Evaluation of a therapy involves comparing a group of patients receiving the intervention with a group of patients who do not receive it (the control group). With a few rare exceptions (such as diseases that currently have 100% mortality), a control group is always required to demonstrate that any improvement observed after treatment is not simply due to the natural course of the illness. There are a number of key elements in the design of these studies that will determine whether the findings are valid and generalisable.

SELECTION AND ALLOCATION OF STUDY PARTICIPANTS

Patients are selected to a trial by a process of recruitment that usually involves identification of potential participants, assessment of eligibility using inclusion and exclusion criteria, followed by a request for consent to participate. Selection can occur at any of these stages to influence the constitution of the study population. This is obviously a necessary process in assembling the study population, but selection can influence the interpretation of the findings.

Selection of patients for a trial will clearly affect generalisability. The results of the trial will only be generalisable to patients who resemble the selected study population. If most eligible patients are identified and recruited, then the results will be generalisable to the wider population. If recruitment is highly selective, then findings may not be easily generalisable. A high disease prevalence in the study population suggests a highly selected cohort.

Once patients have been selected into a trial, they are allocated to intervention or control treatment. Bias may result if patients, carers or researchers can influence the process of allocation. For example, patients may choose a treatment that they think will be beneficial. This will result in certain types of patient being allocated to certain treatments, leading to bias. The more that patients, carers and researchers can influence allocation to treatment group, the …