Table 1

Appraisal of design options for exploratory trial in phase II
	Pros	Cons

Experimental designs
Traditional randomised controlled trial	Gold standard way to understand a difference between intervention and control	Concerns regarding recruitment, patients/staff may not be willing to take part if some patients do not get intervention, some staff had ethical concerns
Cluster randomisation	Reduce problem of disappointment of no service and contamination	Need extremely large sample and number of clusters, analysis required at level of cluster
Patient preference randomisation	Makes explicit problem of patients who have strong preference for one type of service	Difficult for patients to have a preference when they know little about service, large sample size needed, potential for staff or others to advise patients to have a particular preference
Delayed intervention randomised trial	All patients will eventually receive service, uses a gold standard methodology, it is common in this condition for patients to wait 3 months for appointments, longer survival means patients likely to actually receive service	Some staff not happy for patients to wait 3 months, effect of service must be apparent before 3 months (i.e. before control group receive intervention)
Quasi-experimental designs
Geographical comparison	No problems of randomisation, potential to increase sample size by study in an area where no service	Biases involved in variations in service provision between areas
Historical controls	No problems of randomisation	Biases in data collection and potentially in sample selection
Matched controls	No problems of randomisation	Biases in patient selection, difficulty of matching
Observational study	No problems of randomisation	No comparison group, only comparison with how patients were at referral, problems of regression to the mean, interviews and inclusion in study may have effect in itself.

Higginson et al. BMC Palliative Care 2006 5:7 doi:10.1186/1472-684X-5-7