The Risky Business of Public Health Research

Steven Milloy

Copyright © 1995 by Steven J. Milloy. All rights reserved. First edition. Published by the Cato Institute, 1000 Massachusetts Avenue, N.W., Washington, D.C. 20002. Library of Congress Catalog Number: 95-72177. International Standard Book Number: 0-9647463-2-8.

Chapter 2

2-4-6-8 What Can We Associate!

Epidemiology is your key to success. Remember, it's the study of real people in the real world... or at least that's the way you should play it. Epidemiology is very convincing to the public — even though it's often no more reliable than a shaky alibi. So you need to be careful.

Epidemiology is the study of, that's right, epidemics. Arising from the Scientific Revolution of the 17th century, epidemiologic studies have been responsible for many genuine advancements in public health. It's how scurvy among 19th Century sailors was linked with vitamin C deficiency, how cholera outbreaks in 19th Century London were associated with untreated drinking water and how typhoid fever was found to be contagious. Epidemiology has a celebrated history, and its distinguished coattails can take you a long way.

There are two basic types of epidemiology studies that you can perform — cohort and case-control. Avoid cohort studies. They involve following a specific group of people into the distant future. Although cohort studies are the better type of epidemiologic study, they can take 20 years or more to complete. You would have to put your ambitions on hold. By the time your results are in, the general public may have wised up and called a halt to the public health gold rush.

On the other hand, case-control studies are preferred because they're fast. Instead of following a group of people into the future, you simply scrounge up a group you can look at in retrospect. It's like Monday morning quarterbacking, only better. At the end of this game, you can adjust the score almost anyway you want.

All you need is a group of people with the disease you're interested in (the cases) and another group of persons without the disease (the controls). Survey the cases and controls (we'll talk about how to do this in a later chapter) to determine who has been exposed to the risk you're studying. If the prevalence of exposure among the cases is greater than the prevalence of exposure among the controls, you may have a winner. For our purposes, we'll call the statistical representation of this comparison the relative risk.

Rate of exposure among cases
Relative risk = ---------------------------------------------------------
Prevalence of exposure among controls

What does relative risk mean? Let's say you've studied the association between high fat diet and lung cancer. You've calculated a relative risk of 6. The correct interpretation of this relative risk is that the incidence of high fat diets in the study population was six times greater among those with persons with lung cancer than those without lung cancer. Now is that boring or what? This interpretation will take you nowhere fast.

You need to reword and generalize this interpretation to give it some sex appeal. A risk assessor on the make would say something like "this study shows the risk of lung cancer is six times greater among persons with high fat diets." Notice how we've replaced "incidence" with "risk," two very different concepts and used the word "shows."

"Incidence" means we merely observed the reported result in our study. "Incidence" does not imply, one way or the other, that a high fat diet is associated with lung cancer. By replacing "incidence" with "risk," however, we communicate that a high fat diet causes lung cancer. Our study didn't really say that, but don't worry. That's a small detail that the general public won't notice. Finally, use of the word "shows" implies the study proves the risk. In fact, with a single epidemiologic study, it's impossible to prove anything except the limited observations of that study.

The size of your relative risk is,, critically important. The basic rule is simple: the higher the relative risk, the more convincing the association you want to prove.


Relative risk Interpretation (career implications)

Greater than 3

Strong association (jackpot!)

Between 2 and 3

Weak association (may need life support)

Between 1 and 2

Very weak association (call the coroner)


No association (sorry)

Less than 1

Negative association (whoops!)

Now remember, technically a relative risk is only statistical association. It's an apparent relationship between the exposure and disease of interest. Notice the word "apparent" has been struck out. This is not a typo. It's just that you should pretend you never read it. As a matter of science, we really don't know whether the statistical associations identified through epidemiology are real or not. After all, we've only identified them through statistics, and statistics are not science. If science is the sun, statistics are Pluto. In fact, all sorts of wacky associations can be identified through statistics, as shown by the following chart. Tap water and miscarriages, for instance, or whole milk and lung cancer.


Exposure and disease Reported relative risk (by size)

Environmental tobacco smoke and lung cancer


Consuming olive oil and breast cancer


Vasectomy and prostate cancer


Obesity in women and premature death


Sedentary job and colon cancer


3 cups of coffee per week and premature death


Birth weight of 8+ pounds and breast cancer


Baldness in men under 55 and heart attack


Eating margarine everyday and heart disease


Drinking tap water and miscarriage


Regular use of mouthwash and mouth cancer


Abortion and breast cancer


Eating yogurt and ovarian cancer


Drinking whole milk and lung cancer


Obesity in nonsmoking women and premature death


Eating red meat and advanced prostate cancer


Chlorinated drinking water and bladder cancer

2 to 4

Douching and cervical cancer


Workplace stress and colorectal cancer


Eating 12+ hot dogs per month and leukemia


Wearing a brassiere all day and breast cancer


Now between you and me, if you start worrying whether associations you identify through epidemiology make sense, you'll never cut it in risk assessment. A well-developed conscience is not necessary here. So for your purposes, you shouldn't really care whether an association is fact or fiction, only that you've found it. But there is this thing called biological plausibility that you will need to remember.

In addition to just finding a statistical association between exposure and disease, you're supposed to show the statistical association is biologically plausible. That is, it should make sense from a biological standpoint. For example, it is generally accepted as biologically plausible that too much exposure to the sun's ultraviolet rays is associated with an increased risk of skin cancer. However, it is not biologically plausible that too much sun is associated with cavities. So you wouldn't even try to make that association, would you?

Public health researchers have had such a difficult time convincing people electromagnetic fields are harmful because, to date, there's been no convincing evidence the association is biologically plausible.

How do you get biological plausibility? Short of having lots of highly credible epidemiology, laboratory experiments with animals may be necessary. We'll talk more about this in the chapter on biological plausibility.

If you don't have any supporting animal experiments, you're going to have to be creative... maybe even extremely creative. However, biological plausibility is not the equivalent of biological "truth" or "reality" and no one expects it to be (although you must proclaim it such). At best, it means a biological explanation that enjoys the firm possibility of a definite maybe.

Fortunately, it's likely no one will be able to prove you wrong. But your biological explanation should still pass the "red face" test. Depending, that is, on how confident you are no one will challenge you on this.

Click here for Chapter 3: The Significance of Significance

Click here to return to the home page

Copyright © 1996 Steven J. Milloy. All rights reserved. Site developed and hosted by WestLake Solutions, Inc.