Data Source and Sample
We conducted a retrospective cross-sectional analysis using hospital
records from January 1, 2002 through December 31, 2017 that were
contained in the Nationwide Inpatient Sample (NIS). The NIS is made
available by the Healthcare Cost and Utilization Project (HCUP) and is
currently the largest all-payer, publicly available inpatient database
in the United States. (Healthcare Cost and Utilization Project (HCUP))
The two-staged cluster sampling design ensures that hospitalizations in
the NIS are representative of the population on important factors
including month of admission, primary reason for hospitalization,
hospital size, location, ownership, and teaching status.
Hospitalization-level weights are provided with each annual database
which allow national estimates to be generated. In 2017, the NIS
contained approximately seven million inpatient hospitalizations each
year (35 million when weighted) from 47 participating states. HCUP
transitioned from ICD-9-CM to ICD-10-CM format on October 1, 2015.
Our study sample included pregnancy-related hospitalizations among women
aged 15 to 49 years, identified using an HCUP-created variable
“NEOMAT” which captures maternal diagnosis records with diagnosis and
procedure codes for pregnancy and delivery in the ICD-9-CM era. In the
ICD-10-CM period, diagnoses codes ‘Oxx.x’ were used to identify
pregnancy and delivery related hospitalizations. To assess the study’s
primary exposure, we first scanned ICD-9-CM codes (the principal
diagnosis and up to 29 secondary diagnoses) in each woman’s discharge
record for an indication of gestational diabetes mellitus (GDM)
(ICD-9-CM: 648.8x, ICD-10-CM: O24.xx or O99.81) and/or pre-existing
(i.e., diagnosis of type 1 or 2 diabetes prior to the pregnancy)
(ICD-9-CM:249.xx, 250.xx, 648.0x; ICD-10-CM: E10.xx, E11.xx, E13.xx).
The outcome of interest was stillbirth (ICD-9-CM: 656.4x, V27.1x,
V27.3x, V27.4x, V27.6x, V27.7x, 768.0, 768.1, 779.9, 632; ICD-10-CM:
O36.4x, P95.xx, P96.9, Z37.1, Z37.3, Z37.4, Z37.6, Z37.7).
Individual-level sociodemographic and behavioral characteristics were
also extracted from the NIS databases. Maternal age in years was
classified into three categories: 15-24, 25–34, and 35-49 years.
Self-reported maternal race/ethnicity was first based on ethnicity
(Hispanic or non-Hispanic [NH]), and the NH group, further
subdivided by race (White, Black, or other). As the ‘other’ racial group
is not clearly defined in the dataset, our analyses were limited to
NH-White, NH-Black and Hispanics pregnant women. Median household
income, which served as a proxy for socio-economic status, was estimated
using the patient’s zip code and subsequently grouped into quartiles. We
classified the primary payer for hospital admission into following
categories: Medicare, Medicaid, private (commercial carriers, private
health maintenance organization [HMOs], and preferred provider
organization [PPOs]); and self-pay. We also considered several
hospital characteristics including teaching status (teaching vs.
non-teaching), location (urban vs. rural), and US region (Northeast,
Midwest, South, or West).
Information for this study was obtained from a retrospective secondary
de-identified data source. Therefore, patient involvement and core
outcome set requirements are not applicable to this analysis.