Science

Why Epidemiology Rules

Last updated: April 9, 2026

Crissman LoomisMay 9, 2024

Previously, I discussed how Hazard Ratios (HR), often for the risk of premature death, are the best metric for measuring the effectiveness of interventions. Now, let’s look at epidemiological studies (ES), one of the best sources for HR. ES use observational data to find correlations between actions and health outcomes.

Benefits of Epidemiological Studies

Big Data

Premature death, or all-cause mortality, is the primary outcome of interest for HR. Fortunately, even older adults don’t die often. For example, a Norwegian HIIT study, which looked at 70-77-year-olds, despite having an effect size of nearly a 50% hazard ratio and over 8,000 person-years of data, did not reach significance for the HR of premature death.¹ Reaching significance, even for a strong effect with older participants, generally requires over ten thousand person-years of data.

Ten thousand years of data is challenging for a randomized, controlled trial, but many extensive surveys can be searched for correlations for ES. The USA has the National Health and Nutrition Examination Survey with over 100,000 person-years of data, while the UK Biobank has nearly ten million person-years. These resources are sufficient to detect significant results, even for minor effects.

Ethical Approach to Risky Interventions

Perhaps the most famous use of ES is the association between smoking and lung cancer. Given strong indications that tobacco causes lung cancer, conducting a Randomized Controlled Trial (RCT) where people smoke until death would be unethical. Instead, scientists used ES to show the correlation between smoking and lung cancer.

Ability to Examine Long-Term Effects

Because ES requires large data to find statistically significant correlations, they typically span years, not just weeks or months, like RCTs. This is especially important when dealing with the impacts of diet and exercise. It may take months for the body to adjust to changes in diet or activity. For instance, the immediate response to skipping breakfast is weight loss, as the body gets fewer calories than it is used to. RCTs on skipping breakfast confirm this, with an average loss of half a kilogram (a pound) over nine weeks.² In the long term, though, the body adapts by reducing metabolism or increasing hunger, and ES show that breakfast skippers have a nearly a 50% higher risk of overweight or obesity.³

Similarly, according to the book Burn⁴, metabolic adaptations to changes in weekly exercise volume take three to five months. Most RCT durations last for weeks and miss these metabolic effects.

Epidemiological Challenges

Reliability of Self-reported Data

A frequent challenge to the findings of epidemiological studies based on surveys is that people may not answer truthfully. It is true that if you estimate how many calories someone eats in a day by asking them what foods they ate and how much of each, you’ll end up with much less calories than expected. People don’t eat meals with scales, and estimating the weight or amount of servings of food is not a common task. That said, asking someone, “How many servings of fish do you eat weekly?” yields reasonable numbers. Scientists have verified this by comparing food diaries—record the food you ate immediately after eating—with food frequency-type questionnaires (FFQ), and they find some differences. The FFQ responses are close to the food diary responses.⁵

Confounding Confounders

Some people question whether epidemiological studies can be trusted because perhaps food, exercise, or habit doesn’t affect longevity, but rather, people who do those things are “healthy users” (or “unhealthy users”) who distort the data with other habits that do affect longevity. An oft-cited example of health user bias is flu vaccine studies, which can show reduced premature death before flu season starts. Having reviewed those studies, the issue there is not “healthy user” but “financial funding bias”—they’re funded by flu vaccine makers. Independent Cochrane studies don’t show this issue.⁶ Disregard studies funded by vendors, folks!

Unintended correlations with the study objective are called confounders, and epidemiological studies control for the effects of confounders by adjusting for factors that impact health, such as age, gender, socio-economic status, BMI, or physical activity. These factors are added as a separate variable as the study objective, and their influence is removed from the analysis. Once these factors are controlled, the test results will unlikely change from something correlated with the study control and outcome but not with the confounders.

It’s unlikely, but not impossible. It’s educational to look at the one factor I have seen affect the results of epidemiological studies, even when the researchers tried to control its impact as a confounder: smoking. Smoking is correlated with coffee drinking and also reduces the effect of caffeine. Although coffee studies controlled for a yes/no smoking confounder, the increased smoking correlation with increased coffee consumption made it appear that drinking too much coffee was unhealthy. It’s not, but smoking too much certainly is!

Comparison with Randomized Controlled Trials

A key question in health research is how well epidemiological studies compare to the gold standard of randomized controlled trials (RCTs). A meta-epidemiological study addressed this question by comparing healthcare outcomes assessed in observational studies with those in randomized trials.⁷ The results were encouraging for epidemiological research:

• On average, the estimates from observational studies were only about 6% different from those of RCTs. For instance, an observational study might find a correlation of 15%, but the RCT might find a 14% or 16% causality.
• This small difference suggests that well-designed observational studies can provide results that are quite close to those of randomized trials.
• The study found no systematic tendency for observational studies to overestimate or underestimate treatment effects compared with randomized trials.

This finding reinforces the value of epidemiological studies, especially in situations where RCTs are impractical, unethical, or too short-term to capture important health outcomes.

The Bottom Line