Re: Comparison of perinatal outcomes in small-for-gestational-age infants classified by population-based versus customised birth weight standards

Dear Editor,

In their recent work, Cha et al.[1] present a comparison of perinatal outcomes in newborns classified as either SGA or non-SGA using a locally developed customised growth standard. We are writing to point out some concerns regarding the validity of this standard. Their prediction formula for birth weight is a natural log function (‘Ln (BWT)’), which indicates that the birth weights were log transformed before undergoing regression analysis. Log transformation of a variable is justified if the data is heavily skewed. However, in our experience with Western populations, the skew is only modest, and birth weights follow fairly closely a normal distribution.[2] The authors did not present their rationale for using log transformation. On entering their population averages for the non-SGA infants, primiparas and female infants, the predicted log birth weight according to their formula is:

• 

Hence, the predicted birth weight would be EXP (1.36623) or 3.92 kg. This weight is much higher than that expected from table 1, that is, 3.3 ± 0.5 kg, and suggests an error in the formula. Another concern is that the derivation of the percentiles could be distorted by exponentiation. The authors wrote that the lower and upper qth percentiles were computed as follows:

• Lower qth percentile = EXP (predicted birthweight – SD × Z_1−q)
• Upper qth percentile = EXP (predicted birthweight +SD × Z_1−q)

For a median of 3.92 kg, these yield a 10th percentile of 3.25 kg and a 90th percentile of 4.73 kg – hence, the distance from the median to the 90th percentile is almost double the distance from the median to the 10th percentile. This implies a far greater degree of skewing than would be expected in a birth weight population, and thus, the percentile calculations from this model would be very inaccurate. We suggest that a repeat of their analysis with correction of the methodology may reach completely different results.

The authors derived the fetal growth function from birth weights, and the resulting growth curve shows a marked deceleration pattern at term (fig.1). This phenomenon is a typical artefact of pregnancies that have been dated by menstrual dates; growth at term in ultrasound-dated populations is almost linear.[3]

Furthermore, birth weight data from preterm deliveries should not be used in deriving a growth standard because iatrogenic as well as spontaneous preterm delivery is often associated with pathology including growth restriction and pre-eclampsia.[4] The customised growth charts available from the West Midlands Perinatal Institute use intra-uterine growth curves derived from ultrasound data of healthy pregnancies, and these are forced through the expected birth weights using the proportionality principle.[5]

The true value of customised centiles becomes apparent when subgroups of the population, for whom the adjustment is made, are assessed in terms of outcome. This was illustrated in a article not included in the authors review, where a similar fetal weight standard was used to compare a population based and an individually adjusted standard. That article furthermore shows the need to look at population attributable risk as well as odds ratio when comparing the performance of two standards.[6]