Recurrent miscarriage (REMIS) study: how should data from women who do not become pregnant be handled?

The Recurrent Miscarriage (REMIS) study is a double-blind, multicenter, randomized clinical trial designed to evaluate the efficacy of immunization with paternal leukocytes in the prevention of miscarriages in women who have had three or more unexplained pregnancy losses. Women entering the study are immunized with their husband's leukocytes or with a saline control before they become pregnant. After becoming pregnant, they receive weekly ultrasound examinations and psychological support during the first trimester and are followed until a successful delivery or a miscarriage occurs. The primary analysis for the study will be an intent-to-treat analysis in which we shall compare the proportion of successes in the two groups, defining a "success" as a pregnancy achieved within 12 months of randomization that results in a viable offspring. We shall count both miscarriages and nonpregnancies as failures, owing to the possibility of very early losses prior to the detection of pregnancy. In a secondary analysis, we shall exclude women who do not become pregnant within the alloted 12 month period. We compared the test size and power of these two approaches under various configurations for the true rates of nonpregnancy, miscarriage, and delivery in the two groups. Although the analysis excluding nonpregnant women achieves greater power for alternatives in which pregnancy rates are equal and live birth rates higher in the treated group, the rejection rate is not adequately controlled when pregnancy rates differ but live birth rates are unaffected by treatment. It can also lead to a reduction in power if the treatment prevents early as well as later losses. We conclude that the intent-to-treat analysis should remain the primary analysis for the trial.