By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
Peter Attia MD270 ‒ Journal club with Andrew Huberman: metformin, power of belief, & how to read scientific papers
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
CHAPTERS
- 0:02 – 6:38
Why a Journal Club: Dreams, Yerba Mate, and Choosing Papers
Andrew and Peter explain why they want to discuss papers in real time, so listeners can learn how to dissect scientific literature. They banter about Peter's dream involving dew and Andrew's yerba mate habit, and Andrew briefly mentions a transformative dream.
- •Goal is to show how to pick through a paper, including the good, the bad, and the ugly
- •Peter's practice used to run a monthly internal journal club
- •Peter dropped a more technical paper on ATP binding cassettes in cholesterol absorption in favor of a simple epidemiologic paper
- •Yerba mate and Peter's dream about Andrew's special drink
- 6:38 – 13:45
Metformin and Why the Bannister Paper Mattered
Peter introduces the 2022 metformin paper by Keyes and colleagues and explains it revisits the 2014 Bannister paper. He then describes what metformin is and how it inhibits complex one of the mitochondria and reduces hepatic glucose output.
- •Bannister's 2014 paper sparked excitement about geroprotection with metformin
- •Metformin has been used as a first-line type 2 diabetes drug for decades
- •It weakly inhibits mitochondrial complex one, unlike cyanide, which blocks the chain completely
- •Hepatic glucose output is a core problem in type 2 diabetes, and the circulation holds only about five grams of glucose
- •Metformin changes how the body partitions fuel rather than depriving it of calories
- 13:45 – 20:43
Insulin Resistance, Its Causes, and Diabetes Drugs
Peter explains how insulin triggers glucose transporters in muscle and how intracellular fat disrupts that signal, causing insulin resistance. He names inactivity, sleep deprivation, cortisol and energy imbalance as causes, and compares metformin with other diabetes drugs.
- •Insulin triggers a transporter to be placed in the cell membrane so glucose can flow in
- •Rising intramuscular fat means more insulin is needed; high insulin is the early warning sign
- •Inactivity is probably the most important cause; sleep restriction to four hours for a week cut glucose disposal by about half
- •Different drugs target different parts of the chain: sulfonylureas, insulin, metformin, GLP-1 agonists
- •Berberine is described as a poor man's metformin
- 20:43 – 23:32
Peter's Metformin Experiment and Defining Geroprotection
Peter recounts starting metformin in 2011 at the full dose and getting severely nauseous at the Berkshire Hathaway meeting. He then defines geroprotection as targeting hallmarks of aging rather than one disease.
- •Peter started metformin in May 2011 without titrating
- •Early appetite suppression and nausea were side effects
- •Geroprotective agents target hallmarks of aging rather than a specific disease
- •Metformin, rapamycin and NAD precursors are cited as examples
- 23:32 – 32:49
The 2014 Bannister Study and Informative Censoring
Peter describes how Bannister matched metformin-treated diabetics to controls from a UK registry and reported a 15% relative reduction in all-cause mortality. He flags informative censoring as a significant limitation and explains crude death rates per 1,000 person years.
- •Bannister matched metformin-only type 2 diabetics with non-diabetic controls
- •Informative censoring excluded patients who stopped metformin, were lost to follow-up, or progressed to other drugs
- •Crude death rates were 14.4 vs 15.2 per 1,000 patient years, and the hazard ratio implied a 15% reduction
- •Type 2 diabetes shortens life by about six years on average
- •Proponents suggested off-target effects such as mTOR inhibition and reduced inflammation
- 32:49 – 37:57
The Keyes Replication: Danish Cohort, Twins, and Table One
Peter presents the Keyes paper, which used about half a million Danish registry subjects and added a discordant-twin analysis and a censoring sensitivity analysis. They note the TAME trial and examine table one, where medication use was far from matched.
- •The TAME trial aims to test metformin prospectively with random assignment
- •Keyes used singletons plus discordant same-sex twins
- •Table one shows baseline characteristics and should always be checked
- •Medication use differed greatly, e.g. lipid-lowering drugs 45.6% vs 15.4%
- •Confounders can never be fully removed without randomization
- 37:57 – 49:21
Mortality Rates, Person-Years, and Kaplan-Meier Curves
They go through table two's crude mortality rates and unpack what per-1,000-person-years means. They then read the Kaplan-Meier curves and discuss confidence intervals, error bars, and the hazard-ratio model.
- •Crude mortality was higher in the metformin diabetic groups, roughly 25 vs 17 per 1,000 person years in singletons
- •Person-years normalize for different follow-up times
- •Kaplan-Meier curves plot cumulative survival over time and only go down or stay flat
- •Shaded bands show 95% confidence intervals from a Cox model smoothing the data
- •Metformin users with diabetes had a major health issue that controls lacked
- 49:21 – 54:20
Statistical Significance, Power, and Sample Size
Triggered by the error bars discussion, Andrew and Peter cover common misconceptions about overlapping error bars, p-values, and statistical power. They explain how larger samples reduce variability and why epidemiology relies on huge cohorts given the cost of randomized trials.
- •Overlapping error bars do not automatically mean non-significance
- •Failing to reach significance only means something relative to what the study was powered to detect
- •Power tables determine the sample size needed for an expected difference
- •Larger samples reduce variability, and epidemiology compensates for bias with scale
- •The Women's Health Initiative, a roughly billion-dollar trial, illustrates the cost of randomized studies
- 54:20 – 1:05:20
Hazard Ratios Reveal Higher Mortality on Metformin
Peter walks through table four: hazard ratios showed metformin-treated diabetics had higher mortality than non-diabetic matches, including 1.48 unadjusted in singletons and 2.15 in twins. Censoring changed results only modestly, and Peter concludes metformin did not undo the disadvantage of diabetes, though a true effect can't be determined without a randomized trial.
- •A hazard ratio above one means increased risk; Bannister's 0.85 meant a 15% reduction
- •Singleton adjustments moved the ratio from 1.48 to about 1.32-1.33
- •Twins showed 2.15 unadjusted, falling to about 1.7-1.8 after adjustment
- •Removing censoring versus censoring moved 48% to 39% in singletons and 115% to 97% in twins
- •Meat-eater epidemiology is used as an example of uncontrollable confounders
- 1:05:20 – 1:16:07
Animal Lifespan Data, Peter Stopping Metformin, and Glucose-Lowering Agents
The conversation turns to the Interventions Testing Program, where rapamycin and some other agents extended lifespan but metformin did not. Peter explains he stopped metformin after finding elevated resting lactate, and Andrew shares his berberine experience, followed by acarbose and Peter's use of an SGLT2 inhibitor.
- •The ITP found rapamycin, 17 alpha-estradiol, canagliflozin and acarbose worked, while metformin failed
- •Peter stopped metformin around 2018 after seeing resting lactate of about 1.6 mmol
- •Andrew used berberine to buffer cheat-day carbohydrate loads
- •Acarbose extended mouse lifespan without lower weight
- •Peter's only current glucose-related drug is an SGLT2 inhibitor
- 1:16:07 – 1:24:58
Caloric Restriction, Fasting, and Longevity Biomarkers
Peter describes how energy balance matters more than macronutrients, and says he will reconsider metformin if TAME shows benefit. Andrew asks whether short fasts help longevity; Peter says probably not, recounting his own fasting regimen and its muscle loss, and argues the field lacks aging biomarkers.
- •Glucose control can be uncoupled from calorie deficit through sleep, exercise, and thoughtful eating
- •Severe caloric restriction is unrealistic for most people
- •Peter would reconsider metformin if TAME shows a geroprotective benefit
- •Peter's repeated multi-day fasts may have cost him muscle mass, with unknown longevity benefit
- •Lack of validated biomarkers of aging is the major barrier in geroscience
- 1:24:58 – 1:30:10
How Peter and Andrew Read Scientific Papers
Andrew shares the four questions he taught students to ask of a paper, and both describe their reading order. Peter adds that supplemental material often holds key analyses.
- •Four questions: what question, what approach, what they found, and whether conclusions are substantiated
- •Look up unfamiliar methods rather than mastering every one
- •Papers often need rereading and going figure by figure
- •Peter reads figures first, then methods if the topic is unfamiliar
- •Important analyses can sit in the supplemental section
- 1:30:10 – 1:43:03
Belief Effects vs. Placebo and the Nicotine Brain Circuitry
Andrew introduces his paper on belief effects, distinguishing them from binary placebo effects with examples from Alia Crum's work on stress, milkshakes, and hotel workers. He then explains how nicotine acts on thalamic and prefrontal circuits and how it relates to attention and ADHD drugs.
- •The Perl paper is a bioRxiv preprint, not yet peer reviewed
- •Belief effects scale with information, unlike binary placebo effects
- •Crum's examples: stress mindset, ghrelin after a labeled milkshake, hotel worker weight loss
- •Nicotine boosts signal-to-noise via the thalamus and engages reward and ventromedial prefrontal circuits
- •Peter asks how nicotine can be both focusing and calming; Andrew says central focus and peripheral muscle relaxation
- 1:43:03 – 2:02:52
The Perl Experiment: Same Dose, Different Beliefs
Smokers vaped the same low nicotine dose but were told it was low, medium, or high while undergoing fMRI during a trading task. Brain activity scaled with belief, particularly in the thalamus-to-prefrontal pathway, while the pair discuss fMRI limits, figures, statistical power, and missing controls.
- •All participants received the same low dose but were told different strengths
- •Subjective feeling matched the instruction, and thalamic activation and thalamus-prefrontal connectivity scaled with belief
- •Reward-pathway response did not differ by belief
- •Peter notes a weak correlation of 0.27 and small sample sizes, with about 20 per group
- •No zero-nicotine control group was included
- 2:02:52 – 2:11:12
Implications of Belief Effects and a Medical School Tangent
Andrew and Peter discuss implications of dose-dependent belief effects for ADHD dosing, quitting nicotine, blood pressure, GLP-1 drugs, and side-effect framing, including the nocebo effect. The episode closes with medical school anecdotes about zebras and rare conditions and a plan to meet again in Austin.
- •Beliefs may modulate drug effects, with implications for dosing and quitting nicotine
- •Peter notes that trials say you might get the drug, unlike this experiment where participants were told they definitely did
- •Framing side effects as evidence the drug works may reduce distress and improve perceived relief
- •Nocebo effects can make some people unable to tolerate medications
- •Medical students over-think rare diagnoses like sarcoidosis and situs inversus
