Predictive modelling of stillbirth risk factors in Ghana: a retrospective study

Abstract

Background: Stillbirth remains a devastating and unacceptably common occurrence. Quantifying stillbirth risk can support healthcare decision-making. While advancements in prenatal care have led to a global decline in stillbirth rates, millions of babies are stillborn each year. This public health burden is particularly heavy in low-and middle-income countries, where access to quality healthcare may be limited. Objective: This study investigated the predictors and a predictive model of stillbirth at the Korle-Bu Teaching Hospital. Methods: Secondary data were extracted from the obstetric department database at Korle-Bu Teaching Hospital and subjected to predictive analysis using R statistical software. The participants in this study were all births in the obstetrics department from 2015 to 2017, excluding terminations. After all exclusions, the final study population consisted of 9253 live births and 293stillbirths. The logistic model focused on identifying major predictors of stillbirth, with variables including maternal age, hypertension, malaria during pregnancy, placental abruption, and gestation at delivery. The model’s overall goodness-of-fit was tested, with statistical significance set at 0.05. Results: Out of 9,546 births, 293 (3.1%) were stillbirths, including both fresh and macerated types. Predictors of stillbirth included maternal age ≥35 years (OR 1.34, 95% CI 1.02–1.77, p=0.035), abruptio placentae (OR 5.83, 95% CI 3.41–9.97, p<0.001), malaria in pregnancy (OR 2.59, 95% CI 1.08–6.24, p=0.034), and lower gestational age at delivery (OR 0.76 per week, 95% CI 0.73–0.79, p<0.001). The model showed moderate discrimination (AUC 0.71, 95% CI 0.67–0.75) and excellent calibration (slope 1.00, intercept ~0, Brier score 0.028). At the optimal probability cutoff, sensitivity was 61.4%, specificity 74.6%, and negative predictive value 98.4%. Conclusion: The predictive model successfully identified key risk factors for stillbirth, offering insights into targeted interventions for high-risk groups. These findings contribute to the standardised approach to stillbirth analysis and reporting, underscoring the importance of improved prenatal care and screening practices in reducing stillbirth incidence. This study fills a crucial research gap by providing a validated predictive model for stillbirth in a resource-limited setting.