Problems of sample size and selection in analyses of pregnancy outcomes

Abstract

Hingson et al.s article Maternal cigarette smoking psychoactive substance use and infant Apgar reflects the dual problem of sample size and sample selection that plague analyses of pregnancy outcomes. It is difficult with small samples to confirm that a given maternal characteristic is a risk factor and equally difficult to prove that it is not. Additionally samples of mixed ethnic origin and samples that include many younger women further complicate analyses. The sample number of Hingson et al. is 1690 and is indicative of careful planning and considerable hard work yet with only 9% of low 1 minute Apgar scores a sample of that size might fail to show the effects of maternal smoking (arranged by 4 levels of smoking). A table demonstrates this with predicted results assuming no smoking effect and results similar to those obtained by these investigators. Since the differences would involve from 1-12 neonates with low Apgar scores in any 1 cell a smoking effect might well be missed. Their sample mix (59% black 22% Hispanic and 22% white) and the proportion of mothers under age 20 add to the problem. Since black mothers tend to weigh more for their age than white mothers yet produce smaller babies for the same weight gain during pregnancy the data should not be pooled without adjustment. Since younger mothers have more low weight babies and since their babies may also differ in the distribution of Apgar scores the age structure of the Boston sample may obscure a moderate smoking effect. There is also an age effect in regard to smoking affecting both prepregnancy weight (PPW) and gain in weight during pregnancy. With increasing maternal age the weight disparity between smokers and nonsmokers increases and so does the difference in weight between their neonates in both blacks and whites. Since the Boston sample is heavily weighted with young mothers (who tend to smoke less and who show a smaller smoking effect) these results should not be generalized to a rejection of a smoking effect in all mothers. Low Apgar scores which are in the minority do not reflect maternal smoking as dramatically as do low birth weights. These investigators disagree as to whether low Apgar scores are more frequent in the progeny of mothers who smoke. Given the problems of sample size age distribution and ethnicity it is questionable whether the Boston sample truly rejects what the far larger Collaborative Perinatal Project of the National Institute of Neurological and Communicative Disorders sample clearly shows.