Novel Approach to Identify Severe Maternal Morbidity Clusters: A Latent Class Analysis

Abstract

Objective

Whether clusters exist within severe maternal morbidity (SMM), a set of life-threatening heterogeneous conditions, is not known. Our primary objective was to identify SMM clusters using a data-driven clustering technique, their associated predictors and outcomes.

Study Design

From 2008 to 2017, we used a delivery database supplemented by state data and medical record abstraction from a single institution in Pennsylvania. To identify SMM clusters, we applied latent class modeling that included 23 conditions defined by 21 Centers for Disease Control SMM indicators, intensive care unit (ICU) admission, or prolonged postpartum length of stay. Logistic regression models estimated risk for SMM clusters and associations between clusters and maternal and neonatal outcomes.

Results

Among 97,492 deliveries, 2.7% (N = 2,666) experienced SMM by any of the 23 conditions. Four clusters were identified as archetypes of SMM. Deliveries labeled as Hemorrhage (37.7%, N = 1,004) were characterized by blood transfusions and sickle cell anemia; Critical Care (28.1%, N = 748) by ICU admission and amniotic embolism; Vascular (24.5%, N = 654) by cerebrovascular conditions; and Shock (9.8%, N = 260) by ventilatory support and shock. Hypertensive disorders of pregnancy, depression, and Medicaid insurance were associated with Shock cluster. People in all clusters had a high risk of maternal death within 1 year (odds ratio: 12.0, 95% confidence interval: 6.2–23). Infants born to those in the shock cluster had the highest odds of neonatal death, low Apgar scores, and neonatal ICU admission.

Conclusion

We identified four novel SMM clusters that may help understand the collection of conditions defining SMM, underlying pathways and the importance of comorbidities such as depression and social determinants of health markers that amplify the well-established risk factors for SMM such as hypertensive disorders of pregnancy.

Key Points

• A total of 2.7% of deliveries experienced SMM events.

• There are four distinct SMM clusters: Hemorrhage, Critical Care, Vascular, and Shock.

• Not all SMM clusters bear the same risk for adverse perinatal outcomes.

Keywords

maternal morbidity - clusters - hypertensive disorders of pregnancy - depression

Severe maternal morbidity (SMM) is a set of heterogeneous conditions that pose a high risk for adverse effects on maternal health, including death.[1] Despite ongoing improvements in obstetric care, SMM cases have risen 75% in the last decade in the United States, affecting more than 50,000 individuals annually.[2] The classical single-disease paradigm might not be adequate for people who experience heterogeneous SMM events. Therefore, understanding clusters of SMM remains key to accurately identify cases and to guide research and clinical efforts to reduce maternal morbidity.

Using administrative data, prior clinical approaches have clustered SMM by the number of SMM Centers for Disease Control (CDC) indicators, surgical or anesthesia complications, and other events.[3] [4] True SMM cases are those determined by physician adjudication and chart review; however, this is labor-intensive and not efficient for large-scale SMM surveillance. To address these limitations in diagnosing SMM, we take a data-driven approach to identify clinically interpretable clusters of SMM from a larger collection of related conditions that may be linked to underlying disease burden and thus may offer a novel approach to surveil and understand SMM cases.

Latent class analysis (LCA) is a statistical method used to identify hidden subgroups within a population. In clinical research, LCA can help finding distinct groups of patients who share similar characteristics or behaviors, even when these groups are not obvious. LCA methods have been widely used in social sciences and clinical practice, for instance, in the identification of clusters of acute respiratory distress syndrome and asthma that have distinct biological characteristics, outcomes, and implications for treatment.[5] [6] [7] Therefore, LCA has several important features; it can support clinical interpretation regarding the number of classes and their dominant features and through model fit statistics inform the appropriate number of classes.

Notably, it is unknown whether clusters exist for SMM. The present study addressed this gap by conducting an LCA of SMM indicators. We hypothesized that people with SMM exhibit a few distinct clusters that may help identify SMM cases. These efforts may yield clinically meaningful clusters to facilitate the development of strategies to reduce rates and complement existing SMM surveillance tools.

Materials and Methods

Study Design, Population, and Setting

We included people who delivered (livebirth and stillbirth) at Magee-Women's Hospital (MWH) from 2008 to 2017. MWH is a single tertiary care referral institution with nearly 11,000 deliveries per year, an obstetric and general intensive care unit (ICU) and level III neonatal ICU (NICU). Delivery data were extracted from the electronic medical records (EMR) after birth into the Magee Obstetric Maternal and Infant delivery (MOMI) database. This study was approved by the University of Pittsburgh Institutional Review Board (STUDY20050303); data were deidentified and no consent was required.

Maternal Characteristics and Comorbidities

From MOMI, we evaluated clinical comorbidities using the International Classification of Diseases codes. Characteristics from the EMR included self-reported race (African-American vs. Caucasian; other races were too few to analyze, [<5% of deliveries]), presence of hypertensive disorders of pregnancy (HDP), gestational diabetes, substance abuse, depression, Medicaid insurance coverage, and prepregnancy body mass index to classify obesity status. To include markers of social determinants of health (SDH) not typically found in the EMR, deliveries were linked to state birth records to categorize maternal and partner education and the receipt of food assistance.

Severe Maternal Morbidity Indicators and Model Selection

We defined 23 SMM indicators during a delivery hospitalization as the presence of any of the 21 CDC SMM conditions,[8] ICU admission during their delivery hospitalization,[4] or prolonged postpartum length of stay (PPLOS), defined as more than 3 standard deviations above the mean PPLOS by mode of delivery.[9] Deliveries with SMM indicators were identified from the EMR, each delivery in our database had one or more SMM indicators. We performed sensitivity analyses on the stability of clustering without combinations of any blood transfusion, medical intensive care unit (MICU), and PPLOS. To strengthen our results, we randomly selected 244 deliveries across each of 5 years positively screened for SMM by 23 indicators for manual chart review and physician adjudication (K.H.) following previously described methods[10] by the ACOG criteria of true SMM cases. We calculated the sensitivity of our screening approach with administrative data. Although SMM can arise before and after delivery, our study focused on events occurring during the delivery hospitalization consistent with most SMM research using administrative data.

Maternal and Neonatal Outcomes

State mortality records were linked to MOMI, and outcomes were defined as maternal death within 1 year after delivery and neonatal death within the first 28 days of life; preterm delivery (<37 weeks of gestation) and severe preterm (<32 weeks of gestation); small for gestational age (SGA) and severe SGA defined as birth weight <10^th percentile and < 3^rd percentile accounting for gestational age, respectively;[11] 5-minute Apgar scores <7; and admission to NICU at delivery hospitalization and NICU stay of more than 48 hours.

Statistical Analysis

The 23 SMM indicators were considered as binary variables in the LCA models. Next, we fitted a series of latent class models using deliveries that had one or more SMM indicators. To determine the optimal number of SMM classes, we identified models with the fewest clusters that are parsimonious, by identifying the model that generated the minimum class membership >5% of SMM cases, low entropy, and high Bayesian information criterion.[12] [13]

Each SMM delivery was assigned to the latent class with the highest membership posterior probability. Deliveries without SMM events were included in the SMM-free group. Once we established the number of clusters (classes), maternal characteristics associated with the derived SMM cluster were analyzed using descriptive statistics, a chi-square test for categorical variables and analysis of variance for continuous variables as appropriate. Using multinominal logistic regression, we then examined if a priori variables such as HDP, gestational diabetes, depression, and SDH factors[14] [15] [16] were associated with the likelihood of the SMM clusters using the SMM-free group as the reference. Next, we compared the proportions of maternal and neonatal outcomes across the SMM cluster.

We labeled our classes based on the combination of SMM indicators with probabilities of occurring >40% within the class. A p-value < 0.05 was considered significant. Statistical analyses were performed using SAS software, version 9.4 and R (R Core Team, 2020), version 3.5.1.[17]

Results

Severe Maternal Morbidity Exposure and Maternal Characteristics

A total of 97,492 deliveries were evaluated. SMM events were identified in 2,666 deliveries (2.7%). The most frequent SMM indicators were blood transfusion (46.1%), followed by ICU admission (43.3%) and PPLOS (22.4%). The LCA model with four latent classes achieved our selection criteria ([Supplementary Fig. S1], available in the online version). We examined the frequency of SMM indicators in each class and the posterior probability of each indicator belonging to the SMM class ([Fig. 1] and [Supplementary Table S1], available in the online version).

Fig. 1 Probability of SMM indicators in each SMM cluster, Magee Obstetric Maternal and Infant delivery database, 2008 to 2017. The latent classes and probabilities of each SMM indicator were identified using latent class analysis. Labels for each cluster were based on the combination of SMM indicators that have probabilities >40% within the latent class. All (N = 97,492), Hemorrhage (N = 1,004), Critical Care (N = 748), Vascular (N = 654), Shock (N = 260). SMM, severe maternal morbidity.

Four SMM classes were labelled according to the SMM clinical features: Hemorrhage (N = 1,004, 37.7%) by blood transfusions (81.5%) and sickle cell anemia (55.0%); Critical Care (N = 748, 28.1%) by ICU admission (64.8%) and amniotic embolism (66.7%); Vascular (N = 654, 24.5%) by puerperal cerebrovascular disorders (77.8%), aneurysm (100.0%); and Shock (N = 260, 9.8%) by shock (79.4%) and ventilatory support (81.0%), disseminated intravascular coagulation (61.1%), and sepsis (47.8%).

All individuals with SMM were more likely to receive food assistance, have Medicaid coverage, and to have high rates of HDP, gestational diabetes, and maternal comorbidities than those SMM-free. There were, however, differences in some characteristics between SMM clusters ([Table 1]). People in the Critical Care cluster had the highest frequency of HDP (63.6%) and gestational diabetes (10.6%). The Shock cluster was noteworthy as it included people with the highest frequencies of cardiopulmonary and neurological conditions. Additionally, these people had the highest rates of less than high school education (46.5%) and high rates of diagnosed depression (26.2%).

Table 1 Maternal characteristics of the derived severe maternal morbidity clusters, Magee Obstetric Maternal and Infant delivery database, 2008 to 2017

Maternal characteristics[a]	All	SMM-free	SMM clusters				SMM free- SMM clusters p-value	SMM clusters P-value
			Hemorrhage	Critical Care	Vascular	Shock
	N = 97,492	N = 94,826	N = 1,004	N = 748	N = 654	N = 260
African-American	19,687 (20.2%)	18,885 (19.9%)	303 (30.2%)	211 (28.2%)	197 (30.1%)	91 (35.0%)	<0.001	0.2
Age, y	29 (6)	29 (6)	29 (6)	29 (6)	29 (6)	30 (6)	0.1	0.5
Prepregnancy BMI, kg/m2	26.1 (6.5)	26.0 (6.4)	27.3 (7.5)	29.2 (8.7)	28.1 (7.5)	28.6 (8.8)	<0.001	0.001
Parity	1.0 (1.3)	1.0 (1.2)	1.2 (1.6)	1.0 (1.4)	1.0 (1.5)	1.5 (1.9)	<0.001	<0.001
Gravida	2.4 (1.6)	2.4 (1.6)	2.7 (2.0)	2.5 (1.9)	2.7 (2.1)	3.2 (2.5)	<0.001	<0.001
Gestational age at delivery	39.2 (7.2)	39.3 (7.1)	38.6 (10.4)	36.2 (9.0)	37.2 (9.4)	36.2 (7.5)	<0.001	<0.001
HDP	15,516 (15.9%)	14,262 (15.0%)	347 (34.6%)	476 (63.6%)	316 (48.3%)	115 (44.2%)	<0.001	<0.001
Gestational diabetes	5,483 (5.6%)	5,268 (5.6%)	49 (7.5%)	79 (10.6%)	15 (5.8%)	72 (7.2%)	<0.001	0.02
Cesarean delivery	29,107 (29.9%)	27,494 (29.0%)	693 (69.0%)	438 (58.6%)	305 (46.6%)	177 (68.1%)	<0.001	<0.001
Smoking during pregnancy	11,198 (13.1%)	10,840 (13.1%)	124 (14.5%)	107 (16.5%)	95 (16.7%)	32 (14.7%)	<0.001	0.6
Maternal comorbidities
Neurological disease	2,052 (2.1%)	1,874 (2.0%)	43 (4.3%)	70 (9.4%)	40 (6.1%)	25 (9.6%)	<0.001	<0.001
Cardiovascular disease	966 (1.0%)	753 (0.8%)	32 (3.2%)	72 (9.6%)	66 (10.1%)	43 (16.5%)	<0.001	<0.001
Pulmonary disease	8,970 (9.2%)	8,475 (8.9%)	140 (13.9%)	138 (18.4%)	104 (15.9%)	113 (43.5%)	<0.001	<0.001
Substance abuse	3,683 (3.8%)	3,532 (3.7%)	44 (4.4%)	43 (5.7%)	45 (6.9%)	19 (7.3%)	<0.001	0.1
Depression	10,138 (10.4%)	9,629 (10.2%)	160 (15.9%)	172 (23.0%)	109 (16.7%)	68 (26.2%)	<0.001	<0.001
Social determinants of health characteristics
Food assistance	26,453 (29.5%)	25,608 (29.3%)	313 (35.6%)	241 (35.9%)	216 (36.8%)	75 (34.9%)	<0.001	1.0
Medicaid	35,030 (35.9%)	33,728 (35.6%)	465 (46.3%)	360 (48.1%)	336 (51.4%)	141 (54.2%)	<0.001	0.1
Mother education (<HS)	25,835 (28.8%)	24,914 (28.5%)	325 (37.0%)	279 (41.5%)	217 (37.0%)	100 (46.5%)	<0.001	0.03
Partner education (<HS)	24,159 (31.7%)	23,427 (31.5%)	263 (37.8%)	210 (43.7%)	190 (42.7%)	69 (43.7%)	<0.001	0.2

Abbreviations: BMI, body mass index; HDP, hypertensive disorders of pregnancy; SMM, severe maternal morbidity; <HS, high school or less.

^a Numbers provided are frequency (percent), mean (SD, standard deviation). A p-value of <0.05 is considered statistically significant.

After accounting for maternal and SDH characteristics, the occurrence of HDP was associated with increased risk of all SMM clusters but was particularly high for the Critical Care and Vascular groups (adjusted odds ratio [aOR]: 7.3 [95% confidence interval, CI: 5.9–9.1]; 4.0 [95% CI: 3.2–4.9], respectively; [Table 2]). African-American race and maternal education were associated with membership in the Hemorrhage cluster (aOR: 1.6 [95% CI: 1.3–2.0]; 1.3 [95% CI: 1.1–1.7]). Gestational diabetes was associated with both the Critical Care and Vascular clusters. The risk of belonging to the Shock cluster was particularly high for people with depression (aOR: 2.7 [95% CI: 1.8–4.2) and Medicaid coverage (aOR: 1.9 [95% CI: 1.0–3.5]). Smoking and receipt of food assistance were found to be associated with a lower likelihood of membership in this group (aOR: 0.5, 0.6, respectively).

Table 2 Multivariable multinominal logistic regression analysis of the risk of severe maternal morbidity clusters by maternal and socioeconomic characteristics[a], Magee Obstetric Maternal and Infant delivery database, 2008 to 2017

Maternal characteristics	Hemorrhage	Critical Care	Vascular	Shock
African-American vs. Caucasian	1.6	1.1	1.2	1.6
	(1.3–2.0)	(0.9–1.5)	(0.9–1.6)	(0.9–2.6)
Hypertensive disorder of pregnancy	2.1	7.3	4.0	3.0
	(1.8–2.6)	(5.9–9.1)	(3.2–4.9)	(2.1–4.3)
Gestational diabetes	1.1	1.4	1.4	0.6
	(0.8–1.6)	(1.0–2.0)	(1.0–2.0)	(0.3–1.5)
Smoking during pregnancy	0.8	1.1	0.9	0.5
	(0.6–1.1)	(0.8–1.6)	(0.6–1.2)	(0.3–1.0)
Substance abuse	0.8	1.1	1.3	1.5
	(0.5–1.4)	(0.6–1.8)	(0.7–2.2)	(0.8–3.1)
Depression	1.6	1.6	1.1	2.7
	(1.3–2.1)	(1.2–2.1)	(0.8–1.6)	(1.8–4.2)
Mother education (<HS)	1.3	1.1	0.9	1.7
	(1.1–1.7)	(0.9–1.5)	(0.7–1.2)	(0.9–2.9)
Partner education (<HS)	0.9	1.2	1.1	0.8
	(0.7–1.1)	(0.9–1.5)	(0.9–1.4)	(0.5–1.4)
Food assistance	1.0	1.0	1.0	0.6
	(0.8–1.3)	(0.7–1.3)	(0.7–1.2)	(0.3–0.9)
Medicaid	1.0	0.9	1.5	1.9
	(0.7–1.2)	(0.6–1.2)	(1.2–2.0)	(1.0–3.5)

Abbreviations: SMM, severe maternal morbidity; <HS, high school or less.

^a Estimates are exponentiated coefficients from a single multinomial regression analysis for each SMM class. The dependent variable is a discrete variable with five distinct values: (1) no SMM during the delivery hospitalization; (2) Hemorrhage; (3) Critical Care; (4) Vascular; (5) Shock clusters. Independent variables included in the final model were African-American race, hypertensive disorders of pregnancy, gestational diabetes, mother and partner education less than high school, smoking history, substance abuse, depression, food assistance, Medicaid, and gestational age. Values represent odds ratios (95% confidence interval).

Association of Severe Maternal Morbidity Clusters with Maternal and Neonatal Outcomes

Maternal mortality within 1 year of delivery was 5 per 10,000 deliveries overall. Individuals with SMM had a 12-fold increase in the odds of experiencing maternal mortality compared with those SMM-free (odds ratio: 12.0, 95% CI: 6.2–23). Maternal and infant mortality rates were high among people with SMM, regardless of the SMM cluster. When comparing SMM clusters, neonates born to people in the Shock cluster had the highest rates of low Apgar scores, NICU admission, and an NICU stay longer than 48 hours and being delivered severely preterm, whereas those in the Critical Care cluster had higher frequencies of SGA, severe SGA, and preterm birth ([Table 3]).

Table 3 Maternal and neonatal outcomes based on severe maternal morbidity clusters, Magee Obstetric Maternal and Infant delivery database, 2008 to 2017

Clinical outcomes	All	SMM-free	Hemorrhage	Critical Care	Vascular	Shock	SMM free- SMM clusters p-value	SMM clusters p-value
	N = 97,492	N = 94,826	N = 1,004	N = 748	N = 654	N = 260
Maternal outcomes
Preterm[a]	11,492 (11.8%)	10,330 (10.9%)	317 (31.6%)	426 (57.0%)	281 (43.0%)	138 (53.1%)	<0.001	<0.001
Severe preterm[b]	3,195 (3.3%)	2,724 (2.9%)	129 (12.8%)	171 (22.9%)	105 (16.1%)	66 (25.4%)	<0.001	<0.001
Maternal death[c] (per 10,000)	5	4	30	40	61	77	<0.001	0.71
Neonatal outcomes
Infant death[d]	468 (0.5%)	414 (0.4%)	22 (2.2%)	12 (1.6%)	13 (2.0%)	7 (2.7%)	<0.001	0.67
5-min Apgar < 7	1,724 (1.8%)	1,526 (1.7%)	58 (6.1%)	66 (9.2%)	40 (6.3%)	34 (14.9%)	<0.001	<0.001
NICU[e]	12,915 (13.2%)	11,853 (12.5%)	308 (30.7%)	355 (47.5%)	274 (41.9%)	125 (48.1%)	<0.001	<0.001
NICU > 48 h	2,2520 (23.1%)	21,165 (22.3%)	393 (39.1%)	451 (60.3%)	350 (53.5%)	161 (61.9%)	<0.001	<0.001
SGA[f]	10,394 (10.7%)	9,966 (10.5%)	113 (11.3%)	153 (20.5%)	126 (19.3%)	36 (13.9%)	<0.001	<0.001
Severe SGA[g]	2,980 (3.1%)	2,866 (3.0%)	24 (2.5%)	37 (5.2%)	41 (6.4%)	12 (5.1%)	<0.001	0.001

Abbreviations: NICU, neonatal intensive care unit; SGA, small for gestational age; SMM, severe maternal morbidity.

^a Preterm is defined as delivery at <37 weeks of gestation.

^b Severe preterm as delivery <32 weeks of gestation.

^c Maternal death is any death within the first year of delivery.

^d Infant death is any death within the first 28 days of life.

^e NICU admission and NICU > 48 hours represent those who had an NICU stay longer than 48 hours.

^f SGA.

^g Severe SGAs are defined as birth weight <10th percentile and <3rd percentile accounting for gestational age, respectively. A p-value < 0.05 indicates statistical significance.

Validation of Severe Maternal Morbidity Screening Indicators and Latent Clustering

SMM cases screened by 23 indicators (MICU + PPLOS + 21 CDC SMM indicators) detected 64% true SMM cases. Screening for SMM using the CDC definition alone only detected 40% of true SMM cases.[9] This underscores that our study SMM definition criteria were able to detect SMM cases at a higher rate ([Fig. 2A]).

Fig. 2 Sensitivity of SMM screening using electronic medical records and ACOG criteria. (A) Sensitivity of detecting SMM deliveries using different combinations of screening indicators. (B) Sensitivity across SMM clusters using a 23-indicator screening. The solid line indicates the average sensitivity of a 23-indicator screen. Dashed line the average sensitivity using only the 21 CDC SMM indicators. All adjudicated cases (N = 244), Hemorrhage (N = 95), Critical Care (N = 91), Vascular (N = 47), Shock (N = 24). ACOG, American College of Obstetricians and Gynecologists; CDC, Centers for Disease Control; SMM, severe maternal morbidity.

Excluding any blood transfusion indicator from this screening definition for SMM reduces the sensitivity to 49 and 20%, respectively. The proportions of true SMM cases according to SMM clusters were also individually higher than the CDC criteria: Hemorrhage, 60%; Critical Care, 70%; Vascular, 49%; and Shock, 100% ([Fig. 2B]).

We also performed sensitivity analyses restricted to the first pregnancy and including the 21 CDC SMM indicators only ([Supplementary Table S2], available in the online version). Additionally, we fit an LCA model excluding deliveries that only received blood transfusions without additional SMM indicator (N = 1,892), which stably identifies the Critical Care, Vascular, and Shock clusters and excludes most of the cases from the Hemorrhage cluster ([Supplementary Fig. S2], available in the online version).

Discussion

We identified four distinct clusters of SMM using LCA based on the presence of 23 SMM indicators. Including MICU or prolonged length of stay to the 21 CDC indicators increases sensitivity to detect true SMM. We report four SMM clusters: Hemorrhage, Critical Care, Vascular, and Shock and examine main differences between SMM clusters in terms of clinical and SDH characteristics. Social, structural, and clinical factors appear to play a key role in the likelihood of experiencing SMM conditions, and therefore, understanding how SMM indicators cluster—and their predictors—is vital to developing new obstetric protocols targeting those at highest risk. Future research is needed to assess long-term consequences in this population to determine those in need of follow-up care.

In clinical practice, there is a need to develop robust and efficient protocols to surveil SMM. Current SMM surveillance focuses mainly on administrative data. Our new approach reveals a novel method of surveilling by using common diagnosis clustering within single person rather than considering individual diagnosis independently. This approach may improve health system approaches to identifying and examining SMM clusters as part of obstetrical quality initiatives to improve the detection of true SMM cases and design prevention and improvement strategies.

Although preliminary, underlying mechanisms could explain the clustering between the SMM indicators. Pregnancy is considered a stress test that can unmask poor cardiovascular physiology[18] affecting multiorgan systems. For that reason, those who clustered in the Vascular group could have preexisting vascular problems such as endothelial dysfunction, arterial stiffness, or subclinical atherosclerosis and share a common mechanism that predisposes them to SMM. In support of this paradigm, several studies have reported biomarkers (e.g., proangiogenic vascular endothelial growth factor) associated with generalized endothelial dysfunction, resulting in hypertension, renal and cerebral endotheliosis.[19] [20] Thus, identifying a set of SMM indicators in the clusters may lead to a better understanding of clinical complexity in pregnancy and form a basis for considering targeted therapies.

In addition to examining classic obstetric comorbidities as SMM risk factors, we were interested in evaluating depression as this is likely an important driver of SMM. Perinatal mood disorders contribute to adverse maternal outcomes[21] and our results extend these findings and demonstrate that depression is an important risk factor for SMM, especially cases in the Shock cluster, perhaps related to a proinflammatory state.[21] Only 50% of pregnant individuals[22] receive treatment for depression pointing to another important clinical care opportunity to address SMM risk.

We included the Critical Care cluster in our final analysis given the evidence that 45% of individuals who die in the peripartum period were admitted to the ICU.[4] Our data along with others reveal that HDP can contribute to up to 70% of ICU admissions,[3] perhaps due to the need for critical care management among hypertensive individuals that can range from invasive monitoring to mechanical ventilation. We also report that gestational diabetes was associated with an increased likelihood of assignment to this cluster, emphasizing the importance of frequent monitoring to address lifestyle modifications to improve outcomes among affected individuals. The reasons for ICU admission, not available in our data, merit closer examination as they could help identify underlying factors that likely contribute to this distinct cluster.[23] [24]

Racial disparities in SMM events have been reported, with African-American individuals experiencing higher rates of SMM events.[25] Our data indicated this may be especially true for events in the Hemorrhage cluster. This is also consistent with a prior study reporting that African-American individuals have a high risk of postpartum hemorrhage even when adjusting for comorbidities.[26] Race is a social construct and a proxy to measure racism; thus, this observed disparity requires further evaluation of other factors at the institutional level, including hospital quality, culture of leadership, use of evidence-based practices, patient safety, and perceived racism and discrimination that might explain these findings.[22]

Our study evaluated additional SDH markers including education, food stability, and access to affordable care.[27] While markers of disadvantage including lower maternal educational levels and Medicaid insurance coverage were associated with SMM clusters in our data as in other studies,[28] individuals who smoked or needed food assistance were less likely to experience SMM events in the Shock cluster. Our findings regarding smoking and food assistance in this cluster are surprising. Food security could be enhanced via food assistance perhaps providing some protection from SMM events, but this is speculative and warrants further examination.

SMM events have been associated with adverse maternal and neonatal outcomes. Maternal mortality in the United States was reported to be 32.9 per 100,000 live births in 2021, whereas our study observed a higher rate.[29] This can be explained by the time-frame of follow-up; we included all deaths within 1 year after delivery, whereas the CDC restricts to deaths while pregnant or within 42 days after pregnancy. In addition, our tertiary institution cares for people with high-risk pregnancies and accepts referrals from other hospitals, and thus, rates may be expected to be higher. Of note, rates of maternal death were high across all SMM clusters, despite varying levels of alignment with adjudicated SMM cases.

Our study has strengths and limitations worth noting. Using a single-center cohort may limit generalizability. However, this is the first study to our knowledge that has grouped SMM indicators using an unbiased clustering approach. Although there is evidence that the higher the number of SMM indicators the greater the likelihood of maternal death,[4] our study expands this and distinguishes SMM clusters derived from patterns of these indicators. We also evaluated risk factors and maternal and neonatal outcomes associated with each cluster.

SMM indicators rely on administrative data, which are subject to miscoding and have limited data granularity. For example, those who required ICU admission might have needed different levels of ICU care ranging from telemetry to cardiopulmonary support, and thus, true severity may vary. Similarly, any blood transfusion may not be considered a true SMM marker; for instance, the ACOG considers transfusion SMM if ≥4 blood units are given.[9] Despite these inherent limitations of using administrative data, we demonstrated that our administrative database was able to identify true SMM cases (range: 49–100%). We also observed elevated rates of maternal and infant mortality detected within each SMM cluster compared with SMM-free deliveries, further demonstrating the increased disease burden that these clusters carry. When removing blood transfusions without additional SMM indicators, we robustly reproduced clusters aligned with Critical Care, Vascular, and Shock.

The use of an administrative database that is only linked to the delivery hospitalization limits the possibility of evaluating SMM before or after the delivery; therefore, future studies should consider merging databases that include the full scope of maternity care. Our study data were observational thus preventing our ability to identify casual relationships. Diagnosis definitions have changed over time. For example, severe preeclampsia coding began only in 2012, and it is possible that certain indicators were not considered before that time and were undercaptured.

Conclusion

Our results offer a new approach to understanding SMM and demonstrate the importance of comorbidities such as depression and SDH markers that amplify the well-established risk factors such as HDP. The Shock cluster was found to be driven by disadvantage and depression risk factors and SMM cases belonging to this class had the highest odds of adverse neonatal outcomes. Our findings can inform future research on pathways and patterns of risk amenable to intervention to prevent SMM and its devastating maternal and infant consequences.

Conflict of Interest

None declared.

• References

• 1 Callaghan WM, Mackay AP, Berg CJ. Identification of severe maternal morbidity during delivery hospitalizations, United States, 1991-2003. Am J Obstet Gynecol 2008; 199 (02) 133.e1-133.e8
• 2 Callaghan WM, Creanga AA, Kuklina EV. Severe maternal morbidity among delivery and postpartum hospitalizations in the United States. Obstet Gynecol 2012; 120 (05) 1029-1036
• 3 Kilpatrick SK, Ecker JL. American College of Obstetricians and Gynecologists and the Society for Maternal–Fetal Medicine. Severe maternal morbidity: screening and review. Am J Obstet Gynecol 2016; 215 (03) B17-B22
• 4 Ray JG, Park AL, Dzakpasu S. et al. Prevalence of severe maternal morbidity and factors associated with maternal mortality in Ontario, Canada. JAMA Netw Open 2018; 1 (07) e184571-e184571
• 5 Soussi S, Sharma D, Jüni P. et al; FROG-ICU, CCCTBG trans-trial group study for InFACT—the International Forum for Acute Care Trialists. Identifying clinical subtypes in sepsis-survivors with different one-year outcomes: a secondary latent class analysis of the FROG-ICU cohort. Crit Care 2022; 26 (01) 114
• 6 Depner M, Fuchs O, Genuneit J. et al; PASTURE Study Group. Clinical and epidemiologic phenotypes of childhood asthma. Am J Respir Crit Care Med 2014; 189 (02) 129-138
• 7 Calfee CS, Delucchi K, Parsons PE, Thompson BT, Ware LB, Matthay MA. NHLBI ARDS Network. Subphenotypes in acute respiratory distress syndrome: latent class analysis of data from two randomised controlled trials. Lancet Respir Med 2014; 2 (08) 611-620
• 8 National Institute of Health Office of Research on Women's Health. Maternal morbidity and mortality. What do we know? How are we addressing it?. 2020 . Accessed July 6, 2022 at: www.nih.gov/women/maternalhealth
• 9 Main EK, Abreo A, McNulty J. et al. Measuring severe maternal morbidity: validation of potential measures. Am J Obstet Gynecol 2016; 214 (05) 643.e1-643.e10
• 10 Himes KP, Bodnar LM. Validation of criteria to identify severe maternal morbidity. Paediatr Perinat Epidemiol 2020; 34 (04) 408-415
• 11 Alexander GR, Kogan M, Bader D, Carlo W, Allen M, Mor J. US birth weight/gestational age-specific neonatal mortality: 1995-1997 rates for Whites, Hispanics, and Blacks. Pediatrics 2003; 111 (01) e61-e66
• 12 Nylund KL, Asparouhov T, Muthén BO. Deciding on the number of classes in latent class analysis and growth mixture modeling: a Monte Carlo simulation study. Struct Equ Modeling 2007; 14 (04) 535-569
• 13 Weller BE, Bowen NK, Faubert SJ. Latent class analysis: a guide to best practice. J Black Psychol 2020; 46 (04) 287-311
• 14 Zhang J, Meikle S, Trumble A. Severe maternal morbidity associated with hypertensive disorders in pregnancy in the United States. Hypertens Pregnancy 2003; 22 (02) 203-212
• 15 Grigoriadis S, VonderPorten EH, Mamisashvili L. et al. The impact of maternal depression during pregnancy on perinatal outcomes: a systematic review and meta-analysis. J Clin Psychiatry 2013; 74 (04) e321-e341
• 16 Metzger BE, Lowe LP, Dyer AR. et al; HAPO Study Cooperative Research Group. Hyperglycemia and adverse pregnancy outcomes. N Engl J Med 2008; 358 (19) 1991-2002
• 17 Wilson A, Norden N. The R Project for Statistical Computing. 2015 . Accessed December 8, 2021 at: https://www.r-project.org/
• 18 Craici I, Wagner S, Garovic VD. Preeclampsia and future cardiovascular risk: formal risk factor or failed stress test?. Ther Adv Cardiovasc Dis 2008; 2 (04) 249-259
• 19 Ali SM, Khalil RA. Genetic, immune and vasoactive factors in the vascular dysfunction associated with hypertension in pregnancy. Expert Opin Ther Targets 2015; 19 (11) 1495-1515
• 20 Shah DA, Khalil RA. Bioactive factors in uteroplacental and systemic circulation link placental ischemia to generalized vascular dysfunction in hypertensive pregnancy and preeclampsia. Biochem Pharmacol 2015; 95 (04) 211-226
• 21 McKee K, Admon LK, Winkelman TNA. et al. Perinatal mood and anxiety disorders, serious mental illness, and delivery-related health outcomes, United States, 2006-2015. BMC Womens Health 2020; 20 (01) 150
• 22 Farr SL, Hayes DK, Bitsko RH, Bansil P, Dietz PM. Depression, diabetes, and chronic disease risk factors among US women of reproductive age. Prev Chronic Dis 2011; 8 (06) A119
• 23 Rossi RM, Hall E, Dufendach K, DeFranco EA. Predictive model of factors associated with maternal intensive care unit admission. Obstet Gynecol 2019; 134 (02) 216-224
• 24 James AH. The prevention and management of thrombosis in obstetrics and gynecology. Clin Obstet Gynecol 2018; 61 (02) 203-205
• 25 Creanga AA, Bateman BT, Kuklina EV, Callaghan WM. Racial and ethnic disparities in severe maternal morbidity: a multistate analysis, 2008-2010. Am J Obstet Gynecol 2014; 210 (05) 435.e1-435.e8
• 26 Gyamfi-Bannerman C, Srinivas SK, Wright JD. et al. Postpartum hemorrhage outcomes and race. Am J Obstet Gynecol 2018; 219 (02) 185.e1-185.e10
• 27 Crear-Perry J, Correa-de-Araujo R, Lewis Johnson T, McLemore MR, Neilson E, Wallace M. Social and structural determinants of health inequities in maternal health. J Womens Health (Larchmt) 2021; 30 (02) 230-235
• 28 Gray KE, Wallace ER, Nelson KR, Reed SD, Schiff MA. Population-based study of risk factors for severe maternal morbidity. Paediatr Perinat Epidemiol 2012; 26 (06) 506-514
• 29 Hoyert DL. Maternal mortality rates in the United States, 2020. Natl Cent Health Stat 2021; (03) 1

Publikationsverlauf

Eingereicht: 26. Oktober 2023

Angenommen: 11. September 2024

Artikel online veröffentlicht:
08. Oktober 2024