Statistical Analysis
Joinpoint regression was used to identify and describe temporal changes
in the rates of PDM and GDM among pregnancy-related hospitalizations
during the 15-year study period. This type of statistical regression
analysis is valuable in identifying key periods that denote a
statistically significant change in the rate of events over
time. 24 The iterative model-building process
began by fitting the annual rate data to a straight line with no
joinpoints, which assumed a single trend best described the data. Then a
joinpoint, reflecting a change in the trend, was added to the model and
a Monte Carlo permutation test assessed the improvement in model fit.
The process continued until a final model with an optimal (best-fitting)
number of joinpoints was selected, with each joinpoint indicating a
change in the trend, and an annual percent change (APC) estimated to
characterize how the rate was changing within each distinct trend
segment.
Descriptive statistics, including frequencies and percentages, were used
to describe the distribution of pregnancy-related hospitalizations
across patient- and hospital-level characteristics, stratified by
exposure group (PDM and GDM) across racial/ethnic groups: NH-White,
NH-Blacks and Hispanics. Since national estimates were desired, all
statistical analyses were weighted using an HCUP-provided
discharge-level weight that accounted for the sampling design and
appropriately generated variance estimates. Furthermore, we calculated
the stillbirth rates in women with PDM and GDM across various
racial/ethnic groups.
Multivariable survey logistic regression was also used to produce
adjusted odds ratios (OR) that quantified the magnitude of the
association between the exposures, PDM and GDM, and the outcome
stillbirth, across various racial/ethnic groups. Statistical analyses
were performed with R version 3∙5∙ 1 (University of Auckland, Auckland,
New Zealand) and R Studio Version 1∙1∙ 423 (Boston, MA). We assumed a
5% type I error rate for all hypothesis tests (two-sided). Due to the
de-identified, publicly available nature of NIS data, the analyses
performed for this study were considered exempt by the Baylor College of
Medicine Institutional Review Board.