Efficiency and bias in studies of early pregnancy loss.

Abstract

Recent advances in laboratory techniques for measuring the pregnancy hormone human chorionic gonadotropin have allowed detection of occult very early pregnancy losses based on assaying urine. We used data from the Early Pregnancy Study carried out in North Carolina to develop a simplified collection protocol for reducing costs and improving compliance in similar studies conducted in less-selected populations of women. Collection of specimens only during menses would have caused some decrease in sensitivity but would nevertheless have allowed detection of at least 75% of the losses detected in the North Carolina study. Reduced-collection protocols would also result in some loss in specificity. Simulations suggest that moderate losses in sensitivity and specificity, as would result from reduced-collection protocols, produce only moderate loss of power for detecting associations between risk factors and early pregnancy loss. A "fertility bias" exists, however, which can produce seriously misleading results when specificity is less than perfect. We therefore recommend use of a baseline group of sterile women, so that criteria for "pregnancy" can achieve specificity close to 1.0.