Prediction of low birth weight at term in low resource setting of Gulu City, Uganda : a prospective cohort study

Abstract

Introduction: despite the widespread poverty in Northern Uganda resulting in undernutrition, not all mothers deliver low birth weight babies. Therefore, we developed and validated the risk prediction models for low birth weight at term in Northern Uganda from a prospective cohort study. Methods: one thousand mothers were recruited from 16 - 24 weeks of gestation and followed up until delivery. Six hundred and eighty-seven mothers delivered at term. The others were either lost to follow-up or delivered preterm. Used proportions to compute incidence of low birth weight at term, build models for prediction of low birth weight at term in RStudio. Since there were few low birth weight at term, were generated synthetic data using ROSE-package in RStudio by over-sampling low birth weights and undersampling normal birth weights, and evaluated the model performance against the synthetic data using K (10) - fold cross-validation. Results: mean age was 26.3 years with an average parity of 1.5. Their mean body mass index was 24.7 and 7.1% (49 of 687) had lateral placenta. The incidence of low birth weight was 5.7% (39 of 687). Predictors of low birth weight were gravidity, level of education, serum alanine aminotransferase (ALT), serum gamma-glutamyl transferase (GGT), lymphocyte count, placental location, and enddiastolic notch in the uterine arteries. This predicted low birth weight at term by 81.9% area under the curve (AUC), 76.1% accuracy, 72.9% specificity, and 79.1% sensitivity. Conclusion: a combination of gravidity, level of education, serum ALT, serum GGT, lymphocyte count, placental location, and end-diastolic notch in the uterine arteries can be used for screening for low birth weight in prenatal clinics for screening low birth weight at term.