Interpretable machine learning to predict adverse perinatal outcomes: examining marginal predictive value of risk factors during pregnancy.

BACKGROUND: The timely identification of nulliparas at high risk of adverse fetal and neonatal outcomes during pregnancy is crucial for initiating clinical interventions to prevent perinatal complications. Although machine learning methods have been applied to predict preterm birth and other pregnancy complications, many models do not provide explanations of their predictions, limiting the clinical use of the model. OBJECTIVE: This study aimed to develop interpretable prediction models for a composite adverse perinatal outcome (stillbirth, neonatal death, estimated Combined Apgar score of <10, or preterm birth) at different points in time during the pregnancy and to evaluate the marginal predictive value of individual predictors in the context of a machine learning model. STUDY DESIGN: This was a secondary analysis of the Nulliparous Pregnancy Outcomes Study: Monitoring Mothers-to-be data, a prospective cohort study in which 10,038 nulliparous pregnant individuals with singleton pregnancies were enrolled. Here, interpretable prediction models were developed using L1-regularized logistic regression for adverse perinatal outcomes using data available at 3 study visits during the pregnancy (visit 1: 6 0/7 to 13 6/7 weeks of gestation; visit 2: 16 0/7 to 21 6/7 weeks of gestation; visit 3: 22 0/7 to 29 6/7 weeks of gestation). We identified the important predictors for each model using SHapley Additive exPlanations, a model-agnostic method of computing explanations of model predictions, and evaluated the marginal predictive value of each predictor using the DeLong test. RESULTS: Our interpretable machine learning model had an area under the receiver operating characteristic curves of 0.617 (95% confidence interval, 0.595-0.639; all predictor variables at visit 1), 0.652 (95% confidence interval, 0.631-0.673; all predictor variables at visit 2), and 0.673 (95% confidence interval, 0.651-0.694; all predictor variables at visit 3). For all visits, the placental biomarker inhibin A was a valuable predictor, as including inhibin A resulted in better performance in predicting adverse perinatal outcomes (P<.001, all visits). At visit 1, endoglin was also a valuable predictor (P<.001). At visit 2, free beta human chorionic gonadotropin (P=.001) and uterine artery pulsatility index (P=.023) were also valuable predictors. At visit 3, cervical length was also a valuable predictor (P<.001). CONCLUSION: Despite various advances in predictive modeling in obstetrics, the accurate prediction of adverse perinatal outcomes remains difficult. Interpretable machine learning can help clinicians understand how predictions are made, but barriers exist to the widespread clinical adoption of machine learning models for adverse perinatal outcomes. A better understanding of the evolution of risk factors for adverse perinatal outcomes throughout pregnancy is necessary for the development of effective interventions.

Social Determinants of Health Needs and Perinatal Risk in Socially Vulnerable Pregnant Patients.

OBJECTIVES: To understand perinatal risks associated with social needs in pregnancy Methods. Multivariable log-binomial regression analyses adjusting for age, parity, and insurance were used to evaluate the relationship between any social need (e.g., housing, transportation, food, and intimate partner violence) and adverse perinatal outcomes (stillbirth, prematurity, maternal morbidity) in a cohort of English and Spanish-speaking patients who obtained prenatal care and birthed at our institution during a one-year period. RESULTS: Of 2,435 patients, 1,608 (66%) completed social needs screening at least once during prenatal care. The cohort was predominantly non-Hispanic Black (1,294, 80%) and publicly insured (1,395, 87%). Having one or more social need was associated with three-fold increased risk of stillbirth (aRR 3.35, 95%CI 1.31,8.6) and 14% reduction in postpartum care attendance (aRR 0.86, 95%CI 0.78-0.95) and was highest in individuals reporting transportation needs. CONCLUSIONS: Social needs during pregnancy were associated with increased risk of stillbirth.

Timely treatment of severe hypertension and risk of severe maternal morbidity at an urban hospital.

BACKGROUND: Hypertensive disorders of pregnancy have been identified as a leading contributor to severe maternal morbidity and mortality. Pregnant persons with hypertensive disorders who develop severe hypertension at delivery admission have been shown to experience higher rates of severe maternal morbidity relative to those without severe hypertension. Current guidelines recommend prompt treatment of severe hypertension given known associated maternal and fetal risks; however, only 1 previous study has described an association between timeliness of antihypertensive therapy and risk of severe maternal morbidity. OBJECTIVE: This study aimed to characterize how development of severe intrapartum hypertension and its timely treatment affect the risk of severe maternal morbidity. STUDY DESIGN: We conducted a population cohort study of deliveries with and without hypertensive disorders of pregnancy at a single urban hospital between 2016 and 2018. Among deliveries of persons with hypertensive disorders of pregnancy, we identified those with persistent severe hypertension (defined as blood pressure ≥160/105 mm Hg sustained over ≥15 minutes) and further classified individuals with severe hypertension as having received timely (within 60 minutes) or delayed treatment. Severe maternal morbidity was identified using a composite measure developed by the Centers for Disease Control and Prevention. We calculated overall and indicator-specific rates of severe maternal morbidity for 4 categories of deliveries: without hypertensive disorder of pregnancy, with hypertensive disorder of pregnancy without severe hypertension, with severe hypertension and timely treatment, and with severe hypertension and delayed treatment. We assessed the association between hypertensive disorder of pregnancy, severe hypertension, timeliness of treatment, and severe maternal morbidity using multivariable robust Poisson regression, adjusting for demographic and clinical characteristics. RESULTS: Of 3723 delivery hospitalizations within the study time frame, 32.3% (1204/3723) were complicated by presence of a hypertensive disorder without severe hypertension and 5.7% (211/3723) by presence of a hypertensive disorder with severe hypertension. Among those with severe hypertension, 48.8% (103/211) received timely treatment. Compared with deliveries not complicated by a hypertensive disorder, severe maternal morbidity risk was increased for hypertensive disorder of pregnancy without severe hypertension (124.4/1000 vs 52.0/1000; adjusted risk ratio, 1.84; 95% confidence interval, 1.41-2.40), severe hypertension with timely treatment (233.0/1000; adjusted risk ratio, 3.81; 95% confidence interval, 2.45-5.92), and severe hypertension with delayed treatment (305.6/1000; adjusted risk ratio, 5.38; 95% confidence interval, 3.75-7.73). CONCLUSION: Patients with hypertensive disorders of pregnancy are at an elevated risk of severe maternal morbidity, and development of severe hypertension further increases this risk. Timely antihypertensive treatment is associated with lower risk of severe maternal morbidity among those with severe hypertension. These findings emphasize the importance of provider education and quality improvement efforts aimed at expediting treatment of severe hypertension.

Delivery outcomes after induction of labor among women with hypertensive disorders of pregnancy.

OBJECTIVE: Induction of labor is known to be safe and highly effective in low-risk women. However, only limited research considers the relative success rates of induction of labor among women with one or more obstetric comorbidities. Our objective was to determine if the risk of cesarean delivery after induction of labor (IOL) is increased in women with a spectrum of hypertensive disorders of pregnancy compared to women with normotensive pregnancies. STUDY DESIGN: We analyzed data from 1842 women undergoing IOL occurring at Grady Memorial Hospital in Atlanta, Georgia 2016-2018. We used multivariable log binomial models to estimate unadjusted and adjusted risk ratios (aRR) describing the association between hypertensive disorder diagnosis (preeclampsia with or without severe features, gestational hypertension, and chronic hypertension) and cesarean delivery, adjusting for demographics, pre-pregnancy conditions, and gestational age at delivery. RESULTS: Overall, 44% (n = 808) of women in our study were diagnosed with any hypertensive disorder. Among women with hypertensive disorders, 74% had a successful vaginal delivery after IOL as compared to 82% of women without a hypertensive disorder. In the fully adjusted model, women with preeclampsia with severe features (aRR: 1.6, 95% CI: (1.3, 2.0)) and chronic hypertension had the largest risk for cesarean delivery (aRR 1.3, 95% CI: 0.9, 1.7)) compared with women with a normotensive pregnancy. CONCLUSION: Our study suggests that while patients with certain hypertensive diagnoses may be at increased risk for cesarean delivery following IOL, most patients with hypertensive disorders were still able to undergo a successful vaginal delivery following IOL.

Severe maternal morbidity and mortality during delivery hospitalization of class I, II, III, and super obese women.

BACKGROUND: Previous studies show that obesity predisposes patients to higher risks of adverse pregnancy outcomes. Data on the relationship between increasing degrees of obesity and risks of severe maternal morbidity, including mortality, are limited. OBJECTIVE: We examined the association of increasing classes of obesity, especially super obesity, with the risk of severe maternal morbidity and mortality at the time of delivery hospitalization. STUDY DESIGN: Using New York City linked birth certificates and hospital discharge data, we conducted a retrospective cohort study. This study identified delivery hospitalizations for singleton, live births in 2008-2012. Women were classified as having obesity (class I, II, III, or super obesity), as opposed to normal weight or overweight, based on prepregnancy body mass index. Cases of severe maternal morbidity were identified based on International Classification of Diseases, Ninth Revision diagnosis and procedure codes according to Centers for Disease Control and Prevention criteria. Multivariable logistic regression was used to evaluate the association between obesity classes and severe maternal morbidity, adjusting for maternal sociodemographic characteristics. RESULTS: During 2008-2012, there were 570,997 live singleton births with available information on prepregnancy body mass index that met all inclusion criteria. After adjusting for maternal characteristics, women with class II (adjusted odds ratio, 1.14; 95% confidence interval, 1.05-1.23), class III (adjusted odds ratio, 1.34; 95% confidence interval, 1.21-1.49), and super obesity (adjusted odds ratio, 1.99; 95% confidence interval, 1.57-2.54) were all significantly more likely to have severe maternal morbidity than normal and overweight women. Super obesity was associated with specific severe maternal morbidity indicators, including renal failure, air and thrombotic embolism, blood transfusion, heart failure, and the need for mechanical ventilation. CONCLUSION: There is a significant dose-response relationship between increasing obesity class and the risk of severe maternal morbidity at delivery hospitalization. The risks of severe maternal morbidity are highest for women with super obesity. Given that this is a modifiable risk factor, women with prepregnancy obesity should be counseled on the specific risks associated with pregnancy before conception to optimize their pregnancy outcomes.

Gestational Weight Gain and Severe Maternal Morbidity at Delivery Hospitalization.

OBJECTIVE: To examine whether women who varied from recommended gestational weight gain guidelines by the Institute of Medicine (IOM, now known as the National Academy of Medicine) were at increased risk of severe maternal morbidity during delivery hospitalization compared with those whose weight gain remained within guidelines. METHODS: We conducted a retrospective cohort study using linked 2008-2012 New York City discharge and birth certificate data sets. Cases of severe maternal morbidity were identified using International Classification of Diseases, 9th Revision, Clinical Modification codes based on the Centers for Disease Control and Prevention criteria, which consists of 21 indicators of possible life-threatening diagnoses, life-saving procedures, or death. Multivariable logistic regression was used to evaluate the association between gestational weight gain categories based on prepregnancy body mass index (BMI) and severe maternal morbidity adjusting for maternal demographics and socioeconomic status. The analysis was stratified by prepregnancy BMI categories. RESULTS: During 2008-2012, there were 515,148 term singleton live births in New York City with prepregnancy weight and gestational weight gain information. In 24.8%, 35.1%, 32.1%, and 8.0% of these births, women gained below, within, 1-19 lbs above, and 20 lbs or more above the IOM guidelines, respectively. After adjusting for maternal demographic and socioeconomic characteristics, women who had gestational weight gain 1-19 lbs above (adjusted odds ratio [AOR] 1.08, 95% CI 1.02-1.13) or 20 lbs or more above the IOM recommendations (AOR 1.21, 95% CI 1.12-1.31) had higher odds of overall severe maternal morbidity compared with women who gained within guidelines. Although the increased odds ratios (ORs) were statistically significant, this only resulted in an absolute rate increase of 2.1 and 6 cases of severe maternal morbidity per 1,000 deliveries for those who gained 1-19 and 20 lbs or more above recommendations, respectively. Women with gestational weight gain 20 lbs or more above recommendations had significantly higher ORs of eclampsia, heart failure during a procedure, pulmonary edema or acute heart failure, transfusion, and ventilation. CONCLUSION: Women whose gestational weight gain is in excess of IOM guidelines are at increased risk of severe maternal morbidity, although their absolute risks remain low.

Severe cardiovascular morbidity in women with hypertensive diseases during delivery hospitalization.

BACKGROUND: Cardiovascular disease is the leading cause of pregnancy-related death in the United States. Identification of short-term indicators of cardiovascular morbidity has the potential to alter the course of this devastating disease among women. It has been established that hypertensive disorders of pregnancy are associated with increased risk of cardiovascular disease 10-30 years after delivery; however, little is known about the association of hypertensive disorders of pregnancy with cardiovascular morbidity during the delivery hospitalization. OBJECTIVE: We aimed to identify the immediate risk of cardiovascular morbidity during the delivery hospitalization among women who experienced a hypertensive disorder of pregnancy. MATERIALS AND METHODS: This retrospective cohort study of women, 15-55 years old with a singleton gestation between 2008 and 2012 in New York City, examined the risk of severe cardiovascular morbidity in women with hypertensive disorders of pregnancy compared with normotensive women during their delivery hospitalization. Women with a history of chronic hypertension, diabetes mellitus, or cardiovascular disease were excluded. Mortality and severe cardiovascular morbidity (myocardial infarction, cerebrovascular disease, acute heart failure, heart failure or arrest during labor or procedure, cardiomyopathy, cardiac arrest and ventricular fibrillation, or conversion of cardiac rhythm) during the delivery hospitalization were identified using birth certificates and discharge record coding. Using multivariable logistic regression, we assessed the association between hypertensive disorders of pregnancy and severe cardiovascular morbidity, adjusting for relevant sociodemographic and pregnancy-specific clinical risk factors. RESULTS: A total of 569,900 women met inclusion criteria. Of those women, 39,624 (6.9%) had a hypertensive disorder of pregnancy: 11,301 (1.9%) gestational hypertension; 16,117 (2.8%) preeclampsia without severe features; and 12,206 (2.1%) preeclampsia with severe features, of whom 319 (0.06%) had eclampsia. Among women with a hypertensive disorder of pregnancy, 431 experienced severe cardiovascular morbidity (10.9 per 1000 deliveries; 95% confidence interval, 9.9-11.9). Among normotensive women, 1780 women experienced severe cardiovascular morbidity (3.4 per 1000 deliveries; 95% confidence interval, 3.2-3.5). Compared with normotensive women, there was a progressively increased risk of cardiovascular morbidity with gestational hypertension (adjusted odds ratio, 1.18; 95% confidence interval, 0.92-1.52), preeclampsia without severe features (adjusted odds ratio, 1.96; 95% confidence interval, 1.66-2.32), preeclampsia with severe features (adjusted odds ratio, 3.46; 95% confidence interval, 2.99-4.00), and eclampsia (adjusted odds ratio, 12.46; 95% confidence interval, 7.69-20.22). Of the 39,624 women with hypertensive disorders of pregnancy, there were 15 maternal deaths, 14 of which involved 1 or more cases of severe cardiovascular morbidity. CONCLUSION: Hypertensive disorders of pregnancy, particularly preeclampsia with severe features and eclampsia, are significantly associated with cardiovascular morbidity during the delivery hospitalization. Increased vigilance, including diligent screening for cardiac pathology in patients with hypertensive disorders of pregnancy, may lead to decreased morbidity for mothers.