Up to two million babies are stillborn annually in late gestation (ζ28 weeks) worldwide. This is thought to be an underestimate of the true burden given variations in the definitions of stillbirth across settings and poor reporting in many countries [1,2]. While high-income countries have registered significant reductions in stillbirth rates in the past two decades, the narrative is different for sub-Saharan Africa: stillbirth rates have instead increased. For example, in Uganda, there was an 8.6% increment between 2000 and 2019 [2].

The causes of stillbirths are heterogeneous- congenital anomalies and placental dysfunction being key contributing factors [3,4]. Significant decreases in stillbirths will require a well-rounded solution to all known preventable causes. Globally, 7.4% of stillbirths are attributed to congenital abnormalities, though regional disparities exist [5]. In the United States, birth defects account for over 14% of stillbirths [3]. Anomalous fetuses have a significantly elevated stillbirth risk (55 per 1,000 compared to 4 per 1,000 for non-anomalous fetuses; (aOR 15.17, 95% CI: 11.03-20.86)) [6]. The largest proportion of stillbirths, caused by congenital anomalies, can be managed and prevented by early detection [7,8].

Among small non-anomalous fetuses, the common underlying mechanism leading to stillbirths is placental dysfunction attributed to various placental pathologies, maternal vascular malperfusion being the most common and understood lesion. The underlying pathophysiological mechanisms that lead to stillbirths in normal-sized babies are less well understood and a high proportion are classified as unexplained [4]. Existing evidence indicates that placental dysfunction may also have a vital role in this subgroup [9–11]. Related to this, biomarkers of placental function such as human growth hormone (hGH)/chronic somatomammotropin hormone (CSH) genes and placental proteins; soluble fms-like tyrosine kinase 1/placental growth factor (sFlt-1, PlGF) have shown promise [12–14]. For example, growth-restricted babies have low levels of hGH/CSH gene expression [12].

During pregnancy, the maternal cardiovascular system undergoes structural and functional modifications to provide adequate uteroplacental perfusion and nutrition to facilitate normal fetal growth and development [15]. Maternal cardiac adaptation is correlated with physiological and morphological adaptation in the placenta. For instance, uterine spiral artery remodeling physiologically reduces maternal systemic vascular resistance [10]. First-trimester studies have demonstrated that abnormal maternal hemodynamic profiles indicate increased risks of clinical manifestations of fetal growth restriction (FGR) and preeclampsia (PE). Increasing evidence also suggests that underlying maternal cardiac dysfunction is associated with placental dysfunction. Therefore, further exploration of maternal hemodynamics in all trimesters could provide insights into the mechanisms that lead to stillbirth, thus providing an opportunity to improve management outcomes [16–19].

Preventing stillbirth requires timely identification of risk factors to enable effective interventions. At present, no effective screening strategy to detect deprived fetuses at risk of stillbirth exists. Prognostic models to more precisely estimate a woman’s individual risk of stillbirth have been proposed as a potential solution, but there is a shortage of robust, clinically applicable models in low-resource settings [20]. Iatrogenic early birth is often initiated in fear of the worst, however without proper risk stratification resulting in a shift towards neonatal and infant morbidity and mortality. To date, potential markers of clinical importance to predict and prevent stillbirths in low- and middle-income countries (LMICs) have not been well studied.

We have undertaken several studies to address this gap: first, our review of available literature on the prognostic accuracy of Doppler ultrasound for adverse perinatal outcomes revealed a lack of evidence to guide how Doppler ultrasound should be used in LMICs [21]. The review informed a large prospective cohort study which showed that low cerebroplacental ratio (CPR) was strongly associated with stillbirth (OR 4.82, 95% CI 1.09–21.30). CPR and middle cerebral artery (MCA) PI <5^th percentile were independently associated with adverse perinatal outcomes [22,23]. A model combining maternal characteristics with MCA PI or CPR had AUCs of 0.78 (95% CI: 0.67 - 0.87) and 0.78 (95% CI: 0.65 - 0.87), respectively, for predicting perinatal death in the development set [24]. The bootstrap corrected AUC was 0.71 with a slope of 0.70 [24], only moderate and below clinically relevant threshold, but this was based on a small sample with low number of events.

Existing models have not been validated in low-resource settings and have low predictive performance [20,25,26]. Most include maternal and ultrasound features, whereas those including biochemical markers have the highest predictive performance [20,25,26]. Large scale multi-modal stillbirth prediction studies are required. Such complex models may benefit from machine learning modeling techniques given their ability to handle a large number of datasets and high-dimensional data, and complex interactions and none-linear relationships between variables [27].

Therefore, the iTECH project aims to use traditional statistical and machine learning methods to develop and validate prediction models for stillbirth and pathologies that lead the stillbirth (PE and FGR) based on widely accessible and cost-effective markers. Previously, we found that spousal involvement may promote acceptance and use of advanced technologies like Doppler ultrasound, but healthcare providers required more capacity building on how to use it for the benefit of mothers [28]. Stakeholder engagements will continue as an important component of the iTECH project.