The association between severe maternal morbidity and mortality and race/ethnicity with community type in Maryland

Study design

We performed a cross-sectional analysis on hospital delivery records gathered through the Maryland Health Service Cost Review Commission (HSCRC) from October 2015 to June 2020. The Maryland HSCRC is a database created to improve the quality of health and healthcare, and to reduce the cost of care for residents of Maryland. ¹⁷ In Maryland, out-of-hospital deliveries only accounted for about 2.5% of all births. ¹⁸ Our analysis therefore accounted for the majority of deliveries occurring in Maryland during that time period. Deliveries that took place prior to October 2015, before International Classification of Disease, the Tenth Edition, Clinical Modification codes (ICD-10) became available, were excluded. ¹⁹ Of those individuals who were admitted for in-hospital delivery, we included 204,140 residents of Maryland from the ages of 15 to 54 years. Ages 0–14 and 55 and above were excluded due to being outside the age of biological fecundity and outside of the age recommended for the use of obstetrics ICD-10 codes (see Figure 1). ¹² Given that most current data and study findings are on women, adults who self-identified as women were included in this study. This is not to exclude individuals other than self-identifying women who this subject matter may affect, but to simply provide consistency with current data. This study was reviewed by the Johns Hopkins School of Medicine Institutional Review Board (IRB) and was classified as exempt from continuing IRB review under the DHHS regulation 45 CFR 46.101(b)(4), which allows exemption for publicly available data or data that are de-identified in some manner. A waiver of consent was granted based on the following criteria: the study involves very minimal risk to human subjects, the waiver will not have an adverse effect on subjects, and this study could not be conducted without the waiver. The IRB number is IRB00082345. This study also utilized The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for cross-sectional studies.

OPEN IN VIEWER

Figure 1.

Study population.

Race/ethnicity was our main predictor and was defined within the database based on self-reported information. ¹⁷ Multi-racial as well as races/ethnicities other than Non-Hispanic White, Non-Hispanic Black, Non-Hispanic Asian, and Hispanic were included in an Other/Unknown category to ensure that most patients were captured.

Community type was used to determine effect modification. It was a binary variable defined by counties. Rural communities were defined based on 18 jurisdiction recognized by the State of Maryland and defined by the Rural Maryland Council as counties that “share common characteristics that set them apart from their suburban and urban counterparts, such as geographic isolation, lack of transportation, and lack of access to and availability of health care.” ²⁰ The counties were: Allegany, Calvert, Caroline, Carroll, Cecil, Charles, Dorchester, Frederick, Garrett, Harford, Kent, Queen Anne’s, Somerset, St. Mary’s, Talbot, Washington, Wicomico, and Worcester. The remaining counties were deemed urban communities, which included suburban areas as well. ²⁰

The main outcome of interest was a binary, composite variable comprising indicators (or defining diagnoses) that are classified as SMM and also included mortality. SMM was defined according to criteria established by the CDC and the variables were generated by using recommended ICD-10 codes. It is defined as having any of the following 21 indicators in the peripartum period: acute myocardial infarction, aneurysm, acute renal failure, adult respiratory distress syndrome, amniotic fluid embolism, blood products transfusion, cardiac arrest/ventricular fibrillation, conversion of cardiac rhythm, disseminated intravascular coagulation (DIC), eclampsia, heart failure/arrest during surgery/procedure, puerperal cerebrovascular disorders, pulmonary edema/acute heart failure, severe anesthesia complications, sepsis, shock, sickle cell disease with crisis, air and thrombotic embolism, hysterectomy, temporary tracheostomy, and ventilation. ¹² The peripartum period in our study, given the information provided by the database, was the time in which the woman was hospitalized. This could include any length of time in the antepartum, intrapartum, and postpartum period. Because the units of blood transfused were not included in the database, blood transfusion was excluded as one of our indicators of SMM, hence 20 indicators. Some data have shown routine transfusion of one unit of blood preventatively, whereas two units or more may be used in the case of a postpartum hemorrhage—defined by American College of Obstetrics and Gynecology as a blood loss of one liter or more or clinical signs of hypovolemia. The exclusion of blood transfusion is to avoid over estimation of SMM and some data have also begun to report SMM with and without blood transfusion as a result.^21,22 Mortality, another component of our composite variable, was defined by patient disposition at discharge.

Covariates

Age was included in the database as a categorical variable binned in 5-year intervals. Primary payers were defined by the Maryland HSCRC as Medicaid, Private, Self-Pay, and Other—which included Medicare. Comorbidities and cesarean delivery were identified and included as covariates as they were identified as increasing the risk of SMM and mortality a priori. ²³ The comorbidities included the following: diabetes, hypertension, obesity, chronic respiratory disease, chronic renal disease, chronic heart disease, HIV/AIDS, substance use disorder (i.e. tobacco use and substance abuse disorders), depression, hypertensive disorders of pregnancy (which includes preeclampsia and gestational hypertension), and gestational diabetes. The diagnosis and procedure codes used to define these covariates were based on the multiple sources: the codes used to generate Elixhauser Comorbidity Index scores, a study looking at hypertensive disorders of pregnancy by Fingar et al., ²⁴ a study by Admon et al. ²⁵ looking at comorbidities in pregnancy, Current Procedural Terminology (CPT) codes and ICD10Data.com—to find ICD-10 or convert ICD-9 codes.^26–28

Statistical analysis

We performed descriptive analyses on our categorical demographic and clinical characteristics data. Pearson’s chi-square or Fisher’s exact tests were used to assess unadjusted differences by race/ethnicity for categorical covariates. Based on Centers of Medicare and Medicaid Services (CMS), any value between 1 and 10 were not reported in order to ensure confidentiality. ²⁹

Multivariable logistic regression models were used to assess the effect of community type on the relationship between race/ethnicity and the binary outcome of having at least one indicator of SMM and/or mortality. Our first and main model used race/ethnicity as the primary predictor to assess disparities in SMM and/or mortality when comparing races/ethnicities to non-Hispanic White women—reference group. To test whether any of the racial/ethnic disparities in SMM and/or mortality varied by community type, we ran two additional models that were modifications of our main model: one was a stratification of the main model by community type—in order to provide odds ratios (ORs) of SMM and/or mortality comparing each race/ethnic group to Non-Hispanic White women in urban communities and ORs of SMM and/or mortality comparing each race/ethnic group to Non-Hispanic White women in rural communities—and another model added an interaction term (race/ethnicity and community type) to our main model —to assess whether community type serves as an effect modifier. The interaction model included race/ethnicity, community type, and their interaction. All models were adjusted for using the aforementioned confounding variables (i.e. age category, payer, comorbidities, cesarean delivery).^23–27 Measures of association were expressed as OR, as mentioned. We used an alpha level of 0.05 for both bivariate descriptive analyses as well as all logistic regression models, including the interaction term. All data, except the post hoc power analysis, were analyzed using Stata version 15.1 (StataCorp, College Station, TX). Post hoc power analysis was conducted using ClinCalc Statistical software (version © 2023- ClinCalc LLC, clincalc.com). ³⁰

Principal findings

Given that there are known relationships between both rural communities and race/ethnicity with maternal outcomes, we designed this study to investigate if residing in rural communities increased the racial/ethnic disparities in SMM and mortality in Maryland. Our study showed that though there was a significantly higher odds of SMM and mortality for Non-Hispanic Black and Non-Hispanic Asian women when compared to Non-Hispanic White women, residing in rural communities did not increase the racial/ethnic disparity. Regardless of known and perceived barriers that may exist when living in a rural community, race/ethnicity plays a major role in SMM and mortality as seen in previous studies.^{4,11,13,14,31–36}