Data source and population
In this population-based cohort study, data were collected on all women of childbearing age (15–49 years) who had delivered children at medical facilities from 2003 to 2018, using the National Health Insurance (NHI) database, which consists of health-care utilization data, physical checkup data, sociodemographic data, and mortality data. The NHI, as the only health insurer in South Korea, stores cohort data collected during the claims process and includes records of hospitalizations, outpatient care, and drug prescriptions. This database also stores information related to health-care utilization, such as age, sex, residential area, insurance type, income, diagnostic codes, procedure codes, prescription drugs, individual medical expenses, and information on the hospitals covered by the NHI. The NHI cohort data can be used to continuously track the characteristics of patients, clinical records, and health-care providers; indicate the epidemiologic causes of disease; and provide information on the development of health-care policies. These data are anonymized by assigning unique number codes so that personal patient information remains unidentifiable [14]. The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Institutional Review Board (SMWU-1808-HR-076) of Sookmyung Women’s University.
An analysis period of at least 310 days (280 days of full-term pregnancy + 30 postpartum days) is necessary to check for maternal comorbidities during pregnancy and health status during the puerperal period; therefore, only women who had delivered at hospitals during the period from January 1, 2003 to December 1, 2018 and who had hospitalization records were defined as the study population. Postpartum women were defined as women who delivered at a hospital and who had hospitalization records that included an electronic data interchange code consisting of a diagnostic code; recorded as a single vaginal delivery (O80), a single delivery with forceps and vacuum extraction (O81), a single delivery by cesarean section (O82), a single delivery by any other supportive device (O83), or multiple births (O84); and a procedural code, recorded as normal delivery, breech extraction, cesarean section, or forceps or vacuum extraction. In total, 6,421,091 mothers who had delivered children from 2003 to 2018 were included in the study analysis.
SMM indicators
With SMM as the dependent variable, the analysis was performed using the following SMM indicators: (1) the SMM algorithm, defined by the US-CDC; (2) the new gold standard guideline for SMM, defined by the ACOG; (3) Zwart et al.’s. established SMM criteria in the Netherlands; and (4) the EURONET-severe acute maternal morbidity (EURONET-SAMM) index in eight European countries.
The US-CDC’s SMM algorithm defined SMM as the occurrence of at least one of a possible total of 21 indicators, consisting of 16 diagnostic codes and six procedural codes during the postnatal hospital stay [7, 15].
SMM criteria, as defined by the ACOG, are as follows: (1) the occurrence of at least one of 21 SMM indicators, as defined in the US-CDC’s SMM algorithm; (2) prolonged length of postnatal hospital stay; (3) intensive care unit (ICU) admission; (4) transfusion of ≥ 4 units of packed red blood cells; and (5) hospital readmission within 30 days of discharge [8]. In this study, all ACOG’s gold standard guidelines were adopted for analysis except the indicator on length of postnatal hospital stay because of the different criteria for postnatal hospital stays in the US and South Korean medical delivery systems. In the US, postnatal hospital lengths of stay > 3 days and ≥ 6 days are classified as SMM events for vaginal delivery and cesarean section, respectively, while the median length of postnatal hospital stays in South Korea is 3–4 days for vaginal delivery and 6–7 days for cesarean section. Therefore, application of the US standard to Korean cases would classify > 50% of all deliveries as SMM cases. Applying US standards to Korean women might be overestimated; accordingly, among the ACOG criteria, the length of hospital stay was excluded from the scope of analysis of this study.
In the Netherlands, Zwart et al. established SMM criteria, in agreement with the Dutch Maternal Mortality Committee of the Dutch Society of Obstetrics and Gynaecology, as the occurrence of one or more of the following postpartum events: (1) admission to the ICU, (2) uterine rupture, (3) eclampsia, (4) transfusion of ≥ 4 units of packed red blood cells, and (5) other SMM, according to the opinion of the treating obstetrician [9].
Another European standard is the EURONET-severe acute maternal morbidity (EURONET-SAMM) index, established through extracting and comparing the SAMM data of eight European countries, namely, Finland, France, Italy, Portugal, Switzerland, England, Scotland, and Wales, using their respective national hospital discharge records and determining the final codes, which consisted of diagnostic and procedural codes. Patients with any one of five SAMM indicators (eclampsia, septicemia during pregnancy, pregnancy-related hysterectomy, hysterectomy associated with a diagnosis of obstetric hemorrhage, and red blood cell transfusion associated with a diagnosis of obstetric hemorrhage) are classified as having SAMM [10].
Of the international indicators identified through this literature review, those recognizable as claims data were preselected as diagnostic and procedural codes, and the final SMM codes were established with the involvement of obstetrician-gynecologist, medical record administrators, and data scientists.
Covariates
Personal, obstetric, and provider factors were set as covariates. Personal factors included maternal age (range: 15–19, 20–24, 25–29, 30–34, 35–39, 40–44, and ≥ 45 years), household income (divided into quartiles: Q1 [low]), Q2, Q3, and Q4 [high]), type of health insurance (coverage according to region for those who were self-employed; coverage according to workplace for employees; and medical aid), and residential area (Seoul, metropolitan cities, small cities, and rural areas). Obstetric factors included mode of delivery (vaginal delivery, instrumental delivery, and caesarean section), preterm births [No (≥ 37 weeks’ gestation) and Yes (< 37 weeks’ gestation)], parity (primiparous and multiparous), and multiple births (single and multiple embryos). Prenatal care was estimated using Kessner’s adequate prenatal care index [16]. Prenatal care was rated adequate when a woman began prenatal care in the first trimester and had nine prenatal care visits for a normal-length pregnancy [16]. Prenatal care was determined inadequate if a woman began prenatal care after the third trimester and had less than four prenatal care visits. All the other situations in between were classified as intermediate prenatal care [16]. Maternal comorbidities were defined according to Howell’s criteria, [17] thus including cardiac disease, renal disease, musculoskeletal disease, digestive disorder, blood disease, mental disorders, CNS disease, rheumatic heart disease, placentation disorder, chronic hypertension, pregnancy hypertension, lupus, collagen vascular disorder, rheumatoid arthritis, diabetes, diabetes complicating pregnancy, obesity, and asthma/chronic bronchitis. Provider factors included the type of hospital according to number of beds (> 500 beds, 100–499 beds, 30–99 beds, and < 30 beds) and hospital location (Seoul, metropolitan city, small city, and county).
Statistical analysis
Based on the customized datasets obtained from the NHI system, all women who delivered children from 2003 to 2018 were analyzed. Pearson’s Chi-square tests were performed to ascertain differences in sociodemographic characteristics and their distribution between SMM cases and the study population, during labor and delivery hospitalization, based on the established codes. The frequencies and fractions of sub-indicators from each of the four SMM indicators were calculated and compared using basic statistics. Finally, the adjusted relative risk (RR) and 95% confidence intervals (CIs) were calculated using a generalized estimating equations model at a significance level of P < 0.05 to estimate the relationships between each SMM indicator and the demographic, obstetric, and provider factors. Data analysis was performed using SAS 9.4 software (SAS Institute, Inc., Cary, NC, USA).