Study design is vitally important as poorly designed studies may give misleading results. Large amounts of data from a poor study will not compensate for problems in its design. In this chapter and in Chapter 13 we discuss some of the main aspects of study design. In Chapters 14–16 we discuss specific types of study: clinical trials, cohort studies and case–control studies.
The aims of any study should be clearly stated at the outset. We may wish to estimate a parameter in the population such as the risk of some event (Chapter 15), to consider associations between a particular aetiological factor and an outcome of interest, or to evaluate the effect of an intervention such as a new treatment. There may be a number of possible designs for any such study. The ultimate choice of design will depend not only on the aims but also on the resources available and ethical considerations (see Table 12.1).
Experimental or Observational Studies
- Experimental studies involve the investigator intervening in some way to affect the outcome. The clinical trial (Chapter 14) is an example of an experimental study in which the investigator introduces some form of treatment. Other examples include animal studies or laboratory studies that are carried out under experimental conditions. Experimental studies provide the most convincing evidence for any hypothesis as it is generally possible to control for factors that may affect the outcome (see also Chapter 40). However, these studies are not always feasible or, if they involve humans or animals, may be unethical.
- Observational studies, e.g. cohort (Chapter 15) or case–control (Chapter 16) studies, are those in which the investigator does nothing to affect the outcome but simply observes what happens. These studies may provide poorer information than experimental studies because it is often impossible to control for all factors that affect the outcome. However, in some situations, they may be the only types of study that are helpful or possible. Epidemiological studies, which assess the relationship between factors of interest and disease in the population, are observational.
Defining the Unit of Observation
The unit of observation is the ‘individual’ or smallest group of ‘individuals’ which can be regarded as independent for the purposes of analysis, i.e. its response of interest is unaffected by those of the other units of observation. In medical studies, whether experimental or observational, investigators are usually interested in the outcomes of an individual person. For example, in a clinical trial (Chapter 14), the unit of observation is usually the individual patient as his/her response to treatment is believed not to be affected by the responses to treatment experienced by other patients in the trial. However, for some studies, it may be appropriate to consider different units of observation. For example:
- In dental studies, the unit of observation may be the patient’s mouth rather than an individual tooth, as the teeth within a patient’s mouth are not independent of each other.
- In some experimental studies, particularly laboratory studies, it may be necessary to pool material from different individuals (e.g. mice). It is then impossible to assess each individual separately and the pooled material (e.g. that in the well of a tissue culture plate) becomes the unit of observation.
- A cluster randomized trial (Chapter 14) is an example of an experimental study where the unit of observation is a group of individuals, such as all the children in a class.
- An ecological study is a particular type of epidemiological study in which the unit of observation is a community or group of individuals rather than the individual. For example, we may compare national mortality rates from breast cancer across a number of different countries to see whether mortality rates appear to be higher in some countries than others, or whether mortality rates are correlated with other national characteristics. While any associations identified in this way may provide interesting hypotheses for further research, care should always be taken when interpreting the results from such studies owing to the potential for bias (see the ecological fallacy in Chapter 34).
Multicentre Studies
A multicentre study, which may be experimental or observational, enrols a number of individuals from each of two or more centres (e.g. hospital clinic, general practice, etc.). While these centres may be of a different type and/or size, the same study protocol will be used in all centres. If management practices vary across centres, it is likely that the outcomes experienced by two individuals within the same centre will be more similar than those experienced by two individuals in different centres. The analysis of a multicentre study, which is usually performed in a single coordinating centre, should always take account of any centre ‘effects’, either through an analysis suitable for clustered data (Chapter 42), or by adjustment for the centre in a multivariable regression analysis (Chapter 33).
Assessing Causality
In medical research we are generally interested in whether exposure to a factor causes an effect (e.g. whether smoking causes lung cancer). Although the most convincing evidence for the causal role of a factor in disease usually comes from randomized controlled trials (Chapter 14), information from observational studies may be used provided a number of criteria are met. The most well-known criteria for assessing causation were proposed by Hill1.