Yellow fever vaccination and risk of spontaneous abortion

Sirs, In a case-control study of pregnant women in Brazil, Nishioka et al. (1998) found a more than twofold increased risk of abortion after yellow fever (YF) vaccination. Even though the association was not statistically significant, an increased risk of this magnitude would have serious consequences for YF vaccination policies, both during epidemics and for travellers to endemic areas. However, their finding could well be explained by bias and uncontrolled confounding. The authors state that recall bias was unlikely as the subjects were blinded for the research question. This is surprising, as pregnant women were explicitly excluded from the vaccination campaign for safey reasons. A woman experiencing spontaneous abortion can be expected to recall any event branded as potentially harmful, whereas this does not necessarily apply to the women in the control group with uncomplicated pregnancies, particularly not since they were unaware of the hypothesis under study and were interviewed up to several months later than the cases. There is thus considerable potential for overestimation of the risk posed by YF infection. The reported risk was adjusted for confounding by logistic regression analysis. The model did, however, not adjust for differences in gestational age at the time of the vaccination campaign, although this explained a considerable part of the observed association in the stratified (bivariate) analysis (crude OR for YF vaccination 2.49, adjusted OR 1.76). This may be because the effect of gestational age was compensated by the confounding effect of other variables in the multivariate analysis. Alternatively, it may be due to the use of an automated stepwise procedure, in which the decision to select variables for the model is based on the p-value for excluding a variable from the model rather than on the strength of the confounding effect (Hosmer & Lemeshow 1989). The fact that the model presented only includes variables that show significant association with the risk of abortion, but have virtually no confounding effect in the stratified analysis, strongly suggests that this happened, and that confounding by gestational age remained uncontrolled for. This would again result in overestimation of the observed risk. One may wonder why gestational age at the time of the vaccination campaign had a confounding effect. Apart from the obvious association with the vaccination status, it implies a direct association between the calendar time of conception and the risk of spontaneous abortion, suggesting that part of the abortions was due to exposure with a time-dependent, e.g. seasonal or epidemic occurrence. As the vaccination campaign was held during an outbreak of dengue, and a recent history of dengue fever – although having no confounding effect – increased the risk of abortion, whereas the exposure to insecticides used for mosquito control and a recent history of exenthematous disease did not, recent or atypical dengue virus infection would seem a likely candidate. Frank G. J. Cobelens, Division of Infectious Diseases, Tropical Medicine and AIDS, Academic Medical Centre, Amsterdam