RESUMO
OBJECTIVE: to assess the accuracy of paired estimated fetal weights (EFWs) to predict three levels of twin birth weight discordance (>15%, >20% and >25%). METHOD: a cohort of twin pairs underwent ultrasound examinations within 2 weeks from birth. We calculated the frequency of under- and overestimation of the actual birth weight (< or >10%, respectively) in the entire cohort as well in the subset of mono- and dichorionic pairs. RESULTS: discordance was largely underestimated (observed 10.4 ± 0.8% compared to actual 19.2 ± 1.1%, P=0.001) because the larger twin was more frequently underestimated [30.6 vs. 17.7%, odds ratio (OR) 2.0, 95% confidence interval (CI) 1.1, 3.9 in the entire cohort, and 34.2 vs. 13.1%, OR 3.4, 95% CI 1.4, 8.4 in the dichorionic pairs]. Overall, the specificity for detecting the three levels of discordance was adequate (91.5-94.2%) but the sensitivity was poor (11.1-17.8%) and tended to decrease with increasing discordance level. CONCLUSION: the poor ability of paired EFWs to diagnose birth weight discordance results from underestimation of the larger twin.