By Guest Author 24/03/2016 1


Kathryn Snow, University of Melbourne

Would you volunteer to become vegetarian for the next three decades for the sake of science? What if you were asked to run at least 50 kilometres per week, or live through a natural disaster?

Granted, these are extreme requests. Researchers conducting randomised controlled trials often ask volunteers to make far smaller changes to their behaviour: exercise a bit more, eat less sugar or try a new medication.

During these trials, scientists randomly allocate the medicine, treatment or activity being studied to a group of people, and a different intervention or placebo to another group. Then they look for differences in participant outcomes.

Purists believe experiments like this are the only way to gain valuable knowledge, and popular conception of science is intimately connected to experimentation.

Yet some of the most critical scientific questions we face today can’t be investigated through experiment. For instance, we can’t determine whether greenhouse gas emissions are really causing climate change by not producing them for several decades and recording the results.

Likewise, many important medical questions either can’t or shouldn’t be settled experimentally. A chasm separates the controlled conditions of the laboratory from the messy reality of life. Sometimes, studying participants in real conditions through observational studies is the best way to find answers.

‘Only an observational study…’

Epidemiology, broadly defined, seeks to understand the causes of disease.

A chasm separates the controlled conditions of the laboratory from the messy reality of life. from shutterstock.com

An early example of observational epidemiology was John Snow’s discovery that cholera was spreading throughout Victorian-era London not through bad air, as was commonly thought, but through contaminated water from the Thames. He did this by mapping the location of affected households which revealed they clustered around specific water sources.

Almost a century later in the 1950s, Richard Doll and Austin Bradford Hill were the first to observe the link between smoking and lung cancer by surveying doctors about their tobacco use and health. Smoking is now widely recognised as one of the most important modifiable risk factors for early death.

These contributions are often unrecognised by science journalists and even by other researchers. Newspaper articles on the latest finding from observational research often include some variation on the phrase: “only an observational study”, as if this type of scientific inquiry is not to be trusted.

But each study should be evaluated on its own merits – not just its broad design.

In randomised controlled trials, randomisation is used to break the connection between characteristics to identify the true cause of a disease or the most effective cure. For instance, people who exercise frequently may have other healthy habits. These might be the reason for their lower risk of heart attacks, rather than the exercise itself.

Randomisation helps ensure people receiving a particular health intervention are a mixed group and the only thing they definitely have in common is the intervention itself.

Physician John Snow found how cholera was spreading through London using observational epidemiology. Wikimedia Commons

Observational researchers can often use statistical techniques to identify the true causes of disease, even when different relevant factors are clustering together.

For instance, if we are worried people who exercise are less likely to smoke and this might explain their lower risk of heart disease, we can restrict our analyses just to non-smokers. Then if we still see a difference between people who exercise and those who don’t, we can be sure it isn’t due to smoking.

Instead of randomising, observational studies investigate how people live in their natural circumstances – how they behave, their genetic profiles, what’s happened to them in the past, and so on. So many factors that have an impact on health can’t be randomly allocated.

The value of observational research

The repetition of the “only an observational study” mantra ignores the fact that randomised studies are often impossible – for example, if we want to study the impact of genes, long-term patterns in diet or physical activity, personal experiences like childhood trauma or incarceration, or natural disasters.

Obviously, researchers can’t randomly assign these traits or experiences to participants in a trial.

Observational studies have been used to identify the link between those who have the BRCA gene variants and their higher risk of breast cancer.

Now women with these gene variants can take some measures to protect themselves from advanced breast cancer. This contribution joins a long list that began with controlling cholera in London and continued with identifying the harms of smoking.

The complexity of human beings means that medical researchers can’t say with the perfect certainty of physicists that X causes Y, but the world can’t always wait for perfect certainty.

Observational epidemiologists design studies with the greatest degree of rigour possible given the messy reality of life, and we offer our findings up in the hope of protecting public health. Every so often, that can be the difference between life and death.

The Conversation

Kathryn Snow, Epidemiologist, University of Melbourne

This article was originally published on The Conversation. Read the original article.

Featured image: Flickr CC, Alex Frag.


One Response to “In defence of observational science”

  • Excellent Observation (Pardon the pun).

    I think it is also worth noting that observational study design should be used more when examining the effects of nutritionals, such as vitamins and supplements. While, the pharmaceutical/RCT study model is much less effective.

    The reasons are seated in these two basic truisms for nutritionals, I) standardization across more natural products is very very difficult and ii) the mechanisms, pathways, and health conditions which are implicated with more natural products are myriad. For these reasons and others too many to mention here, does not suit the pharmaceutical/RCT design model. But observational study can be useful. Especially if we don’t desire a direct and singular cause and effect relationship with a clinical outcome. Patterns, trends, and correlations can be extremely useful information to better inform the use and study of nutritional based therapeutics. And as in your blog examples, it may be the only method for obtaining useful information.

    One last point is our newly found access to Big Data or even More Data. As we move towards n=100 every step towards this increases the strength and value of information/data obtained under an observational model. I argue that at some point of inflection along that road to n=100, observational information will be more value than RCT derived information. I will leave the determination of that point of inflection to the statisticians. But let us hail the analysis of the many-to-a-few vs. the few-to-the-many.