Horses for courses

In part 1 of this blog post series, we tackled how to ‘ask the research question’. You have now come up with a research question with enough detail. How do you go about choosing the right study design to answer it?

The two main clinical study designs include randomised controlled trials (RCTs) and observational studies.

What are randomised controlled trials (RCTs)?

RCTs utilise the simple but brilliant idea of randomly allocating participants to different interventions to make a fair comparison and is appropriate for most causal questions. Depending on the “estimand” (see part 1), there are many options in the RCT toolbox to help tackle the problem.

Some examples include:

parallel trials, where one/several interventions are initiated in parallel groups;
crossover trials, where participants are allocated to a sequence of interventions;
factorial trials, where participants are randomised to multiple interventions at the same time;
cluster randomised trials, where “clusters” of participants are randomised to interventions rather than the individual participant.

You could also design different trials to test whether one intervention is better than another (superiority); comparable to one another (equivalence); or no worse than another (non-inferiority).

Randomisation does not remove all types of bias, and therefore care is still needed to design a trial that minimises bias that can arise from the randomisation process, deviations from the intended intervention (also related to the estimand), missing outcome data, measurement of the outcome, or selected reporting of the results.

What sort of resources are available for setting up an RCT?

Care and expertise from an experienced team, such as support offered from a Clinical Trials Unit (CTU), is needed to tailor the right design and subsequent analysis to the chosen estimand, and to help conduct the trial by ensuring that ethical and regulatory frameworks are followed. A large “explanatory” trial often needs to be preceded by smaller pilot studies to assess the feasibility of the trial design such as recruitment, administration of the intervention, or outcome measurement; and/or to obtain an initial estimate of treatment effect size to calculate the sample size for the eventual large RCT.

Aside from CTU support, you will need to be able to obtain a source of funding, commonly from charities, public funding bodies such as MRC, NIHR (see BAD funding opportunities page) or from the industry (for example a pharmaceutical company), to carry out the trial, and to identify centres to recruit your participants from. RCTs are therefore a significant collaborative undertaking which will take up several years of your professional life but may come up with results that are potentially transformative for many patients.

Additional forms of support are available. The UK Dermatology Clinical Trials Network is a support organisation made up of clinicians, methodologists, and patients, and is an ideal source of support for clinical researchers to navigate some of the more difficult challenges listed above and help develop your ideas to become a fully funded study.

What are observational studies?

Observational studies do not allocate any intervention to participants but simply observes what happens when people receive interventions, usually in routine clinical settings. This means that they are often cheaper to run, easier to recruit to, and can include more participants than RCTs.

They can be more reflective of the target population that is useful for clinical decision making. They tend to be useful tools for estimands that aim to reflect routine clinical practice, or for outcomes that are rare. However, they can be potentially affected by a lot more biases compared to RCTs.

Commonly seen biases in observational studies include:

selection bias, where the selection process of recruitment for the intervention may introduce a bias in favour of one or more interventions;
confounding, where factors that lead to the introduction of the intervention also lead to the outcome (a particularly problematic one involves confounding by indication where the clinician who tries to tailor the treatment choice to the patient characteristic introduces a bias, e.g. choosing dupilumab for people with a higher risk of infection over upadacitinib where the study evaluates the risk of pneumonia between these two treatments);
immortal time bias, where the choice of the start of the study does not align with the intervention start date and thus may introduce observed study time when the outcome cannot occur.

A classic example for immortal time bias includes studies that looked at hydroxychloroquine for Covid which had cohort entry at hospitalisation but the exposure as treatment with hydroxychloroquine at any time during follow-up or within 48 hours of hospitalisation. This creates immortal time between hospitalisation and the start of treatment as the participant would have needed to survive until they could receive the drug.

There are three main observational study designs for groups of participants.

Cross-sectional studies look at a snapshot in time; they are good for generating hypotheses about associations between exposures and/or interventions but cannot tell us about causal relationships.

Case-control studies start from cases, who are participants who had the outcome, and controls, who are participants who did not have the outcome but are from the same underlying population and look backwards in time to see what exposures are associated with the outcome. They are good for epidemiological studies of rare outcomes and can look at multiple exposures at the same time, but these studies can be particularly challenged by recall bias.

Finally, cohort studies compare interventions or exposures and look forward in time and are the most analogous to RCTs.

All three studies can be retrospective or historical – that is data has already been collected at the time of conducting the study, which is common for studies using routinely collected data; or prospective, where the study is set up to collect data specifically to answer a question, and this does not necessarily map directly to how time is handled in the study itself. Like CTUs for RCTs, getting help from a friendly epidemiologist or statistician is helpful to choose the right observational study design and to apply for a source of funding.

What kind of supporting tools exist for observational studies?

We are fortunate to have bespoke disease registers in British dermatology run by the BAD in psoriasis (BADBIR) and atopic eczema (A-STAR), and due to have ones starting in vitiligo (VIRTUAL-UK), alopecia areata (GRASS-UK), and hidradenitis suppurativa (H-STRONG).

These are types of cohort studies that might be good databases which collect the data to answer your research question – do look to see whether they would be good options for you and apply to get involved!

An ending note of encouragement

It is beyond the scope of two blog posts to cover all the design choices and considerations that would go into a clinical research study, but I hope these articles give you food for thought to develop the big idea into an actionable study and change care for patients with skin problems both locally and internationally for the better.

Author’s note

My heartfelt thanks for Professor Beth Stuart for acting as peer reviewer for these blog posts.

Further resources

The NIHR Associate Principal Investigator Scheme supports early career researchers by aiming to develop health and care professionals into future Principal Investigators.

A set of online dermatology research resources providing a range of information about courses and opportunities for getting hands-on-experience in a breadth of research domains are available on the BAD website.