Maternal Morbidity Predicted by an Intersectional Social Determinants of Health Phenotype: A Secondary Analysis of the NuMoM2b Dataset

Abstract

Maternal race, ethnicity and socio-economic position are known to be associated with increased risk for a range of poor pregnancy outcomes, including maternal morbidity and mortality. Previously, researchers seeking to identify the contributing factors focused on maternal behaviors and pregnancy complications. Less understood is the contribution of the social determinants of health (SDoH) in observed differences by race/ethnicity in these key outcomes. In this secondary analysis of the Nulliparous Pregnancy Outcomes Study: Monitoring Mothers-to-Be (nuMoM2b) dataset, latent mixture modeling was used to construct groups of healthy, nulliparous participants with a non-anomalous fetus in a cephalic presentation having a trial of labor (N=5763) based on SDoH variables. The primary outcome was a composite score of postpartum maternal morbidity. A postpartum maternal morbidity event was experienced by 350 individuals (6.1%). Latent class analysis using SDoH variables revealed six groups of participants, with postpartum maternal morbidity rates ranging from 8.7% to 4.5% across groups (p<0.001). Two SDoH groups had the highest odds for maternal morbidity. These higher-risk groups were comprised of participants with the lowest income and highest stress and those who had lived in the U.S. for the shortest periods of time. SDoH phenotype predicted MM outcomes and identified two important, yet distinct groups of pregnant people who were the most likely have a maternal morbidity event.

Introduction

The United States’ maternal mortality rate in 2019 was more than double that of most other wealthy countries[1] at 20.1 deaths/100,000 live births [2], [3]. Maternal mortality in the U.S. reflects a significant racial and ethnic disparity, whereby the rate for Black populations is 2.5 times the rate for non-Hispanic white populations and 3.5 times the rate of Hispanic populations. In addition to high maternal mortality rates, severe maternal morbidity (including near miss morbidity) is rising, complicating approximately 60,000 births per annum [4]. Disparities in maternal morbidity are also reported by race/ethnic subgroups, as well as by socioeconomic divides [5], [6]. Researchers and clinicians seeking to study and address U.S. maternal mortality rates and improve disparities have traditionally considered individual behaviors (weight gain) [7]–[9] or examined biological markers of risk like blood pressure [10] or maternal age [3] within or across race or ethnicity-based groups. These approaches have not led to reductions in racial/ethnic disparities in maternal morbidity nor mortality because they do not address structural factors, including historic or current racial discrimination [11], access to quality health care [12], [13] and other social determinants of health (SDoH) as key mechanisms underpinning poor maternal health inequities [14]. Moreover, this approach perpetuates previous research findings (i.e. existence of disparities by race or income) without providing context for interpreting the data within the social condition, or suggesting interventions to improve disparities [15]. Rather than suggest solutions, this narrow research lens has the effect of placing blame for poor morbidity and/or mortality outcomes and disparities on pregnant individuals and their families [16].

The primary structural factor underlying U.S. racial disparities in maternal morbidity/mortality is systemic racism [17]–[19]. Systemic racism is defined as the ways that societies enact discrimination based on race or ethnicity through the social, economic, environmental, and educational circumstances in which people are born and live their lives [20]. These circumstances are conceptualized under the umbrella term SDoH [21]. Differences in SDoH are theorized to explain disparities in perinatal outcomes within and between racial / ethnic populations[11], [22] and help explain disparities along socio-economic and rural/urban divides [23], [24]. However, it is challenging to analyze the relationship between SDoH and health inequities due to the intersectional nature of people’s multiple social conditions [25]. Despite these challenges, multiple professional organizations and regulatory bodies have charged clinicians, policy-makers and researchers to examine SDoH and racial equity in their work [26]–[30]. A SDoH framework is also a key component of the Healthy People 2030 initiative’s objectives and strategies [31].

Therefore, the purpose of this study was to apply a latent mixture modeling approach to categorize the intersection of multiple SDoH in predicting the occurrence of maternal morbidity among subset of nulliparous participants of a prospective cohort study (the Nulliparous Pregnancy Outcomes Study: Monitoring Mothers-to-Be study, or nuMoM2b) [32] who were healthy upon entering pregnancy. We hypothesized that maternal morbidity events would differ by phenotypes representing the intersectional social determinants of health, as would the frequency of health behaviors, pregnancy conditions, and labor/birth circumstances associated with morbidity. Guided by a theoretical model where most pregnancy conditions, behaviors, and labor/birth characteristics are found along a causal pathway between the intersectional social phenotype and maternal morbidity events, we used moderation and mediation modeling to identify pregnancy variables that modified the effect or direction of the relationship between maternal morbidity and SDoH grouping. Our intent, approach and hypothesis focused specifically on participants who did not have significant co-morbidities prior to pregnancy, highlighting the importance of examining SDoH even for those who present as healthy at pregnancy onset.

Methods

Sample

This study was a secondary analysis of the nuMoM2b dataset, which was collected during the Nulliparous Pregnancy Outcomes Study: Monitoring Mothers-to-Be from 2010–2015. Funded by the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) and the National Institutes of Health Office of Research on Women’s Health, nuMoM2b was a prospective cohort study which enrolled a racially, ethnically, and geographically diverse group of 10,038 nulliparous people with singleton pregnancies from hospitals affiliated with eight clinical research centers in the United States [33]. The primary outcomes of interest for the original study were preterm birth, preeclampsia, and fetal growth restriction, and investigators collected a range of maternal characteristics, environmental factors, and pregnancy/birth characteristics on each participant [34]. Participants had four study visits during pregnancy, ranging from 6 weeks, 0/7 days’ gestation to the time of birth, with information collected from personal interviews, self-administered questionnaires, clinical measurements, medical record abstraction, and biological specimens. A total of 9023 participants completed the study and were included in the final dataset. More detailed information on the nuMoM2b study can be obtained in the protocol publication [32].

For this secondary analysis of nuMoM2b, we obtained the de-identified dataset using the NICHD Data and Specimen Hub (DASH) [35] under a study-specific data use agreement for both the investigators’ institutions. Emory University Institutional Review Board review (#00001356) determined that this project was exempt under 45 CR 46.104(d). We limited our final sample to individuals with a trial of labor who had no significant co-morbid conditions at enrollment and had a non-anomalous fetus. We also excluded non-cephalic presentations and births occurring at less than 24 0/7 weeks of gestation, resulting in a final sample of N=5763 cases (Figure 1).

Figure 1.

Participant Selection and Grouping for Secondary Analysis of New Moms to Be Study: Social Determinants of Health Phenotype Prediction of Maternal Morbidity

Theoretical Framework

At study outset, we developed a theoretical framework for examining the influence of SDoH variables on maternal morbidity outcomes within the larger context of pregnancy and baseline characteristics. We examined subgroups of variables characterized as either health behaviors (i.e. diet [Modified Block 2005 Food Frequency Questionnaire] [36], tobacco use in preconception period, physical activity [37], sleep quality), pregnancy conditions (i.e. body mass index pre-pregnancy, gestational age at birth, maternal age, neonatal sex, maternal mental health, pregnancy complications such as hypertension or gestational diabetes, placental abruption, antenatal steroid or progesterone use), or labor/birth characteristics (i.e. mode of labor onset, premature rupture of membranes, cervical dilation at labor admission, amniotomy, labor augmentation, epidural use, episiotomy use, chorioamnionitis, labor duration, operative vaginal birth attempt) (Supplemental Figure 1). In this framework, SDoH variables were considered separately from variables describing behaviors, pregnancy conditions, or labor/birth characteristics to more clearly consider the independent influence of the SDoH phenotype on maternal morbidity outcomes. Furthermore, in this framework most pregnancy conditions, behaviors, and labor/birth characteristics are found along a causal pathway between the intersectional social phenotype and maternal morbidity events.

Latent Class Indicators: Social Determinants of Health

The selection of nuMoM2b variables to include as latent class indicators of SDoH for this study were informed by the Institute of Medicine’s (IOM) 2014 recommendations of core domains of the social determinants of health for inclusion in electronic medical records and clinical care [21]. The IOM defined five overarching domains of the SDoH: sociodemographic, psychological, behavioral, individual-level social relationships/living conditions, and neighborhood and community (Table 1). Several of these domains were not available in the nuMoM2b dataset, such as many characteristics of the neighborhood/community. Moreover, we did not include the IOM behavioral domain in our SDoH latent class models. The IOM SDoH Behavioral domain includes dietary patterns, physical activity, use of alcohol and other substances, sexual practices, and risk-taking behaviors. The choice to not include these with other SDoH variables was made with the rationale that pregnancy behaviors are often a focus of educational interventions to improve outcomes, and as such may vary depending on the quality of antepartum care received by the participant. Therefore, we examined the influence of pregnancy behaviors separately from the other SDoH on study outcomes. We also examined the influence of depression and anxiety separately from other SDoH in these analysis with the rationale that, like behavioral variables, mental health conditions are the focus of screening and various interventions during the antenatal course, and thus were evaluated separately as a pregnancy condition. Overall, 14 SDoH variables were considered in the latent models. For questionnaires administered multiple times during pregnancy, the earliest administration score was used for latent class analysis construction to best reflect the condition of the participant during the majority of their pregnancy. Permissions to use these questionnaires were obtained by both the nuMoM2b investigators and by the authors of this secondary analysis.

Table 1.

Comparison of Institute of Medicine’s 2014 Key Domains of Social Determinants of Health with NuMoM2b Variables Used for Latent Class Analysis.

Domains of the Social Determinants of Health	IOM 2014 Variable	NuMoM2b Variable
Sociodemographic	Sexual orientation	n/a
	Gender identity	n/a
	Race/ethnicity	participant’s presenting race/ethnicity (as reported by the participant)
	Country of origin	number of years lived U.S. English ability (speaks very well vs. less than very well)
	Education	level of education attainment
	Employment	employment status (unemployed throughout study vs. employed consistently)
	Finance resource strain	insurance type (government, commercial) percent of federal poverty level
Psychological	Stress	stress (Perceived Stress Scale, PSS-10)
	Negative mood and affect (depression, anxiety, hostility, anger, hopelessness)	Considered separately from SDoH as Pregnancy Conditions in study analyses
	Psychological assets (coping, positive affect, life satisfaction, conscientiousness, engagement, optimism, self-efficacy)	resilience (Connor-Davidson Resilience Scale, CD-RISC),
Behavioral	Dietary patterns	Considered separately from SDoH as Pregnancy Behaviors in study analyses
	Physical activity
	Tobacco use/exposure
	Substance/alcohol use
	Exposure to firearms, risk-taking behaviors
Individual-Level Social Relationships and Living Conditions	Social connections or isolation	experiences of discrimination (Experiences of Discrimination scale, EOD)
	Exposure to violence
	Social support (emotional, instrumental, other)	partnered status social support (Multidimensional Scale of Perceived Social Support, MSPSS),
	Work conditions	n/a
	History of incarceration	n/a
	Military service	n/a
	Community norms of health decision making	n/a
Neighborhoods and Communities	Neighborhood socioeconomic and racial/ethnic characteristics	n/a
	Neighborhood and community contextual characteristics (air pollution, allergens, nutritious food options, transportation, parks, open spaces, health care, social services, educational and job opportunities)	participants’ perceptions of prenatal care (quality and positivity) derived from two items on the Pregnancy Experiences Scale (PES-Brief)

The 14 indicators of SDoH were tested for a best fitting class structure using methods described by Ram & Grimm (2009) within the sample [38]. For this analysis, we used an exploratory and confirmatory approach. We first split the sample into two sub-samples determined by a random number generator function. We tested for the best-fitting class structure in the first subsample by serially testing the application of 2–7 latent classes, comparing the fit statistics including AIC/BIC, posterior probability, entropy, and the significance of likelihood ratio tests. We then repeated these procedures in the second subsample to confirm that the best fitting class number was found using the second subgroup. Then, the entire sample was rejoined and used for generating the final models, after repeating all steps listed above. Latent class solution procedures include performance metrics for the best-fitting class structure (entropy, posterior probabilities, Vuong Lo Mendell Rubin LRT, Lo Mendell Rubin adjusted LRT and parametric bootstapped LRT. Class number was then assigned to each participant/birth and is referred to as the SDoH phenotype or class.

Outcome: Postpartum Maternal Morbidity

Our primary outcome was postpartum maternal morbidity (MM), which was defined by one or more of the following subcategories: hemorrhage (postpartum hemorrhage requiring transfusion, severe postpartum anemia, or hysterectomy), abnormal coagulation event (postpartum pulmonary embolus or deep vein thrombosis), cardiovascular/cardiomyopathy (postpartum cardiomyopathy or cerebral vascular accident), infection (postpartum endometritis, wound infection or dehiscence, pyleonephritis, urinary tract infection, maternal sepsis, or any other maternal postpartum infections within the first 14 postpartum days), hospital readmission in first 14 days postpartum and/or obstetric anal sphincter injury (OASI), defined as experiencing a 3^rd or 4^th degree perineal laceration. We created a binary postpartum MM outcome (presence or absence of any of the morbidity events), and also created variables indicating the presence of MM by subcategories (see above).

Moderator or Mediator Variables: health behaviors, personal characteristics and pregnancy conditions

Pregnancy variables related to maternal morbidity were grouped into categories consistent with our theoretical model (categories: pregnancy behaviors, pregnancy conditions and labor/birth characteristics, Supplemental Figure 1) and assessed with bivariate tests as appropriate to determine those that were associated with MM and also with SDoH. We included the variable in the final models if it was significantly related to both the exposure class and the MM outcome (p<0.05).

Statistical Approach

After determining the best-fitting classes, each participant was assigned to one SDoH class (Supplemental Table 1). We first created Poisson regression models to estimate incident rate ratios (IRR) for postpartum MM by SDoH Class. Next, we considered the variables that were significantly related to both the SDoH exposure and MM outcome as either moderators or mediators of the relationship between SDoH and MM. We considered if the variables were theoretically found along the causal pathway between SDoH and MM (i.e SDoH class representing economic and educational level etc. was leading the healthy eating index score, not the reverse). Therefore, in these moderation and mediation tests, we determined if the presence of any of these variables would modify the effect or direction of the relationship between MM and SDoH class.

To identify moderators of the relationship between SDoH class and MM outcome, we performed interaction analyses using Poisson regression with robust standard errors [39]. Next, we identified mediators of the relationship between SDoH class and MM outcome using generalized structural equation models with Poisson distribution. We created a mediation model (gsem) (StataCorp.,2021) estimating the IRR for postpartum MM, using SDoH class as the primary predictor and testing variables that met criteria for inclusion into the mediation model. The Baron & Kenny approach was used to determine full or partial mediation [40]. The final SEM model examined MM using SDoH class along with the influence of any mediators. Analyses were conducted using R statistical software (Version 1.2.5019), STATA MP (Version 17.0), and Mplus (2019). Significance was set at a p-value of 0.05.

Results

Of the 5,763 healthy participants in the final sample, 350 individuals experienced a postpartum maternal morbidity event (6.1%, Supplemental Table 2). Latent class analysis revealed that six SDoH classes best fit the data (Figure 2) with the largest SDoH class representing 49.1% of the sample, and the smallest SDoH class representing 2.9% of the sample. Entropy (0.94) and posterior probabilities were high (>0.9), and the Vuong-Lo-Mendell-Rubin LRT, Lo-Mendell-Rubin adjusted LRT and parametric bootstrapped LRT were significant (indicating that our solution compared to the n-1 class was better fitting (p<0.001). AIC/BIC were 290064.77 and 290979.46 respectively. Although two of these SDoH classes represented less than 5% of the sample, we retained them with the rationale that each contained over 150 persons and were part of a latent class solution with robust overall fit statistics. Below, we describe each SDoH phenotype according to their indicators, starting with the largest class and moving to smallest (Figure 2, Supplemental Table 1). Bivariate statistical analyses demonstrated significant differences across pregnancy characteristics/behaviors and labor/birth characteristics for the SDoH classes (Table 2) and are selectively discussed along with the SDoH descriptions below.

Figure 2.

Latent Class Phenotypes of Social Determinants of Health among Healthy Participants in the New Moms to Be Dataset with Trial of Labor with Non-anomalous Fetus (N=5763) (top). Distribution of Race and Ethnicity (Identified by Participants as How They Present to Others) (bottom)

Table 2.

Pregnancy, Labor, Birth, and Neonatal Characteristics Distribution in Social Determinants of Health Groupings within the Healthy Participants of the New Moms to Be Study (N=5763)

Social Determinants of Health Grouping	1	2	3	4	5	6	p-value
Sample (n)	198	366	480	2831	166	1722
Pregnancy Behaviors
2010 Healthy Eating Index (mean)	60.63	66.22	60.14	67.28	58.57	54.33	<0.001
No smoking in 3 months Prior to Pregnancy (%)	79.7%	93.1%	77.5%	90.2%	75.9%	69.6	<0.001
No exercise In the Month Prior to Visit 3 (%)	40.3%	39.9%	31.5%	20.4%	28.5%	40.7%	<0.001
Sleep Quality in the Month Prior to Visit 3 (%)							<0.001
Very Sound/Restful	12.7%	10.7%	11.3%	9.0%	11.0%	16.4%
Sound/Restful	33.5%	27.2%	24.1%	26.9%	22.6%	22.6%
Average Quality	37.0%	48.9%	39.2%	42.9%	41.8%	40.7%
Restless	15.6%	9.2%	23.1%	18.8%	20.5%	16.6%
Very Restless	1.2%	4.0%	2.4%	2.4%	4.1%	3.7%
Gestational Weight Gain, kg (Median/IQR)	13.33 [10.41, 16.77]	12.95 [10.00, 16.34]	13.15 [10.21, 16.42]	13.61 [10.89, 16.90]	12.27 [8.16, 15.16]	13.52 [9.71, 17.64]	<0.001
Pregnancy Conditions
BMI Pre-pregnancy, kg/m2 (Median/IQR)	24.83 [22.67, 29.44]	22.96 [21.12, 25.92]	25.30 [22.27, 30.59]	23.93 [21.84, 27.40]	26.31 [22.47, 33.62]	25.08 [21.80, 29.99]	<0.001
Gestational Age at Birth, weeks (mean)	39.25	39.41	39.32	39.49	39.14	39.15	<0.001
Maternal Age, years (mean)	26.61	26.80	26.11	29.18	27.03	21.57	<0.001
Edinburgh Postpartum Depression Scale Score >=10 at Visit 3 (%)	18.5%	18.0%	20.3%	9.5%	24.2%	22.6%	<0.001
State-Trait Anxiety Inventory Category, Visit 1 (%)							<0.001
Least Anxious	25.3%	24.6%	24.2%	31.2%	19.9%	18.7%
Middle 50% Anxious	46.2%	52.5%	46.5%	54.1%	41.8%	47.5%
Most Anxious	28.5%	22.9%	29.2%	14.6%	38.4%	33.8%
Absence of Diabetes based on Pregnancy Testing (%)	99.0%	96.2%	97.3%	97.6%	92.8%	97.9%	0.001
Intrauterine Growth Restriction Diagnosis (%)	3.0%	3.0%	2.3%	1.6%	3.6%	3.4%	0.003
Oligohydramnios/Anhydramnios Diagnosis (%)	2.0%	6.0%	4.6%	3.7%	6.0%	4.4%	0.129
Abruption Diagnosis (%)	0.5%	1.4%	0.2%	0.7%	0.6%	0.6%	0.497
Hypertensive Disorders of Pregnancy (%)							<0.001
Eclampsia	0.0%	0.3%	0.2%	0.0%	0.0%	0.1%
Severe Pre-eclampsia	7.7%	1.9%	5.8%	3.3%	3.0%	4.7%
Received Antenatal Steroids (%)	4.5%	4.4%	4.0%	2.4%	8.4%	3.9%	<0.001
Received Antenatal Progesterone (%)	1.5%	2.5%	1.9%	1.5%	2.4%	1.0%	0.287
Labor & Birth Characteristics
Mode of Labor Onset (%)							0.012
Not Augmented	17.7%	20.3%	16.7%	20.2%	17.5%	16.7%
Augmented	44.9%	48.2%	49.8%	45.9%	48.2%	45.8%
Induced	36.4%	29.6%	32.9%	33.4%	33.1%	36.8%
Rupture of Membranes Prior to Labor (%)							0.031
Yes	49.0%	46.6%	46.7%	48.7%	53.9%	44.8%
No	48.5%	50.7%	51.9%	50.2%	43.6%	53.3%
Unknown	2.5%	2.7%	1.5%	1.1%	2.4%	1.9%
Cervical Dilation on Hospital Admission, cm (mean)	2.98	2.99	2.93	3.10	2.72	2.82	0.001
Amniotomy (%)	48.5%	50.7%	51.9%	50.2%	43.6%	53.3%	0.031
Epidural (%)	79.8%	77.0%	81.9%	80.3%	78.2%	84.1%	0.006
Episiotomy (%)	13.8%	12.8%	9.0%	11.3%	6.7%	10.6%	0.138
Chorioamnionitis diagnosis (%)	6.1%	7.7%	9.2%	6.1%	7.8%	7.7%	0.123
Shoulder Dystocia (%)	0.0%	0.8%	3.5%	0.7%	0.9%	2.4%	0.012
Operative Vaginal Delivery Attempt (%)	11.2%	9.2%	9.4%	10.6%	6.7%	8.5%	0.161
Neonatal Outcomes
Stillbirth (%)	0.5%	0.3%	0.0%	0.1%	0.0%	0.2%	0.684
Birthweight (mean)	3311.7	3253.00	3269.16	3372.07	3209.47	3223.90	<0.001
Sex of Baby Assigned at Birth (%)							0.688
Male	55.2%	53.2%	53.5%	50.9%	55.9%	49.4%
Female	44.8%	46.8%	46.5%	49.1%	44.1%	50.6%
APGAR Score >7 at 5 min (%)	80.0%	89.5%	94.1%	92.5%	88.9%	94.7%	0.82
APGAR Score >7 at 10 min (%)	80.0%	89.5%	94.1%	92.5%	88.9%	94.7	0.82
Highest Level of Neonatal Care Needed (%)							0.001
Died in Labor and Delivery	0.0%	0.0%	0.2%	0.0%	0.0%	0.0%
Immediate Transfer	0.0%	0.3%	0.0%	0.1%	0.6%	0.1%
Well Baby Nursery	88.0%	83.7%	83.3%	85.9%	83.2%	83.8%
Intermediate Care/Stepdown	2.6%	3.4%	2.2%	3.3%	3.7%	1.4%
NICU/Acute Care	9.4%	12.7%	14.3%	10.7%	12.4%	14.8%

¹p-values determined by chi-square (categorical), ANOVA (normally-distributed continuous), or Mann-Whitney U Test.

²Postpartum morbidity outcomes considered include: Hemorrhage (postpartum hemorrhage requiring transfusion, severe postpartum anemia, or hysterectomy), Clotting disorders (postpartum pulmonary embolus or deep vein thrombosis), Cardiovascular/Cardiomyopathy (postpartum cardiomyopathy or cerebral vascular accident), Infection (postpartum endometritis, wound infection or dehiscence, pyleonephritis, urinary tract infection, maternal sepsis, or any other maternal postpartum infections within the first 14 postpartum days), maternal hospital readmission within first 14 days postpartum, or perineal laceration (3^rd or 4^th degree).

Class 4 (n= 2831, 49.1%) was the largest class, consisting of a highly-educated, mostly White-presenting and economically secure group of nuMoM2b participants. This class of individuals appeared to have the most socio-economic stability and social privilege, along with high self-reported support and positive prenatal care experiences. Compared to other SDoH groups, this group had the highest resilience and social support scores, while stress and discrimination scores were the lowest in this group. These participants had the highest Healthy Eating Index (HEI) score overall, were older (average age 29.2 years) than those in other classes and gave birth at a higher mean gestational age and neonatal birthweight. They also had the lowest reported smoking, depression, anxiety scores and also the lowest unplanned cesarean birth rate (18.0%) compared to members of other SDoH classes. Only 4.5% (n=128) experienced a postpartum MM outcome, which was among the lowest compared to other classes.

Class 6 (n= 1722, 29.9%) consisted of individuals with the lowest income and educational attainment levels, compared to other groups. This group was split roughly by thirds as Black-presenting, White-presenting, and Hispanic-presenting. Most members of this group used government sponsored health insurance for their pregnancy care, and 14% were not partnered during their pregnancy. Interestingly, 15% reported that their prenatal visits did not make them feel happy or positive. Mean perceived stress levels were among the highest in this group. This group had the lowest mean scores on the Healthy Eating Index, second lowest percentage of non-smokers before pregnancy, youngest mean age (21.6 years) and among the highest depression and anxiety scores, as well as the highest percentage of newborns needing NICU or acute care after birth. Unplanned cesarean was used in 20.4% of their births, which positioned them in the middle compared to other groups. Class 6 had the highest number of postpartum MM events, occurring after 8.7% (n=149) of births.

Class 3 (n= 488, 8.3%) consisted of individuals who were neither the highest nor lowest in terms of stress, support, and resilience; however, they did report the second highest level of experiences with discrimination among the groups. This group was also split roughly in thirds by Black-, White-, and Hispanic-presentation, but unlike Class 6, Class 3 individuals were economically stable, with the majority having at least some college education and being consistently employed. Most other characteristics for Class 3 were not the highest or lowest among all the classes, though shoulder dystocia occurred most frequently in this class (3.5% of births) and NICU admission in neonates born to members of this group was the second highest at 14.3%. Pregnancy outcomes in this group were also neither the highest nor lowest among all classes. Unplanned Cesarean made up 24.6% of births and postpartum maternal morbidity was identified in 5.6% of Class 3 members (n=27).

Class 2 (n=366, 6.2%) consisted of individuals who had most recently immigrated to the United States. The majority of this group were Hispanic (58%) or Asian (9%) presenting, and more than half reported that they did not speak English well. Income and education in this group were in the middle range of all classes, while social support was among the highest and stress scores were among the lowest among all classes. However, nearly a third of individuals in this group reporting having “some” or “a great deal” of concern or worry about their quality of their health care. Mean maternal age was 26.8 years. This class had the lowest epidural rate for labor and the second highest episiotomy use (12.8%). This class had the second highest rate of postpartum MM outcomes at 7.9% of the class (n = 29). This was noted despite this class having the second highest Healthy Eating Index, the highest non-smoking rates for the sample, the lowest mean BMI pre-pregnancy, and the lowest rate of any hypertensive disorder/diagnosis during pregnancy or birth. However, this group had the highest use of antenatal progesterone at 2.5% and had highest oligohydramnios diagnoses (6.0%) as well highest rate of placental abruption (1.4%). Unplanned Cesarean occurred in 25.1% of births, which was the second highest rate among all classes.

Class 1 (n= 198, 3.4%), stood apart as having significantly lower perceived social support than other groups (mean PSS score of 18.3 vs. >75 in other classes). Additionally, 10% of this group reported that their prenatal visits “not at all positive”, making them one of the worst in terms of their perceptions of prenatal care. Most members of this group were employed, and most had more than some college education and a high level of income. In this class we found the highest episiotomy utilization as well as the highest operative vaginal delivery attempts. This class had one of the lowest rates of postpartum MM at 4.5% (n = 9), and an unplanned cesarean rate of 23.2%, which was in the middle of the range for other classes.

Finally, Class 5 (n=169, 2.9%) reported the highest level of experiences of discrimination, lower levels of support, and highest stress levels compared to other groups. This group of nuMoM2b participants were over 51% Black-presenting and 25% Hispanic-presenting. Most members of this group had at least some college education, and about ¼ of them had concerns about the quality of their medical care. Additionally, this class had the second highest level of exercise in the third trimester, the lowest mean gestational weight gain, and the highest pre-pregnancy BMI. Mean maternal age was 27.0 at delivery, which was the second oldest. These participants had the highest depression scores and reported the most anxiety, along with the highest percentage with gestational diabetes. These participants had the highest use of antenatal steroid administration for lung maturation, highest growth restriction, lowest mean gestational age, and lowest neonatal birth weights. Postpartum maternal morbidity was found in 4.8% of members in this class (n = 8), which was one of the lowest rates compared to other SDoH classes. However, class 5 was notable for the highest unplanned Cesarean birth rate at 28.3%.

Pregnancy Variables Associated with Postpartum Maternal Morbidity or SDoH Class

In bivariate statistics evaluating the relationship between postpartum maternal morbidity with pregnancy health behaviors, pregnancy conditions, and labor/birth variables, the following factors were significant at p < 0.05: Health Eating Index score, exercise in third trimester, body mass index pre-pregnancy, maternal age, hypertensive disorders in pregnancy (particularly preeclampsia with severe features/eclampsia), mode of labor onset, operative vaginal delivery attempt, cervical dilation on admission, chorioamnionitis diagnosed in labor, unplanned cesarean birth, and neonatal intensive care admission (Supplemental Table 2). Among these variables, the following also had a significant relationship at p < 0.05 with SDoH class (Table 2): Health Eating Index score, third trimester exercise, pre-pregnancy body mass index, maternal age, mode of labor onset (induction vs. spontaneous labor), cervical dilation on admission (spontaneous / induced labor analyzed separately), hypertensive disorder of pregnancy, and unplanned cesarean birth. This final group of variables were considered in mediation and moderation analyses of the maternal morbidity by SDoH class.

Models of Maternal Morbidity Outcomes

Rates of postpartum maternal morbidity were significantly different across SDoH groups, ranging from a low of 4.5% (Class 4) to a high of 8.7% (Class 6) (p<0.001) (Supplemental Table 2 and Figure 3). Using composite postpartum MM as an outcome, unadjusted Poisson regression identified Class 2 and Class 6 (ref Class 4) as having the highest odds for MM, (Class 2 IRR) 1.75; 95% Confidence Interval (CI) [1.19–2.38], Class 6 IRR 1.91; 95% CI [1.52–2.40]).

Figure 3.

Distribution of Social Determinants of Health Groups (Classes) on Maternal Morbidity and Other Key Pregnancy Outcomes

To evaluate for moderation and mediation of the relationship between SDoH class and maternal morbidity, we first considered variables that were significantly related to postpartum MM and SDoH class in bivariate analyses. These included health behaviors (Healthy Eating Index score, third trimester exercise), pregnancy/personal characteristics (maternal age, pre-pregnancy BMI, hypertensive disorder of pregnancy) and labor/birth characteristics (mode of labor onset, mode of delivery, cervical dilation on admission). In analyses to evaluate for moderation effects (Table 3), IRR for main effects for Classes 2 and 6 compared to Class 4 were significantly higher for the following variables: spontaneous labor onset, closed cervix on admission (spontaneous or induced labor), absence of hypertensive disorders, and vaginal birth (class 6 only). However, the only interaction term (SDoH class x variable) which was significant for Class 2 was greater dilation on admission (labor induction only), which was associated with lower MM (0.12, 0.02–0.66) (IRR, 95%CI). In Class 6, the only significant interaction term was greater dilation on admission for spontaneous labor and lower MM (IRR 0.78, 95% CI [0.65–0.93]). Neither the main effects nor interactions were significant for maternal age, pre-pregnancy BMI score, Healthy Eating Index score, or exercise in the third trimester (data not shown). In mediation analyses (Table 4), we found partial mediation of the direct relationship between SDoH class and maternal morbidity in Class 2 (vs. Class 4) with unplanned cesarean birth and hypertensive disorders of pregnancy. In Class 6 (vs. Class 4), no mediators were found. In the models, health behaviors and maternal age were significantly related to the SDoH class but not to the outcome of MM.

Table 3.

Moderation models of the Relationship between Social Determinant of Health Grouping and Maternal Morbidity within the Healthy Participants of the New Moms to Be Study (N=5763)

Maternal Morbidity Outcome IRR (95%CI)	Class 1	Class 2	Class 3	Class 5	Class 6
Unadjusted/No interaction	1.00 (0.52–1.94)	1.75 (1.19–2.58)	1.24 (0.83–1.86)	1.07 (0.53–2.14)	1.91 (1.52–2.40)
Mode of labor onset
Spontaneous Labor	0.82 (0.31–2.23)	2.15 (1.35–3.43)	1.05 (0.59–1.86)	1.18 (0.48–2.85)	1.91 (1.40–2.60)
Class x Labor induction	1.49 (0.39–5.65)	0.54 (0.22–1.31)	1.51 (0.67–3.39)	0.83 (0.19 –3.48)	0.99 (0.62–1.57)
Dilation on Admission (SOL)
Closed cervix	5.89 (1.16–28.69)	3.39 (1.29– 8.89)	1.79 (0.65–4.90)	2.24 (0.64–7.80)	4.25 (2.18–9.29)
Class x cervical dilation (cm)	0.51 (0.27 –0.94)	0.89 (0.69 –1.14)	0.85 (0.66–1.11)	0.83 (0.61–1.11)	0.78 (0.65–0.93)
Dilation on Admission (IOL)
Closed cervix	1.20 (0.34–4.25)	3.82 (1.68–8.69)	2.38 (1.03– 5.48)	1.77 (0.48–6.51)	2.64 (1.52 –4.58)
Class x cervical dilation (cm)	1.03 (0.62–1.69)	0.12 (0.02–0.66)	0.75 (0.49 –1.15)	0.57 (0.24–1.34)	0.81 (0.61– 1.07)
Hypertensive disorder
No hypertensive disorder	0.96 (0.39–2.34)	1.96 (1.25–3.08)	1.58 (0.99–2.52)	1.34 (0.59–3.01)	2.03 (1.52–2.71)
Class x hypertensive disorder	1.01 (0.27–3.80)	0.87 (0.35–2.18)	0.43 (0.17–1.11)	0.46 (0.09 – 2.26)	0.82 (0.51–1.31)
Mode of birth
Vaginal birth	0.84 (0.35–2.03)	1.39 (0.82–2.38)	0.92 (0.51–1.62)	0.86 (0.32–2.30)	1.58 (1.19–2.11)
Class x Cesarean	1.43 (0.38–5.38)	1.50 (0.69–3.28)	1.78 (0.79–4.03)	1.37 (0.34–5.52)	1.59 (0.99–2.56)

Bolded= significance <0.05

SOL = spontaneous onset of labor

IOL= induction of labor

Reference Group is Class 4

Table 4.

Mediation Models for Social Determinants of Health Grouping and Maternal Morbidity within the Healthy Participants of the New Moms to Be Study (N=5763)

	Effect mediator on MM outcome IRR(95%CI)		Effect SDoH on mediator IRR(95%CI)	mediation effect
Mediator
Body Mass Index (pre-pregnancy)	1.00 (0.98–1.02)	Class 2 Class 6	0.95 (0.93–0.97) 1.05 (1.04–1.07)	No mediation
Unplanned Cesarean Birth	1.81 (1.36–2.41)	Class 2 Class 6	1.39 (1.17–1.74) 1.13 (0.99–1.30)	Partial mediation (Class 2)
Hypertensive Disorder in Pregnancy	1.43 (1.09–1.89)	Class 2 Class 6	0.56 (0.42–0.76) 1.07 (0.94–1.21)	Partial mediation (Class 2)
Mode of Labor Onset	1.06 (0.78–1.43)	Class 2 Class 6	0.89 (0.74–1.09) 1.10 (0.99–1.22)	No mediation
Cervical Dilation on Hospital Admission	0.95 (0.88–1.03)	Class 2 Class 6	0.98 (0.92–1.06) 0.91 (0.88–0.94)	No mediation
Healthy Eating Index score	0.99 (0.99–1.00)	Class 2 Class 6	0.98 (0.97–0.99) 0.81 (0.80–0.81)	No mediation
Exercise	0.81 (0.61–1.06)	Class 2 Class 6	0.75 (0.65–0.87) 0.75 (0.69–0.80)	No mediation
Age	0.99 (0.97–1.03)	Class 2 Class 6	0.92 (0.89–0.94) 0.74 (0.73–0.75)	No mediation

In the final SEM, which included partial mediators of both hypertensive disorders and cesarean birth, we found Class 2 individuals had a 71% increased rate of MM compared to class 4 (IRR, 1.71, 95%CI, 1.34–2.56) and Class 6 individuals had an 85% increased rate of MM (IRR, 1.85, 95% CI, 1.46–2.34). Both hypertension (IRR 1.51, 95% CI, 1.21–1.89) and Cesarean (IRR 2.36, 95% CI, 1.90–2.94) were also associated with MM in the model. In sum, these data support the hypothesis that SDoH may underpin differences in MM outcomes.

Discussion

In this study, we applied latent mixture modeling to empirically determine subgroups of people according to their complex and interrelated social determinants of health factors, then evaluated how these groupings predicted maternal morbidity following birth. These findings help contextualize the discussion of maternal health inequities in the United States and contribute to the growing body of research documenting worse health outcomes among racially marginalized groups, rural residents, and socially isolated communities (refugee or immigrant). However, in our analysis, we also found that the intersectional SDoH phenotype predicted maternal morbidity and identified two important, yet distinct groups of individuals who were more likely to have a postpartum maternal morbidity event.

Maternal morbidity is often studied and discussed as a function of pre-pregnancy comorbidities, including factors like maternal age, prior uterine surgery, and body mass index[8], [41], [42]. Race and ethnicity are also identified as ‘risk-factors’ for adverse outcomes [5], often without attention to the mechanisms of how marginalized social experiences underpin the association of ‘race’ with morbidity/mortality [23] (i.e. current discrimination or racism, the legacy of historic discrimination, and limitations in social or economic mobility). To address this issue, our intent with this analysis was to specifically examine the interplay of social (presenting race/ethnicity, nativity, partner support), economic (income, education, insurance) and lived experiences (stress, discrimination, feelings about pregnancy care), rather than hold their effects constant among all other confounders (i.e. controlling for insurance/education).

Our study specifically highlights worse postpartum maternal morbidity outcomes for two important groups— young people living close to the federal poverty level who have lower levels of educational attainment (Class 6) and people with limited proficiency in the English language who have limited experience living in the United States (Class 2). Although these two groups largely consisted of individuals from communities of color (using self-reported presenting race/ethnicity), the SDoH phenotype approach used in this study helped reveal the intersection of race and ethnicity within the context of nuMoM2b participant’s pregnancy risks and socio-economic environments. Class 6, for example, was split nearly equally between Black, White, Hispanic presentation, but individuals in this group also possessed lower levels of education attainment (likely due to young overall age), high poverty levels, and high reported stress scores compared to other groups. By contrast, the SDoH group with the lowest frequency of maternal morbidity were predominantly White-presenting, but were also more highly educated than other groups, with stable employment, commercial insurance, and higher levels of social support. We saw that the intersectional influence of presenting race, ethnicity, income, education, social support, and stress was more predictive of maternal morbidity than race, ethnicity, or any of the other SDoH factors alone.

This research supports the importance of maternal morbidity interventions that are individualized to people’s circumstances. For example, maternal morbidity reduction interventions indicated by this research include different strategies; one for recently immigrated people with very limited English (Class 2) and another for pregnant adolescents with limited partner support or financial means (Class 4). Supporting this idea, we conducted a sub-analysis (data not shown) using within-class step-wise regressions of member’s lifestyle and intrapartum experiences, which painted a very different picture of the factors associated with maternal morbidity in Class 4 (healthiest group), 2, and 6. In Class 4, in analyses including all pregnancy behaviors, the only one associated with maternal morbidity was higher gestational weight gain, while in Class 2 disrupted sleep was associated, and in Class 6 no pregnancy behavior was associated with morbidity. For all of these groups, development of hypertension during pregnancy was associated with maternal morbidity, but there were big differences in the intrapartum variables associated with maternal morbidity within each group. For example, Class 4 members had higher odds of maternal morbidity if they had labor induction or augmentation, cesarean birth, developed chorioamnionitis, or had an operative vaginal birth. Like Class 4, Class 6 and Class 2 had significant associations between maternal morbidity and chorioamnionitis, but the odds ratios for this association were twice as high as Class 4 among Class 6 members and four times as high among Class 2 members, after adjusting for other intrapartum factors.

Also in this study, we conceptualized common pregnancy ‘risk factors’ of maternal morbidity as occurring along the causal pathway between SDoH phenotype and maternal morbidity. Considering causal paths more carefully has been promoted in recent years, to properly distinguish confounding variables from mediators in maternal health research [43], [44]. In moderation analyses, we found that neither the magnitude nor direction of the relationship between social determinants of health class and maternal morbidity in this sample were affected by mode of labor onset, mode of delivery or hypertensive disorders. However, cervical dilation at the time of hospital admission was associated with lower maternal morbidity in both Class 2 (among participants who were induced) and Class 6 (among participants with spontaneous labor onset). Thus, our research provides more nuanced guidance for clinicians on which groups of people might best benefit from interventions to prevent maternal morbidity. Strategies that promote admission during the active phase of labor might be one strategy to prevent MM. Indeed researchers have found that active phase labor admission is associated with fewer birth-related interventions (including cesarean birth) as well as lower costs [50]. Use of a home-like waiting area where early labor can be supported without being admitted to the unit (i.e. labor lounge) may be one specific (albeit low-tech) intervention for maternal morbidity reduction [51], [52]. Other options to help support younger expectant parents throughout pregnancy and prepare for birth include use of virtual visits and home nurse visiting programs. In mediation analyses, our finding indicate that interventions targeting unplanned cesarean birth would be potentially useful in healthy nulliparous people who are similar to Class 2 members. Strategies for reducing primary Cesarean birth vary along the spectrum from promoting physiologic birth and continuous labor support among healthy individuals [45]–[47] to considering labor induction in the early term gestation without medical indication [48], [49].

Importantly, BMI did not mediate the relationship between SDoH and MM, but SDoH did predict higher BMI (which has been independently linked to higher Cesarean use or abnormal labor progress) [53], [54]. Given the relatively young age of the members of Class 6, interventions would require promoting pre-pregnancy health among early adolescents and young adults as part of an overall public health strategy. Furthermore, given that higher BMI, earlier in life, has been associated with socio-economic factors in particular, broader community-based interventions may be necessary [55]. Given the lower diet quality scores in Class 6, support could also come in the form of counseling from a dietician, supplemental food programs or smartphone-based applications that support dietary decision-making, geared toward adolescents.

Given that participants in our sample were healthy at pregnancy onset, our finding that approximately 1 in 16 births involved a postpartum maternal morbidity complication draws attention to this important and understudied group for maternal morbidity reduction interventions. Importantly, Class 6, which was composed of the youngest participants with lowest income and highest stress, had higher odds for postpartum MM than any other group. These findings are fairly consistent with other data reporting increased odds of certain peripartum complications (postpartum hemorrhage) for individuals giving birth who are under 20 years of age [56]. While advanced maternal age (≥35) has been discussed widely and implicated in rising MM outcomes [42], [57], maternal age itself did not emerge in our data as a moderator nor mediator of the outcome. Overall, birth rates among adolescents and young adults are falling in the United States; however, these individuals remain a vulnerable birthing population. During the nuMoM2b study period of 2010–2015, births to individuals under 20 years went from 9.3% to 5.8% of overall births, and births to persons aged 20–24 went from 23.8% to 20.4% according to CDC vital statistics reports [58]. Despite these changes with time, greater than 1:5 births occur to mothers of younger aged individuals overall, with over 30% of individuals in this younger age group being non-Hispanic Black and Hispanic. Thus, younger aged populations remain an important focus, particularly considering the intersection of these individuals age with other SDoH forces.

Similar to our findings for Class 6 individuals, the association between Class 2 SDoH assignment (U.S. immigrants with lower levels of English language use) and postpartum MM has also been reported in other studies; however, many of these studies focused on preterm birth[59]. However, in analyses of births from New York and Washington state, disparate maternal morbidity or unplanned cesarean birth outcomes among immigrant or migrant pregnant individuals were also reported [60], [61]. Although investigators of a California study did not find differences in maternal outcomes by US vs. foreign-born Hispanic history [62], other wealthy countries have noted maternal morbidity disparities among migrant/immigrant pregnant individuals [63], [64]. Social isolation or other factors (i.e. access to interpretive services) may play a role in these outcomes, especially relating to our observation that chorioamnionitis was an especially strong predictor of maternal morbidity among Class 2 members. Possibly, better communication with pregnant people similar to Class 2 members could improve prenatal education on signs/symptoms of membrane rupture or reduce the use of cervical examinations during labor—both of which could contribute to lower rates of chorioamnionitis diagnoses and downstream maternal morbidity events.

Strengths

The strengths of this study include the use of a large, prospectively collected dataset that was broadly representative of the population of the United States and also the SDoH framework and method of analysis using latent mixture modeling. Another key strength of this study is the use of presenting race/ethnicity, as reported by each participant, for the SDoH groups. Unlike many other investigations, the NuMo2b team collected information on participants’ racial and ethnic background, their self-identified race/ethnicity, and the race/ethnicity that others most often perceived them to be [32]. Our use of others’ perceptions of participants’ race/ethnicity in this analysis rather than racial background allows the opportunity to consider discrimination based on these participants’ physical attributes; which may, in turn, have contributed to levels of stress or other measured variables in the SDoH phenotype.

Limitations

This secondary analysis, like all other studies with this design, is limited by the variables and design of the original study. Although NuMoM2b investigators collected a range of variables describing the social circumstances of participants, there were several missing areas. For example, the nuMoM2b dataset did not include data on neighborhood-level SDoH factors like access to healthy food options, neighborhood violence, etc. Increasingly, health disparities researchers stress the importance of collecting community-level information from participants, in addition to individual- and interpersonal- level variables [21], [65]. Many neighborhood descriptors can be quantified using participants’ full zip code [66]. Although funded by the NIH, the nuMoM2b dataset available for secondary analyses did not include variables describing community (zipcode/built environment) influences. In addition, this analysis was limited by the lack of information on lifecourse factors like measures of childhood adversity in the nuMoM2b dataset. It is crucial that researchers prioritize collection of the full scope of SDoH variables when developing future studies.

Conclusions

While latent class analysis has been used to examine health disparities and the influence of psychosocial stress in other perinatal contexts [67], we believe this is the first study to examine maternal morbidity by individuals’ intersectional SDoH phenotypes. In a previous analysis by our team, latent class analysis had proven a useful approach when examining maternal pregnancy outcomes in relation to complex groups of predictors [68], [69]. Future research and interventions aimed at improving maternal health should consider the social, economic, stress and educational environment in which pregnancy is occurring to better prevent maternal morbidity, particularly in those individuals who appear healthy at the onset of pregnancy.