This post will give you insight into the underbelly of studies, how they are performed, and which studies should be more trustworthy. You will understand how scientific knowledge is gained under the framework of the current science paradigm.
If one of the reasons why you have never read them was because you did not completely understand them, then you have come to the right place. This post is meant for people who would like to get a better understanding how studies about nutrition are done, and how they should be interpreted.
If you dislike smartypants talk, I suggest skimming the flowchart at the end of the post, as this really is something I would like everyone to know, remember, and use.
All the posts made on this page are based on scientific studies (and not old wives tales, anecdotal evidence, or “common sense” knowledge). They include a lot of links to other pages, such as this, that base information about nutrition on scientific studies.
All studies that are posted in peer reviewed journals must go through a meticulous process of being reviewed by fellow scientists. They have to approve the content and quality of a particular study.
Throughout history there have been a lot of animal studies. Nutritional science is not excluded from this. However, because the science of human nutrition is a very specific topic, we cannot close the effect a certain eating pattern would have on our species, based on an experimental design done on animals.
Still, specific information can be extracted about certain topics from animals that have unique properties. For example – the Watanabe rabbit has defective LDL receptor functioning. That is why this animal model is useful for studying the role of eating patterns on a arterial disease related to LDL receptors.
But let us stick with human studies. I would like to focus mostly on different design types. We must know that not all studies are created equal. For a long time now, there has been a ton of articles in the media about certain issues or foods that have been linked to certain outcomes (eating red meat to cancer). Why this is bad and most of the time gets blown out of proportion will be explained further down the post.
- Nutritional epidemiology
- Experimental human nutrition
Nutritional epidemiology uses observational studies to show possible connections between certain eating patterns, our lifestyle (exercise, smoking etc.) and the outcomes. These are usually different diseases or similar health parameters.
Experimental human nutrition has a different approach. They take a hypothesis, or several of them, and try to understand the nature of a relationship between a certain macro- or micronutrient, and the specific outcome that follows after increasing or decreasing that nutrient.
Example: How would adding 200g of red meat affect your blood pressure, HDL and LDL cholesterol etc. If you started consuming it for 3 weeks on a daily basis.
They would gather information about those parameters before and after you would start eating those amounts of meat – the effects would then be attributed to the consumption of meat (this is a very simple model of an experiment – actual experiments dive deeper into controlling other important characteristics that could affect the outcome too).
With enough evidence that has been repeated by different studies (this is very important and is one thing that really gets on my nerve with media misinterpreting and blowing a certain study out of proportions), we can make safer assumptions about the (un)healthiness of certain foods. Experimental nutritional epidemiology will then also check population level interventions and how they can affect the prevalence of certain diseases.
Experimental research tries to understand the mechanism and is usually connected to samples of smaller size, while the epidemiological research will use larger numbers to try and decide what impact a certain eating pattern intervention will have on the majority of people.
We come in all shapes, sizes and colors. That is why there is a staggering amount of variability when it comes to genetics, our environment, our lifestyle, eating patterns prior to the study, current health, physical status, and a lot of other variables that influence experiment outcomes in a significant way.
Not only that, if we would like to manipulate a certain eating pattern fully – say, we would like the people in our study to start eating healthy in a way they never ate, or never would eat (think – making their food intake pure protein powder or something similarly bizarre), we could extrapolate certain effects from such eating patterns on different metabolic measures, but they would not be relevant as nobody eats that way (at least to my knowledge).
Studies must still have their practical applications, despite having complete control over the eating patterns of people.
On the other hand, relying solely on self-report questionnaires is not really something you want to do. Ok yes, people will try and write honest answers (well, not everyone), but we do not have perfect memories, nor are we completely rational. There is under/over reporting (full pdf) and there is a failing memory. I doubt many of you even remember what you ate last week, including me, let alone last year. Which is why such studies are very prone to higher levels of errors.
The best design would be where we could exclude (or control) as many possible variables that could affect our results, and at the same time still keep our study as practically applicable as possible, which brings me to the different study design types.
A randomized clinical trial is the golden standard. It provides us with information about a cause-effect relationship between foods and diseases (or variables in general). Here we get the long lost CAUSATION, from the sentence – correlation does not imply causation.
The best randomized clinical trials will take in account the effect of a certain food type (nutrient) on different parameters of the body, while taking in account differences in age, gender and more specific measures (alcohol drinking, exercise, smoking, health status etc.) – these things often greatly influence studies and are a source of error that is not accounted for.
Randomized clinical trials can use something called a minimization scheme. With this technique, people can be assigned into different groups by how similar they are to each other in important characteristics.
Nutritional epidemiology then integrates the knowledge gained in this smaller scale nutrition research. It reexamines the same parameters and relationships on whole populations.
It serves as another safety net for the conclusiveness of evidence. Therefore, we can understand the role of our eating patterns in the cause, and in prevention, of certain illnesses more easily.
Epidemiological study designs vary as well. They can be divided into two broad categories.
- Nonexperimental (observational)
In experimental epidemiological designs, people are exposed to a certain treatment (they are presented with a particular food choice/eating pattern) over a time period. This treatment is assigned so that both groups – the treated and untreated (the experimental and control) group are maximally comparable in regard to all important characteristics. They also have to be randomly assigned into either of the two groups.
In such research, the primary exposure of interest are usually the eating patterns of people. The outcome usually involves prevalence or occurrence of disease and/or nutritional status indicators.
In general, experimental studies where a person gets randomly assigned to a group, give us the strongest evidence for our effects. They are also the type of studies that should be the “go-to for certain information” standard.
Randomized controlled trials are one of the study designs that dominate the area of population studies. The second are crossover studies.
In crossover studies, people get randomly assigned to the placebo or treatment group. Then we compare changes in certain health indicators or disease status and compare between the two groups at the end of the experiment to find the effect of the exposure.
A crossover design is based on the same principle as a repeated measure design common in basic science research – all people in the study receive the placebo and treatment for equal periods and a “washout” (buffer) period in between. The order of receiving the treatment is random for each person.
Generally speaking – experimental epidemiological study designs are well suited to find causal relationships between specific eating patterns/nutrients and indicators of health/disease status. But, being 100% about a causal link is impossible, the sheer number of variables is always enormous and can never be controlled fully.
Randomized controlled trials, crossover studies and epidemiological research are some of the most important types of designs when it comes to nutrition, but they are not the only ones.
There are also
Single trials; comparing only one nutrient to a placebo and observing effects.
Quasi experiments (intervention trials); the scientist decides who gets the nutrients that are included in the study. People are not randomly assigned to groups.
Primary prevention trials (field trials); a form of randomized clinical trials where healthy people are included with the aim of preventing the onset of a particular disease.
Secondary prevention trials; here we have people who are suffering from a particular disease and are randomly assigned for the nutrient intake or a placebo.
Non Experimental (observational)
The main difference between different observational study designs is the time when the received nutrients and outcomes are measured.
Cross-sectional studies (prevalence studies); they give us information on nutrition and gather data about a certain population; how healthy are they eating – what are they missing, what do they have enough of. They record the characteristics of eating patterns of a particular population and certain health outcomes at the same time. They are the base on which health promotion and disease prevention programs can lean on in a particular time period.
Case-control studies; they get information on past eating patterns and nutrient intake, then they measure the current health status. Usually, they compare the dietary and lifestyle patterns of people with a certain illness with people, who can be matched by gender and age.
They get information on their earlier eating patterns with interviews, questionnaires or even medical record reviews.
Cohort studies; they see the eating patterns at the moment and make predictions about the most probable outcome for certain populations, based on how (un)healthy those eating patterns are.
Meta-analysis – they systematically combine results from published randomized trials to give more definitive information about certain eating patterns.
This was an overview of possible study designs in nutrition. When reading we should strive to take epidemiological studies with more reserve as they only give us connections (correlations). If a certain connection is statistically significant; p value is under 0,05 (if the p value in the results is over 0,05, then the connection has most probably happened by chance and is not important; it is statistically insignificant), that does not mean that there is not a third variable that greatly affects this relationship.
But, as I do not wish to dismiss correlations totally, here is a nice quote:
While not giving us any proof, when such a connection arises, it warrants further investigation.
The preferred study to interpret, if we want to be very sure about what we are looking for, is a randomized controlled trial, preferably more of them and more of them in a meta-analysis, where they were probably also corrected for publication bias (another funny part of science).
By now you should have added a few new words into your vocabulary when it comes to scientific studies about nutrition. Next time you talk about nutrition and people start claiming wild stories, ask them how many randomized controlled trials/intervention trials have measured such effects. Then, witness the jaw dropping/mouth shutting effect of such a question.
Below is a flowchart that shortly summarizes what I have written about (right click and open in new window to get the full picture).