Learning objectives

Outline the general idea of the clinical trial.

Explain the concept of randomisation, and why it is important, and demonstrate that you can use a random number table to perform a simple block randomisation.

Describe the concept of blinding, and what it is intended to achieve.

Outline and compare the design of the parallel and cross-over randomised controlled trials, and summarise their respective advantages and shortcomings.

Explain what intention-to-treat means.

Be able to choose an appropriate study design to answer some given research question.

Hey ho! Hey ho! It’s off to work we go

There are two main threads here. First, the study design question, and second, the data collection question. Study design embraces issues like:

What is the research question? What are we hypothesising?

Which variables do we need to measure?

Which is our main outcome variable (the variable we are most interested in)?

How many subjects need to be included in the study?

Who exactly are the subjects? How should we select them?

How many groups do we need?

Are we going to make some form of clinical intervention or simply observe?

Do we need a comparison group?

At what stage are we going to take measurements? Before, during, after, etc.?

How long will the study take? And so on.

Study design is a systematic way of dealing with these issues, and offers a good-practice blueprint that is applicable in almost all research situations.

Second, the data collection question. Having decided an appropriate study design, we then have to consider the following:

How are we going collect the data from the subjects?

How do we ensure that the sample is as representative as possible?

I want to start with the data collection question. First, though, a brief mention of what we mean by a population.

Collecting the data – types of sample

Now the data collection question. There are many books wholly dedicated to the various methods of collecting sample data. I am going to do little more than mention a couple of these methods by name. Those interested in more details of the methods referred to should consult other readily available sources.

The simple random sample and its offspring

The most important consideration is that any sample should be representative of the population from which it is taken. For example, if your population has equal numbers of male and female babies, but your sample consists of twice as many male babies as female, then any conclusions you draw are likely to be, at least, misleading. Generally, the most representative sample is a simple random sample. The only way that a simple random sample will differ from the population will be due to chance alone.

For a sample to be truly random, every member of the population must have an equal chance of being included in the sample. Unfortunately, this is rarely possible in practice, since this would require a complete and up-to-date list (name and contact details) of, for example, every lowbirthweight baby born in the UK in 2007. Such a list is called a sampling frame. In practice, compiling an accurate sampling frame for any population is hardly ever going to be feasible!

This same problem applies also to two close relatives of simple random sampling – systematic random sampling, and stratified random sampling. In the former, some fixed fraction of the sampling frame is selected, say every 10th or every 50th member, until a sample of the required size is obtained. Provided there are no hidden patterns in the sampling frame, this method will produce samples as representative as a random sample. In stratified sampling, the sampling frame is first broken down into strata relevant to the study, for example men and women; or nonsmokers, ex-smokers and smokers. Then each separate stratum is sampled using a systematic sampling approach, and finally these strata samples are combined. But both methods require a sampling frame.

Contact or consecutive samples

The need for an accurate sampling frame makes random sampling impractical in any realistic clinical setting. One common alternative is to take as a sample, individuals in current or recent contact with the clinical services, such as consecutive attendees at a clinic. For example, in the study of stress as a risk factor for breast cancer (Table 1.6), the researchers took as their sample 332 women attending a clinic at Leeds General Infirmary for a breast lump biopsy.

Alternatively, researchers may study a group of subjects in situ, for example on a ward, or in some other setting. In the nit lotion study (Table 2.1), researchers took as their sample all infested children from a number of Parisian primary schools, based on the high rates of infestation in those same schools the previous year.

If your sample is not a random sample, then the obvious question is, ‘How representative is it of the population?’ And, moreover, which population are we talking about here? In the breast cancer study, if the researchers were confident that their sample of 332 women was