1) Geographic and demographic characteristics
The study was conducted in Yunnan province, which located in southwest China and bordered Myanmar in the west and Laos and Vietnam in the south. Yunnan is distinct from other provinces of China for a very high level of ethnic diversity and is the only province including all of China’s 55 minorities [21]. Over 38% of the province population are members of ethnic minorities, including the Yi, Bai, Hani, Dai, Miao, and so on.
2) The NFPHEP project in Yunnan
The National Free Preconception Health Examination Project (NFPHEP) was a nationwide prospective population-based study, implemented by the Chinese National Health and Family Planning Commission and Ministry of Finance in 220 pilot counties in 30 provinces of China, to offer free preconception health examinations to rural married couples who planned to conceive within the next 6 months. General information including parental characteristics, medical and reproductive history, living habits and other exposing status related to adverse birth outcomes were recorded by local health workers. The pregnancy outcomes were identified by local hospitals. Detailed design and implementation of the NFPHEP are described elsewhere [22, 23].
This study was based on the data on couples enrolled in the NFPHEP from Yunnan province during 2010–18. By 27 August 2018, there were 1,140,417 families enrolled, of which 223,422 women conceived within 6 months were closely followed. Final records showed that 217,070 women had already delivered, 3,668 remained undelivered until the end of the study and 2,684 dropped out of the last follow-up. Our study focused on 217,070 gestations with definite outcomes, of which the top five ethnic groups accounted for 88.25%, where Han accounted for 61.34%, Yi 16.96%, Dai 4.47%, Miao 2.77% and Hani 2.71%. The detailed statistics of participated ethnic groups were displayed in Supplementary Table A, Additional Files.
3) Definition and Assessment of variables
For stillbirth, we used the definition of fetal loss occurred on or after 20 weeks or birth body weight over 500 g if gestational age (GA) was unavailable [24,26]. Fetus loss referred to those deaths prior to the complete expulsion or extraction from its mother, irrespective of the duration of pregnancy [1]. The gestation age was primarily determined by the interval between the first day of the last menstrual period and the date of delivery, and birth weight was measured within the first hour after delivery [1]. Ethnicity information was collected based on identification card of participants.
A variety of factors previously reported to be associated with racial or ethnic disparities had been incorporated in this paper with the following: maternal sociodemographic characteristics including maternal age, education, BMI, height, occupation (farmers and non-farmer), economic stress [15,11,8,25,27,28]. Economic stress in our study was self-reported according to the question “how much stressed do you feel in your economic situation?” with the options “never”, “slightly”, “considerable” ; pregnancy-associated characteristics including parity, adverse pregnancy history (induced abortion, natural abortion and stillbirth) and birth order [8,15], and maternal diseases (any report on maternal health reports including maternal hypertension, thyroid disease, syphilis, hepatitis B, anemia, diabetes, renal diseases and epilepsy) [8,15]; maternal substance use before and during early pregnancy including tobacco, folate and Intrauterine devices (IUD) [11]; and fetal characteristics including fetal sex, gestation age and the diagnosis of congenital anomaly [8,15]. Continuous variables were further categorized as follows based on a previous risk stratification. Maternal age was divided into 3 groups: <20 years, 20-35 years, >35 years [26-27]. BMI (kg/m2) were grouped as underweight (<18.5), normal weight (18.5-24.9), overweight (25-29.9), obese (≥30) according to conventional World Health Organization (WHO) [28]. Maternal height was dichotomized as short maternal height (<150 cm) and non-short maternal height( 150 cm) [8]. Education level was classified into 3 groups: low (completed primary school or lower), middle (completed middle school), high (completed high school or higher), to avoid categories with small number of participants. For categorical factors whose missing value proportion were over 20%, we treated their missing values as a separate category rather than discarded them directly in our analyses.
4) Statistical analysis
We mainly incorporated the Han majority and other four main minorities including Yi, Dai, Miao, Hani into the analysis. Principal analysis was limited to singleton gestations that delivered during 20-42 weeks’ gestation, with the best clinical estimate of gestation age [29]. Pregnancies with missing plurality and gestation age would be firstly removed. The possible selection bias resulted from missing plurality and gestation age would be identified by the sensitivity analysis, which would be presented in the discussion section.
Comparisons concerning maternal and fetus characteristics among five ethnic groups were performed by chi-squared tests. The analyses to evaluate the influence of maternal age, education and birth order on stillbirth were stratified by ethnicity. Ethnic disparities were examined in the subsets of gestational age fell in 20-23 weeks, 24-27 weeks, 28-31 weeks, 32-36 weeks, 37-40 weeks, and 41-42 weeks, respectively. Stillbirth hazards were estimated by life-table analysis as the number of stillbirths occurring during different intervals divided by the number of ongoing pregnancies at the beginning of the corresponding intervals minus half of the total live births in this interval. The relative rate (RR) with 95% confidence intervals (95% CI) of stillbirth hazard (the Hans were the reference group) was calculated in each gestational age interval [30].
Multivariable logistic regressions were modeled by sequentially controlled several sets of covariates. In model 1, we examined the association between maternal ethnicity and stillbirth risk after maternal age, education level, and birth order controlled which had been considered as important confounders by prior researches [7,12,25]. In model 2, we added smoking status, BMI, height, occupation, economic stress, folate use and IUD use, adverse pregnancy history. In model 3, we introduced the preterm birth variable (gestation age <37 weeks) known to be an important mediator of risk of stillbirth [12]. Compared with the Han majority, the excess stillbirth risk (ESR) in minorities (%) were computed as follows:
Excess stillbirth risk = (RR - 1)/RR,
where RR = adjusted relative risk of stillbirth. The adjusted odds ratios were used to approximate the RR. We used the ESR to reflect the magnitude of ethnic disparities in stillbirth risk. The analysis was repeated again in the group of women excluding those with maternal diseases and the group of deliveries excluding those congenital anomalies, respectively. All analysis was performed by R statistical software (R.3.6).