Diet quality, alcohol thresholds, and chronic stress

The four factors covered in the other guides — smoking, BMI, exercise, sleep — carry the model's largest weights and rest on its firmest evidence. The three here are different: smaller effects, weaker or more contested data, and harder measurement. We've grouped them partly for that reason. Where the evidence is thinner, showing the reasoning matters more.

Diet quality: what the score measures

Nutrition research is the most contested area in this calculator, and for good reason. Randomised dietary trials are expensive, hard to blind, and nearly impossible to run for the decades that healthspan outcomes require. Most evidence is observational, and dietary self-report is notoriously unreliable — people misremember, underestimate portions, and systematically under-report things they feel judged about.

Given that, the model deliberately avoids scoring macronutrients. It does not care about your carbohydrate ratio, your fat intake, or your protein timing, because the evidence for any specific macronutrient split producing better long-run health outcomes is weak and heavily contested.

What it scores instead is dietary pattern, on something close to a Mediterranean-style index. This approach has the most consistent evidence behind it, including from the PREDIMED trial — one of the few large randomised dietary intervention trials with hard cardiovascular endpoints. The components such indices typically measure:

• High vegetable and fruit intake
• Legumes and whole grains rather than refined
• Fish and seafood in preference to red and processed meat
• Olive oil as the primary added fat
• Nuts and seeds regularly
• Low intake of ultra-processed food, added sugar, and processed meat

What's notable is how much these indices don't specify. There's no calorie target, no macro ratio, no meal timing, and no prohibition on any single food. The pattern is defined by what predominates rather than by what's forbidden — which is also, plausibly, why it's more sustainable than most prescriptive approaches.

Ultra-processed food

The most active current question in nutritional epidemiology is whether industrial processing itself carries risk independent of nutrient content. Observational studies consistently associate high ultra-processed food intake with adverse cardiometabolic outcomes, and a small but influential controlled feeding trial found that people ate several hundred more calories per day on an ultra-processed diet than a matched unprocessed one, despite the diets being matched for presented macronutrients.

Whether the mechanism is caloric — hyper-palatable food is easier to overeat — or something more specific involving additives, emulsifiers, or food matrix disruption, remains genuinely open. What is reasonably well supported is the association itself, which is why "low in ultra-processed food" appears in our description of a high-quality diet even though the causal pathway isn't settled.

How the model scores diet

The calculator asks you to self-rate diet quality on a five-point scale, from Poor to Excellent, and applies modifiers from −3 to +2 years.

We should be straightforward about the weakness here: this is a self-assessment, and self-assessments of diet are optimistic. Most people rate their diet as better than a food diary would suggest. A more rigorous model would ask a dozen frequency questions about specific food categories and compute an index. We chose the single-question version for length and completion rate, and the cost is real precision.

If you want a more honest input, rate yourself against the specific list above rather than against a general impression — how many of those six components genuinely describe most of your weeks?

Alcohol: where the curve bends

The most cited threshold comes from a 2018 Lancet analysis by Wood and colleagues, pooling individual-participant data from nearly 600,000 current drinkers across dozens of prospective studies. It found that all-cause mortality risk was lowest at intakes below roughly 100g of alcohol per week — about seven standard drinks — and rose progressively above that.

The analysis also found something useful about which risks rise. Higher intake was associated with elevated stroke, heart failure, fatal hypertensive disease, and fatal aortic aneurysm. Interestingly, myocardial infarction showed a different pattern — which is part of why the older cardio-protective story took so long to unravel.

Why the old J-curve collapsed

For decades, the received wisdom was that moderate drinking was better than abstaining — the "J-curve," with light drinkers at lower risk than both abstainers and heavy drinkers. That finding is now widely regarded as substantially an artefact, and understanding why is a useful lesson in reading health research generally.

Three problems drove it:

• Sick-quitter bias. The "abstainer" reference group in older studies included former drinkers who had stopped because of illness — sometimes illness caused by drinking. Comparing moderate drinkers against a group enriched with sick ex-drinkers makes moderate drinking look protective.
• Confounding by socioeconomic status. In many Western populations, moderate drinking correlates with higher income, education, and healthcare access. Those factors independently predict better outcomes.
• Mendelian randomisation. Studies using genetic variants that affect alcohol metabolism as a natural experiment — an approach far less vulnerable to these confounders — have generally failed to find a protective effect at low intake, and tend to show risk rising monotonically.

The current consensus position is that there's no established cardiovascular benefit from alcohol that outweighs its risks, and that risk increases from low levels of intake upward. Several national guidelines have been revised downward accordingly.

How the model scores alcohol

Standard drinks / week	Modifier
0 – 1	0
2 – 7	−1 yr
8 – 14	−2 yrs
15 – 21	−3 yrs
Over 21	−5 yrs

Note that abstaining is the reference state scoring zero — there is no bonus for moderate drinking anywhere in the model, which follows directly from the collapse of the J-curve.

One measurement note that materially affects your input: a standard drink is a specific quantity — roughly 14g of pure alcohol in US terms, about 10g in UK units. A restaurant wine pour is commonly 1.5 to 2 standard drinks, and a strong craft beer can be two or more. Counting glasses rather than standard drinks is the single most common way people under-report here, often by a factor approaching two.

Chronic stress: real but hard to measure

Stress carries the smallest weight in the model — a range of just −2 to +1 — and that reflects genuine uncertainty rather than a judgment that stress doesn't matter.

The mechanistic case is reasonably strong. Sustained activation of the hypothalamic-pituitary-adrenal axis elevates cortisol, which in chronic excess is associated with visceral fat deposition, insulin resistance, hypertension, and impaired immune function. Chronic psychological stress has been linked to shortened leukocyte telomere length in several studies, and to elevated inflammatory markers. There's also a well-documented indirect pathway: chronic stress reliably degrades the other factors in this model, worsening sleep, diet, exercise adherence, and alcohol use.

The problem is measurement. Unlike cigarettes or hours of sleep, "stress" is a subjective construct measured by questionnaires that vary between studies and aren't directly comparable. It's also strongly confounded with the things that cause it — poverty, insecure work, caregiving burden, discrimination, illness — each of which independently affects health. Isolating an effect of stress itself from an effect of stressful circumstances is genuinely difficult, and the literature reflects that.

Some evidence also suggests that the appraisal of stress matters, not just exposure — that people who interpret stress as harmful show worse outcomes than similarly stressed people who don't. That's an interesting finding but a fragile one, and it shouldn't be over-read into advice that people should simply think differently about difficult circumstances.

How the model scores stress

Self-rated on five points, from Very low to Very high, with modifiers from +1 (very low) to −2 (very high). The small range is intentional: we're reasonably confident the effect is real and directionally negative, and not at all confident about its magnitude. Where we're unsure, we prefer a small weight to a bold one.

The practical framing we'd offer is that stress is worth attending to at least as much for its downstream effects as for its direct ones. If chronic stress is degrading your sleep and your exercise consistency, its true cost in this model is larger than the −2 the stress factor itself applies — it's showing up in the other rows too.