Clinical research: from design to registration

Clinical research is the effort to measure systematically what an intervention does in people. Whether that measurement can be trusted rests not on statistics alone but on the fit between design and question, on an ethical framework being in place, and on the study having been registered in advance.

Matching design to question

Not every question is answered by the same design:

• Randomised controlled trial — gives the strongest answer to a question of causation, and suits measuring the effect of an intervention.
• Cohort study — follows outcomes over time where randomisation is not ethical or practical.
• Case-control study — looks backwards from the outcome, useful for rare events.
• Cross-sectional study — measures frequency at a point in time and cannot establish causation.

In observational designs confounding is the central problem: an unmeasured variable may be the real reason behind an apparent association. Statistical adjustment reduces that risk without removing it.

The ethical framework

The founding principles of research in humans are set out in the Declaration of Helsinki: the wellbeing of the participant takes precedence over scientific and societal interests, informed consent must be genuinely obtained, and an independent ethics committee must review the study. Good Clinical Practice guidelines translate these principles into the conduct of a trial.

Informed consent is a process rather than a signature: a participant has to have understood the purpose of the study, its foreseeable risks, the alternatives, and that withdrawal is possible at any time.

Prospective registration

Registering a study in a public registry before enrolment begins makes any later change to the primary endpoint or analysis plan traceable. Registration also offers some protection against publication bias: a study left unpublished because its result was unfavourable can at least be seen to exist.

Sample size

A sample size calculation determines in advance how large a difference a study is capable of detecting. Where a study is underpowered, the absence of a significant result does not show that there is no effect; it shows only that the study was not adequate to demonstrate one. The distinction is frequently blurred when results are interpreted.

Collecting safety data

Safety is as much a result of a study as efficacy. How adverse events are defined, the severity scale used to grade them, and how their relationship to the intervention is judged should all be specified in advance. Rare adverse effects usually cannot be detected at the sample sizes of pre-authorisation studies, which is why post-authorisation surveillance is required as a separate stage.

Reporting

The value of a study depends on how it is reported as much as on what it found. CONSORT defines what a randomised trial should contain, from participant flow through to the analysis set. A study reported incompletely cannot be appraised independently, however well it was conducted.