Sex-specific mediating effect of gestational weight gain between pre-pregnancy body mass index and gestational diabetes mellitus

Study design

This study established an observational cohort to monitor the weight of pregnant women from the beginning of pregnancy until the day of their GDM screening, typically between 24 and 28 gestational weeks. We analyzed the relationship between BMI and GWG as well as the risk of GDM. Potential confounding factors, including maternal age, multiparity, active smoking, education, and family history of diabetes, were adjusted for in the multivariable analysis.

Population and data collection

This cohort included all singleton pregnant individuals aged 18–45 who were registered with the Tianjin Women and Child Health Care Network between January 1 and December 31, 2015. Pregnant women were excluded if they had: (1) been diagnosed with diabetes or primary hypertension before the current pregnancy, (2) had a hypertensive disease during pregnancy, (3) started prenatal care later than 13 weeks+6 days, or (4) terminated their pregnancies before 24 weeks+0 days (see Fig. 1).

Fig. 1: Study flow chart.figure 1

GDM gestational diabetes mellitus.

Prenatal examination data were collected from the Tianjin Women and Children Health Care Network database, a government-administered public health system covering all communities in Tianjin, China and with antenatal care coverage rates of the local pregnant population exceeding 95%. At registration, each pregnant woman received a unique identification number that linked the antenatal care information recorded by different care providers. Basic information included date of birth, ethnicity, gravidity, parity, last menstrual period, history of chronic disease, family history of chronic and genetic diseases, and routine prenatal measurements and tests, such as height, weight, blood pressure, routine complete blood counts, urine test, blood glucose, liver, and kidney function at the first prenatal visit. Maternal weight, blood pressure, pregnancy complications, and medical treatments were continuously recorded. Antenatal care information was anonymized exported from the database, with only unique identification numbers used. The study was approved by the Ethics Committee for Clinical Research of Tianjin Women’s and Children’s Health Center. The need for written informed consent was waived due to this was retrospective analysis of data routinely collected from participants.

Screening and diagnosis of gestational diabetes mellitus

All pregnant women were screened for diabetes between 24 and 28 weeks of gestation. After the publication of the 2010 International Association of Diabetes in Pregnancy Study Group (IADPSG) recommendations, our antenatal care system considered adopting the IADPSG criteria of interpreting the oral glucose tolerance test (OGTT) results. However, due to local conditions and an attempt at minimizing changes in public health management, the previous two-step testing method was maintained, which did not comply with the IADPSG one-step criteria [22]. In this study, a 50-g 1-h glucose-challenge test was used to screen pregnant women for diabetes at community hospitals. Women with plasma glucose (PG) ≥ 7.8 mmol/L were then referred to the Tianjin Women and Children’s Health Center for a 75 g OGTT. Participants were diagnosed with GDM and defined as the GDM group if their PG with a 75 g OGTT met one or more of the following criteria: (1) fasting PG ≥ 5.1 mmol/L, (2) 1-h PG ≥ 10.0 mmol/L, or (3) 2-h PG ≥ 8.5 mmol/L [1, 2]. Pregnant women whose PG did not reach these cutoff points were defined as the non-GDM group.

Calculation of weight gain

GWG was defined as the difference between the final and baseline weights. Self-reported pre-pregnancy or measured weight in the first trimester is usually used to calculate pre-pregnancy BMI and GWG [11]. Although the accuracy of self-reported pre-pregnancy weight is uncertain, it can be easily collected or retrieved from medical records. In the present study, the mean differences between the two weight measurements were under 2 kg, which had little impact on BMI classification and GWG calculation [23]. Thus, it seemed to be a suitable choice to define the baseline weight. The final weight was measured on the day of diabetes screening tests that took place between 24 and 28 weeks of gestation. In addition, the gestational age of weight was adjusted for in the analysis.

BMI was calculated by dividing weight in kilograms by the square of height in meters. Participants were divided into four groups based on pre-pregnancy BMIs using the World Health Organization’s (WHO) BMI classification criteria: underweight (BMI < 18.5 kg/m2), normal weight (BMI 18.5–24.9 kg/m2), overweight (BMI 25.0–29.9 kg/m2), and obesity (BMI ≥ 30.0 kg/m2).

Statistical analysis

IBM SPSS Statistics for Windows (Version 21.0. Armonk, NY: IBM Corp), R statistical software (R version 4.0.3, Comprehensive R Archive Network), and GraphPad Prism 8 (San Diego, CA: GraphPad Software) were used for data analysis and figure drawing. Normally distributed continuous variables were presented as means (standard deviations) and were compared between two groups using a t-test of independent samples. Moreover, non-normally distributed continuous variables were presented as medians (interquartile ranges), and an independent sample Mann–Whitney U test was performed to compare the two groups. Categorical variables were presented as frequencies (percentages) and were compared using the chi-square test. Restricted cubic spline (RCS) analysis was used in the logistic regression to assess nonlinear associations of BMI or weight gain with GDM, while linear regression analysis was performed to analyze the effects of pre-pregnancy BMI on GWG. Logistic regression analysis was performed to analyze the effects of BMI and GWG on GDM. A two-tailed P value of less than 0.05 was considered statistically significant, and multiple imputations were performed for missing values.

Furthermore, mediation analysis was conducted to determine whether GWG could mediate the relationship between pre-pregnancy BMI and GDM. Theoretically, if the independent variable X has a certain influence on the dependent variable Y through a certain variable M, then M plays a mediating role between variables X and Y and would therefore be the mediator. Mediation analysis can help explain mechanisms in the relationship between independent and dependent variables. In the most common mediation model, X, Y, and M are continuous variables. Using linear regressions, the mediation analysis equations were as follows:

$$}}}^ = }}}^ }}} + }}} + e_3$$

(3)

c is the total effect of X on Y, a × b is the mediating effect through mediator M, and c’ is the direct effect. When only one mediator exists, c = c’+a × b. The mediating effect was measured using the equation c-c’ = a × b.

In this study, the dependent variable Y was GDM, the independent variable X was pre-pregnancy BMI, and the mediator M was GWG. Consequently, logistic regressions were used in place of the standard linear regression [24], where we applied the mediation model with X and M as continuous variables and Y as a binary variable with the following equations:

$$}}}^ = i_4 + }}} + e_4$$

(4)

$$}}} = i_5 + aX + e_5$$

(5)

$$Y^ = i_6 + }}}^ }}} + }}} + e_6$$

(6)

$$Y^ = LogitP\left( }}}X} \right) = \ln \frac}}}X} \right)}}}}}X} \right)}}$$

(7)

$$Y^ = LogitP\left( }}}M,X} \right) = \ln \frac}}}M,X} \right)}}}}}M,X} \right)}}$$

(8)

As the dependent variable Y was binary, a logistic regression was adopted for Eqs. 4 and 6. The mediator M was continuous; therefore, linear regression was adopted in Eq. 5 [25, 26]. Here, the regression coefficient a came from the regression of the continuous variable M to X (the scale of the continuous variable), while regression coefficient b came from the regression of the binary dependent variable Y to M, X (the scale of logit). Therefore, the two regression coefficients were not on the same scale and were not comparable. For the regression coefficients to have the same scale, Lacobucci proposed the Sobel method [25]. A t-test was used to test the significance of regression coefficient a in the linear regression, and the tested statistic was t = a/SE(a). Typically, when sample sizes increase to more than 30 degrees of freedom, the t-test can be viewed as a z-test, which can be written as Za = a/SE (a). In the logistic regression, the significance of the regression coefficient b was tested using the Wald χ 2 test, and the test statistic was calculated as χ2 = b/SE(b)2. The square root of the test statistic is b/SE(b), which is the t-test statistic. When sample sizes increase to more than 30 degrees of freedom, Zb = b/SE(b). After the regression coefficients a and b were converted into Za and Zb, they were on the same scale. Therefore, the size of the mediating effect of this model with binary dependent variables was Za × Zb, and a significance test of the mediating effect was used to test the significance of Za × Zb. The statistics were calculated as follows:

$$Z = \frac}}}}}(Z_)}} = \frac})}} = \frac} }}$$

(9)

MacKinnon and Cox suggested applying the distribution-of-the-product method to build confidence intervals (CIs) for the mediating effect. The RMediation software package with R software was automatically operated to obtain the asymmetric CIs of Za × Zb [27]. A significant mediating effect was defined as a CI that did not include zero (asymmetric interval).

留言 (0)

沒有登入
gif