Michael Nelson

Statistics in Nutrition and Dietetics


Скачать книгу

circumstances into account [8]. The existing guidance is due to be updated in 2019.

Randomized placebo‐controlled cross‐over trial displaying a box labeled Population with an arrow connecting to Selection of eligible subjects, to Matching and randomization, etc. leading to Final measurements.

      1.6.4 Epidemiological Studies

      Epidemiological studies examine relationships between exposures and health‐related outcomes in populations. In the context of nutritional epidemiology, exposures might include individual diet, community health intervention programmes, supplementation, food advertising, dietary advice, or other nutrition‐related variables. Outcomes can include changes in nutrition‐related blood biochemistry (e.g. cholesterol levels, haemoglobin), clinical outcomes (e.g. xerophthalmia, obesity), or morbidity or mortality statistics relating to nutrition.

      Epidemiological studies fall into three categories.

      Descriptive Studies

      Descriptive studies in epidemiology include ecological studies, cross‐sectional studies, and time trend analysis. They are useful for generating hypotheses. Measurements can be made in individuals at a given point in time (cross‐sectional studies) or accumulated over time in groups of people (ecological studies). They are used to relate measures of exposure and outcome in groups of people that share common characteristics (e.g. vegetarians versus omnivores) or to compare regions or countries. For example, they might compare diet and disease patterns between countries (are heart disease rates lower in countries where people eat lots of oily fish?) or between subgroups (do vegetarians have lower risk of heart disease compared to non‐vegetarians?).

      There are two main problems with this type of study. First, there may be other factors that could explain an observed association or changes in the population over time. For example, populations with higher oily fish consumption may be more active or less obese. Second, not everyone in the population or subgroup is exposed at the same level: some individuals in the population may eat lots of oily fish, while others may eat very little. Are the people with low oily fish consumption the ones that have higher rates of heart disease?

      Analytical Studies

      These include cohort and case‐control studies. Their primary characteristic is that they relate exposures in individuals (factors that are likely to influence the occurrence of disease or mortality within the population) to outcomes (disease or mortality rates). Analytical studies are usually based on observations relating to large numbers of people in the population (hundreds or thousands). They provide much stronger evidence of diet–disease relationships than descriptive studies. In terms of the Bradford Hill model of causality (Box 1.4), they provide evidence of temporal association and dose‐response. If blood or urine samples are collected, they may also provide evidence of a plausible physiological mechanism for causation. Detailed descriptions of these designs and their statistical analysis are given in epidemiological texts such as Rothman [9] or Margetts and Nelson [10].

      The risk of developing disease (or related outcomes) in the exposed group is compared with the risk of developing disease in the unexposed group. This is known as the Relative Risk.

Diagram displaying an ellipse labeled Population with an arrow pointing to a box labeled People without the disease having arrows to Exposed and Unexposed with vertical two-headed arrow in between labeled Compare, etc.

      In nutritional epidemiology, the meanings of ‘Exposed’ and ‘Unexposed’ are different from the meanings in infectious disease or occupational epidemiology. In infectious disease epidemiology, for example, a subject either is or is not exposed to a particular bacterium that can cause illness. In occupational epidemiology, a subject may be classified either as exposed or not exposed to a potential risk factor (e.g. asbestos dust). In nutritional epidemiology, in contrast, variables are usually continuous – no one has a diet that does not contain energy or iron, for example. Subjects are therefore classified into bands of exposure. ‘Exposed’ might mean someone in the top half of the distribution of intake, and ‘Unexposed’ might mean someone in the bottom half. Of course, degrees of exposure may also be ascertained in other spheres of epidemiology, but in nutritional epidemiology it is the norm and forms the basis for most analysis of risk.

      Measures of exposure in the past are usually determined by asking cases and controls about their habits in the past using a questionnaire (although sometimes there are records of exposure, for example the type of work someone did, or hospital records of birth weight). The past exposures of the cases are then compared with the past exposures of the controls. The relevant statistical outcome is the Odds Ratio: what are the chances of being exposed and in the disease group compared with being exposed and in the non‐disease group. This is the best estimate of the Relative Risk, which cannot be measured directly in case‐control studies [9, 10].

      Case‐control studies are much cheaper to carry out than cohort studies, but accurate measurements of past exposure are often difficult to collect. For example, asking someone what their diet was like 10 or 15 years ago is likely to be heavily influenced by the type of diet they consume now.

Diagram of case-control study design with arrows from boxes labeled Cases (with disease) and controls (without disease) heading to Measure previous exposures and two-headed arrow labeled Compare in between the boxes.

      Experimental Studies

      In epidemiology, experimental studies take two forms: clinical trials and community