Skip to content
Tutorials Center
My learning

Topics

Lesson 35 of 47 ยท Diet, Disease and Evidence

Chapter 7 review: diet, disease and evidence

Review study designs, confounding, absolute versus relative risk, diet and heart disease, DASH, PREDIMED, the DPP and how to judge claims, with a ten-question quiz.

This chapter taught a skill as much as a set of facts: how to move from a question about food and health to a defensible conclusion. The summary below places the chapter's landmark studies, biases and rules of thumb in one view, followed by a quiz that mixes all three lessons.

Four steps: ask a clear question, understand the study design and its biases, weigh effect size, consistency and plausibility, then advise at the population level with humility
Caption: chapter 7 in four steps, from a question to a weighed conclusion.

What you will learn

  • The landmark studies and the lesson each taught.
  • The biases and statistical pitfalls to check for first.
  • A short list of the best-supported diet-disease statements.

Recap table

Topic Key ideas
Study designs Ecological, case-control, cohort, RCT, meta-analysis; each has typical biases
Confounding A hidden factor linked to exposure and outcome; HRT looked protective in cohorts but not in the WHI trial
Absolute vs relative risk Relative risk needs the baseline; NNH = 1 / absolute increase
Heart and blood pressure Trans fat harmful; replace saturated with unsaturated fat; sodium raises BP; DASH lowers BP; fiber is protective
Patterns PREDIMED: Mediterranean diet with olive oil or nuts, about 30% fewer major events; DIETFITS: low-fat and low-carb similar
Type 2 diabetes DPP: lifestyle cut incidence by about 58% vs placebo, metformin by about 31%
Claims and fads Checklist; multiple comparisons; ATBC, CARET and SELECT showed harm from high-dose supplements

Key terms to know: cohort, RCT, systematic review, confounding, reverse causation, residual confounding, relative risk, absolute risk, NNH, DASH, PREDIMED, glycemic index, multiple comparisons, p-hacking, placebo effect.

Best-supported statements (as of this writing)

  1. Industrial trans fats are harmful and have been eliminated by regulation in many countries.
  2. Lowering sodium and eating more fruit, vegetables and potassium lowers blood pressure.
  3. Higher fiber and whole grain intake is associated with lower cardiovascular disease, type 2 diabetes and colorectal cancer.
  4. Replacing saturated fat with unsaturated fat from plants reduces cardiovascular events.
  5. Healthy dietary patterns (Mediterranean, DASH, healthy plant-based) are associated with better outcomes.
  6. In prediabetes, weight loss and physical activity can prevent most progression to diabetes.
  7. High-dose isolated antioxidant supplements do not prevent disease and sometimes cause harm.

Common mistakes to avoid

  • Taking one headline as settled science.
  • Confusing "statistically significant" with "important".
  • Treating relative risk as absolute risk.
  • Judging a diet by its anecdotes rather than its trials.

A worked example

A news story says that a drink raises the risk of a disease by 25 percent. Walk through the chapter's questions. First, what is the baseline risk? Suppose it is 4 in 1,000 over ten years; then a 25 percent increase means 5 in 1,000, one extra case per 1,000 people, and the number needed to harm is 1,000. Second, what was the design? If a cohort study, ask about confounding: people who drink a lot of it may smoke more or exercise less. Third, was the exposure measured well, and was the result replicated? Fourth, does a plausible mechanism exist, and do trials of the exposure agree? Only after these steps does the headline become a calibrated statement: perhaps "a small association, uncertain whether causal".

Try it

Use the calculator to turn a headline into numbers: give it a baseline risk and a relative risk and see what the claim means for 1,000 people.

JS Headline to numbers

Press Run (or Ctrl + Enter) to see the console output here.

No output.

Check your understanding

Quick check ยท 10 questions

of 10 answered You scored / 10 ยท perfect!

  1. 1. Which design best protects against confounding by chance assignment?

    Randomization balances both measured and unmeasured factors on average.
  2. 2. In the confounding simulation, why did supplements look protective?

    Lifestyle drove both supplement use and low disease risk.
  3. 3. A disease has a baseline risk of 4 in 100 and a relative risk of 1.25. The exposed risk is...

    4 x 1.25 = 5 in 100.
  4. 4. What is the DASH diet designed to lower?

    DASH means Dietary Approaches to Stop Hypertension.
  5. 5. Which result came from the Diabetes Prevention Program?

    Rates were 4.8, 7.8 and 11.0 cases per 100 person-years.
  6. 6. What did ATBC and CARET show for beta-carotene in smokers?

    Both trials suggested harm.
  7. 7. Which problem arises when 20 foods are tested against one outcome?

    At the 5 percent level, about one in twenty null tests looks significant.
  8. 8. What is residual confounding?

    Statistical adjustment cannot remove what was not measured well.
  9. 9. Which advice is best supported for reducing cardiovascular risk?

    These measures have consistent evidence from trials and cohorts.
  10. 10. Why is PREDIMED's retraction and republication instructive?

    Science corrects itself; the central finding held.

Summary

  • Judge the design, the biases, the absolute effect and the replication.
  • The best-supported advice is consistent across many kinds of evidence.
  • Healthy skepticism is a tool, not cynicism.

Next: Chapter 8 overview: food systems and policy