Prediction and prevention of stillbirth: dream or reality

Machine Learning for Predicting Stillbirth: A Systematic Review

Stillbirth is a major global issue, with over 5 million cases each year. The multifactorial nature of stillbirth makes it difficult to predict. Artificial intelligence (AI) and machine learning (ML) have the potential to enhance clinical decision-making and enable precise assessments. This study reviewed the literature on predictive ML models for stillbirth highlighting input characteristics, performance metrics, and validation. The PubMed, Cochrane, and Web of Science databases were searched for studies using AI to develop predictive models for stillbirth. Findings were analyzed qualitatively using narrative synthesis and graphics. Risk of bias and the applicability of the studies were assessed using PROBAST. Model design and performance were discussed. Eight studies involving 14,840,654 women with gestational ages ranging from 20 weeks to full term were included in the qualitative analysis. Most studies used neural networks, random forests, and logistic regression algorithms. The number of predictive features varied from 14 to 53. Only 50% of studies validated the models. Cross-validation was commonly employed, and only 25% of studies performed external validation. All studies reported area under the curve as a performance metric (range 0.54-0.9), and five studies reported sensitivity (range, 60- 90%) and specificity (range, 64 - 93.3%). A stacked ensemble model that analyzed 53 features performed better than other models (AUC = 0.9; sensitivity and specificity > 85%). Available ML models can attain a considerable degree of accuracy for prediction of stillbirth; however, these models require further development before they can be applied in a clinical setting.

Prenatal prediction of adverse outcome using different charts and definitions of fetal growth restriction

OBJECTIVE

Antenatal growth assessment using ultrasound aims to identify small fetuses that are at higher risk of perinatal morbidity and mortality. This study explored whether the association between suboptimal fetal growth and adverse perinatal outcome varies with different definitions of fetal growth restriction (FGR) and different weight charts/standards.

METHODS

This was a retrospective cohort study of 17 261 singleton non-anomalous pregnancies at ≥ 24 + 0 weeks' gestation that underwent routine ultrasound at a tertiary referral hospital. Estimated fetal weight (EFW) and Doppler indices were converted into percentiles using a reference standard (INTERGROWTH-21st (IG-21)) and various reference charts (Hadlock, Fetal Medicine Foundation (FMF) and Swedish). Test characteristics were assessed using the consensus definition, Society for Maternal-Fetal Medicine (SMFM) definition and Swedish criteria for FGR. Adverse perinatal outcome was defined as perinatal death, admission to the neonatal intensive care unit at term, 5-min Apgar score < 7 and therapeutic cooling for neonatal encephalopathy. The association between FGR according to each definition and adverse perinatal outcome was compared. Multivariate logistic regression analysis was used to test the strength of association between ultrasound parameters and adverse perinatal outcome. Ultrasound parameters were also tested for correlation.

RESULTS

IG-21, Hadlock and FMF fetal size references classified as growth-restricted 1.5%, 3.6% and 4.6% of fetuses, respectively, using the consensus definition and 2.9%, 8.8% and 10.6% of fetuses, respectively, using the SMFM definition. The sensitivity of the definition/chart combinations for adverse perinatal outcome varied from 4.4% (consensus definition with IG-21 charts) to 13.2% (SMFM definition with FMF charts). Specificity varied from 89.4% (SMFM definition with FMF charts) to 98.6% (consensus definition with IG-21 charts). The consensus definition and Swedish criteria showed the highest specificity, positive predictive value and positive likelihood ratio in detecting adverse outcome, irrespective of the reference chart/standard used. Conversely, the SMFM definition had the highest sensitivity across all investigated growth charts. Low EFW, abnormal mean uterine artery pulsatility index (UtA-PI) and abnormal cerebroplacental ratio were significantly associated with adverse perinatal outcome and there was a positive correlation between the covariates. Multivariate logistic regression showed that UtA-PI > 95th percentile and EFW < 5th percentile were the only parameters consistently associated with adverse outcome, irrespective of the definitions or fetal growth chart/standard used.

CONCLUSIONS

The apparent prevalence of FGR varies according to the definition and fetal size reference chart/standard used. Irrespective of the method of classification, the sensitivity for the identification of adverse perinatal outcome remains low. EFW, UtA-PI and fetal Doppler parameters are significant predictors of adverse perinatal outcome. As these indices are correlated with one other, a prediction algorithm is advocated to overcome the limitations of using these parameters in isolation. © 2023 The Authors. Ultrasound in Obstetrics & Gynecology published by John Wiley & Sons Ltd on behalf of International Society of Ultrasound in Obstetrics and Gynecology.

Identifying risk of stillbirth using machine learning

BACKGROUND

Previous predictive models using logistic regression for stillbirth do not leverage the advanced and nuanced techniques involved in sophisticated machine learning methods, such as modeling nonlinear relationships between outcomes.

OBJECTIVE

This study aimed to create and refine machine learning models for predicting stillbirth using data available before viability (22-24 weeks) and throughout pregnancy, as well as demographic, medical, and prenatal visit data, including ultrasound and fetal genetics.

STUDY DESIGN

This is a secondary analysis of the Stillbirth Collaborative Research Network, which included data from pregnancies resulting in stillborn and live-born infants delivered at 59 hospitals in 5 diverse regions across the United States from 2006 to 2009. The primary aim was the creation of a model for predicting stillbirth using data available before viability. Secondary aims included refining models with variables available throughout pregnancy and determining variable importance.

RESULTS

Among 3000 live births and 982 stillbirths, 101 variables of interest were identified. Of the models incorporating data available before viability, the random forests model had 85.1% accuracy (area under the curve) and high sensitivity (88.6%), specificity (85.3%), positive predictive value (85.3%), and negative predictive value (84.8%). A random forests model using data collected throughout pregnancy resulted in accuracy of 85.0%; this model had 92.2% sensitivity, 77.9% specificity, 84.7% positive predictive value, and 88.3% negative predictive value. Important variables in the previability model included previous stillbirth, minority race, gestational age at the earliest prenatal visit and ultrasound, and second-trimester serum screening.

CONCLUSION

Applying advanced machine learning techniques to a comprehensive database of stillbirths and live births with unique and clinically relevant variables resulted in an algorithm that could accurately identify 85% of pregnancies that would result in stillbirth, before they reached viability. Once validated in representative databases reflective of the US birthing population and then prospectively, these models may provide effective risk stratification and clinical decision-making support to better identify and monitor those at risk of stillbirth.

Middle cerebral arterial flow redistribution is an indicator for intrauterine fetal compromise in late pregnancy in low-resource settings: A prospective cohort study

Objective

We aimed to determine the prevalence of abnormal umbilical artery (UA), uterine artery (UtA), middle cerebral artery (MCA) and cerebroplacental ratio (CPR) Doppler, and their relationship with adverse perinatal outcomes in women undergoing routine antenatal care in the third trimester.

Design

Prospective cohort.

Setting

Kagadi Hospital, Uganda.

Population

Non‐anomalous singleton pregnancies.

Methods

Women underwent an early dating ultrasound and a third‐trimester Doppler scan between 32 and 40 weeks of gestation, from 2018 to 2020. We handled missing data using multiple imputation and analysed the data using descriptive methods and a binary logistic regression model.

Main outcome measures

Composite adverse perinatal outcome (CAPO), perinatal death and stillbirth.

Results

We included 995 women. The mean gestational age at Doppler scan was 36.9 weeks (SD 1.02 weeks) and 88.9% of the women gave birth in a health facility. About 4.4% and 5.6% of the UA pulsatility index (PI) and UtA PI were above the 95th percentile, whereas 16.4% and 10.4% of the MCA PI and CPR were below the fifth percentile, respectively. Low CPR was strongly associated with stillbirth (OR 4.82, 95% CI 1.09–21.30). CPR and MCA PI below the fifth percentile were independently associated with CAPO; the association with MCA PI was stronger in small‐for‐gestational‐age neonates (OR 3.75, 95% CI 1.18–11.88).

Conclusion

In late gestation, abnormal UA PI was rare. Fetuses with cerebral blood flow redistribution were at increased risk of stillbirth and perinatal complications. Further studies examining the predictive accuracy and effectiveness of antenatal Doppler ultrasound screening in reducing the risk of perinatal deaths in low‐ and middle‐income countries are warranted.

Tweetable abstract

Blood flow redistribution to the fetal brain is strongly associated with stillbirths in low‐resource settings.

Blood flow redistribution to the fetal brain is strongly associated with stillbirths in low‐resource settings.

This article includes Author Insights, a video abstract available at https://vimeo.com/bjogabstracts/authorinsights17115.

Predicting singleton antepartum stillbirth by the demographic Fetal Medicine Foundation Risk Calculator-A retrospective case-control study

OBJECTIVE

To assess the risk of singleton intrauterine fetal death (IUFD) in women by the demographic setting of the online Fetal Medicine Foundation (FMF) Stillbirth Risk Calculator.

METHODS

Retrospective single-centre case-control study involving 144 women having suffered IUFD and 247 women after delivery of a live-born singleton. Nonparametric receiver operating characteristics (ROC) analyses were performed to predict the prognostic power of the FMF Stillbirth risk score and to generate a cut-off value to discriminate best between the event of IUFD versus live birth.

RESULTS

Women in the IUFD cohort born a significantly higher overall risk with a median FMF risk score of 0.45% (IQR 0.23-0.99) compared to controls [0.23% (IQR 0.21-0.29); p<0.001]. Demographic factors contributing to an increased risk of IUFD in our cohort were maternal obesity (p = 0.002), smoking (p<0.001), chronic hypertension (p = 0.015), antiphospholipid syndrome (p = 0.017), type 2 diabetes (p<0.001), and insulin requirement (p<0.001). ROC analyses showed an area under the curve (AUC) of 0.72 (95% CI 0.67-0.78; p<0.001) for predicting overall IUFD and an AUC of 0.72 (95% CI 0.64-0.80; p<0.001), respectively, for predicting IUFD excluding congenital malformations. The FMF risk score at a cut-off of 0.34% (OR 6.22; 95% CI 3.91-9.89; p<0.001) yielded an 82% specificity and 58% sensitivity in predicting IUFD with a positive and negative predictive value of 0.94% and 99.84%, respectively.

CONCLUSION

The FMF Stillbirth Risk Calculator based upon maternal demographic and obstetric characteristics only may help identify women at low risk of antepartum stillbirth.