal range while underlying metabolic dysfunction worsens. By the time blood sugar rises, he says, a person may have been progressing toward metabolic disease for years.

He also disputes the idea that a laboratory reference range is necessarily a health target. Reference ranges describe what is statistically common in a population, Hyman says, not what is optimal. If excess weight is common, it may be statistically normal without being healthy. He says a laboratory insulin reference range may extend to 18, while he considers an optimal level below 10 or even 5.

Continuous glucose monitors offer a way for people to see individual variation rather than rely entirely on generalized food rules, Hyman says. He does not think people need to wear one forever, but believes a monitor can show how a particular person responds to particular foods. Genetics and the microbiome, he says, contribute to different glucose responses to the same food.

Jason Calacanis describes his own experiments: cereal sharply spiked his glucose, while Häagen-Dazs caused a smaller spike. Hyman attributes the difference to the fat and protein in ice cream versus rapidly absorbed processed starch in cereal. The composition and order of a meal affect glucose dynamics, they argue. Hyman endorses eating carbohydrates after protein rather than first and walking after meals; Calacanis began parking several blocks from restaurants to make a post-meal walk routine.

Measurement does not settle what intervention should follow. Calacanis describes losing weight from 213 pounds to 168 with GLP-1 medication and intermittent fasting after years of gaining two or three pounds annually. Hyman calls GLP-1 drugs a useful tool and says they can be lifesaving for people who have struggled with weight in an environment dominated by processed food. But he rejects the idea that they are a universal answer or an intervention for someone who merely wants to lose five pounds.

Hyman says altered microbiomes and diets that fail to support them may be part of the obesity problem because the body produces its own GLP-1 and a healthy microbiome supports endogenous production. He says GLP-1 drugs have tradeoffs: about 4% of users experience serious adverse effects, in his characterization, and about 75% experience digestive issues.

His primary concern is muscle loss. A person who loses muscle while losing weight loses what Hyman calls the body’s metabolic engine. Muscle takes up glucose and supports metabolic function; less muscle means fewer calories burned at the same body weight. If weight returns after medication stops—as Hyman says most people regain it—the returning weight may come back as fat.

Hyman calls the resulting condition “skinny fat,” or TOFI: thin on the outside, fat on the inside. His recommendation is that GLP-1 drugs be prescribed alongside adequate protein, nutrition guidance, and strength training.

He is especially cautious about microdosing. Calacanis says microdosed GLP-1s have become popular among people who are already visibly fit. Hyman says there is not much data on the practice and that its long-term benefits and consequences are unknown. People may report positive experiences, he says, but he does not think it is personally necessary for someone who understands how to eat and exercise in a way that maintains body composition.

That view leads to Hyman’s broader insistence on resistance training. He calls muscle “the currency of longevity.” Rucking, carrying weights, bands, kettlebells, bodyweight work, and conventional strength training can all serve the purpose, he says. Even 20 minutes three times a week covering major muscle groups can matter. Walking is valuable for people who do little else; weighted walking can add strain that Hyman says supports bone density and metabolic health.

Hyman says he began strength training at 59 after years as a thin runner and cyclist, and that it transformed his body. The concern is age-related muscle loss, which he associates with frailty, disability, dysfunction, and decline. In this model, medication can help create an opening for weight loss, but maintaining metabolic capacity still depends on what happens to muscle.

Jason Calacanis worries that psychiatry has become too quick to turn distress, academic difficulty, or behavioral problems into diagnoses and long-term prescriptions. He describes friends who spent years on SSRIs, Klonopin, Adderall, or multiple psychiatric medications and later felt they had lost a decade. He is particularly concerned about stimulant use in children.

Mark Hyman agrees with the premise. “We’ve medicalized psychiatry,” he says. Hyman does not dispute that the country has a mental-health crisis. His argument is that conventional psychiatry too often labels someone as depressed, anxious, ADD, OCD, or otherwise disordered and treats the label with drugs without adequately investigating potentially contributing biology.

He gives the example of a Function patient whose results showed low omega-3s, low vitamin D, low B vitamins indicated through homocysteine testing, and insulin resistance. Hyman says findings such as those, along with low magnesium, can contribute to symptoms including depression and anxiety. He points to nutritional psychiatry at Harvard and metabolic psychiatry at Stanford as areas examining the relationship between nutrition, metabolism, and psychiatric symptoms.^†

Hyman characterizes ultra-processed diets as a major driver of the mental-health crisis and says the relationship is well established in the scientific literature. He cites a juvenile-detention intervention in which serving whole food and removing processed food reduced violent crime by 97%, use of restraints by 75%, and suicide by 100% among teenage boys.

The practical framework shared by Hyman and Calacanis is diet, exercise, sleep, meditation, and socialization. Calacanis says that when he feels anxious, depressed, or “blue,” he is usually doing poorly across those five categories; when he feels well, he is managing three, four, or five of them. Hyman agrees that lifestyle and foundational health measures can be important in addressing many psychiatric problems. He frames that as a needed rethinking of mental health.

The same demand for a deeper understanding of causes shapes Hyman’s interest in psychedelic treatments. He sees what he calls a two-pronged revolution in psychiatry: nutritional and metabolic psychiatry on one side, psychedelic psychiatry on the other. Calacanis recounts reports of ketamine helping people with severe depression and of people with PTSD describing ibogaine as changing their relationship to traumatic memories. He is also alarmed by people using ketamine casually and unsupervised, including people who take it simply because they are bored.

Hyman agrees that casual use is “not good.” Early data on psilocybin, MDMA, and ibogaine is promising, he says, but the field needs more data and is not ready for broad expansion to the general population. He supports more research and investment, noting Texas’s investment in ibogaine trials.^† His position is that research should proceed carefully, rigorously, and scientifically rather than through underground experimentation.

Hyman says these compounds may do more than provide symptomatic relief, potentially involving brain repair and neuroplasticity. He has seen reports of improved MRI scans after ibogaine treatment for brain trauma. The possibility that psychedelics might, as Calacanis puts it, “rewrite the software code” of the brain is precisely why Hyman argues for serious research rather than casual adoption.