The limits of small-for-gestational-age as a high-risk category

Abstract

SGA (small for gestational age) is widely used to identify high-risk infants, although with inconsistent definitions. Cut points range from 2.5th to 10th percentile of birthweight-for-gestational age. We used receiver operator characteristic curves (ROC) to identify the cut point with maximum sensitivity and specificity (Youden Index), and the area under the curve (AUC), which assesses overall discriminating power. Analysis was conducted on 3,836,034 US births (2015) and 292,279 Norwegian births (2010–14). Birthweight percentiles were calculated using exact birthweights at each week of gestational age, and then summarized across gestational ages. We also conducted a companion analysis of gestational age itself to consider the predictive power of gestational week of birth. Outcomes were neonatal mortality and cerebral palsy, both strongly associated with birthweight. Birthweight percentiles performed poorly in all analyses. The AUC for birthweight percentiles as a discriminator of neonatal mortality was 60% (where 50% is no better than a coin-toss). At such low discrimination, the Youden Index provides no useful SGA cut point. Results in Norway were virtually identical, with an AUC of 58%. The AUC with cerebral palsy as the outcome was even lower, at 54%. In contrast, gestational age had an AUC of 85% as a predictor of neonatal mortality, with < 37 weeks as the optimum cut point. SGA provides surprisingly poor identification of at-risk infants, while gestational age performs well. Similar results in two countries that differ in mean birthweight, percent preterm, and neonatal mortality suggest robustness across populations.