What are the best wearables? And why does Peter use a continuous glucose monitor? (AMA #1)

Attia's personal CGM use has continued for three years at time of recording. The stickiness comes from both diagnostic value (still learning from it) and behavioral feedback (real-time accountability). He spot-checks his G6 against fingerprick once daily by choice even though the device doesn't require it — a habit from the G5 era. The key insight: unlike step-count wearables where you quickly learn everything the metric will tell you, glucose is variable enough meal-to-meal and day-to-day that the data stream remains informative indefinitely.

Attia's empirical observation is that wearables abandoned at the two-week mark were all ones that either told you something irrelevant or told you something relevant but inaccurately or without a feedback loop you could act on. The step counter example is particularly instructive: you learn your baseline step count within a few days, you can't really change it easily, and it doesn't predict disease outcomes with useful specificity. By contrast, glucose is highly actionable — eat differently, exercise differently, manage stress — and the CGM gives you immediate feedback on those interventions.

Attia uses LP(a) as his canonical example: it is a codominant inherited gene, meaning you can trace it through a pedigree like a dominant trait. When he sees three generations of premature cardiovascular disease in a family tree, he predicts elevated LP(a) with high confidence and is almost never wrong. The family history effectively pre-empts the need for genome sequencing in most cases — because the genome will confirm what the pedigree already showed. For cancer screening, Attia notes the relationship requires double-clicking into specifics: a first-degree relative with bladder cancer may or may not be relevant to the patient's prostate cancer risk, and that nuance cannot be captured by a 23andMe result.

Attia explains the step-count failure explicitly: 'I know how many steps I take, why do I care?' For a metric to earn wearable real estate, it has to be clinically relevant, not just quantifiable. The Libre failed criterion two (inaccurate, no forced calibration) and criterion three (no phone interface, required a separate reader). The Dexcom G5 passed all three but required twice-daily calibration, creating friction for non-diabetic users. The G6 prototype that Attia had at the time of this AMA passed all three with the friction removed: plug-and-play insertion, no mandatory calibration, direct phone integration with real-time glucose readings.

The Libre was bought by Abbott and marketed as a simpler alternative — no calibration required, inserted on the back of the arm. But Attia's clinical experience was that it was 'categorically useless — so inaccurate you can't force a calibration, and it doesn't interact with your phone.' The G6 resolves all three of these failure modes. Insertion is as easy as the Libre. Accuracy rivals the G5 (Attia was 'blown away'). Phone interface is 'second to none.' The only concession is that the G6 is calibration-optional rather than calibration-free — Attia still spot-checks his once a day voluntarily.

Attia frames the trade-off clearly: with CGM plus food logging you 'get to see a month of someone in their real environment eating.' A patient eating poorly during that month will almost certainly generate a revealing glucose response. The limitation is that you lose the insulin dimension — hyperinsulinemia could be present without clearly abnormal glucose. But the breadth of behavioral and environmental data captured over 30 days offsets that gap for most practical diagnostic purposes. He adds the caveat explicitly: 'it's still not a complete substitute for that hyperinsulinemia, so it's not perfect.'

The strongest illustration Attia gives is LP(a) (Lp-little-a): a cardiovascular risk marker with codominant inheritance. When he takes a history and sees 'a violent streak of heart disease' through three generations of a family — dad had it, dad's mom had it, dad's mom's mom had it — he knows with high confidence the patient will have elevated LP(a) before a single blood test has returned. The genome sequencing would confirm what the family pedigree already told him. For cancer, the relationship is even more nuanced because first-degree cancer relationships are cancer-type-specific: a first-degree relative with bladder cancer has different implications than a first-degree relative with prostate cancer, and no genome scan resolves that — only a detailed family history does.

Attia's granola story makes the mechanism concrete: he didn't eat the granola not because of willpower, not because he was afraid of long-term health consequences, but because he could see in his mind the glucose spike trace he would have to look at on his phone afterwards. 'It just pisses me off, so it's like I'm not gonna do it.' He speculates that without the CGM he probably would have eaten the whole box. This is a classic implementation of the behavioral accountability loop: immediate visible feedback converts a distant health consequence (metabolic disease over decades) into a proximate emotional consequence (frustrated by a bad glucose trace on your phone in 30 minutes), which is neurologically much more effective at modifying behavior.