Pseudo cluster randomization performed well when used in practice

In the Dutch EASYcare Study, pseudo cluster randomization (PCR) randomized clinicians in two groups (H and L) with a high or a low proportion of the patients of the clinician randomized to intervention or to control arm accordingly. We used PCR because cluster randomization risked selection bias and individual randomization risked contamination. We evaluated the performance of PCR.

Clinicians were asked about treatment arm preferences, recruitment behavior, possible contaminating behavior, and what they thought the allocation ratio was. We compared patients' baseline characteristics and clinicians' recruitment rates.

The groups were comparable at baseline. Clinicians favored the intervention arm (Visual Analogue Scale 14.5 [SD 15.6]; 0-100; 0=strongly favoring intervention arm, 100=strongly favoring usual care arm) and 58% said they would have recruited fewer patients had every participant been allocated to the control group. Sixty five percent of clinicians used intervention elements in control patients. Sixty seven percent of clinicians estimated that a 50:50 allocation ratio was used.

The assumptions underlying PCR largely applied in this study. PCR performed satisfactorily without signs of unblinding or selection bias.