Common trial designs
There are a number of commonly used alternative randomised trial designs rather than a standard design. Roughly put, alternative trial designs involve changing the randomisation options, what is randomised, and/or having multiple analyses that might lead to a change in the initial design (e.g., stopping early or dropping a randomisation option). The latter type of designs are often referred to as “adaptive trials", as the design allows for pre-planned adaptation. A very brief summary of the standard design and some of the more common designs are given below.
A further set of study designs are what might be called supra-trial designs, as they are related to setting up a single study that in essence allows for multiple RCTs to be carried out over time. Use of an alternative trial design has implications for the design (including the sample size), conduct, analysis, and reporting. More extensive introductions to the designs considered here, and variations of them, can be found in the literature (see for example Cook et al). A list of less common alternative (randomised) trial designs is given here.
Standard trial design
Figure 1: Standard two-arm individually randomised parallel-group trial
A standard randomised trial design can be formally stated to be a two-arm individually randomised parallel group trial. They probably make up around half of all RCTs carried out and are what may instinctively come to mind when thinking of an RCT.
Figure 1 alongside shows what a standard trial design might look like, with two treatments, A and B, and a single common follow-up period.
Multi-arm trial
Figure 2: Four-arm parallel-group trial evaluating treatments A, B, C, and D
The most common way to change the design from the standard is to move to three or more arms; i.e., random allocation is to three or more options. The options can be various things, including variations of a new intervention or a completely different approach. Moving to three or more arms also opens up the possibility of combinations of interventions and evaluating interventions in isolation.
Figure 2 shows a simple form of multi-arm trial: a four-arm parallel-group trial evaluating treatments A, B, C, and D.
Crossover trial
Figure 3: Cross-over trialIn a crossover design, an individual receives a randomly allocated series of treatments over multiple time periods. The simplest design is a two-period two-treatment design (2x2 design), where there are two treatments (A and C). This design is also often called an AB/BA design as these are the only possible sequences of treatments in this design (see Figure 3).
A gap is often used between the end of follow-up of the first period and the start of the second treatment period. This gap is commonly referred to as a “washout” period, reflecting the typical use of this design for comparing two drugs. A period of time is needed where the outcome is not observed to allow for the first drug to leave the participant's system.
Within-person trial
Figure 4: Within-group trialWhere two or more units belonging to the same person can be randomised to different treatments, a within-person design can be used. This design has been used to asses treatments for a number of different conditions, such as dental interventions (using a “split-mouth” design), eye drops, and orthopaedic interventions (e.g., a different operation for each knee). The outcomes need to be unit-specific.
The simplest within-person trial is illustrated in Figure 4. Each individual randomised has only two sub-units (e.g., knees). One receives A and the other receives B. The only possible options are AB and BA, similar to the crossover trial, although what A and B are differs.
Cluster trial
A cluster trial involves randomisation of groups of units (e.g., patients). The clusters (or groups of units) may already exist (e.g., hospitals) or may be an artificial construct for the purpose of the trial (e.g., creating equal-sized geographical areas in a city). All units within a cluster receive the cluster's allocation.
A two-arm parallel-group cluster randomised trial with twelve clusters of ten units each is shown in Figure 5 below. Note here the follow-up lines indicate follow-up of individual units within the respective cluster unlike the previous figures where the randomisation options are indicated.
Cluster randomised trials are typically more difficult to conduct and are at greater risk of bias because the randomisation is only at the cluster level.
Figure 5: Cluster-randomised trial
Adaptive trials
Many alternative trial designs are what might be described as adaptive or having a “flexible” design. The general value of adaptive designs are nicely covered in these papers. A list of some of the more common and interestingly named adaptive trial designs (along with other fixed designs) is given here.