Pregnant women included in this study were a subcohort of the ongoing Shanghai Preconception Cohort Study (SPCC; ClinicalTrials.gov NCT02737644) [17], who were enrolled between March 2016 and December 2018 from one of the study sites—a tertiary maternity hospital where HbA_1c and fasting blood glucose (FBG) levels were routinely examined at the first antenatal visit for all pregnant women. This maternity hospital is one of the largest delivery hospitals (with >20,000 births per year) during the study period and accounts for over 20% of the annual deliveries in the city. Participants were enrolled at their first antenatal visit at early pregnancy clinics, and each woman had only one medical record for this study. Biochemistry and SPL diagnosis data were extracted from the hospital’s electronic medical record system. Among all pregnant women during the study period (N=13,129), 10,773 were eligible for the primary analysis after the exclusion of those who met any one of the following criteria: missing medical records after the first antenatal visit, missing information regarding HbA_1c or FBG levels at entry, having received artificial abortions, self-reported diabetes before pregnancy (ie, an HbA_1c level of ≥6.5% or FBG level of ≥7.0 mmol/L at the first antenatal visit), or having received insulin treatment during pregnancy or taking oral hypoglycemic drugs before or during pregnancy. We defined this group as pregnant women without diabetes.

Ethics approval for this study was sought from the institutional Ethics Committee of the Children’s Hospital of Fudan University (2016–49). Written informed consent was obtained from all participants before recruitment. All data were anonymously analyzed.

We treated the fasting HbA_1c levels in early pregnancy as the main exposure in this study. Given that FBG levels are also an index of glycemic control, both HbA_1c and FBG levels were abstracted from the medical records for analyses. According to the routine practice of the hospital where this subcohort was recruited, HbA_1c and FBG levels were measured in 2 hours using a venous blood sample after overnight fasting (>8 hours of fasting) at their first antenatal care visit. Venous blood for the HbA_1c test was collected in an EDTA-containing tube and determined using high-performance liquid chromatography (Bio-Rad) in the hospital’s certified standard clinical examination center following standard protocols.

A series of variables regarding known or suspected risk factors for SPL were considered covariates in the association analysis [18]. As described elsewhere, demographic characteristics and pregnancy history of the participants upon enrollment were collected through a prespecified standard self-administered questionnaire and an interview with the obstetric nurse during the first antenatal visit. Maternal preconception BMI (pre-BMI) was calculated using self-reported measures of prepregnancy body weight and categorized as normal weight or overweight. We used both the Chinese standard (≥24 kg/m²) and the international standard (≥25 kg/m²) to define the overweight status [19,20]. We defined a family history of diabetes as having at least 1 first-degree relative diagnosed with diabetes. Smoking exposure and alcohol drinking were defined as smoking cigarettes or having been exposed to second-hand cigarette smoke, and as consuming any alcoholic beverages within 3 months before or during the current pregnancy, respectively. Folic acid supplementation (FAS) was defined as having a regular intake of pure folic acid tablets or multivitamins containing folic acid before or during early pregnancy. Gestational weeks at enrollment were routinely determined by the last menstruation period and confirmed through a routine ultrasonographic examination. Participants were defined as having a history of adverse pregnancy outcomes if they had abortions, preterm delivery, stillbirths, or ectopic pregnancy in previous pregnancies.

The recorded SPL cases from the hospital’s electronic medical system were fetal deaths occurring before 28 gestational weeks in accordance with the Chinese clinical guidelines [4]. Stillbirth (fetal deaths after 28 gestational weeks), and artificial abortions due to ectopic pregnancies, molar pregnancies, or any clinically recognized disorders were not considered SPL in this study. Trained staff verified these diagnoses through a personal telephone interview with pregnant women or their husbands before the analysis.

Continuous variables were reported as mean (SD) values for a normal or approximate normal distribution and median (IQR) values for a skewed distribution, and 2-tailed unpaired Student t tests or Mann-Whitney U tests were used for comparisons between the SPL and non-SPL group, respectively. Normality was visually inspected using frequency histograms. Categorical variables were summarized as frequencies and percentages, and chi-square tests were used for the group comparisons.

Our primary aim was to investigate the associations of HbA_1c and FBG levels as continuous variables with SPL risk. We used generalized linear models with binomial family and log link functions treating HbA_1c (rescaled through dividing by 0.5) and FBG levels as continuous variables to estimate crude and adjusted risk ratios (aRRs) and 95% CIs. A 0.5% absolute increment was chosen for the HbA_1c level because it reflects a clinically important change [21]. Considering that a HbA_1c level of 5.9% in early pregnancy has been suggested as the cutoff for identifying women at increased risk of adverse pregnancy outcomes [11], we further assessed the associations based on dichotomous HbA_1c (coded as 1 for a HbA_1c level of ≥5.9% and 0 for an HbA_1c level of <5.9%). We used a restricted cubic spline (RCS) regression model with 3 knots (5th, 50th, and 95th percentile levels) fitted in R (rms package) to assess the potential nonlinear dose-response relationship of HbA_1c levels with SPL risk. Adjusted covariates included maternal age, education level, pre-BMI, gestational week at HbA_1c examination, gravidity, history of adverse pregnancy outcomes, family history of diabetes, FAS, and exposure to smoking and drinking during the periconception period.

We conducted an exploratory analysis to investigate the HbA_1c cutoff indicating pregnant women at an increased risk of SPL by using a series of regression models using HbA_1c levels from the median level of the study population (5.1%) to 6.0% in a 0.1% interval without adjustment for multiple testing.

As there were 2147 participants—1360 (10.5%) with missing medical records after the first antenatal visit and 787 (6.1%) with missing HbA_1c data—we conducted 2 sensitivity analyses to test the robustness of the main results after multiple imputation with chained equations based on a missing at random assumption. First, we added 1360 pregnant women with missing SPL data after imputation (sensitivity analysis 1: n=10,773+1360). Second, we included 787 pregnant women with missing HbA_1c levels at enrollment after imputation (sensitivity analysis 2: n=10,773+787). To ensure that our main results are free from potential bias from unmeasured confounders, we further repeated the primary association analyses within subgroups with a potentially low risk of SPL, including pregnant women younger than 35 years, without overweight, a family history of diabetes, a history of adverse pregnancy outcomes, or smoking or alcohol drinking exposures. Statistical analyses were performed by Stata (version 16.0; StataCorp) and R package (version 3.6.1; The R Foundation), and all statistical tests were 2-sided at a significance level of .05. A post hoc power analysis was performed for the main association analyses of continuous HbA_1c levels with SPL risk, which revealed that with the current sample size, associations with a risk ratio (RR) lower than 0.84 or greater than 1.19 for continuous exposures of interest will ensure a power of ≥80% at an α level of 5%.

A total of 10,773 eligible pregnant women were included in the main analysis (Figure 1), with a mean age of 30.5 (SD 4.0) years and an average of 10.8 (SD 1.7) weeks of gestation at enrollment (Table 1). The majority of participants (n=9866, 91.6%) had a college or higher level of education, 4394 (40.9%) had more than 1 gravidity, 2056 (19.1%) had overweight before pregnancy, and 3874 (36.3%) had a family history of diabetes. Smoking and alcohol consumption were reported by 1043 (9.8%) and 857 (8.0%) pregnant women, respectively. Besides, 2817 (26.4%) pregnant women had a history of adverse pregnancy outcomes. The distribution of HbA_1c and FBG levels approximately met a normal distribution, with a mean value of 5.09% (SD 0.30%) and 4.49 (SD 0.78) mmol/L, respectively. The 2 markers were weakly correlated (r=.19; P<.001) and overlapped in quartiles (Figure S1 in Multimedia Appendix 1). We summarized the characteristics of the 10,773 eligible pregnant women and 2147 women with missing medical records or HbA_1c data (Table S1 in Multimedia Appendix 1), which were overall similar.

The incidence rate of SPL was 2.5% among 10,773 pregnant women in this study. Compared to pregnant women without SPL, those with SPL were 1.6 years older (mean ages of the SPL and non-SPL groups were 32.0, SD 4.4 vs 30.4, SD 3.9 years, respectively) during pregnancy, were more likely to have overweight before pregnancy (23.8% vs 19.0%), and had higher levels of HbA_1c (5.15% SD 0.33% vs 5.09% SD 0.30%; albeit in the clinically normal range) and FBG (4.64, SD 0.39 vs 4.48, SD 0.47 mmol/L) at the first antenatal visit.

As shown in Table 2, maternal HbA_1c levels showed significant positive associations with SPL risk with an unadjusted RR of 1.34 (95% CI 1.10-1.63) per 0.5% increase in HbA_1c levels, and the association remained significant after adjusting for covariates (aRR 1.23, 95% CI 1.01-1.50; P=.04). The RCS model showed a linear association of SPL risk with increasing HbA_1c levels through the whole range of HbA_1c values (P=.77 for the nonlinearity test; Figure 2). No significant association was found between a HbA_1c level of ≥5.9% and SPL risk after adjusting for covariates. Every 1 mmol/L increment in maternal FBG levels was associated with a >2-fold higher risk of SPL (aRR 2.12, 95% CI 1.61-2.80; P<.001).

In the first sensitivity analysis including 1360 pregnant women with missing medical diagnosis records, the association between HbA_1c levels and SPL risk did not substantially change (aRR per 0.5% increment 1.25, 95% CI 1.02-1.53; P=.03; Table 2). Similar results were also observed in the second sensitivity analysis upon including pregnant women with missing data on HbA_1c levels (aRR per 0.5% increment 1.25, 95% CI 1.01-1.55; P=.04).

Our exploratory analyses showed that the strength of the associations increased markedly from below 1.2 to 1.6 at an HbA_1c level of 5.6% (aRR 1.60; 95% CI 1.01-2.54; P=.048) and increased further at higher cutoff levels, although significance was not achieved at HbA_1c levels of 5.9% and 6.0% (Figure S2 in Multimedia Appendix 1). In further subgroup analyses including low-SPL risk populations without overweight, alcohol drinking, a family history of diabetes, and a history of adverse pregnancy outcomes, the associations of every 0.5% increase in HbA_1c levels with SPL risk remained very similar compared to those in the main analysis (aRR 1.32, 95% CI 1.07-1.63; aRR 1.18, 95% CI 1.01-1.38; aRR 1.24, 95% CI 1.03-1.48; and aRR 1.73, 95% CI 1.26-2.36; respectively; Table S2 in Multimedia Appendix 1).

In this large prospective cohort study, we provide solid evidence that among pregnant women without diabetes, HbA_1c levels in early gestation within the clinically normal range were associated with an increased risk of SPL in a linear dose-response manner. Although far below the recommended threshold for diagnosing gestational diabetes, HbA_1c levels may indicate an increased subsequent SPL risk in pregnant women in general. Our findings in a prospective cohort are novel and deepen our understanding of the important pathophysiologic role of impaired maternal glycemic metabolism in the development of SPL.

Several studies have investigated the adverse effect of elevated HbA_1c levels on SPL risk only among pregnant women with diabetes and have reported conflicting findings [12-14,22]. A case-control study including 432 control women and 386 women with type 1 diabetes found that the rate of SPL did not significantly differ between the 2 groups (16.1% vs 16.2%, respectively) [12]. However, another case-control study among women with type 1 diabetes reported that elevated HbA_1c levels were associated with SPL risk [13]. Another cohort study of 573 women with type 1 diabetes observed a linear association between HbA_1c levels and the risk of adverse pregnancy outcomes including SPL when HbA_1c levels were >7.0% [14]. This is the first prospective cohort study demonstrating the link between maternal HbA_1c levels in early pregnancy and SPL risk among pregnant women without diabetes. These associations were still significant in the population at a low risk of SPL. We also observed a highly overlapped distribution of FBG levels among HbA_1c categories, and consistent associations with SPL were also observed with regard to FBG levels. However, regarding their clinical application, FBG level is a less ideal marker than HbA_1c level because the former is less stable, requires a fasting state for examination, and has relatively greater intraindividual variability [23]. Our findings suggest that attention to glycemic control should not be limited to pregnant women with diabetes but should also include those with high HbA_1c levels within the clinical range during early gestation for the related increased risk of SPL.

Our study addresses 2 important issues with important clinical and research implications. First, our findings expand on the literature on risk factors for SPL, an adverse pregnancy outcome whose modifiable determinants remain poorly understood. Second, we address a risk factor that is known to be modifiable through lifestyle and pharmacological interventions. Measurement of HbA_1c levels during pregnancy is conventionally used to monitor glycemic control in pregnant women with diabetes [24]; this marker is superior to FBG levels, mainly for its greater stability but lesser variability among individuals and its nonrequirement of fasting [23]. Given the special physiological status of pregnancy, the target HbA_1c level for ideal glycemic control remains inconclusive. In China, based on recommendations for the general population, an HbA_1c level less than 6.5% is recommended to indicate ideal glycemic control before conception [25]. An HbA_1c level of ≥5.9% is recommended by the ADA’s guidelines as an indicator for screening pregnant women with a higher risk of preeclampsia, macrosomia, shoulder dystocia, and perinatal death [11]. Our findings add new evidence for the linear relationship between glucose metabolism markers and SPL risk when HbA_1c levels are clinically normal [11], and the magnitudes of the associations are even stronger in those exposed to smoking or those who have a history of adverse pregnancy outcomes.

The mechanism underlying the association between maternal HbA_1c levels and SPL in women without diabetes remains unclear. Animal studies found that poor glycemic control may facilitate premature programmed cell death of key progenitor cells of the blastocyst and promote pregnancy loss [26]. In addition, poor glycemic control may interfere with implantation by inhibiting trophectoderm differentiation and increasing oxidative stress, thus affecting the expression of critical genes essential for embryogenesis [27,28]. Further population studies exploring the biological mechanisms underlying the association between HbA_1c levels and SPL in pregnant women without diabetes are warranted.

The findings of our exploratory analysis are clinically relevant, indicating that pregnant women with HbA_1c levels above 5.6% need attention for owing to a potentially increased risk of SPL. Given that HbA_1c levels are 0.5%-0.6% lower in early pregnancy than in nonpregnant women [29], HbA_1c levels over 5.6% at early gestation may reflect an elevated HbA_1c status that occurred before or at pregnancy (ie, prediabetes status), which accounts for over 13% of the population [6]. Another large cohort study and systematic review have reported associations between an HbA_1c level of <6.4% with increased severe maternal morbidity and subsequent gestational diabetes mellitus [30,31]. Taken together, based on the aforementioned evidence and our findings, we propose that monitoring of HbA_1c levels in early pregnancy is necessary general pregnant women, and HbA_1c levels exceeding 5.6% might be considered an indicator of high risk for SPL and additional medical care.

The merits of our study include the prospective nature of the data, the large sample size, and the consistent results from robust analyses, which make our findings convincing. However, several limitations exist. First, residual confounding, such as uterine abnormalities, chromosomal abnormalities, antiphospholipid syndrome, and thyroid disorders, cannot be completely ruled out due to the nature of observational studies. Data on important biomarkers such as insulin levels during early gestation were not available. Second, in the exploratory analyses, we did not correct for multiplicity. Prospective multicenter cohort studies investigating whether a maternal HbA_1c level of 5.6% in early pregnancy could indicate an increased risk of SPL are warranted in the future. Third, compared with the rate of clinically registered pregnant women ending in SPL (11%-20%), the SPL rate in this cohort was much lower. Given the nature of the study population, which was recruited at their first antenatal care visit (with a median of 10 weeks of gestation), SPL occurred earlier than our observation was inevitably missed, selection bias exists, and our findings are not generalizable to the entire general population. The study population we selected, comprising pregnant women without diabetes, may include some cases of mild gestational diabetes diagnosed midgestation and without insulin treatment. Moreover, potential biases such as outcome misclassification could not be ruled out from the association analyses based on imputation.

This study is the first to document that maternal HbA_1c levels in early pregnancy are associated with the subsequent risk of SPL in a dose-response manner among pregnant women without diabetes. Our findings support the need to monitor HbA_1c levels for identifying high risk of subsequent SPL in pregnant women in general and expand on the growing literature linking overall metabolic health to reproductive and pregnancy health among otherwise healthy women.