1 December 2019


Low carb diets. Low fat diets. Owning a dog. Eating leafy greens. Fermented foods. Dairy foods. Eggs. Salt. Sugar. We are bombarded with conflicting messages about food, diet and health in the media and on-line. Universities and research organisations keen to promote their staff and their findings in turn bombard the media with often inflated press releases. How do we work out who or what to believe? In Food for Thought, Dr Alan Barclay explains the different kinds of evidence and how compelling it is, or isn’t.

Evidence pyramid

1. RANDOMISED CONTROLLED TRIALS (RCTS) ARE CONSIDERED THE “GOLD STANDARD”, providing the highest level of evidence, as they can prove that intervention A improves health outcome B, while all other known factors (known as confounders such as age, gender, body mass index, etc.) have been accounted for by the randomisation process. The process of studying people tends to improve their health independent of the intervention itself, because people know that they are being monitored and are more conscious of their health, and are consequently being more careful about what they eat and drink, so having a control group is vital. Only randomised controlled trials are able to show that a particular intervention causes a particular outcome.

2. OBSERVATIONAL STUDIES PROVIDE MEDIUM LEVEL EVIDENCE because scientists are simply observing and measuring people’s behaviour at a point in time, or over a particular time frame, without randomising them to groups and providing different dietary interventions. The best epidemiological evidence comes from large prospective cohort studies where large groups (typically thousands) of people have a medical check-up, their dietary patterns are measured, and they are followed up regularly for long periods of time (e.g., 5–25 years).

Observational studies can’t provide as high a level of evidence as RCTs can as it is not possible to control for all confounders (e.g. people who are already overweight may drink more beverages than those who aren’t, as fluid requirements are proportional to body size, and being overweight is an independent risk factor for developing many chronic diseases), and our tools of observation (e.g. a food frequency questionnaire for measuring a person’s usual food and drink intake) are imperfect. Observational studies are only able to prove that event A is associated with outcome Z. It’s possible that unknown or unmeasured intermediary factors (B, C, D, E, etc) are involved. They are not able to prove that event A causes outcome Z – only RCTs can.

3. ANIMAL STUDIES ONLY PROVIDE LOW LEVELS OF SCIENTIFIC EVIDENCE, however, they can be used to generate hypotheses that can be tested in human populations (either using RCTs or observational studies) and to investigate hypothesised physiological mechanisms in experiments that cannot be ethically conducted in people. They are also used to determine the toxic dose of novel ingredients, like food additives, for example, and results are extrapolated to people using a large safety factor (typically 100 x).

HOW SCIENTISTS REVIEW EVIDENCE: 3 THINGS YOU NEED TO KNOW 1. SYSTEMATIC LITERATURE REVIEWS are based on careful searches of scientific databases (e.g., PubMed, EMBASE, CINAHL, and Cochrane Library) with pre-determined search terms looking for all of the research published on a particular topic over a long period of time (ideally with no time constraints). Once all studies have been identified, researchers then go through each paper’s reference lists to make sure as best as possible that they have not missed any additional evidence. The data from each paper is then extracted and the results summarised in a table. The quality of each study is also rated or graded. Strong conclusions can be drawn from the summarised data.

2. META-ANALYSES can be performed when three or more similarly designed studies on a particular health outcome have been published in scientific journals. The outcome data from each study is entered into specialised software and weighted according to the study size and statistical significance. A final summary statistic is given that indicates whether an intervention is effective, and if so, how effective.

3. SYSTEMATIC LITERATURE REVIEWS AND META-ANALYSES OF RANDOMISED CONTROLLED TRIALS are considered the highest level of evidence. Cochrane reviews are a good example of this method. You can also do systematic reviews and meta-analyses of observational studies. However, because the underlying study design is not as robust as the randomised controlled trials, they are not considered to be as high a level of evidence as a Cochrane review, for example.

If the latest study broadcast in the news is: 

  • A systematic literature review and meta-analysis of randomised controlled trials then the results are worth taking notice of if the people involved are similar to you, and live under similar circumstances. 
  • A systematic literature review and meta-analysis of observational studies then the results are interesting, but a randomised controlled trial in humans that studied the same effect would be necessary to prove that the relationship was causal.
  • A systematic literature review and meta-analysis based on an animal study or in vitro (test-tube) study, then more research in people is needed to prove the hypothesis.
Source: www.dralanbarclay.com