Study population and data collection
We used 2014 Fetal Death and Live Birth data to conduct a population-based cross-sectional study. We included foetal deaths of U.S. residents in New York City, the District of Columbia and the 41 states (eTable 1) that adopted the Standard Report of Fetal Death 2003 revision by January 1, 2014. This geographic population represents 88% of stillbirths in 2014 .
We obtained Fetal Death micro-data files via NVSS  and aggregate Live Birth data via the Centers for Disease Control and Prevention (CDC) Wonder natality online databases . Fetal Death data files included information on select maternal sociodemographic, behavioural, medical, and obstetric factors, along with birth characteristics that were potential risk factors for stillbirth. Gestational age at delivery in these data files is based on the obstetric estimate, the best estimate of the infant’s gestation in completed weeks based on the birth attendant’s final estimate of gestation.
The geographic areas mentioned above reported 24,032 stillbirths to U.S. residents, excluding induced terminations of pregnancy . We retained singleton births and excluded observations with gestational ages at delivery < 24 weeks (n = 6951) or > 43 weeks (n = 4). The former due to variation across states in the reporting requirements of foetal deaths, and likely underreporting at the lower limit of the required reporting period for each state . The latter due to possible implausibility.
We included aggregate data on live births occurring in New York City, the District of Columbia and the 41 states previously mentioned. These areas reported 3,316,293 singleton live births born 24–43 weeks’ gestation.
The outcome variable was stillbirth by timing of foetal death with respect to labour. Estimated timing of death in the NVSS data was categorized as follows: foetus alive at initial assessment and died in labour (intrapartum stillbirth), foetus not alive at initial assessment and not in labour (antepartum stillbirth), foetus not living during labour and no initial assessment was performed (possible intrapartum or antepartum death since the timing of demise with respect to labour onset is unknown), and timing of death not known. Although we included observations from vital statistics jurisdictions that indicated collecting data on estimated timing of foetal death, these data were unknown for some observations. For the purposes of this study, we combined the latter two groups as “unknown timing”. We created a categorical variable for stillbirth: intrapartum stillbirth (n = 453), antepartum stillbirth (n = 6200), unknown timing (n = 5403). Of those where timing of foetal death, with respect to labour, was not known, 19.6% (1061/5403) were in labour at the time when foetal death was first diagnosed and could have been either antepartum or intrapartum stillbirths.
We created categorical variables of the potential risk factors. Maternal sociodemographic factors: maternal age at delivery (< 20, 20–24, 25–34, ≥35 years), educational attainment (<high school diploma, high school diploma/GED, some college of Associate’s degree, ≥Bachelor’s degree, unknown), and race/ethnicity (white, Black, American Indian/Alaska Native, Asian/Pacific Islander, Hispanic). Maternal behavioural factors: timing of first prenatal care visit (first trimester, after first trimester, no care, unknown) and self-reported cigarette smoking at any time during pregnancy (yes, no, unknown). Medical and obstetric factors: parity (primipara, multipara, unknown), pre-pregnancy body mass index (BMI, kg/m2), pre-pregnancy diabetes, gestational diabetes, pre-pregnancy hypertension, gestational hypertension, and hypertension eclampsia (yes, no, unknown) . Healthcare provider factors: place of delivery (in-hospital, out of hospital, unknown) and attendant (Doctor of Medicine, Doctor of Osteopathy, Certified Nurse Midwife/Other, unknown). We included foetal sex (male, female) as a foetal characteristic.
We created a combined variable indicating whether pathological examination (either autopsy or histological placental exam) had been performed or planned, versus neither. Initiating causes of death were reported in the death file using International Classification of Diseases tenth revision (ICD-10) classification. The certifier selected one cause of death from the list of conditions and diseases and reported it separately as the initiating cause of death . We created a variable, informed generally by the ICD-PM (Perinatal Mortality) system, grouping the causes by timing of foetal death .
In this secondary data analysis, we examined the distribution of potential risk factors among stillbirths. We then fit a multivariable log-binomial regression model with each risk factor (listed in the Exposures section above) as an exposure and timing of stillbirth as the outcome (intrapartum versus antepartum), adjusting for all other listed risk factors as confounders. We estimated risk ratios (RRs) and corresponding 95% confidence intervals (CIs).
Gestational age-specific stillbirth hazard was calculated by timing of foetal death. We approximated the hazard as the number of stillbirths occurring at a specific gestational week, divided by the number of ongoing pregnancies at that gestational week (i.e., live births and stillbirths occurring during that gestational week, plus all foetuses still in utero) . We reported this approximate hazard (incidence density) per 10,000 foetus-weeks.
We calculated the conditional probabilities of intrapartum and antepartum stillbirth for each gestational week. For the conditional probability of intrapartum stillbirth, the numerator was the number of intrapartum stillbirths and the denominator included all births (live births and stillbirths), except antepartum stillbirths, occurring at a specific gestational week. Similarly, for the conditional probability of antepartum stillbirth, the numerator was the number of antepartum stillbirths and the denominator excluded intrapartum stillbirths .
We examined the frequency of pathological examination (autopsy and histological placental exam) performed or planned by timing of foetal death overall and separately in stillbirths without congenital malformations. We also examined the initiating causes of stillbirth, by timing of foetal death in a subset of the foetal death data with information on the initiating cause of death (N = 9024).
The NVSS public-use micro-data file provided already imputed maternal race and age. We performed multiple imputation (MI), with 100 imputations, using fully conditional specification to handle the remaining missing data  in the main analyses. We assumed the missing data mechanism was ignorable; that is, missing at random (MAR)  based upon the observed data. Continuous variables (in the original data files) were modelled using a linear model, and categorical variables were modelled using a logistic model; MI is robust to non-normality . After imputation, we excluded observations originally missing information on timing of foetal death . The multivariable log-binomial regression model was run using each imputed dataset. Results were combined using Rubin’s rules .
We conducted a sensitivity analysis imputing for missing outcome values. We performed additional sensitivity analyses including 1) collapsing the category of unknown foetal death timing with antepartum death since the majority of foetal deaths occur antepartum, and 2) excluding foetal deaths affected by congenital anomalies. Obstetric interventions are often withheld in cases of known lethal or serious malformations, potentially affecting foetal monitoring and likelihood of planned pathological examination.
All analyses were conducted using Stata 15.1 (College Station, TX, USA).