Chapter 12 — Alcohol: The Honest Assessment — Cardiovascular "Protection," Cancer Risk, and the J-Curve That Probably Isn't

The Hook: "But isn't red wine good for you?"

We were most of the way through Theo's plan when he asked it, and he asked it the way people do — half as a question and half as a hopeful statement.

"But isn't red wine supposed to be good for your heart? I've read that for years."

He drinks two beers most evenings. Twenty-seven grams of ethanol a day, about a hundred and ninety calories, roughly fourteen drinks a week. He does not consider himself a heavy drinker and by the standards of everyone he knows, he isn't.

It is also, looking at the lipid panel from Chapter 9, the single largest modifiable lever on the page. His triglycerides are 186. Alcohol raises triglycerides reliably and dose-dependently. His sleep is six hours and fragmented. Alcohol does that too. It's 6% of his energy intake, arriving with no satiety return whatsoever.

So I had to answer the red wine question honestly, and honestly is uncomfortable, because the answer is:

The evidence that moderate drinking protects your heart has largely fallen apart over the last decade, and most people — including a lot of clinicians — haven't caught up.

I want to be careful about how this chapter reads, because there is a version of it that is a lecture, and lectures about alcohol are both useless and slightly obnoxious. This is not a temperance chapter. I'm not going to tell you what to do. Alcohol is woven into how humans have marked occasions for thousands of years, it is genuinely enjoyable, and adults get to make trade-offs about their own risk — as they do with driving, climbing, and every other thing that is pleasant and imperfectly safe.

What I owe you is the actual numbers, which are worse than you've been told and less catastrophic than the loudest recent headlines suggest, and a framework for deciding for yourself.

Then it's your call.

🏃 Fast Track: §12.4 (why the J-curve is probably an artefact), §12.6 (cancer), and §12.10 (the honest framework). Twenty minutes.

🔬 Deep Dive: §12.5 (what Mendelian randomization added — this is Chapter 2 §2.10's convergence criterion doing real work) and §12.9 (why the guidelines moved) are where students and clinicians should spend time.

12.1 What alcohol is, and what your body does with it

Ethanol is a small molecule that your body treats as a priority.

It provides 7 kcal per gram — between carbohydrate/protein (4) and fat (9) — and unlike the three macronutrients, you have no storage form for it and no way to excrete meaningful amounts. A small fraction leaves in breath and urine; the rest must be metabolized.

Because acetaldehyde — the first breakdown product — is toxic, your body clears ethanol ahead of everything else. That priority is the source of several of its effects.

The pathway

Ethanol  --[alcohol dehydrogenase, ADH]-->  ACETALDEHYDE  --[ALDH2]-->  Acetate  -->  CO2 + water

Acetaldehyde is the important molecule in this chapter. It's toxic, it damages DNA and interferes with DNA repair, and it is the reason alcohol is a carcinogen (§12.6). It's also what makes you feel terrible — the flushing, the nausea, much of a hangover.

Normally acetaldehyde is cleared quickly by ALDH2. But roughly a third to a half of people of East Asian descent carry a variant of the ALDH2 gene that clears it slowly — producing the alcohol flush reaction: facial flushing, rapid heartbeat, nausea after small amounts.

Hold on to that variant. It becomes the most important piece of evidence in the chapter (§12.5).

Three metabolic consequences

1. Fat oxidation stops. While your liver is clearing ethanol, it substantially suppresses fat oxidation — because acetate floods in as a readily available fuel and the liver's redox state shifts. You don't stop burning fuel; you switch to burning the alcohol (Chapter 6 §6.6's mixing board, with one slider pushed hard).

2. Triglycerides rise. Alcohol metabolism promotes hepatic triglyceride synthesis, which is why Theo's are 186 and why alcohol reduction moves that number faster than almost anything else.

3. Very little ethanol becomes body fat directly. This is true and it is the basis of a misleading claim, addressed in §12.2.

12.2 The calories nobody counts

Some arithmetic, because it never gets done.

Theo's two beers: 27 g ethanol, ~190 kcal/day, ~1,330 kcal/week.

Now run Chapter 4 §4.10's arithmetic on that. Against his surplus of +490 kcal/day, the beer is nearly 40% of it, from one decision made once a night without ever appearing in his mental account of what he ate.

🔬 Claim → Evidence → Verdict

The claim: "Alcohol calories don't really count — ethanol isn't stored as fat, and a lot of it's burned off as heat."

Where it comes from: Two genuinely true facts. Very little ethanol is converted directly into body fat — the conversion is metabolically expensive and quantitatively minor. And ethanol's thermic effect is relatively high (Chapter 4 §4.5), somewhere in the region of 10–30%, so not all of the 7 kcal/g is available.

What the evidence actually shows: Both facts are true and the conclusion doesn't follow, for three reasons. First, direct conversion was never the mechanism. Alcohol contributes to fat gain by suppressing fat oxidation while you're metabolizing it (§12.1) — the fat you'd otherwise have burned is spared and stored, while the alcohol is burned instead. The books balance; the sparing is the mechanism. Second, alcohol has essentially no satiety return — it doesn't reduce subsequent eating, and there's decent evidence it increases it by reducing inhibition. Third, the thermic-effect discount is real and modest: 190 kcal becomes perhaps 140–170, not zero.

📉 Evidence quality: Well-established metabolic studies.

Verdict: 🟠 Probably false. The calories count. They just count through a different mechanism than people assume, which is why the "it isn't stored as fat" fact is technically correct and practically useless — a shape you've now seen a dozen times in this book.

12.3 The J-curve: where it came from

For about thirty years, the standard picture looked like this.

📊 Diagram (described). Picture a graph: alcohol intake across the bottom, from zero on the left to heavy on the right; risk of death up the side.

The line does not rise steadily from left to right. It starts at a moderate height on the far left — at zero drinks — then dips downward as intake rises to roughly one or two drinks a day, reaching its lowest point there, before turning and climbing steeply through heavy drinking.

A shallow bowl with a long right-hand tail. A J, tipped slightly.

The visual message is unmistakable and it is what a generation absorbed: the safest place to be is not zero. Non-drinkers do worse than moderate drinkers. A drink or two a day appears protective.

This shape appeared in study after study, across countries and decades, for all-cause mortality and for cardiovascular disease specifically. It was one of the most reproducible findings in observational epidemiology.

Mechanisms were proposed and were plausible: alcohol raises HDL cholesterol; it has modest antiplatelet effects; red wine contains polyphenols including resveratrol.

Public health messaging followed. So did an enormous amount of marketing. "A glass of red wine a day is good for your heart" became something everybody knew.

And it is now, on the best available evidence, probably wrong.

12.4 Why the J-curve is probably an artefact

Three problems, and the first one is the killer.

Problem 1: who is in the "non-drinker" group?

Sick-quitter bias, and it is the cleanest example of confounding in this entire book.

When a study compares "moderate drinkers" to "non-drinkers," who are the non-drinkers?

Some are lifelong abstainers. But a substantial fraction are former drinkers who stopped — and people stop drinking for reasons: a liver problem, a cancer diagnosis, a heart attack, a new medication, a doctor's instruction, developing frailty, or a history of dependence.

The abstainer group is systematically enriched with sick people. So the reference category against which moderate drinkers look healthy is not "healthy people who don't drink." It's "everyone who doesn't drink, including a lot of people who stopped because something was wrong."

Push the reference group's risk up, and the drinkers below them look protected. That's the dip in the J.

Problem 2: moderate drinkers are healthier in every other way

Chapter 2 §2.2's healthy-user bias, arriving on cue. In most Western populations studied, moderate drinkers — as opposed to heavy drinkers or abstainers — tend to have higher incomes, more education, better healthcare access, more social connection, and better overall diets. Moderate drinking is a marker of a certain kind of stable, sociable, comfortable life.

Every one of those independently predicts lower mortality.

Problem 3: the drinking categories are badly measured

Self-reported alcohol intake is underreported at least as badly as food (Chapter 4 §4.7), and "moderate" is defined by averaging — so someone drinking fourteen units across a week and someone drinking fourteen on a Saturday can land in the same category with very different risk profiles.

What happens when you correct for these

Studies that restrict the comparison to lifelong abstainers, or that adjust more carefully for socioeconomic and health status, generally find the protective dip shrinking or disappearing — while the harm at higher intakes remains.

The J flattens toward something closer to a line rising from zero.

🧩 Productive struggle. Five minutes before reading on.

You want to test whether moderate drinking genuinely protects the heart. A randomized trial is the obvious answer: assign people to drink or not drink for years and count events.

Design it. Then list every reason it will not happen.

What I'd say

The design is straightforward and the obstacles are fatal.

1. Blinding is impossible. People know whether they're drinking. Expectation effects on a behavioural outcome study are unavoidable.

2. Adherence over years is hopeless. You'd be asking abstainers to start drinking and drinkers to stop, for a decade, and measuring compliance by self-report — the instrument Chapter 4 §4.7 already demolished.

3. The ethics are genuinely difficult. Randomizing people to consume a Group 1 carcinogen for years is a hard case to put to an ethics committee, and randomizing people with any history of problem drinking is unconscionable.

4. Funding. A large long-term alcohol trial is expensive, and the obvious funders are the alcohol industry — which creates a conflict severe enough to have derailed at least one high-profile attempt. A major NIH-supported trial of moderate drinking was halted amid concerns about industry involvement in its design and funding, and that episode is worth reading about (§12.13).

5. Duration. Cardiovascular endpoints need years to decades.

So we will probably never have the trial — which is exactly the situation Chapter 2 §2.10 said demands a different tool. And in this case one exists, which is §12.5, and it's the most satisfying thing in this chapter.

12.5 What Mendelian randomization added

This is the section that changed the field, and it is Chapter 2 §2.10's convergence criterion doing exactly what it was supposed to do.

The problem: we can't randomize alcohol. The solution: find something that already randomizes it.

Remember the ALDH2 variant from §12.1 — the one that clears acetaldehyde slowly and causes flushing. People carrying it drink substantially less, because drinking makes them feel unwell.

And critically:

• The variant is allocated at conception, essentially at random with respect to income, education, diet, exercise, and everything else that confounds observational studies.
• It doesn't change with lifestyle.
• It affects mortality risk primarily through alcohol intake, not through other pathways.

That is a natural randomized trial, running for a lifetime, in millions of people, immune to sick-quitter bias and healthy-user bias by construction.

What do these studies find?

Broadly: no cardioprotection. Genetically-predicted lower alcohol intake is generally associated with lower blood pressure, lower stroke risk, and lower cardiovascular risk — with the relationships looking monotonic rather than J-shaped. Less is better, including at the low end.

Similar analyses using other alcohol-related genetic variants in large biobank populations have found much the same: the observational J-curve does not survive genetic instrumentation.

💡 Aha moment. This is the single best worked example in the book of Chapter 2 §2.10's rule: when a new method with different weaknesses agrees with the old evidence, confidence goes up; when it disagrees, that's a serious signal.

Observational cohorts and Mendelian randomization have completely non-overlapping failure modes. Cohorts are vulnerable to sick-quitter bias, healthy-user bias, and reporting error. MR is vulnerable to none of those — it has its own assumptions, which can be violated, but not those.

They disagree. Sharply. About a finding a generation was taught as fact.

That's not a footnote. That's the field's error-correction mechanism working — and it is why Chapter 2 spent so long on why convergence matters. This is what it looks like when it pays off.

🔬 Claim → Evidence → Verdict

The claim: "Moderate drinking — one or two a day — protects your heart. The J-curve shows non-drinkers do worse."

Where it comes from: Thirty years of remarkably consistent observational data, plausible mechanisms (HDL, antiplatelet effects), and enthusiastic amplification by an industry with a great deal at stake.

What the evidence actually shows: The J-curve is substantially explained by sick-quitter bias (the abstainer reference group is enriched with people who stopped because they were ill), healthy- user bias (moderate drinkers are wealthier, better educated, better connected), and exposure misclassification. Analyses restricted to lifelong abstainers attenuate or eliminate the dip. Mendelian randomization — which is immune to all three — finds no cardioprotection and generally monotonic relationships. The HDL mechanism is also weaker than it looked, since Chapter 9 §9.10 established that raising HDL doesn't reliably reduce events.

📉 Evidence quality: Large rung 5 body pointing one way; genetic instrumentation with non-overlapping weaknesses pointing the other. Disagreement of exactly the kind that should move you.

Verdict: 🟠 Probably false. Not "definitely refuted" — MR has its own assumptions, and there remains genuine debate. But the confident cardioprotection claim no longer has evidence behind it proportionate to how often it's repeated.

🔬 Claim → Evidence → Verdict

The claim: "Red wine specifically is protective — the resveratrol and polyphenols."

Where it comes from: Resveratrol is real, does interesting things in cell and animal models (including some of the longevity signalling from Chapter 6 §6.9), and red wine does contain polyphenols. Add the "French paradox" — the observation that French cardiovascular rates seemed low relative to saturated fat intake — and you have a story.

What the evidence actually shows: The resveratrol content of wine is very low. Doses used in the animal studies would require drinking implausible quantities — estimates commonly run to dozens or hundreds of litres a day to match. Human trials of resveratrol supplementation have been underwhelming. And the French paradox has several more prosaic candidate explanations, including differences in how cause of death was recorded.

Meanwhile: red wine contains the same ethanol as everything else, and ethanol is what drives the cancer risk in §12.6.

📉 Evidence quality: Rung 1–2 mechanism at doses unreachable through drinking; null human supplementation trials.

Verdict: ❌ Not supported. If resveratrol were the active ingredient, you would take it as a supplement without the ethanol — and people have tried, and it didn't work. Drink red wine because you like red wine.

12.5b The trial that didn't happen

There is a postscript to §12.4's productive-struggle exercise, and it's worth knowing because it demonstrates Chapter 1 §1.5's incentive analysis operating at the highest level of the research system.

In the mid-2010s, US National Institutes of Health researchers began developing a large randomized trial of moderate alcohol consumption — the kind of trial that would, in principle, have settled the cardioprotection question directly. Participants would be randomized to a moderate daily drink or to abstention, and followed for cardiovascular outcomes.

The trial was halted before completing enrolment, following an internal review that raised serious concerns about the way it had been developed — specifically about interactions between NIH staff and the alcohol industry during its design and funding phase, with a substantial portion of the funding having been solicited from alcohol producers.

I want to be careful about what this episode shows and doesn't.

What it shows: that even at the level of a flagship government trial, funding pressure can shape what gets studied and how a question gets framed. It also shows the correction mechanism working — the concerns were investigated, published, and acted on, at institutional cost.

What it doesn't show: that the trial's designers were dishonest, or that the underlying question was illegitimate. It was a good question. A well-designed trial funded by the industry that profits from one answer is compromised regardless of anyone's integrity, because the design choices — comparator, dose, population, endpoints, duration — are where influence operates (Chapter 2 §2.8).

And the practical consequence for you: the definitive randomized trial on moderate drinking does not exist, is unlikely to be funded independently at the scale required, and would face the obstacles in §12.4 even if it were. Which is precisely why the Mendelian randomization evidence in §12.5 is doing so much work in this chapter — it is the best available substitute for a trial we will probably never have.

🔍 Why this works. It's worth understanding why statistical adjustment can't rescue the J-curve, because "they adjusted for confounders" is the standard reassurance and Chapter 2 §2.3 said it's insufficient. Here's the specific reason.

To adjust away sick-quitter bias, you'd need to identify and remove the former drinkers who quit for health reasons. You can't, for three compounding reasons.

One: the data usually isn't there. Many cohorts asked "do you drink?" at baseline and didn't ask "did you used to, and why did you stop?" You cannot adjust for a variable you never measured.

Two: even where it was asked, the reason is unreliable. People who quit after a health scare often don't attribute it to the health scare, and people who quit for other reasons often had unmeasured early illness anyway.

Three — and this is the killer — the illness may precede the diagnosis by years. Someone with early undiagnosed liver disease, early cancer, or developing frailty may reduce drinking because they feel unwell, long before anything appears on a medical record. No adjustment can capture a reason the person themselves doesn't know.

This is Chapter 2 §2.3's residual confounding in its most concrete form: you can only adjust as well as you measured, and here the thing you'd need to measure is invisible to everyone including the participant. Excluding former drinkers helps, and it doesn't fully solve it, and that is exactly the situation that makes a genetic instrument so valuable.

12.6 Cancer: the part that hasn't wobbled

While the cardiovascular story was falling apart, the cancer story was getting stronger.

Alcohol is classified by IARC as a Group 1 carcinogen — the category for agents with sufficient evidence of carcinogenicity in humans. So is acetaldehyde associated with alcohol consumption.

Chapter 2 §2.6 established what Group 1 means and doesn't: it is a statement about confidence in causation, not about magnitude. That caveat applies here as it did to processed meat. But the mechanism is unusually clear — acetaldehyde damages DNA and impairs repair — and the epidemiology is consistent.

Cancers with established associations:

The breast cancer finding is the one that matters most for this chapter's framing, because it is where "moderate" stops being reassuring. The dose-response for breast cancer appears to begin at low levels of intake — the increase is small in absolute terms at one drink a day, and it is not zero, and there is no identified threshold below which risk is unaffected.

This is why guidance has shifted (§12.9), and why statements like the WHO's 2023 position that no level of alcohol consumption is safe for health are technically defensible even though they read as absolutist.

What that statement means, precisely: there is no identified threshold below which risk is zero. What it does not mean: that any amount is dangerous in a way that should dominate your decisions. Those are different claims and the headlines merged them.

🔄 Check your understanding. Alcohol is Group 1, like tobacco and asbestos. A friend concludes that a glass of wine is comparably dangerous to a cigarette. What's wrong, and what's right?

Answer

What's right: the classification is accurate, and the causal evidence for alcohol and several cancers is genuinely strong. This isn't a technicality.

What's wrong: Group 1 describes confidence that something causes cancer, not how much cancer. Chapter 2 §2.6. Tobacco shifts lung cancer risk by more than an order of magnitude; alcohol at one drink a day shifts several cancer risks by a much smaller relative amount, on small absolute baselines.

The honest framing: a glass of wine a day carries a small, real, dose-dependent increase in several cancer risks — most notably breast cancer, where the increase begins at low intakes. That's a genuine consideration and it is not equivalent to smoking, and saying so is not minimizing it.

And the trap to avoid: if you conclude "wine equals cigarettes," you'll disbelieve the whole thing the moment you notice that people who drink a glass of wine a day are visibly not dropping dead. Overstatement is how accurate warnings get discarded — which is Chapter 1's entire argument about how confident claims destroy trust.

12.7 The harms that don't make headlines

Cancer and cardiovascular disease dominate the discussion. Several other effects are more likely to matter to how you actually feel this week.

Sleep. Alcohol reduces the time it takes to fall asleep, which is why it feels like it helps. It then suppresses REM sleep and fragments the second half of the night, producing lighter, more broken sleep and early waking. People who reduce alcohol frequently report better sleep within days — and per Chapter 3's Case Study 1, this was probably the largest single component of why Theo's juice cleanse made him feel wonderful.

Blood pressure. Alcohol raises it, dose-dependently, and reduction lowers it — one of the more reliable non-pharmacological blood pressure levers available.

Triglycerides. §12.1. Reliable and fast-moving.

Liver. Fatty liver develops with sustained intake and is substantially reversible on reduction. Progression to hepatitis and cirrhosis is dose- and duration-dependent, with large individual variation.

Brain. Heavy long-term use causes measurable atrophy and cognitive impairment. At moderate intakes the evidence is more contested, with some large imaging studies reporting associations with brain volume differences at surprisingly modest intakes — findings that carry the usual observational caveats.

Injury and acute harm. Frequently omitted from nutrition discussions and quantitatively important: falls, road traffic collisions, violence, drowning, and — in older adults — fractures from falls in someone already sarcopenic (Chapter 8).

Medication interactions. ⚠️ Alcohol interacts with a great many drugs. Paracetamol/acetaminophen (hepatotoxicity risk), benzodiazepines and opioids (respiratory depression), metformin (lactic acidosis risk), warfarin, many antidepressants, and others.

Dependence. Alcohol is addictive. Roughly a small but substantial minority of drinkers develop a use disorder, and the risk is not evenly distributed — family history, mental health, trauma, and social context all matter. This is a medical condition, not a moral failure, and §12.11 covers routing.

12.8 What about the social and psychological side?

A chapter that only counted harms would be dishonest, so let's be even-handed.

Alcohol does things people value, and pretending otherwise is how public health messaging loses its audience:

• Social lubrication and connection. Social connection is itself a robust predictor of health and longevity, and for many people alcohol is embedded in how it happens.
• Pleasure. Wine with food, a beer after work, a drink at a wedding. Enjoyment is a legitimate input to a decision.
• Cultural and ritual meaning, in many traditions, over millennia.
• Short-term stress relief, which is real, and which is also the mechanism by which use escalates.

The honest accounting is that these are genuine benefits, that they are not health benefits in the epidemiological sense, and that a person weighing them against a small increase in cancer risk is making a legitimate trade-off rather than committing an error.

What's changed is that the trade-off no longer has a freebie in it. For thirty years the calculation was "pleasure and connection, plus a cardiovascular bonus, versus some risk." The bonus is gone. The pleasure and the connection remain, and they were always the real reason.

🔄 Check your understanding. Two people each drink 14 US standard drinks a week. Person A has two most evenings. Person B has all fourteen on Saturday. Both appear in a study as "moderate drinkers." What does the evidence suggest about their relative risk, and what does the fact that they share a category tell you about the literature?

Answer

Their risk profiles differ substantially, and not in a single direction.

Person B's pattern — binge drinking — carries markedly elevated risk for acute harms (injury, road collisions, violence, arrhythmias, acute pancreatitis) and appears worse for cardiovascular outcomes than the same total spread out. Blood alcohol concentration peaks matter, not just the weekly integral.

Person A's daily pattern carries its own issues — nightly drinking is more strongly associated with dependence and with sustained liver exposure, and daily habitual use is harder to reduce than weekend use because the cue is fixed and frequent.

What it tells you about the literature is the important part. Studies that categorize by weekly average merge these two people, and that merging dilutes the measured harm — some of the risk attributed to "moderate drinking" belongs to binge patterns hiding inside the category, and some of the apparent safety belongs to people whose pattern really is low-risk.

This is exposure misclassification (§12.4's third problem) and it works in both directions, which is why it's rarely emphasized: it isn't a bias with a convenient story attached. It's just noise that makes everything harder to see — and it's part of why the Canadian guidance moved to counting drinks per week rather than per day, and why the UK guidance specifies spreading intake over three or more days rather than just capping the total.

12.9 Why the guidelines moved, and what they say

Guidance has shifted noticeably in the last decade and the direction is consistent.

Two things to notice.

First, they've converged on framing risk as a continuum rather than a threshold. The old model — a "safe limit" you stay under — has been replaced by "less is better, and here's roughly where the risk climbs."

Second, none of them says don't drink. Even the WHO statement is about safety thresholds, not prohibition. The guidance treats this as an informed adult decision, which is the right posture and the one this chapter takes.

Standard drink definitions differ, which causes endless confusion:

So "two drinks a day" means meaningfully different things in different countries. Count grams of ethanol if you want to compare anything to anything.

12.9b How big is the risk, actually?

Chapter 2 §2.6 requires that this book convert relative risks to absolute ones, and it applies here as much as anywhere — particularly because the recent headlines have been alarming and the underlying numbers are more moderate than the coverage.

Take breast cancer, since it's the outcome where the dose-response starts lowest and therefore where "moderate" drinking is least reassuring.

Lifetime breast cancer risk in a typical Western population is roughly 12–13% — about 12 or 13 in 100 women. The commonly cited estimate is around a 7–10% relative increase per 10 g of ethanol per day.

Work it through for one drink a day (about 14 g in the US):

Baseline lifetime risk          ≈ 12.5%
× ~1.10 (roughly, for ~14 g/day) ≈ 13.8%
Absolute difference             ≈ 1.3 percentage points
                                ≈ 1 extra case per ~75 women, lifetime

Roughly one additional case per seventy-five women drinking one a day for life.

Three honest observations.

It is real and it is small per person. That's a genuine increase and it is not the sort of number that should terrify anyone out of a glass of wine at a wedding. It is also the sort of number that, applied to a country, means a great many people.

It scales with dose. Two drinks a day roughly doubles the increment; four drinks a day is a different conversation entirely. The relationship is approximately linear across the moderate range, which means halving your intake captures roughly half the available benefit — the smoothness in §12.10's Step 3.

And it stacks with everything else in §12.6 and §12.7. Breast cancer is one outcome among several, alongside blood pressure, triglycerides, liver, sleep and injury. No single number in this chapter is frightening. The sum is what makes the guidelines move, and the sum is also why "less is better" replaced "under this limit is safe."

💡 Aha moment. Notice that this is structurally identical to Chapter 11's fiber arithmetic — approximately 1 case per 75–100 people, across several stacked endpoints, with a smooth dose-response and no threshold. One is an argument for doing something and one is an argument for doing less of something, and the evidence has the same shape.

Which is a useful calibration: if you found the fiber number persuasive and the alcohol number dismissible, or vice versa, that difference is coming from you rather than from the data.

12.10 So what should you actually do?

Here's the framework. It's a decision aid, not a prescription.

Step 1 — Find out what you actually drink. Not what you'd estimate. Count a normal week, honestly, including the large pours and the ones with dinner. Then convert to grams of ethanol and to calories (§12.2).

Most people find the number is higher than they'd have said. Theo's was 189 g of ethanol a week — about thirteen and a half US standard drinks — and he'd have told you "a couple in the evening."

Step 2 — Know what you're trading. The honest ledger:

Step 3 — Notice that the dose-response is smooth. There is no cliff. Going from 14 drinks a week to 7 captures a large share of the available benefit — this is not an all-or-nothing decision, and framing it as one is why so many people do nothing.

Step 4 — Look at your own situation specifically. The calculus genuinely differs by person (§12.11).

Step 5 — Decide, and don't moralize about it. An adult who reads the above and chooses to keep drinking five nights a week has made a decision, not a mistake. An adult who cuts to twice a week has made a different one. Both are legitimate. What isn't legitimate is not knowing — which is where most people were, because they'd been told about the red wine.

🍽️ On your plate — what Theo did. He went from fourteen drinks a week to four, on weekends only. Not zero. He was explicit that he wasn't giving it up and I was explicit that I wasn't asking him to.

What it delivered: ~135 kcal/day removed (about 28% of his entire surplus, from one decision); triglycerides moved faster than any other change he made; and — the one he actually noticed — sleep. He reported feeling different within about ten days, which is the fastest feedback of anything in his whole plan.

At eighteen months it was one of the three changes out of seven that had survived. It survived because it was a reduction rather than an elimination, because weekends kept the social function intact, and because he'd chosen the number himself.

12.11 When the calculus is different

Not everyone is making the same decision.

⚠️ When to see a professional. If any of these are true, this is a conversation with a physician rather than a self-directed calculation: you have tried to cut down and couldn't · you drink in the morning or to prevent shakes · you have blackouts · people close to you have raised it · you drink alone to cope with mood · you have liver disease, pancreatitis, or an elevated liver enzyme result · you are pregnant or trying to conceive · you take any of the medications above.

And a specific safety point: if you drink heavily and daily, do not stop abruptly without medical advice. Alcohol withdrawal can be dangerous and in some cases life-threatening, and it is genuinely managed medically. This is one of very few places in this book where "just stop" is the wrong advice.

Alcohol use disorder is a medical condition with effective treatments — behavioural, pharmacological, and social. It is not a failure of character, and the single most useful thing this chapter can do for someone in that position is say so and point at help.

12.12 If you decide to drink less

Practical, since "decide" is easy and "do" isn't.

What works, roughly in order of evidence:

1. Set a specific number, not a direction. "Drink less" fails; "four drinks, Friday and Saturday only" survives — Chapter 10's specificity principle.
2. Change the environment, not the resolve. Don't keep it in the house; buy singles rather than multipacks; move the glasses. Chapter 33's argument, arriving early.
3. Break the cue. Most habitual drinking is cue-driven — arriving home, cooking, sitting down at nine. Identify yours and put something else in that slot. The cue is the intervention point, not the drink.
4. Use alcohol-free substitutes deliberately. The category has improved enormously and it preserves the ritual, which per Chapter 11's Case Study 1 is often what's actually being defended.
5. Track it for two weeks. The count itself reduces intake for most people, the same way a food diary does (Chapter 4).
6. Pick reduction over elimination unless you have a reason not to. Reduction targets survive; elimination targets have a binary failure mode — Chapter 10 §10.9's reset-on-failure rule.

🧾 Cost check. Theo's fourteen drinks a week were roughly $28–$45/week — $1,450–$2,340 a year. Cutting to four saves somewhere around $1,000–$1,600 annually, which is comparable to Walt's entire supplement bill and rather more than the cost of every dietary change in this book combined.

🪞 Learning Check-In

Fourth of these, and Part II is now complete. Three minutes.

• This chapter probably contradicted something you were confident about. The red wine finding is the most widely-held nutrition belief that this book overturns. How did it feel? Notice whether your first response was to look for a reason the evidence might be wrong — and notice that the same response is what makes the fat wars (Chapter 9) unresolvable.
• Did you notice the chapter refusing to tell you what to do? That was deliberate, and it's a different posture from the fiber chapter, which pushed hard. Ask yourself whether the difference is justified — is it about the evidence, or about the topic being socially loaded? (I think it's partly both, and I'm not entirely comfortable with that.)
• Part II is done. Six chapters on macronutrients, and the honest summary is that almost none of what matters was about macronutrient ratios. Look back at what actually earned strong verdicts: fiber, protein adequacy in older adults, replacing saturated with unsaturated fat, intact over refined, alcohol. Not one of them is a ratio.

One thing to carry into Part III: the micronutrient chapters are where "more is better" reasoning does the most damage. You've now seen it fail four times — beta-carotene, vitamin E, the fat-burning zone, and the J-curve. Watch for the fifth.

Spaced Review

Answer before reading on.

1. (Chapter 2) Sick-quitter bias is a specific case of a general problem. Which one, and what's the general form?

Confounding by indication / reverse causation in the reference group. The general form: when you compare a group to a "control" group, ask who is in the control group and why. People who don't do a thing may not do it because they're already unwell — which makes the exposed group look artificially healthy. Same shape as Chapter 2's diet soda example.

2. (Chapter 9) The J-curve's proposed mechanism was that alcohol raises HDL. Why is that mechanism weaker than it looked?

Because Chapter 9 §9.10 established that HDL is a marker, not a lever — drugs that raise HDL have repeatedly failed to reduce cardiovascular events, and Mendelian randomization doesn't support a simple causal role. A mechanism that runs through HDL was never as strong as it appeared, and that was knowable before the MR alcohol studies arrived.

3. (Chapter 4) Theo's beer is 190 kcal/day. Why does that matter more than the same calories from food?

Because it arrives with no satiety return — it doesn't reduce subsequent eating, and there's evidence it increases it. Per Chapter 4, energy that doesn't displace other energy is the most efficient possible route to a surplus. Plus §12.1's fat-oxidation suppression, and the fact that it never appears in anyone's mental account of what they ate.

Project Checkpoint: Your Alcohol Audit

Component twelve, and the last of Phase 2 — Analysis. Fifteen minutes, and the instruction that makes it work is: count a normal week, not a good one.

Step 1 — the honest count. Go through the last seven days. Every drink. Include the ones with dinner, the large pours, and the one you had while cooking.

Step 2 — convert to grams of ethanol. This is the step that makes it comparable:

grams ethanol = volume (ml) × ABV% × 0.789 ÷ 100

Example: a 175 ml glass of 13% wine = 175 × 13 × 0.789 ÷ 100 = 18 g — which is more than one US standard drink and more than two UK units, from something most people count as "a glass."

My weekly total: ______ g ethanol · ______ US standard drinks · ______ UK units

Step 3 — the calories. Multiply grams of ethanol by 7. Divide by 7 for a daily average. Compare it to your total intake from Chapter 4.

Alcohol calories: ______ kcal/week = ______ kcal/day = ______% of my intake

Theo: 189 g/week → 1,323 kcal/week → 189 kcal/day → 6% of intake.

Step 4 — the cost. Weekly spend × 52.

Step 5 — your own risk/benefit call. Write, in your own words:

• What do I actually get from this? (Be specific — pleasure, ritual, connection, stress relief, habit. Some of these are worth more than others.)
• What am I trading? (§12.10's ledger, applied to me.)
• Does anything in §12.11 apply to me?
• What number would I choose, if I were choosing rather than defaulting?

Step 6 — decide, and write it down. A number and an occasion, not a direction. Or explicitly decide to change nothing — which is a legitimate output of this exercise, and writing it down deliberately is different from drifting.

Non-tracking alternative. If counting is uncomfortable, answer three questions instead: Do I drink more than I'd like to? What would I lose by drinking less? What's the smallest change I'd actually make? The specific number matters less than the deliberateness.

⚠️ And if this exercise is difficult in a way that concerns you — if you find yourself undercounting deliberately, or if the honest number is frightening — please see §12.11 and talk to a physician. That reaction is information, and it is common, and there is effective help.

Next checkpoint (Chapter 13): your vitamin gap analysis against the DRIs — from food first. Phase 3 begins.**

Chapter Summary

Ethanol: 7 kcal/g, no storage form, cleared as a priority via ADH → acetaldehyde → ALDH2 → acetate. Acetaldehyde is the carcinogen and the hangover. The ALDH2 variant (common in East Asian populations) clears it slowly — causing flushing, and providing this chapter's key evidence.

The calories count — through suppressed fat oxidation rather than direct conversion, plus zero satiety return. Theo's two beers: 27 g ethanol, ~190 kcal/day, ~40% of his entire surplus.

The J-curve — thirty years of consistent data showing a protective dip at 1–2 drinks/day — is probably an artefact of three things:

1. Sick-quitter bias — the abstainer reference group is enriched with people who stopped because they were ill
2. Healthy-user bias — moderate drinkers are wealthier, better educated, better connected
3. Exposure misclassification — underreporting, and averaging that merges very different patterns

Mendelian randomization settles it as far as anything can. Using ALDH2 and related variants — allocated at conception, immune to all three biases — studies generally find no cardioprotection and monotonic relationships. Cohorts and MR have non-overlapping weaknesses and they disagree, which is exactly the signal Chapter 2 §2.10 said to take seriously.

Cancer got stronger while cardiovascular got weaker. IARC Group 1. Oral cavity, pharynx, larynx, oesophagus, liver, colorectum, female breast — where risk increases from low intakes with no identified threshold. That's what "no safe level" means: no threshold below which risk is zero. It does not mean any amount is dangerous enough to dominate your decisions.

Underrated harms: sleep (fastest noticeable improvement on reduction) · blood pressure · triglycerides · liver · injury · medication interactions · dependence.

Guidelines have converged on a continuum of risk, not a safe threshold — and none of them says don't drink. Count grams of ethanol, because a "standard drink" is 14 g in the US, 8 in the UK, 10 in Australia.

This chapter's verdicts:

In absolute terms (§12.9b): one drink a day is roughly 1 extra breast cancer case per 75 women, lifetime — real, small per person, dose-dependent, and stacking with everything in §12.6–§12.7. No single number in this chapter is frightening; the sum is what moved the guidelines. And notice it's structurally identical to Chapter 11's fiber arithmetic — if you found one persuasive and the other dismissible, that's coming from you rather than from the data.

The one thing to remember: the trade-off no longer has a freebie in it. For thirty years the calculation was pleasure and connection plus a cardiovascular bonus versus some risk. The bonus is gone. The pleasure and the connection remain, and they were always the real reason.

What's Next

Part II is complete. Six chapters, three macronutrients plus fiber, the diet wars, and alcohol.

And here is the honest summary of all of it: almost nothing that earned a strong verdict was about a macronutrient ratio. Fiber. Protein adequacy, especially in older adults. Replacing saturated with unsaturated fat. Intact over refined. Alcohol. Not one of those is a ratio, and the arguments that consume the entire public conversation turned out to be about the wrong axis.

Part III drops from grams to milligrams — vitamins, minerals, water, and the $50 billion industry built on top of them.

Chapter 13 carries the threshold concept that governs the whole part: nutrient status is not nutrient intake, and supplementing someone who isn't deficient almost never helps.

You have now watched "more is better" fail four times — beta-carotene, vitamin E, the fat-burning zone, and the J-curve. Part III is where it fails most expensively.