Anticonvulsant teratogenesis 5: observer bias in a cohort study.

Abstract

The most common physical effects of prenatal exposure to the anticonvulsant drugs phenytoin and phenobarbital are midface and digit hypoplasia. These features are subjective in nature, which makes their detection vulnerable to any bias the examiner has, such as an increased expectation that the infant being examined has been exposed prenatally to an anticonvulsant drug. The purpose of this analysis was to determine whether or not there was evidence of observer bias in the detection of presumed anticonvulsant drug-related physical features. The presence or absence of three groups of features were compared: 1) midface hypoplasia, such as anteverted nostrils; 2) digit hypoplasia, e.g., tapered fingers, and 3) other features, e.g., prominent occiput, not known to be affected by exposure to anticonvulsants. These infants were examined by one masked physician examiner, using a study protocol, in three time periods: in two similar time periods, the masked examiner knew that one in four (25%) of the infants to be examined had been exposed to an anticonvulsant drugs; in the third time period, the chance was 50%. The frequencies of the subjective features of midface (odds ratio [OR] = 2.6) and digit hypoplasia (OR = 3.9), but not the other features, were significantly more common in the time periods when the masked examiner knew that there was a 50% chance of exposure to anticonvulsant drugs, compared to the time periods when there was a 25% chance. The findings suggest the effect of observer bias. In addition, there was evidence of context bias, meaning that a study that focuses on determining the presence of specific features will identify them more often.