#54–Kevin Sayer, CEO of Dexcom: Continuous glucose monitors–impact of food, sleep, & stress

Attia's complementary cadence: 7-day report every day for SD and average trend, then 90-day trailing report once a month to see the longer-horizon picture. He frames SD as the metric most likely to predict health outcomes at a population level — more informative than body weight fluctuations, because glucose SD directly captures metabolic variability in a way that weight cannot.

Attia found that on high-cortisol nights his glucose would rise from 93 at bedtime to 105 by morning — a 12-point rise attributable purely to hepatic glucose production driven by elevated cortisol, not insulin resistance. The intervention is whatever reduces nighttime cortisol: stress management, sleep quality improvement, evening cortisol-calming practices. The measurement clarifies the target.

Attia describes two airport incidents: in one, he glanced at his CGM, saw his glucose was fine, and left the trail mix on the shelf. In another (pre-CGM learning), he ate yogurt-covered raisins, hit his only ever >200 reading, and crashed to the low 70s on the rebound. The behavioral change requires the real-time data — a knowledge of glycemic index in the abstract does not produce the same avoidance behavior as seeing a number spike in real time.

Attia's personal data: ~30-point spike during a workout, then ~10-point lower average glucose for the rest of that day compared to a sedentary day. Sayer separately notes that what you eat before exercise versus after produces a 'very different glycemic response' — the physiologic state at the time of eating determines the excursion more than the food alone.

Attia's experience: lab A1c 5.6–6.0 vs CGM-imputed A1c 4.5–5.1. The discrepancy has caused repeated problems with life insurance underwriting. His practical protocol: calibrate the G6 daily with a finger-stick even though calibration is optional, generating a tightly-referenced CGM trace. The 90-day trailing average then gives him a number he trusts more than the laboratory value.

Sayer describes the evolution toward automated insulin delivery (artificial pancreas) as the natural extension of this principle — the CGM provides trend data that the insulin pump algorithm uses to modulate basal and bolus delivery in real time. The Tandem pump Dexcom was partnering with shuts off insulin delivery when glucose falls below threshold and restarts when the trend reverses — essentially operationalizing the trend arrow at hardware speed.

Sayer describes his interest in developing a CGM-based diagnostic alternative to the OGTT: 'a blinded sensor on somebody for a week — you know infinitely better' than a single oral glucose challenge. His daughter-in-law with gestational diabetes during a twin pregnancy had a Dexcom placed; the OB-GYN called it 'remarkable — we should all be on this.' The regulatory path requires a gestational-specific study, which Dexcom intended to run.

Kushner's T1D patients were achieving A1cs that most non-diabetic patients don't hit — 5.6 — while simultaneously reducing hypoglycemic events. The mechanism is that CGM lets patients see trend arrows (rising fast, falling fast) and intervene with micro-corrections rather than waiting for symptoms or finger-stick snapshots. Attia frames this as a paradigm shift: the old standard was set by the limitations of the measurement tool, not by biology. Once you have 5-minute continuous data with trend direction, the ceiling on glycemic control moves dramatically.

Attia's daily CGM practice: he reads the 7-day report every morning, checks average and SD, then about once a month pulls the 90-day trailing average. He describes his actual imputed A1c at tight nutrition as 4.5–5.1 — a full percent lower than what his blood test would show. The SD component is particularly actionable: a rising SD before the average moves tells you variability is increasing before the average glucose has changed. Sayer notes that Dexcom's internal goal ('eliminating the outliers') is precisely about minimizing sensor-to-sensor SD in performance — same concept applied to device quality control.

Attia's protocol: collect overnight urine to measure cortisol output, then overlay the resulting cortisol estimate against the CGM's overnight glucose trace. He found that the degree of cortisol production predicted the overnight glucose rise. The practical clinical translation: a patient presenting with a fasting glucose of 100 who slept poorly the night before is not necessarily insulin-resistant — their cortisol spiked and drove hepatic glucose production. He now counsels patients that fasting glucose 'is not nearly as helpful as people think it is inside of a reasonably physiologic range' — a fasting glucose of 100 vs 90 can easily be cortisol, not pathology.

Sayer explicitly calls CGM 'the technology most likely to change how people eat — more than any other technology I have laid eyes on.' The mechanism is behavioral, not pharmacological: seeing a spike within 30 minutes pairs a food choice with an immediate physiologic consequence. Attia's yogurt-covered raisins story — the only time he hit above 200 mg/dL, rising from baseline to 200+ in 30 minutes then crashing to the low 70s on the rebound — is a vivid example. He has not eaten them since. The 30-minute causal loop is tight enough that even lay users make the connection.

Attia uses this observation routinely in clinical conversations: a glucose spike during a workout is not a problem — it is followed by an extended period of improved glucose disposal. The CGM provides the only way to see both effects in the same data stream. Without it, a finger-stick taken during the workout would look alarming; a finger-stick taken the next morning would look normal but miss the mechanism entirely. The net result is that exercise days have lower average glucose even though they include a higher peak.

Attia's observation: going to bed with high glucose after a high-carb dinner is associated with worse sleep quality. The reverse is also true: nights of poor sleep (whether from cortisol, apnea, or disruption) are associated with higher glucose the following day. The mechanism runs through cortisol, growth hormone secretion patterns during sleep, and reduced insulin sensitivity after sleep restriction. The CGM makes both directions of the relationship visible within the same week of data.

Sayer explains the engineering layers: metal alloy wire + multiple membrane coatings + electrochemical reaction + transmitter algorithm. The G6 automated the needle insertion (in and out faster than a hummingbird wing-flap), eliminating the patient-driven insertion variability that degraded G5 accuracy. New membrane technology + new algorithm together drove G6's accuracy improvements over G5. Calibration is still possible on G6 but no longer required. Acetaminophen, which caused false-high readings on the G5, has no effect on the G6 after a large 1986-era re-validation study.

Attia's lab A1c runs 5.6–6.0 because his RBCs stick around longer and accumulate more glycation than normal-turnover cells at the same ambient glucose. His actual 90-day CGM average, carefully calibrated with daily finger-stick checks, imputes an A1c of 4.5–5.1. The discrepancy causes practical problems: life insurance actuaries see his blood test and classify him as pre-diabetic; he has had to explain the limitation of A1c in hemoglobinopathy to non-specialist actuaries multiple times. CGM is the only tool that bypasses this confound.