This study was supported by the National Free Preconception Health Examination Project (NFPHEP), providing rural married couples who planning to conceive within 6 months with free preconception health examinations in 220 pilot counties in 30 provinces in China since 2010. In this study, we analyzed data registered in Yunnan province during 2010 to 2018, which recorded the general information of 217,070 pregnancies including parental characteristics, disease history, living habits before and during pregnancies, and final pregnancy outcomes. Compared with other provinces or regions of China, Yunnan province is less developed and has a very high level of ethnic diversity, which mostly leads to a relatively low education level.
In this study, analysis limited to all singleton births that delivered between 20 and 42 weeks’ gestation. Of all 217,070 pregnancies, we sequentially excluded 13,359 pregnancies with missing fetus number and 1,103 multiple births, 1,912 births with gestational age at delivery < 20 and > 42 weeks’ gestation, 2 births with missing gestational age. 2,972 pregnancies with missing maternal education information were further removed. After the above exclusions, 197,722 singleton births remained for analysis (Fig. 1). The possible selection bias resulted from missing would be identified by comparisons of characteristics between the included sample and excluded sample with missing information, which would be presented in the discussion section.
Maternal education was obtained by questionnaire, which was classified into illiteracy, primary school, junior middle school, senior middle school, undergraduate, and above, which corresponded to the Chinese current education system. For comparisons with convenience, education level was further categorized by the number of completed years of schooling, which were low education level (less than 6 years) corresponding to illiteracy and primary school, mid (6–9 years) corresponding to junior middle school and high (more than 9 years) corresponding to senior middle school and undergraduate and above.
Our outcomes of interest were presented as follows: stillbirth, neonatal mortality, PTB, and LBW. Stillbirth was defined as a fetal loss occurring at 20 weeks or longer. Neonatal mortality referred to deaths occurred in the first 4 weeks of life. PTB was defined as live births before 37 complete weeks of gestations. LBW was defined as live births with birthweight < 2,500 g.
Covariates included maternal ethnicity (Han, Yi, Dai, and other ethnic groups), parity (0, 1, > 1 previous deliveries), maternal age (< 30, >=30), work (farmers, non-farmers), BMI (< 24.9 kg/m2, >=24.9 kg/m2), economic stress (never, slightly, considerable), smoking status (smoke-free, passive smoking (only), smoker), folate using (unused, irregular used, regular used), IUD use, fetus sex, prior adverse outcomes and maternal diseases including hypertension, thyroid disease, syphilis and hepatitis B. For categorical factors whose missing value proportion were over 20%, we regarded missing value as a separate category rather than excluded cases with missing values. Also, two dichotomous variables, gestational age (< 37, >=37 weeks) and birthweight (< 2,500 g, >=2,500 g) were created for all pregnancies, which needs to be distinguished from the definitions of PTB and LBW that refer to the outcome of live births.
Comparisons concerning above characteristics between women with low, mid, and high educational levels were performed by chi-squared tests. The distribution of maternal characteristics among adverse pregnancy outcomes were also compared. Binary logistic regression analyses were modeled to estimate the effect of maternal education level, with and without control for covariates of maternal ethnicity, parity, maternal age, work, maternal BMI, economic stress, smoking in early pregnancy, folate using, IUD use, and prior adverse outcomes.
Gestational age and birthweight were known to be important mediators for stillbirth, which considered as causal confounders. In this study, we expected maternal education level to act primarily through the effect on the risk of gestational age and birthweight, then impacted the risks of stillbirth and neonatal mortality. To demonstrate this, we made a mediation analysis using difference method, observing whether any considerable difference occurred in the coefficients of education levels in the multivariable regressions without and with gestational age and birthweight variables. The percentage change in odds ratio (ORC) due to adding variables to the model was calculated to measure the mediation effect with the formula:
([OR model-OR model + mediator] / [1-ORmodel] * 100),
on the promise that the adjusted model showed a significant association.
Additionally, consider the heterogeneity among the multi-ethnic population, the multivariable analysis above was run again in three main ethnic groups separately including Han, Yi, and Dai, which totally accounted for 83% of the study population. All statistical analyses were performed using R software 3.6.0. A P-value of < 0.05 was considered statistically significant in all analysis.