What is it?
Comprehensive cohort study (CCS) designs are a type of pragmatic trial which include participants who do not consent to be randomised to the treatment group. This reduces selection bias and improves generalisability.
In a typical randomised controlled trial (RCT), patients who do not consent to be randomised to treatment are excluded from the trial and all analyses, which may cause a selection bias based on patient preference and a lack of generalisability of the trial results to a real-world population. CCS designs allow the patient preference to be incorporated into the trial, through the use of a ‘parallel-preference group’.
At the beginning of a CSS patients’ preferences are elicited before randomisation occurs, and the study design recruits all patients that are eligible regardless of their consent to randomisation. Those who do not consent to randomisation are kept in the study but their treatment choice is made based on preference (often as a combination of patient and physician preference). Patients who consent are randomised to the two treatment choices. The design is illustrated in figure 1.
Figure 1: Elicit patient preference prior to randomisation and follow-up parallel patient preference arms that do not consent to randomisation.
Statistical analysis of CCS designs involves:
- Standard analysis of the RCT sub-cohort: these results have internal validity.
- Comparison of characteristics of randomised patients with non-randomised patients: this aims to measure external validity of the RCT. This is used to assess if there are differences in confounding/prognostic factors between those who consent to randomisation and those who do not. What, if any, factors influence consent to randomisation?
- Comparison of treatment effect between randomised patients and non-randomised patients: this is a central aim of the analysis (Schmoor et al 1996); if there are observable differences between the treatment effect in the randomised and non-randomised patients, it may not be possible to conclude that the trial results are externally valid. However, these differences in treatment effects may be due to unmeasured or unknown confounding.
An alternative design to the CSS has been proposed that both helps understand the relationship between patient preferences and outcome, and retains the randomised design. Patient preference is measured at baseline and patients who consent are randomised. The flowchart in figure 2 represents this process. The patient preference can then be used within the statistical analysis to understand how preference may affect outcomes in non-blinded studies.
Figure 2: Elicit patient preference prior to randomisation and use within statistical analysis, and only follow-up those who consent to randomisation.
Why is it useful?
- Improves generalisability: external validity and generalisability of the trial are improved by retaining participants who have strong preferences – in the real-world – about which treatment they will receive.
- Improves recruitment rates: participants with strong preferences are not excluded from the trial, so more participants are included who might otherwise be excluded from a typical RCT.
When is it suitable?
- CCS design is only suitable under extraordinary circumstances, such as when the proportion of patients with a strong preference is high (Schmoor et al 1996).
What are the limitations?
- May require additional resources: significant resources may be needed to follow-up non-randomised patients. However if follow-up is undertaken as routine practice and data collection done using electronic health records (EHR) costs will be greatly reduced.
- Clinicians’ efforts in encouraging patient consent may be reduced: the option to enter non-randomised patients into the study may reduce the clinicians’ efforts in encouraging patient consent.
- All patients could refuse randomisation: if all patients refuse randomisation the design is reduced to an observational study.
- Not yet widely used: it may be regarded with scepticism until more studies are available.