Adjustment of fetal ponderal growth curves according to a mixture model

In the distribution of birth weight several authors have observed an excess of babies large for early gestational age (GA), suggesting that pregnancy length might be underestimated by the reported LMP. In 102,884 single preterm newborns (18-36 weeks), selected from the babies born in 1990-94 in Italy (ISTAT), we observe a similar pattern, which in some cases shows a clear bimodality. This suggests that the observations in GA classes less than 35 weeks might be not really homogeneous. In order to define reliable standards for fetal ponderal growth, we have modelled the data by a two normal component mixture distribution through maximum likelihood fitting techniques, and have partitioned them according to their posterior probabilities of component membership. For each GA, the "correct" major distribution has been identified and separated from the secondary spurious component. Such secondary component accounts for ~6% of the total distribution at the early GAs, increases up to ~21% at the 30th week, and then vanishes by the 35th week. At 30-32 weeks this distribution of large-for-GA babies has an estimated mean of 3030g and a standard deviation of 350g: this result is consistent with about six weeks underestimation of pregnancy duration. The growth pattern by vitality has been described by means of centile values: the whole pattern in stillborn babies, from 26 to 42 GA weeks, is shifted toward the low values, with the 50th centiles corresponding to the 25th centiles of the liveborns.