When Is “Interaction” Really Interaction?
To the Editor:
Shaw and colleagues1 suggest that periconceptional vitamin use modifies the association between maternal fever and development of conotruncal defects among offspring. Using women with no fever and vitamin use as the referent category, relative risks were 1.8 among women with fever and maternal vitamin use, 1.7 among women with no fever and no maternal vitamin use, and 2.4 among those with maternal fever and no maternal vitamin use. Based on these data, however, there is no evidence of interaction under an additive model (ie, 1.7 + 1.8 − 1.0 = 2.5). That is, the combined effect of fever and no vitamin use was nearly identical to what one would predict based on the observed effects of each risk factor alone. Maternal fever and no vitamin use appear to be independent risk factors but there is no evidence that vitamin use modifies the impact of fever on the development of heart defects.
Shaw and colleagues1 suggest that their findings extend an earlier report by Botto and colleagues2 who found that the risk of heart defects associated with maternal fever was 2.3 among infants whose mothers did not use vitamins and 1.1 among those whose mothers did use vitamins. Conducting a similar stratified approach using the data presented by Shaw and colleagues,1 the risk of heart defects associated with maternal fever was 1.4 (2.4/1.7) among women who did not use vitamins and 1.8 (1.8/1.0) in women among women who did use vi tamins. A stratified approach assumes a multiplicative model. A stratified approach, using the data presented by Shaw and colleagues,1 actually suggests that fever might be somewhat more strongly associated with heart defects in offspring of women who used vitamins than those who did not.
The study by Shaw and colleagues1 illustrates the methodologic complexities of assessing statistical interaction.3 It also shows the importance of reporting the statistical model used to evaluate combined effects, as well as the magnitude and direction of departure from the model, to avoid travel down wrong etiologic roads.
University of Massachusetts
1.Shaw GM, Nelson V, Carmichael SL, et al. Maternal periconceptional vitamins: interactions with selected factors and congenital anomalies. Epidemiology
2.Botto LD, Lynberg MC, Erickson JD. Congenital heart defects in relation to maternal multivitamin use. Am J Epidemiol
© 2003 Lippincott Williams & Wilkins, Inc.
3.Rothman KJ. Epidemiology: An Introduction
. New York: Oxford University Press; 2002.