Aims/IntroductionA debate exists on the relation of adverse pregnancy outcomes with glycemic levels in early pregnancy. We aimed to investigate the association of maternal characteristics including post‐load glucose and first‐trimester HbA1c test results with adverse pregnancy outcomes in women without gestational diabetes mellitus.Materials and MethodsA dataset (January 2011 and September 2017) from a hospital prenatal clinic was explored to find the important predictors of adverse pregnancy outcomes using maternal characteristics and glucose assessments in mothers without gestational diabetes. We used two machine learning algorithms to capture nonlinearity in selecting important maternal characteristics and developed predictive models for each outcome. In total, 1,618 pregnant women were included in the analytic dataset with a mean (SD) age of 26.8 (3.5) years and gravida of 1.7 (0.9).ResultsImportant associations were detected between maternal features and primary cesarean section, fetal distress, premature rupture of membranes, macrosomia, small or large for gestational age, APGAR <7 at 1 or 5 min, hyperbilirubinemia, and poly‐ or oligo‐hydramnios. Overall, the predictive models showed good performance and large areas under the curves (0.732, 0.765, 0.646, 0.651, 0.730, 0.646, 0.684, 0.716, and 0.678, respectively). Specifically, they had high positive likelihood ratios.ConclusionsHigh glucose levels were associated with adverse pregnancy outcomes. Post‐load glucose was the most reliable test for predicting the outcomes. Overall, fasting blood sugar was of more predictive value than HbA1c. Our study showed that further research should account for the nonlinearity and interactions inherent in the data.