Low birth weight risk prediction model: a prognostic study in the Birhan field site in Ethiopia

Pregnancy-related complications remain a global challenge, with low- and middle-income countries bearing the highest burden. Predicting the absolute risk of adverse birth outcomes will facilitate the delivery of early preventative and therapeutic interventions. We aimed to developed and internally validate a risk prediction model for low birth weight (LBW) in Ethiopia.

We conducted a prognostic study using a prospective maternal and child health cohort in the Birhan field site, Amhara region, Ethiopia. We included all pregnant women with a live birth who had enrolled in the Birhan maternal and child health cohort between 2018 and 2021. We analysed data from 2076 live births. We first applied a multivariable logistic regression model to select variables for the risk prediction model, and used a classification and regression tree to select the most potent predictors. We presented the model with a nomogram suited to clinical use. We also calculated measures of risk prediction model accuracy, discrimination, and calibration, and used bootstrapping for internal validation. We assessed the clinical utility of the model using the decision curve analysis.

The incidence of LBW was 9.44% (95% confidence interval (CI) = 8.2, 10.8). We identified seven predictors: previous maternal complication, previous foetal complication, pregnancy induced hypertension, average maternal body weight, average diastolic blood pressure, preterm delivery, and gravidity. The prediction model had an area under the curve (AUC) of 0.67 (95% CI = 0.63, 0.72). After internal validation, the corrected discrimination AUC value was 0.64 (95% CI = 0.59, 0.68). The classification and regression tree identified four predictors: preterm, gravidity, average maternal body weight, and previous foetal complication, with a discriminative ability of 0.65 (95% CI = 0.61, 0.69). The decision curve analysis showed that the prediction model had high net benefit at different threshold probabilities in both the nomogram and the classification and regression tree.

We developed a modestly accurate risk prediction model to identify pregnancies leading to LBW babies that could aid in early decision-making for prevention. This model is a crucial first step towards developing a clinical decision support tool to prompt early referral of women who are at high risk of having a LBW infant.