Beginning in 1977, numerous studies have reported an association betwe
en high levels of maternal serum alpha-fetoprotein, measured in early
pregnancy, and low birthweight and perinatal death. It has been sugges
ted, however, that these findings may be explained by bias. We reviewe
d 21 studies and found that they were unanimous in reporting this asso
ciation and that the association cannot be explained by any of the pro
posed biases. Comparing women whose alpha-fetoprotein levels were 2.5
times the median or more with all other women, the highest relative ri
sk of fetal death was 21.0, with a 95% confidence interval of 8.4-52.2
, and the lowest relative risk of fetal death was 4.4, with a 95% conf
idence interval of 2.8-7.0. For low birthweight (<2,500 grams), the hi
ghest relative risk was 6.4 (95% confidence interval = 3.1-13.1) and t
he lowest relative risk was 1.9 (95% confidence interval = 1.5-2.5).