Influence of prior assignment on refusal rates in a trial of supplemental oxygen for retinopathy of prematurity.

Investigators aware of prior treatment assignments may deliberately or subliminally influence rates of subsequent consent in randomised trials, thus introducing the possibility of bias. As a trial of Supplemental Therapeutic Oxygen for Prethreshold Retinopathy of Prematurity (STOP-ROP) was not masked to health professionals other than ophthalmologists, this possibility was evaluated by comparing the incidence of refusal to participate in relation to the previous assignment. Over 5 years, the STOP-ROP trial enrolled 649 infants, while 394 eligible subjects refused. Enrolled infants were assigned to supplemental oxygen or control groups at 30 centres including 71 hospitals, and stratified into two severity strata of their retinopathy of prematurity. We studied whether the assignment of the previously enrolled infant to either the supplemental arm or the control arm of the study affected the incidence of refusal to participate in the study by the next eligible infant. We also evaluated the possibility that refusal based on the preceding assignment could have influenced either the baseline balance of eye severity and pulmonary status between the two arms of the study or the final outcome of the randomised trial. There was a significantly higher incidence of refusal if the prior enrollee had been assigned to the supplemental oxygen group (44%) as compared with the control group (34%, P = 0.004). This effect from the prior assignment was larger in the later half of the study. Irrespective of the prior enrollee's assignment, refusal was more likely for more mature infants and those with milder baseline disease. Selection bias did not influence the ophthalmic or pulmonary findings at baseline, or the study outcomes. In incompletely masked studies, monitoring the distribution of refusers in relation to the previous assignment can reveal selection bias. Despite evidence of unequal refusal rates, the results of the STOP-ROP trial were not affected.