Placental Abruption and Cardiovascular Event Risk (PACER): Design, Data Linkage, and Preliminary Findings

Abstract

Background:

Obstetrical complications impact the health of mothers and offspring along the life course, resulting in an increased burden of chronic diseases. One specific complication is abruption, a life-threatening condition, with consequences on cardiovascular health that remain poorly studied.

Objectives:

To describe the design, and data linkage algorithms for the Placental Abruption and Cardiovascular Event Risk (PACER) cohort, and to provide descriptive preliminary findings on rates of abruption and fatal and non-fatal cardiovascular disease (CVD) in pregnant persons and their offspring.

Population:

All subjects that delivered in New Jersey, USA, between 1993 and 2020.

Design:

Retrospective, population-based, birth cohort study.

Methods:

We linked the vital records data of fetal deaths and live births to delivery and all subsequent hospitalisations along the life course for birthing persons and newborns. The linkage was based on a probabilistic record-matching algorithm.

Preliminary results:

Over the 28 years of follow-up, we identified 1,877,824 birthing persons with 3,093,241 deliveries (1.1%, n = 33,058 abruption prevalence). The linkage rates for live births-hospitalisations and fetal deaths-hospitalisations were 92.4% (n = 2,842,012) and 70.7% (n = 13,796), respectively, for the maternal cohort. The corresponding linkage rate for the live births-hospitalisations for the offspring cohort was 70.3% (n = 2,160,736). The median (interquartile range) follow-up for the maternal and offspring cohorts was 15.4 (8.1, 22.4) and 14.4 (7.4, 21.0) years, respectively. We will undertake multiple imputations for missing data, and develop inverse probability weights to account for selection bias owing to unlinked records.

Conclusions:

Pregnancy offers a unique window to study chronic diseases along the life course, and efforts to identify the aetiology of abruption may provide important insights into the causes of future CVD. This project presents an unprecedented opportunity to understand how abruption may predispose women and their offspring to develop CVD complications and chronic conditions later in life.

Background

Cardiovascular disease remains the leading cause of death in women worldwide, accounting for up to a third of all deaths in 2019.¹ Events that occur during pregnancy have a lasting impact on the health of mothers and newborns along the life course. Preeclampsia, placental abruption, and small for gestational age (SGA) birth, conditions that constitute the syndrome of “ischaemic placental disease,”^2–4 as well as gestational diabetes and preterm delivery, are obstetrical complications that confer increased risks of maternal chronic diseases later in life.⁵

In uncomplicated pregnancies, placental separation from the uterus occurs immediately after birth, while in pregnancies complicated by abruption, the placenta detaches prematurely.⁶ Placental abruption affects 0.7–1.2% of pregnancies, with 10-fold higher rates of perinatal mortality than normal pregnancies.^7–9 The occurrence of abruption in one pregnancy, however, is associated with substantially increased recurrence.^10–12 Pregnancies diagnosed with abruption end 3–4 weeks earlier than other uncomplicated pregnancies,¹³ with over half of abruption births resulting in preterm delivery.^14–17 It is the strongest known “trigger” of spontaneous labour and premature rupture of membranes,^{13, 18} contributing to high rates of preterm birth;^{7, 19} 55% of the excess perinatal deaths associated with abruption are due to early delivery (<37 weeks), and an additional 9% to intrauterine growth restriction.⁸ Fetal haemorrhage and hypoxia also contribute to increased risk.^20–22 Acute maternal risks associated with abruption include disseminated intravascular coagulopathy, haemorrhage,^23–28 and hysterectomy.^{21, 23} The majority of cases of abruption are idiopathic and not amenable to prophylactic intervention, leading to extreme anxiety regarding its high recurrence in future pregnancies.

The aetiology of abruption remains elusive,²³ but it may be the end result of an acute process, the final culmination of long-standing chronic processes or both.^{23, 29} Mechanical forces to the abdomen can lead to premature placental separation (acute process), or abruption can result from a chronic aetiology (thrombosis, inflammation, infection, and decidual and uteroplacental vasculopathy). Abnormal vascular remodeling, thrombosis/coagulation, infection, and angiogenesis, which often involve inflammation, are essentially intermediate pathophysiological responses that lead to poor or inadequate placentation or accelerated placental detachment^{2, 20} (Figure 1).

Figure 1.

Conceptual model for placental abruption and cardiovascular disease in women and offspring (Adapted and modified, with permission, from Ananth CV and Kinzler WL. UpToDate 2022)

The red pathways indicate the ones that we will specifically examine in the PACER project

We designed this project to (i) estimate the association between abruption and risks of long-term fatal and non-fatal CVD events in women and newborns; (ii) examine the association of abruption and risks of CVD events in high-risk subsets: women with 2 or more abruptions (recurrence); and women that delivered twins; (iii) to evaluate the extent to which the association between abruption and CVD events is mediated through preterm delivery, as well as preterm obstetrical interventions; and (iv) examine whether race/ethnicity and socioeconomic status modify the association between abruption and risks of CVD events.

Methods

Design and cohort composition

Placental Abruption and Cardiovascular Event Risk (PACER) is a retrospective cohort study of all subjects who delivered a fetal death or live birth in the state of New Jersey, USA, between 1993 and 2020. Pregnancies and deliveries were individually linked to mother and newborn delivery hospitalisations as well as all subsequent hospitalisations in New Jersey over the study period (Figure 2).

Figure 2.

Flow chart describing the data organisation and cohort composition: Placental Abruption and Cardiovascular Event Risk (PACER), 1993–2020

The boxes shown in dotted lines indicate that the linkage of these databases is planned. Fetal death files for 2016–2017, and 2019- and 2020 were unavailable at the time of the linkages.

CVD, cardiovascular disease; MI, myocardial infarction

Vital records data

We utilised the NJ electronic birth records for live births (1993–2020), and fetal death (1993–2015, and 2018; the 2016–2017 data are not available from the New Jersey Department of Health, and the 2019–2020 records were unreleased at the time of data linkages) certificates linked to their corresponding hospital delivery discharge records from. These data include all births in the state, ascertained through the 1989 version (1993 to 2013) and 2003 revised (since 2014) live birth and fetal death certificates. Although we received vital records, mortality, and hospitalisation records starting in 1985, the paucity of linkage variables in these early years precluded linkages between 1985–1992. Therefore, the PACER cohort only included data starting in 1993. The cohort construction was restricted to births that occurred at ≥20 completed weeks’ gestation. These include rich and detailed data on maternal and paternal sociodemographic factors, medical conditions, obstetrical complications, labour and delivery characteristics, and the course and outcome of pregnancies.

NJ Hospital Discharge Data Collection System (NJDDCS)

The NJDDCS comprises data from NJ hospitals, including information on admission and discharge dates, hospital code, services rendered, charges, and type of insurer. New Jersey inpatient, emergency department, and other outpatient discharge data files are derived from hospital uniform billing (UB) information. The UB information is used to electronically submit claims for health care provided in an institutional setting to payers as well as to exchange claims information between payers. Data fields contained in the NJDDCS include information about the admission hospital (urban/rural location, number of hospital beds, hospital size).

Mortality data

The Archive Mortality File from the New Jersey Department of Health’s Center for Health Statistics and Informatics contains mortality information confirmed by up to three sources. The source code (source of data) includes both the single and multiple and the out-of-state mortality data downloaded from the State and Territorial Exchange of Vital Events platform (https://www.steve2.org/). Some out-of-state observations may be contained in the single or multiple cause-of-death fields, but not found in out-of-state mortality data during certain time periods. Deaths can have a value ranging from zero (no source code) to all the above three confirming death. Data fields include a clinical estimate of gestation, birth order, congenital anomalies, and if an autopsy was performed.

Linkage Structure

Each entry was provided a unique identifier (PACER-ID) referencing the data file, year, and a randomly assigned 7-digit number. Matching was completed across datasets and years by examining identifiers including first name, middle name or initial, last name, maiden name, year, sex, date of birth, hospital code, county code, and municipality code. To match by hospital codes, the hospital coding scheme was standardised across the NJ vital records and NJDDCS. The NJ Vital records linked hospitals across 3 formats: pre-Electronic Birth Certificate (EBC) 1993–1998, EBC 1998–2013 (partial years of 2014–2015 as combined EBC and VIP), and VIP (Vital Information Platform) 2016 onwards coding systems. The mortality records did not contain a hospital or facility code for the hospital of death. A new PACER hospital code was generated that enabled linking a hospital through hospital name changes and ownership throughout the years. Hospitals within a health system (for example, different campuses associated with Robert Wood Johnson Health Network) were given the same newly generated PACER matching hospital code.

County and municipality coding schemes were different between the NJ Vital Records Data and NJDCCS data but were matched one-to-one on the provided county and municipality names within the data types. Linkage structure by patient name was approached by linking names directly, there was special consideration needed for the fetal death files with a large portion of first names relating to ‘Baby Boy’ or ‘Baby Girl’ variation. Date of birth was approached accounting for the different variables recorded as either date of birth in one variable, or spread across three variables as day, month, and year of birth.

Linkage algorithm

The NJ Department of Health (NJ-DOH) provided us with data on live births, fetal deaths, hospitalisations, and mortality for the years 1993 through 2020 (2018 for fetal deaths). The variables available and the structure of the variables were different both across file types and by year. We reviewed data files and identified a total of 16 different file structures among the 140 different annual files (e.g., live birth data 2004–2013 contained the same variables and variable structures; similarly, hospitalisations in 2008–2020 had the same structure).

Data processing and preparation

With the goal of linking each file type-year against all other year-file types (e.g., live births 2005 versus hospitalisations 2015) we created a standardised set of output variables from each of the 16 different structures which were named LB1-LB6 (live births), FD1-FD4 (fetal deaths), UB1-UB4 (hospitalisations), and MT1-MT2 (mortality). The standardisation included, where necessary, splitting the names of mother and child into separate first, middle initial, and last names; creating a standardised format for date of birth; computing the mother’s age where not available, and setting placeholder names such as “BABY”, “FETUS”, “INFANT” as missing. Given that the coding of several variables was inconsistent between files and years, the standardisation also included recoding these variables so that they used a consistent baseline set of codes. We recoded hospital and municipal IDs and harmonised race such that all 16 file types used the same 5-category race classification (White/Caucasian; Black/African-American; Asian; Native Hawaiian/Other Pacific Islander; or American Indian/Alaska Native).

The final set of standardised and consistent variables included social security number (SSN), patient’s date of birth, birth year, maiden name, last name, middle initial, race, age, residential municipality, county, and ZIP code, hospital, and date of admission (or death). In the case of the live birth data, the final variables also included the child’s date of birth, the child’s first and last name, the child’s middle initial, and sex. Importantly, not all variables were available for each of the 16 file types. For example, SSN was only provided for one file type.

Determination of linkage variables

Given the different availability and completeness of variables across the file types, we developed a schema for which variables would be included in the linkage process including which variables would be linked via string distance (fuzzy matching) and numeric window matched. As ZIP, county, and municipality are nested geographic codes and would provide similar information we opted to use only one of these variables as part of the linkage. Based on a review of the availability of the variables we opted to include the municipality in linkages.

There were two sets of circumstances where linkages were not performed. First, we did not perform linkages that would not logically yield matches – for example, births in 2000 against deaths in 1995. Second, given that the hospital may legitimately not match in many cases and that it, too, could be considered somewhat nested geographically with municipalities, we only included the hospital as a linkage variable if several other variables were not available. In total, we created a schema for 128 file type-to-file type linkages (e.g., LB2 [1999–2003] versus UB2 [1993–2003]).

Establishing linkages

We used the fastLink package in R³⁰ to perform a Fellegi-Sunter probabilistic record linkage model that allows for missing data. In linking records for the mothers, we limited the datasets to females and performed blocking on the year of birth. For child-related linkages, we used sex and year of birth for blocking. Names were matched using Jaro-Winkler string matching^{31, 32} using a threshold of 0.95. We also matched the full YYYYMMDD date of birth using string distance to allow for matches with small variations (e.g., “20120204” would match “20120203”). We used the default 0.1 parameter for upweighting the importance of the first characters (e.g., the “Cha” would get a higher weight than “rlie” in the name “Charlie”). We set a permissive overall match threshold of 0.75 to ensure we captured the maximum possible matches. No partial matching was allowed resulting in three possible results for each variable comparison: 0 representing a non-match, 1 representing a match, or NA if the values were missing.

Manual review of linkage patterns

As an additional validation of the linkage results, we performed a manual review of all linkage patterns identified in the probabilistic match.³³ As an example, a linkage pattern for linking births to hospitalisations might include DOB, last name, first name, middle initial, race, and municipality and result in a pattern of 0, 1, 0, 0, 1, 1, or 1, NA, 1, NA, 1, 1. We developed codes to randomly select pairs of records with a given linkage pattern for a manual review. For all linkage patterns in the results, we reviewed 5–30 examples and assigned a numerical score between 0–1. A total of over 4100 patterns were manually reviewed. Most records were assigned a manual score of 0, 0.25, 0.5, 0.75, or 1 representing obvious non-matches, mostly non-matches, half non-matches/half matches, mostly matches, or obvious matches.

Linkage validation and establishing final health records for individual women and offspring

The rationale for establishing linkages with small variation of DOB was to identify all potential true matches in the presence of miscoded DOB. However, this generated varying proportions of non-matches based on each file type. After a manual review of the linkages, the accuracy of matches was highly dependent on an exact match on the date of birth. This restriction to exact DOB would decrease the number of potential linked records but highly increase the accuracy of true linkages. Therefore, links with mismatched date of birth were not considered a true match while matches with missing date of birth for one or both file types were included. Delivery hospitalisation events (LB – UB / FD – UB) were identified by admission and discharge dates relative to the child’s date of birth or fetal death and pregnancy-related codes (see e-Appendix A table in MacDonald et al.³⁴). Data were processed to ensure each live birth or fetal death was linked to only one delivery hospitalisation and mortality record.

For the maternal cohort, a unique ID was created to assign pregnancies, hospitalisations, and death records by each individual. Any live birth or fetal death records assigned to multiple unique IDs were excluded from both final maternal and offspring datasets. For the offspring cohort, information from birth certificates, child hospitalisation, and corresponding maternal hospitalisation were combined. All matches regardless of probability score and manual match scores were included in the analyses since match scores were dependent on type of file. Maternal matches with fetal death records had lower match scores due to fewer patient identifiers available than that of LB files. Linkages with exact match on DOB were manually reviewed and validated to be true matches regardless of match score.The final datasets were reviewed for the accuracy of individual women and offspring health records across the study period (See Supplement for more details).

Exposure: Placental abruption

All data relating to medical and obstetrical complications, as well as CVD outcomes from the hospitalisation data files were coded based on the International Classification of Disease (ICD) version 9 (ICD-9) and version 10 (ICD-10) codes (Table S1).

Placental Abruption.

The primary exposure was abruption. The diagnosis of abruption is based on clinical findings, including women that present with vaginal bleeding, and accompanied by at least one of the following conditions: fetal distress, uterine tenderness, or uterine hypertonicity. If the placenta shows evidence of a tightly adherent clot consistent with retroplacental bleeding, or sonographic signs of abruption are present, the diagnosis of abruption is recorded.^{35, 36}

Severe placental abruption.

Severe abruption will be defined as abruption accompanied by at least 1 of the following complications: maternal (disseminated intravascular coagulation, hypovolemic shock, blood transfusion, hysterectomy, renal failure, or in-hospital death), fetal (non-reassuring fetal status, IUGR, or stillbirth), or neonatal (preterm delivery or SGA birth) outcomes.³⁵

Outcomes: CVD events

Cardiovascular events.

All heart disease and stroke events will be identified based on the ICD-9 (1993–2015) and ICD-10 (2016–2020) in the linked database. The primary ICD codes to identify the presence of CVD outcomes are as follows. For heart disease outcomes, we will examine ischaemic heart disease, chronic ischaemic heart disease, atherosclerotic heart disease, acute myocardial infarction, heart failure, pulmonary embolism, other pulmonary heart diseases, chronic rheumatic heart disease, cardiomyopathy, and other vascular diseases. Of note, heart failure will be assessed both as a single entity and, in a sensitivity analysis, by heart failure subtype (e.g., systolic, diastolic, and combined), owing to the known challenges with the specificity provided by coding for these subtypes.

Cerebrovascular events.

Stroke outcomes will include all strokes, as well as ischaemic stroke and haemorrhagic stroke types. We will also examine ischaemic stroke without transient ischaemic attack.

Statistical analysis

We undertook a descriptive analysis by comparing the distributions of maternal sociodemographic characteristics, chronic conditions, obstetrical complications, labour, and delivery events, as well as delivery, and infant outcomes in relation to abruption status.

Computing person-time

The PACER dataset involves time-varying exposures (e.g., abruption across multiple pregnancies) and recurrent time-to-event CVD outcomes (Figure 3). To estimate person-time, we will organise the data in a counting process style.³⁷ In the counting process structure the data will be expanded from one record per patient to one record per interval between each event time, per patient. This structure is motivated by the fact that the partial likelihood will include a contribution at each event time. We will update the covariate information at these times, but not in between.³⁸ We will use the survival package³⁹ in R (survSplit function) for this purpose. In the setting of competing events (e.g., death due to other causes), we will use the msprep function in the mstate package.⁴⁰

Figure 3.

Conceptual directed acyclic diagram depicting the pathways for estimating person-time: Placental Abruption and Cardiovascular Event Risk (PACER), 1993–2020

A denotes placental abruption across the first 3 pregnancies; H denotes hospitalisations for cardiovascular disease, and D denotes deaths from cardiovascular disease

Missing data

The PACER includes variables with missing data of two types: (i) Missing data in covariates (distinguished in the tables and figures); and (ii) Missing data because of unlinked vital statistics-hospitalisation records. To address missing covariate data, we will undertake multiple imputations in R using the mice package⁴¹ based on the fully conditional specification approach to create 25 imputed sets and integrate the results based on Rubin’s method.

Reweighting method

To address potential selection bias due to unlinked records, we will model the probability of unlinked records based on likelihood-based methods to derive an inverse probability of weighting (IPW) strategy.^{42, 43} Covariates that we will consider will include year, mother’s age, clinical gestational age as continuous variables, and placental abruption, fetal death, parity (1, 2, ≥3), mother’s race/ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, other), mother’s education (<8, 9–12, 13–16, ≥17 years of completed education), marital status (single, married), smoking status, primary insurance (Medicare, Medicaid, private, self-pay, others), number of prenatal visits (no care, 1–9, 10–12, ≥13), child’s sex (female, male), plurality (singletons, twins, triplets and higher order), diabetes (non-diabetic, pre-pregnancy, and gestational diabetes), and hypertensive disorders (normotensive, chronic hypertension, gestational hypertension, mild preeclampsia, severe preeclampsia, superimposed preeclampsia, eclampsia) as categorical variables. We will consider interaction terms of abruption and all other covariates, mother’s education by marital status, smoking, insurance status, and mothers age by marital status. Assessment of the model fit will be determined based on comparing the change in the Aikike Information Criterion (AIC) and deviance of nested models.

From this model, we will derive the estimated probability of unlinked records and calculate the IPW. This weight will be applied to all statistical analysis. Estimation of IPW will be accomplished using the ipw package⁴⁴ in R.

All programming for data linkages was done in R, data integration in SAS (version 9.4; SAS Institute Inc., NC), and statistical analysis in SAS and R.

Ethics approval

This study was approved by the institutional review boards of the Rutgers University Health Sciences (Pro2020001298), as well as Rowan University, NJ (Pro2020001058).

Results

Linkage characteristics

The layout of the data files, including the vital records data of live births, fetal deaths, and mortality, as well as the hospitalisation records are shown in Figure 2. Although hospitalisation records of CVD procedures, myocardial infarction-related procedures, and stroke registry for select years are available, we did not link these data files to the PACER master file. Figure 4 shows the linkage of the vital records files with the delivery hospitalisation records. We were able to successfully link 92.4% and 70.7% of the live birth and fetal death records, respectively, with their corresponding hospitalisation records.

Figure 4.

Flow chart describing the linkage of vital records with their corresponding hospitalisation data: Placental Abruption and Cardiovascular Event Risk (PACER), 1993–2020

Sociodemographic characteristics

Of a total of 3,093,241 deliveries between 1993 and 2020, 1.1% (n = 33,058) had a diagnosis of abruption (Table 1). While abruption rates were stable across the study period, the rates progressively increased with advancing maternal age at delivery, multiparity (parity ≥3), and persons that were African Americans, smokers, and those with no prenatal care. The distribution of chronic conditions, obstetrical complications, and labor, and delivery events in relation to abruption are shown in Table 2. Compared to singletons, abruption rates were higher among multiple gestations. Compared to normotensive women, the rates of abruption were higher among those with chronic hypertension, preeclampsia, and eclampsia. Abruption rates were also increased among women with pre-pregnancy, but not gestational, diabetes.

Table 1.

Distribution of maternal characteristics and risk factors for placental abruption at delivery: Placental Abruption and Cardiovascular Event Risk (PACER) Project, 1993–2020

	Total cohort	Placental abruption present		Abruption absent
	Number (%col)	Number (%col)	(%row)	Number (%col)
Total deliveries	3,093,241 (100.0)	33,058 (100.0)	(1.1)	3,060,183 (100.0)
Year of delivery
1993–1995	343,126 (11.1)	3493 (10.6)	(1.0)	339,633 (11.1)
1996–2000	564,332 (18.2)	6315 (19.1)	(1.1)	558,017 (18.2)
2001–2005	589,900 (19.1)	6200 (18.8)	(1.1)	583,700 (19.1)
2006–2010	574,084 (18.6)	6040 (18.3)	(1.1)	568,044 (18.6)
2011–2015	531,719 (17.2)	5441 (16.5)	(1.0)	526,278 (17.2)
2016–2020	490,080 (15.8)	5569 (16.8)	(1.1)	484,511 (15.8)
Maternal age (years)
<15	3066 (0.1)	39 (0.1)	(1.3)	3027 (0.1)
15–19	175,291 (5.7)	1793 (5.4)	(1.0)	173,498 (5.7)
20–24	487,942 (15.8)	5035 (15.2)	(1.0)	482,907 (15.8)
25–29	792,651 (25.6)	7787 (23.6)	(1.0)	784,864 (25.6)
30–34	976,944 (31.6)	10,112 (30.6)	(1.0)	966,832 (31.6)
35–39	529,672 (17.1)	6397 (19.4)	(1.2)	523,275 (17.1)
40–44	112,690 (3.6)	1634 (4.9)	(1.4)	111,056 (3.6)
45–49	7611 (0.2)	128 (0.4)	(1.7)	7483 (0.2)
≥50	794 (0.0)	20 (0.1)	(2.5)	774 (0.0)
Unknown	6580 (0.2)	113 (0.3)	(1.7)	6467 (0.2)
Gravida
1	913,882 (29.5)	8564 (25.9)	(0.9)	905,318 (29.6)
2	888,526 (28.7)	8373 (25.3)	(0.9)	880,153 (28.8)
≥3	1,225,872 (39.6)	15,881 (48.0)	(1.3)	1,209,991 (39.5)
Unknown	64,961 (2.1)	240 (0.7)	(0.4)	64,721 (2.1)
Parity
1	1,262,029 (40.8)	12,400 (37.5)	(1.0)	1249,629 (40.8)
2	1,035,411 (33.5)	10,340 (31.3)	(1.0)	1025,071 (33.5)
≥3	758,351 (24.5)	10,165 (30.7)	(1.3)	748,186 (24.4)
Unknown	37,450 (1.2)	153 (0.5)	(0.4)	37,297 (1.2)
Maternal race/ethnicity
Non-Hispanic White	1,465,699 (47.4)	14,213 (43.0)	(1.0)	1,451,486 (47.4)
Non-Hispanic Black	443,257 (14.3)	6904 (20.9)	(1.6)	436,353 (14.3)
Hispanic	709,469 (22.9)	7468 (22.6)	(1.1)	702,001 (22.9)
Other	442,162 (14.3)	4417 (13.4)	(1.0)	437,745 (14.3)
Unknown	32,654 (1.1)	56 (0.2)	(0.2)	32,598 (1.1)
Maternal education (years)
<8	128,828 (4.2)	1353 (4.1)	(1.1)	127,475 (4.2)
9–12	1,097,070 (35.5)	13,460 (40.7)	(1.2)	1,083,610 (35.4)
13–16	1,294,268 (41.8)	13,228 (40.0)	(1.0)	1,281,040 (41.9)
≥17	447,753 (14.5)	4,199 (12.7)	(0.9)	443,554 (14.5)
Unknown	125,322 (4.1)	818 (2.5)	(0.7)	124,504 (4.1)
Marital status
Single	962,734 (31.1)	12,548 (38.0)	(1.3)	950,186 (31.0)
Married	2,074,923 (67.1)	20,443 (61.8)	(1.0)	2,054,480 (67.1)
Unknown	55,584 (1.8)	67 (0.2)	(0.1)	55,517 (1.8)
Maternal smoking
Non-smoker	2,767,915 (89.5)	28,597 (86.5)	(1.0)	2,739,318 (89.5)
Smoker	243,721 (7.9)	4437 (13.4)	(1.8)	239,284 (7.8)
Unknown	81,605 (2.6)	24 (0.1)	(0.0)	81,581 (2.7)
Pre-pregnancy smoking
Absent	461,097 (94.7)	5022 (91.4)	(1.1)	456,075 (94.8)
Present	25,610 (5.3)	472 (8.6)	(1.8)	25,138 (5.2)
Not recordeda	2,606,534	27,564	-	2,578,970
First trimester smoking
Absent	461,097 (96.7)	5022 (93.5)	(1.1)	456,075 (96.8)
Present	15,647 (3.3)	349 (6.5)	(2.2)	15,298 (3.2)
Not recordeda	2,616,497	27,687	-	2,588,810
Second trimester smoking
Absent	461,097 (97.5)	5022 (94.3)	(1.1)	456,075 (97.5)
Present	11,986 (2.5)	305 (5.7)	(2.5)	11,681 (2.5)
Not recordeda	2,620,158	27,731	-	2,592,427
Third trimester smoking
Absent	461,097 (97.6)	5,022 (94.7)	(1.1)	456,075 (97.7)
Present	11,136 (2.4)	280 (5.3)	(2.5)	10,856 (2.3)
Not recordeda	2,621,008	27,756	-	2,593,252
Pre-pregnancy body-mass index (kg/m2)
<18.5	15,569 (3.2)	211 (3.9)	(1.4)	15,358 (3.2)
18.5–24.9	221,787 (45.7)	2514 (46.4)	(1.1)	219,273 (45.7)
25.0–29.9	135,612 (27.9)	1515 (28.0)	(1.1)	134,097 (27.9)
30.0–34.9	66,683 (13.7)	667 (12.3)	(1.0)	66,016 (13.7)
35.0–39.9	28,199 (5.8)	340 (6.3)	(1.2)	27,859 (5.8)
≥40	17,901 (3.7)	170 (3.1)	(0.9)	17,731 (3.7)
Not recordeda	2,607,490	27,641	-	2,579,849
Primary insurance status
Medicare	8768 (0.3)	144 (0.4)	(1.6)	8624 (0.3)
Medicaid	477,491 (15.4)	5824 (17.6)	(1.2)	471,667 (15.4)
Private	1,933,147 (62.5)	20,104 (60.8)	(1.0)	1,913,043 (62.5)
Self-pay	147,586 (4.8)	2170 (6.6)	(1.5)	145,416 (4.8)
Others	191,236 (6.2)	2445 (7.4)	(1.3)	188,791 (6.2)
Unknown	335,013 (10.8)	2371 (7.2)	(0.7)	332,642 (10.9)
Number of prenatal visits
No care	30,744 (1.0)	1197 (3.6)	(3.9)	29,547 (1.0)
1–9	923,086 (29.8)	14,486 (43.8)	(1.6)	908,600 (29.7)
10–12	1,254,031 (40.5)	10,059 (30.4)	(0.8)	1,243,972 (40.7)
≥13	743,047 (24.0)	5363 (16.2)	(0.7)	737,684 (24.1)
Unknown	142,333 (4.6)	1953 (5.9)	(1.4)	140,380 (4.6)
Child sex
Female	1,508,995 (48.8)	15,301 (46.3)	(1.0)	1,493,694 (48.8)
Male	1,582,625 (51.2)	17,737 (53.7)	(1.1)	1,564,888 (51.1)
Unknown	1621 (0.1)	20 (0.1)	(1.2)	1601 (0.1)

^a.Data only available between 2016 and 2020

Table 2.

Chronic conditions, obstetrical complications, and labor, and delivery events in relation to placental abruption: Placental Abruption and Cardiovascular Event Risk (PACER), 1993–2020

	Total cohort	Placental abruption present		Abruption absent
	Number (%col)	Number (%col)	(%row)	Number (%col)
Total deliveries	3,093,241 (100.0)	33,058 (100.0)	(1.1)	3,060,183 (100.0)
Plurality
Singletons	2,966,852 (95.9)	30,153 (91.2)	(1.0)	2,936,699 (96.0)
Twins	118,447 (3.8)	2697 (8.2)	(2.3)	115,750 (3.8)
Triplets and higher order	7257 (0.2)	202 (0.6)	(2.8)	7055 (0.2)
Unknown	685 (0.0)	6 (0.0)	(0.9)	679 (0.0)
Birth order
1	3,004,674 (97.1)	31,385 (94.9)	(1.0)	2,973,289 (97.2)
2	61,562 (2.0)	1415 (4.3)	(2.3)	60,147 (2.0)
≥3	2481 (0.1)	69 (0.2)	(2.8)	2412 (0.1)
Unknown	24,524 (0.8)	189 (0.6)	(0.8)	24,335 (0.8)
Hypertension-related diagnoses
Normotensive	2,607,881 (84.3)	26,558 (80.3)	(1.0)	2,581,323 (84.4)
Chronic hypertension	36,593 (1.2)	712 (2.2)	(1.9)	35,881 (1.2)
Gestational hypertension	112,193 (3.6)	1545 (4.7)	(1.4)	110,648 (3.6)
Preeclampsia, no severe features	57,427 (1.9)	1201 (3.6)	(2.1)	56,226 (1.8)
Preeclampsia, with severe features	31,936 (1.0)	1309 (4.0)	(4.1)	30,627 (1.0)
Superimposed preeclampsia	4511 (0.1)	185 (0.6)	(4.1)	4326 (0.1)
Eclampsia	12,073 (0.4)	370 (1.1)	(3.1)	11,703 (0.4)
Unknown	230,627 (7.5)	1178 (3.6)	(0.5)	229,449 (7.5)
Diabetes mellitus
Non-diabetic	2,885,850 (93.3)	30,855 (93.3)	(1.1)	2,854,995 (93.3)
Pre-pregnancy	35,237 (1.1)	497 (1.5)	(1.4)	34,740 (1.1)
Gestational	130,798 (4.2)	1376 (4.2)	(1.1)	129,422 (4.2)
Unknown	41,356 (1.3)	330 (1.0)	(0.8)	41,026 (1.3)
Mode of delivery
Vaginal	2,094,892 (67.7)	12,493 (37.8)	(0.6)	2,082,399 (68.0)
Cesarean	989,089 (32.0)	20,561 (62.2)	(2.1)	968,528 (31.6)
Unknown	9260 (0.3)	4 (0.0)	(0.0)	9256 (0.3)
Forceps extraction
Absent	2,970,569 (96.0)	31,822 (96.3)	(1.1)	2,938,747 (96.0)
Present	36,297 (1.2)	293 (0.9)	(0.8)	36,004 (1.2)
Unknown	86,375 (2.8)	943 (2.9)	(1.1)	85,432 (2.8)
Vacuum extraction
Absent	2,975,289 (96.2)	32,185 (97.4)	(1.1)	2,943,104 (96.2)
Present	117,868 (3.8)	869 (2.6)	(0.7)	116,999 (3.8)
Unknown	84 (0.0)	4 (0.0)	(4.8)	80 (0.0)
Premature rupture of membranes
Absent	2,890,880 (93.5)	29,511 (89.3)	(1.0)	2,861,369 (93.5)
Present	202,198 (6.5)	3532 (10.7)	(1.7)	198,666 (6.5)
Unknown	163 (0.0)	15 (0.0)	(9.2)	148 (0.0)
Placenta previa
Absent	3,071,537 (99.3)	31,276 (94.6)	(1.0)	3,040,261 (99.3)
Present	21,704 (0.7)	1782 (5.4)	(8.2)	19,922 (0.7)
Unknown	0 (0.0)	0 (0.0)	(0.0)	0 (0.0)
Preterm labor
Absent	2,619,056 (84.7)	17,997 (54.4)	(0.7)	2,601,059 (85.0)
Present	236,252 (7.6)	13,721 (41.5)	(5.8)	222,531 (7.3)
Unknown	237,933 (7.7)	1340 (4.1)	(0.6)	236,593 (7.7)
Induction of labor
Absent	2,486,423 (80.4)	28,249 (85.5)	(1.1)	2,458,174 (80.3)
Present	580,261 (18.8)	4795 (14.5)	(0.8)	575,466 (18.8)
Unknown	26,557 (0.9)	14 (0.0)	(0.1)	26,543 (0.9)
Precipitous labor
Absent	2,962,248 (95.8)	31,250 (94.5)	(1.1)	2,930,998 (95.8)
Present	122,848 (4.0)	1644 (5.0)	(1.3)	121,204 (4.0)
Unknown	8145 (0.3)	164 (0.5)	(2.0)	7981 (0.3)
Prolonged labor
Absent	3,030,915 (98.0)	32,450 (98.2)	(1.1)	2,998,465 (98.0)
Present	55,527 (1.8)	546 (1.7)	(1.0)	54,981 (1.8)
Unknown	6799 (0.2)	62 (0.2)	(0.9)	6737 (0.2)
Seizure during labor
Absent	3,083,255 (99.7)	32,869 (99.4)	(1.1)	3,050,386 (99.7)
Present	9986 (0.3)	189 (0.6)	(1.9)	9797 (0.3)
Unknown	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)

Adverse perinatal outcomes

Rates of adverse perinatal outcomes in relation to abruption are shown in Table 3. Fetal death rates were over 9-fold higher among abruption (5.0%) and non-abruption (0.6%) pregnancies, respectively. Similarly, preterm delivery rates among abruption and non-abruption pregnancies were 52.6% and 9.6%, respectively.

Table 3.

Perinatal outcomes in relation to placental abruption: Placental Abruption and Cardiovascular Event Risk (PACER), 1993–2020

	Total cohort	Placental abruption present		Abruption absent
	Number (%col)	Number (%col)	(%row)	Number (%col)
Total deliveries	3,093,241 (100.0)	33,058 (100.0)	(1.1)	3,060,183 (100.0)
Stillbirth (at ≥20 weeks)
Absent	3,074,448 (99.4)	31,416 (95.0)	(1.0)	3,043,032 (99.4)
Present	18,793 (0.6)	1642 (5.0)	(8.7)	17,151 (0.6)
Gestational age (weeks)
20–27	32,066 (1.0)	3642 (11.0)	(11.4)	28,424 (0.9)
28–31	30,594 (1.0)	3692 (11.2)	(12.1)	26,902 (0.9)
32–33	37,924 (1.2)	2997 (9.1)	(7.9)	34,927 (1.1)
34–36	209,126 (6.8)	7074 (21.4)	(3.4)	202,052 (6.6)
37–38	749,925 (24.2)	6925 (20.9)	(0.9)	743,000 (24.3)
39–40	1,695,777 (54.8)	7431 (22.5)	(0.4)	1,688,346 (55.2)
41	244,620 (7.9)	991 (3.0)	(0.4)	243,629 (8.0)
42–44	28,954 (0.9)	113 (0.3)	(0.4)	28,841 (0.9)
Unknown	64,255 (2.1)	193 (0.6)	(0.3)	64,062 (2.1)
Preterm delivery
<37 weeks	309,710 (10.0)	17,405 (52.6)	(5.6)	292,305 (9.6)
≥37 weeks	2,719,276 (87.9)	15,460 (46.8)	(0.6)	2,703,816 (88.4)
Unknown	64,255 (2.1)	193 (0.6)	(0.3)	64,062 (2.1)
Small for gestational age (%)
Not applicableb	18,793	1642	-	17,151
<3	85,802 (2.8)	1573 (5.0)	(1.8)	84,229 (2.8)
<5	146,357 (4.8)	2471 (7.9)	(1.7)	143,886 (4.7)
<10	293,522 (9.5)	4381 (13.9)	(1.5)	289,141 (9.5)
≥90	2,414,353 (78.5)	24,647 (78.5)	(1.0)	2,389,706 (78.5)
Unknown	366,573 (11.9)	2388 (7.6)	(0.7)	364,185 (12.0)
Birthweight (g)
<1500	58,642 (1.9)	6425 (19.4)	(11.0)	52,217 (1.7)
1500–2499	199,600 (6.5)	8716 (26.4)	(4.4)	190,884 (6.2)
2500–3999	2,531,988 (81.9)	16,853 (51.0)	(0.7)	2,515,135 (82.2)
≥4000	268,943 (8.7)	878 (2.7)	(0.3)	268,065 (8.8)
Unknown	34,068 (1.1)	186 (0.6)	(0.5)	33,882 (1.1)
5-min Apgar score
0–3	9386 (0.3)	1087 (3.5)	(11.6)	8299 (0.3)
4–7	46,832 (1.5)	3959 (12.6)	(8.5)	42,873 (1.4)
8–10	2,930,826 (95.3)	26,295 (83.7)	(0.9)	2,904,531 (95.4)
Unknown	87,404 (2.8)	75 (0.2)	(0.1)	87,329 (2.9)
Not applicableb	18,793	1642	-	17,151
Congenital anomaly
Absent	3,019,622 (97.6)	31,596 (95.6)	(1.0)	2,988,026 (97.6)
Present	59,990 (1.9)	1037 (3.1)	(1.7)	58,953 (1.9)
Unknown	13,629 (0.4)	425 (1.3)	(3.1)	13,204 (0.4)
Hyaline membrane disease
Absent	3,043,152 (98.4)	31,500 (95.3)	(1.0)	3,011,652 (98.4)
Present	9477 (0.3)	966 (2.9)	(10.2)	8511 (0.3)
Unknown	40,612 (1.3)	592 (1.8)	(1.5)	40,020 (1.3)
Newborn seizure
Absent	3,079,420 (99.6)	32,578 (98.5)	(1.1)	3,046,842 (99.6)
Present	1023 (0.0)	93 (0.3)	(9.1)	930 (0.0)
Unknown	12,798 (0.4)	387 (1.2)	(3.0)	12,411 (0.4)
Breech presentation
Absent	2,956,772 (95.6)	30,557 (92.4)	(1.0)	2,926,215 (95.6)
Present	108,519 (3.5)	2294 (6.9)	(2.1)	106,225 (3.5)
Unknown	27,950 (0.9)	207 (0.6)	(0.7)	27,743 (0.9)
Fetal distress
Absent	2,842,380 (91.9)	27,525 (83.3)	(1.0)	2,814,855 (92.0)
Present	222,922 (7.2)	5326 (16.1)	(2.4)	217,596 (7.1)
Unknown	27,939 (0.9)	207 (0.6)	(0.7)	27,732 (0.9)
Meconium
Absent	2,904,515 (93.9)	31,428 (95.1)	(1.1)	2,873,087 (93.9)
Present	181,014 (5.9)	1465 (4.4)	(0.8)	179,549 (5.9)
Unknown	7712 (0.2)	165 (0.5)	(2.1)	7547 (0.2)

Cardiovascular disease outcomes

The median (interquartile range) follow-up for the maternal and offspring cohorts was 15.4 (8.1, 22.4) and 14.4 (7.4, 21.0) years, respectively. Analyses of associations between abruption and CVD events in the women and the offspring are underway.

Unlinked records and missing data

We examined the distributions of covariates between the linked (delivery with hospitalisation records) and unlinked cohorts based on abruption status. The prevalence rate of abruption in the 237,933 unlinked records was 0.6%, in comparison to a rate of 1.1% in the linked records. The prevalence proportions of unlinked records were higher among non-Hispanic Blacks, Hispanics, low education levels, and unmarried persons (Table S2). The distributions of chronic conditions, obstetrical complications, and labor, and delivery events were similar between the linked and unlinked cohorts (Table S3). The proportions of unlinked records were higher among low gestational age at delivery, including preterm delivery (Table S4).

Comment

Principal findings

Evidence from large, population-based cohorts of events that occur before and during pregnancy and delivery with follow-up of persons over the life course to understand how obstetrical complications shape the risks of chronic conditions remains a challenge. Such cohorts include those in many Scandinavian and European countries, and provincial data from British Columbia, and Ontario, Canada, but similar efforts in a US population are scarce. The creation of the PACER cohort of 3.1 million deliveries in the state of New Jersey, USA with 28 years of follow-up affords an opportunity to evaluate how obstetrical complications shape the risks of CVD along the life course. Preliminary findings in PACER indicate that the prevalence of placental abruption is 1.1%. Efforts to understand if and how abruption may confer an increased burden of heart disease and stroke are underway.

Strengths of the study

The PACER cohort is unique in several aspects, including a large number of pregnancies and deliveries linked to every hospitalisation along the life course. Using mortality records, we were successfully able to link deaths along with the primary and secondary cause of death classifications. Although not the focus of this paper, the PACER cohort enables the characterisation of CVD risks in the offspring.

Limitations of the data

A few limitations of the PACER cohort merit some discussion. First, fetal death records were not provided by the NJ Department of Health in 2016–17 and were unavailable to us at the time of data linkage for 2019–20. Second, the successful linkage of vital records with hospitalisation was high for live births (92.4%) and moderate for fetal deaths (70.7%). These differing proportions may introduce the potential for bias, and we address this bias by reweighting the data for all analyses. Third, the linkage of CVD hospitalisations was restricted to those that occurred within the state. Therefore, if a person delivered in New Jersey and was hospitalised in another state or country later, we were unable to capture those hospital records. Therefore, the CVD rates in the PACER cohort may be slightly lower than in comparable population-based studies.

Interpretation

Placental abruption is associated with increased risks of heart disease and stroke mortality,^45–48 as well as non-fatal CVD events. A comprehensive meta-analysis⁴⁹ based on 11 cohort studies comprising 6,325,152 pregnancies (69,759 abruptions), and 49,265 heart disease and stroke events found that abruption was associated with higher mortality from heart disease (risk ratio [RR] 2.64, 95% confidence interval [CI] 1.57, 4.44; I² = 76%; τ² = 0.31) as well as stroke (RR 1.70, 95% CI 1.19, 2.42; I² = 40%; τ² = 0.05). There was evidence of compounding of CVD mortality risks with an increasing number of abruptions in women across multiple pregnancies. DeRoo and colleagues⁴⁷ reported that the hazards ratio [HR] for mortality from CVD among persons with 1, and 2 or more abruptions was 1.6 (95% CI 1.5, 2.2) and 2.2 (95% CI 0.8, 6.0), respectively. Similarly, Ananth and colleagues⁴⁵ reported that the corresponding hazard ratio of stroke deaths was 1.1 (95% CI 0.5, 2.7) and 2.7 (95% CI 1.1, 6.5) for 1 and 2 or more abruptions, respectively. There was no evidence of a dose-response relationship between the increasing number of abruptions and the risk of non-fatal CVD events.⁴⁵

There is scant evidence of an association between abruption and CVD risks in the offspring. A study from Israel⁵⁰ (217,910 deliveries between 1991–2014) reported that the cumulative incidence of long-term CVD morbidity was 1.0% and 0.6% in children born of abruption and non-abruption pregnancies, respectively (adjusted HR 1.12, 95% CI 0.60, 2.11). Given that the maximum age at follow-up was 18 years, the estimated CVD risks may be underestimated.

There are at least four mechanisms through which abruption may confer an increased burden of CVD events. First, patients with abruption and CVD share common abnormal biomarker profiles. A case-control study of 75 women with a history of abruption and 79 women with otherwise uneventful pregnancies from the Netherlands⁵¹ reported higher systolic and diastolic blood pressures, fasting levels of glucose and insulin, and substantially higher levels of total cholesterol, low-density lipoprotein, and triglycerides, and lower levels of high-density lipoprotein. Each of these profiles confers greater CVD risk. Second, shared epidemiologic risk factors have also been reported in patients with abruption and CVD. Third, it is plausible that abruption may not only accelerate the development of cardiometabolic risk factors postpartum, but the vascular and thrombotic insults that accompany abruption may lead to a persistent endothelial and microvascular dysfunction⁵² – all precursors to CVD risk. Fourth, genetics and gene-environment interactions may be similar between these groups of patients.

Conclusions

Despite the strong temporal decline in mortality rates from heart disease and stroke,^53–57 and a slowing of mortality from heart disease decline in recent cohorts,⁵⁷ an estimated 44 million women in the US suffer from CVD.⁵⁸ The PACER project will offer an unprecedented opportunity to understand how sentinel obstetrical complications shape future risks of chronic conditions along the life course. This project is designed to address CVD risks in both the mother and offspring in relation to abruption, with consideration of the severity of abruption, preterm delivery, and other important characteristics that are implicated in premature CVD. Abruption is unpredictable and attempts toward prevention have been disappointing. However, efforts to minimise abruption recurrence through quitting smoking, cocaine, or alcohol use; use of folate supplementation; optimal pregnancy spacing; healthy nutrition and reducing unhealthy lifestyle; and optimising postpartum maternal cardiometabolic health, may be worthy of consideration.⁵⁹ We believe that findings from this large population-based cohort may help provide the much-needed clarity and nuanced CVD risk characterisation along the life course.

Synopsis.

Study question

To describe the study design, and data linkage algorithms for the Placental Abruption and Cardiovascular Event Risk (PACER) cohort, and to provide descriptive preliminary findings on risks of placental abruption and fatal and non-fatal cardiovascular disease (CVD) in pregnant persons and their offspring.

What is already known

The risks of premature chronic diseases, including heart disease and stroke, are two-fold higher among women who experience an abruption.

What this study adds

The PACER project will provide unprecedented opportunities in a US-based cohort to understand the burden of CVD risks in both women and their offspring along the life course and how placental abruption and associated obstetrical complications shape the risks.

Maternal cohort

Match validation and data processing

We combined similar file types (e.g., LB2-UB2, LB3-UB3, LB3-UB4…) to create 7 file types (LB – LB / FD – FD/ LB – FD/ LB – MT / FD – MT/ LB – UB/ FD – UB). We manually reviewed about 20–40 examples from file type by match score. The accuracy of matches was highly dependent on an exact match in the mother’s date of birth. Live birth matches (LB – UB/ LB – MT/ LB – LB) had higher overall match score patterns than fetal death matches (FD – UB/ FD- MT/ FD – FD) due to more patient identifiers in the birth records. With this date restriction, links with low match scores of 0.1 or 0.25 and exact match on DOB were manually reviewed and observed to be true matches. Therefore, all matches, regardless of probability score and manual match scores, were included, and links with mismatched mother’s date of birth were not considered a true match while matches with missing date of birth for one or both sets of matches were included. To identify true deaths, each PACER ID (LB – MT/ FD – MT) was restricted to one mortality match, ensuring the date of death occurred at or after the delivery event, and age was within 10–89 years to exclude few mismatches with infant deaths and elderly deaths born in a different time period. Delivery hospitalisation events (LB – UB / FD – UB) were identified by admission and discharge dates relative to the child’s date of birth or fetal death and delivery-related ICD 9 and 10 diagnosis and procedures codes. All other LB – UB/ FD – UB hospitalisation links were retained as hospitalisation events that occurred before or after the delivery event. Data cleaning was required for LB – LB and FD – FD matches since all possible combinations were present and only the distinct match was needed.

Hospitalisation delivery matches

1. Live birth matches to hospitalisation records were identified in two steps. First, hospital admission date ±5 from the child’s date of birth or fetal death event delivery codes with any delivery code (ICD 9 diagnosis codes 630–670; ICD 10 diagnosis codes O00-O9A) to capture all delivery-related hospitalisations. If a PACER ID was matched to multiple hospitalisations within the date restrictions, the hospitalisation record with pregnancy codes (see Table below) and/or longest length of stay was identified as the delivery event.

2. Second, the remaining live births that were not identified in Step 1 were determined by restricting admission dates occurring at or before the child’s date of birth and discharge dates at or after the child’s date of birth with pregnancy codes shown in the table below.

Fetal deaths matched to hospitalisation records were identified in two steps.

1. Admission and discharge date within the timeframe of the fetal death and codes for stillbirth, fetal death, mixed outcome, and abortion. Any PACER ID that was linked to more than one hospitalisation event was sorted by highest match score and longest length of stay.

2. Then the remaining fetal death records that were not identified in Step 1 were identified with the same date restrictions above and pregnancy-related codes (ICD 9 diagnosis codes 630–670; ICD 10 diagnosis codes O00-O9A).

Establishing final health record by individual

We then created a new unique ID to track pregnancies, hospitalisations, and mortality records by individual. Since live birth records contained more patient identifiers compared to fetal death records, a new unique ID was created based on LB – LB matches. The first step was to identify women who had multiple live births within the cohort. Any PACER IDs that were linked to two unique IDs were excluded (6,476 live births) to remove overlapping linkages between two or more individuals. Any live birth records that were not matched in LB – LB were also assigned a unique ID, and this represented women who only had one live birth during the study period. Next, we identified women who were in the LB – FD cohort and carried over the unique ID from live birth records to fetal records using the LB – FD matches. Similarly, fetal deaths linked to two unique IDs were excluded (247 fetal deaths). These unique IDs were carried over to the FD – FD dataset to identify women who had a live birth and more than one fetal death. Finally, any other remaining fetal deaths that were not linked were considered as an individual who only had 1 fetal during the study and assigned a unique ID.

Final validation and data limitations

After the dataset was constructed, we reviewed 5–15 unique IDs for each scenario (women who had one live birth, one fetal death, multiple live births, multiple fetal deaths, live birth, and fetal death, those who died during the study, those with abruption during pregnancy, those with CVD hospitalisation, and other random unique IDs). The links were accurate in identifying correct health records for each woman across the duration of the study except for individuals with only fetal deaths and hospitalisation records. The latter contained matches that were more ambiguous but could not be considered an incorrect match. Another limitation is that some live births were not linked in the LB – LB dataset due to spelling variations of names across records. This would result in one woman appearing as two different people in the cohort although this may be uncommon.

Offspring Cohort

Match validation and data processing

Using a similar approach as the maternal cohort, we combined file types (e.g., LB2-UB2, LB3-UB3, LB3-UB4…) to create 2 file types (LB – UB/ LB – MT) for the child cohort. Linkages with mismatch child date of birth for live birth, hospitalisation, and mortality records were not considered true matches while matches with missing date of birth for one or both matches were included. To identify true offspring deaths, each offspring was linked to only one death, ensuring date of death occurred at after birth date, and age at death did not exceed maximum follow-up time (28 years). Hospitalisation delivery matches were determined primarily by admission date relative to child’s date of birth, ICD coding, and/or match scores. All other LB – UB links that occurred after child’s date of birth were retained as a hospitalisation event. After conducting these data cleaning procedures, we manually reviewed 25–30 matches for all match scores for both LB-UB and LB-MT files and validated all linkages were true matches regardless of match scores. Therefore, no linkage restrictions were made based off probability or match scores.

Hospitalisation delivery matches

1. Matches with hospital admission date up to 5 days after the child’s date of birth were identified to capture all delivery-related hospitalisations. Admission date prior to and discharge date after child date of birth were not considered a true match. If a PACER ID was matched to multiple hospitalisations within the date restrictions, the hospitalisation record with pregnancy codes (see Table below) and/or highest match score was identified as the delivery event. These links allow for additional information of newborn complications not recorded on birth records.

2. Maternal LB – UB hospitalisation records were combined with the child matches to provide labor and delivery information such as abruption status during delivery.

Final validation and data limitations

After the dataset was constructed, we reviewed 25–35 examples consisting of delivery events, delivery complicated by abruption, CVD and non-CVD-related hospitalisations, infant deaths, deaths after 1 years of age). In these reviews, the links were accurate in identifying correct health records for each offspring across the duration of the study. One limitation is that child delivery hospitalisations had lower proportion of birth certificate and hospitalisation matches possibly due to variations in alias baby names specific to delivery hospitalisations such as “BABY”, “BABYG”, “BABYB” in one or both files. However, a combination of birth certificate, child hospitalisation, and corresponding maternal hospitalisation record helps to provide more comprehensive information at time of delivery.

International Classification of Disease (ICD) 9 and 10 codes to define the end of pregnancy Placental Abruption and Cardiovascular Event Risk (PACER), 1993–2020

Livebirth	ICD-9-CM Diagnosis: 766.0 766.1 765.1x 766.2x V27.0 V27.2 V27.5 V30.xx V31.xx V34.xx V39.xx Diagnosis Related Group: 790 791 792 793 794 795 ICD-10-CM Diagnosis: P070 P0700 P0701 P0702 P0703 P071 P0710 P0714 P0715 P0716 P0717 P0718 P073 P0730 P080 P081 P082 P0821 P0822 Z370 Z372 Z375 Z3750 Z3751 Z3752 Z3753 Z3754 Z3759 Z380 Z3800 Z3801 Z381 Z382 Z383 Z3830 Z3831 Z384 Z385 Z386 Z3861 Z3862 Z3863 Z3864 Z3865 Z3866 Z3868 Z3869 Z387 Z388
Stillbirth	ICD-9-CM Diagnosis: 656.40 656.41 656.43 V27.1x V27.4x V27.7x Current Procedural Terminology: 88016 ICD-10-CM Diagnosis: O364 O364XX0 O364XX1 O364XX2 O364XX3 O364XX4 O364XX5 O364XX9 Z371 Z374 Z377
Mixed Birth	ICD-9-CM Diagnosis: 651.31 651.41 651.51 651.61 V27.3x V27.6x V32.xx V35.xx V36.xx ICD-10-CM Diagnosis: O311 O3111X0 O3111X1 O3111X2 O3111X3 O3111X4 O3111X5 O3111X9 O3112X0 O3112X1 O3112X2 O3112X3 O3112X4 O3112X5 O3112X9 O3113X0 O3113X1 O3113X2 O3113X3 O3113X4 O3113X5 O3113X9 O312 O3121X0 O3121X1 O3121X2 O3121X3 O3121X4 O3121X5 O3121X9 O3122X0 O3122X1 O3122X2 O3122X3 O3122X4 O3122X5 O3122X9 O3123X0 O3123X1 O3123X2 O3123X3 O3123X4 O3123X5 O3123X9 Z373 Z376 Z3760 Z3761 Z3762 Z3763 Z3764 Z3769
Spontaneous Abortion	ICD-9-CM Diagnosis: 632 634.xx 637.xx Current Procedural Terminology: 01965 59812 59820 59821 59830 Diagnosis Related Group: 770 779 ICD-10-CM Diagnosis: O02 O021 O030 O031 O032 O033 O0330 O0331 O0332 O0333 O0334 O0335 O0336 O0337 O0338 O0339 O034 O035 O036 O037 O038 O0380 O0381 O0382 O0383 O0384 O0385 O0386 O0387 O0388 O0389 O039 O3102X3 O3102X4 O3102X5 O3102X9
Elective Termination	ICD-9-CM Diagnosis: 779.6x 635.xx 636.xx ICD-9-CM Procedures: 74.91 75.0x 69.01 69.51 Current Procedural Terminology: 01966 59840 59841 59850 59851 59852 59855 59856 59857 Healthcare Common Procedure Coding System: S0190 S0199 S2260 S2262 S2265 S2266 S2267 ICD-10-CM Diagnosis: O045 O046 O047 O048 O0480 O0481 O0482 O0483 O0484 O0485 O0486 O0487 O0488 O0489 Z332 ICD-10-CM Procedures: 10A00ZZ 10A03ZZ 10A04ZZ 10A07ZX 10A07ZZ 10A08ZZ
Ectopic	ICD-9-CM Diagnosis: 761.4 633.00 633.10 633.20 633.80 633.90 ICD-9-CM Procedures: 66.62 74.3x Current Procedural Terminology: 59120 59121 59130 59135 59136 59140 59150 59151 Diagnosis Related Group: 777 ICD-10-CM Diagnosis: O00 O01 O000 O0000 O001 O00101 O00102 O00109 O002 O00201 O00202 O00209 O008 O0080 O009 O0090 P014 ICD-10-CM Procedures: 10T20ZZ 10T23ZZ 10T24ZZ 10T27ZZ, 10T28ZZ
Unspecified Delivery	ICD-9-CM Diagnosis: 763xx, 7632x, 7633x, 7634x, 7636x, 644.21 645.11 645.21 649.81 649.82 650 651.01 651.11 651.21 651.81 651.91 658.20 658.21 658.30 658.31 662.01 662.11 662.21 662.31 669.51 669.61 669.71 763.0 763.2 763.3 763.4 763.6 768.0 V27 V27.9 V33.xx V37.xx ICD-9-CM Procedures: 72.x 73.0 73.01 73.09 73.1 73.2 73.22 73.3 73.4 73.5 73.51 73.59 73.6 73.8 73.9 73.91 73.92 73.93 73.94 73.99 74.0 74.1 74.2 74.4 74.9 74.99 75.4 Current Procedural Terminology: 01960 01961 01963 01967 01968 01969 59400 59409 59410 59414 59510 59514 59515 59525 59610 59612 59614 59620 59622 99464 Diagnosis Related Group: 765 766 767 768 774 775 ICD-10-CM Diagnosis: Z37 O601 O6012X0 O6012X1 O6012X2 O6012X3 O6012X4 O6012X5 O6012X9 O6013X0 O6013X1 O6013X2 O6013X3 O6013X4 O6013X5 O6013X9 O6014X0 O6014X1 O6014X2 O6014X3 O6014X4 O6014X5 O6014X9 O602 O6022X0 O6022X1 O6022X2 O6022X3 O6022X4 O6022X5 O6022X9 O6023X0 O6023X1 O6023X2 O6023X3 O6023X4 O6023X5 O6023X9 O630 O631 O632 O639 O641 O641XX0 O665 O755 O758 O7582 O80 O82 P030 P032 P033 P034 P035 Z379 ICD-10-CM Procedures: 0U7C7ZZ 0W8NXZZ 10900ZC 10903ZC 10904ZC 10907ZA 10907ZC 10908ZA 10908ZC 10A00ZZ 10A03ZZ 10A04ZZ 10A07Z6 10A07ZZ 10A08ZZ 10D00Z0 10D00Z1 10D00Z2 10D07Z3 10D07Z4 10D07Z5 10D07Z6 10D07Z7 10D07Z8 10D17Z9 10D18Z9 10E0XZZ 10J07ZZ 10S07ZZ 10S0XZZ

ICD-9-CM = International Statistical Classification of Diseases, Ninth Revision, Clinical Modification

Note: This table was adapted from MacDonald SC, Cohen JM, Panchaud A, McElrath TF, Huybrechts KF, Hernandez-Diaz S. Identifying pregnancies in insurance claims data: Methods and application to retinoid teratogenic surveillance. Pharmacoepidemiol Drug Saf. 2019; 28: 1211–1221.

Data sharing

Due to data privacy issues, we are unable to make these data publicly available.