Measures of Association in Epidemiology

In epidemiology, moving from observing a potential link between a factor and a disease to quantifying that relationship is a critical leap. Measures of association are the mathematical tools that allow researchers to do just that, moving beyond speculation to provide numerical evidence of the strength, direction, and public health impact of a relationship. Mastering these measures—relative risk, odds ratio, and attributable risk—is fundamental for interpreting study results, assessing causal evidence, and informing effective health policy and clinical decisions.

From Counts to Comparison: Understanding Risk

Before calculating association, we must define risk itself. In epidemiological terms, risk is the probability that an event (like developing a disease) will occur within a specified period. It’s calculated from cohort study data as the number of new cases (incident cases) in a group divided by the total number of individuals in that group at the start of the study period.

For example, imagine a 10-year cohort study investigating smoking and lung cancer. If 200 out of 1,000 smokers develop lung cancer, the risk among the exposed (smokers) is $200/1000=0.20$ or 20%. If 10 out of 2,000 non-smokers develop lung cancer, the risk among the unexposed is $10/2000=0.005$ or 0.5%. These absolute risks are informative, but the measure of association tells us how these risks compare.

Relative Risk: The Ratio of Risks

Relative risk (RR), also called the risk ratio, is the primary measure of association used in cohort studies and randomized controlled trials. It directly compares the risk of disease in the exposed group to the risk in the unexposed group. The formula is:

$$RR=\frac{Ris{k}_{exposed}}{Ris{k}_{unexposed}}$$

Using our smoking example: $$RR=\frac{0.20}{0.005}=40$$

An RR of 40 means that smokers in this study were 40 times more likely to develop lung cancer over the 10-year period compared to non-smokers. The interpretation hinges on the value:

• RR = 1: No association. Risk is identical in both groups.
• RR > 1: Positive association. The exposure is associated with an increased risk of the outcome (e.g., RR=40 for smoking and lung cancer).
• RR < 1: Negative (protective) association. The exposure is associated with a decreased risk of the outcome (e.g., RR=0.5 for vaccination and disease).

Odds Ratio: The Approximator in Case-Control Studies

Cohort studies measure risk directly, but case-control studies start with people who already have the disease (cases) and without it (controls), making direct risk calculation impossible. Here, we use the odds ratio (OR). The odds of exposure is the ratio of the probability of being exposed to the probability of not being exposed.

In a case-control study, we calculate the odds of exposure among cases and compare it to the odds of exposure among controls. The cross-product formula from a 2x2 table is simplest:

$$OR=\frac{a/c}{b/d}=\frac{a\times d}{b\times c}$$

Imagine a case-control study on smoking (exposure) and pancreatic cancer (outcome). If we have 90 exposed cases, 10 unexposed cases, 30 exposed controls, and 70 unexposed controls: $$OR=\frac{90\times 70}{30\times 10}=\frac{6300}{300}=21$$

We interpret this as: The odds of being a smoker were 21 times higher among pancreatic cancer cases than among controls. When the disease is rare (typically <10% in the population), the OR provides a very good approximation of the relative risk. This is a key reason why the odds ratio is so valuable in epidemiology; it allows us to estimate the strength of an association from the efficient case-control study design.

Attributable Risk: The Public Health Impact

While RR and OR measure strength of association, they do not tell us about the disease burden caused by the exposure in a population. Attributable risk (AR), also called risk difference, addresses this. It is the absolute difference in risk between the exposed and unexposed groups.

$$AR=Ris{k}_{exposed}-Ris{k}_{unexposed}$$

From our first cohort example: $AR=0.20-0.005=0.195$ or 195 per 1000. This means 195 of the 200 cases per 1000 smokers are attributable to smoking. A more policy-relevant measure is the population attributable risk (PAR), which estimates the excess incidence of disease in the total population attributable to the exposure. It depends on the AR and the prevalence of the exposure ($P_{exposed}$) in the population:

$$PAR=Ris{k}_{total}-Ris{k}_{unexposed}$$ or equivalently, $$PAR=AR\times P_{exposed}$$

If smoking prevalence is 20% in our population, the PAR would be $0.195\times 0.20=0.039$ or 39 per 1000. This tells public health officials that if smoking were eliminated, 39 cases of lung cancer per 1000 people in the population could be prevented, providing a powerful metric for prioritizing interventions.

Common Pitfalls

1. Interpreting Odds Ratio as Relative Risk Without Consideration: The most frequent error is treating an OR from a case-control study as an exact RR when the disease is not rare. This can overestimate the association. Always note the study design and disease frequency when interpreting an OR.
2. Confusing Strength with Public Health Importance: A strong relative risk (e.g., RR=15 for a rare genetic mutation and disease) may have a small PAR if the exposure is very rare. Conversely, a modest RR (e.g., RR=1.3 for sedentary behavior and heart disease) can have a huge PAR because the exposure is so common. Effective policy requires looking at both RR and PAR.
3. Ignoring the Time Element in Risk: Risk is a probability over a defined time period. An RR of 2 from a 1-year study is not equivalent to an RR of 2 from a 20-year study. Always consider the follow-up time when comparing risks across studies.
4. Overlooking Confounding: No measure of association is valid if it is distorted by a confounding variable—a factor associated with both the exposure and the outcome. Statistical techniques like stratification and multivariable modeling are required to calculate adjusted measures of association that control for confounders.

Summary

• Measures of association, including Relative Risk (RR), Odds Ratio (OR), and Attributable Risk (AR), are essential for quantifying the relationship between exposures and health outcomes in epidemiological research.
• Relative Risk (RR) is the direct ratio of risks from cohort studies, where an RR > 1 indicates increased risk, RR < 1 indicates decreased risk, and RR = 1 indicates no association.
• Odds Ratio (OR) is the primary measure from case-control studies and approximates the RR when the outcome is rare, providing a crucial tool for studying diseases with long latency or low incidence.
• Attributable Risk (AR) and Population Attributable Risk (PAR) shift the focus from the strength of association to the public health burden, estimating the amount of disease incidence that can be attributed to—and potentially prevented by eliminating—a specific exposure.