Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2023 Mar 1.
Published in final edited form as: Epidemiology. 2022 Mar 1;33(2):278–286. doi: 10.1097/EDE.0000000000001454

Gestational Diabetes in Twin Versus Singleton Pregnancies with Normal Weight or Overweight Pre-Pregnancy Body Mass Index: The Mediating Role of Mid-Pregnancy Weight Gain

Michelle C Dimitris a, Jay S Kaufman a, Lisa M Bodnar b, Robert W Platt a, Katherine P Himes c, Jennifer A Hutcheon d
PMCID: PMC8810679  NIHMSID: NIHMS1762111  PMID: 34907972

Abstract

Background:

Although gestational diabetes might be more common in twin versus singleton pregnancies, reasons for this are unclear. We evaluated the extent to which the relationship is explained by higher mid-pregnancy weight gain within normal weight and overweight pre-pregnancy body mass index (BMI) strata.

Methods:

We analyzed serial weights and glucose screening and diagnostic data abstracted from medical charts for twin (n=1397) and singleton (n=3117) pregnancies with normal or overweight pre-pregnancy BMI delivered from 1998–2013 at Magee Women’s Hospital in Pennsylvania. We used causal mediation analyses to estimate the total effect of twin versus singleton pregnancy on gestational diabetes, as well as those mediated (natural indirect effect) and not mediated (natural and controlled direct effects) by pathways involving mid-pregnancy weight gain.

Results:

Odds of gestational diabetes were higher among twin pregnancies [odds ratios (ORs) for total effect=2.83 (95% CI 1.54; 5.19) for normal weight and 2.09 (95% CI 1.16; 3.75) for overweight pre- pregnancy BMI], yet there was limited evidence that this relationship was mediated by weight gain [ORs for natural indirect effect=1.21 (95% CI 0.95; 1.54) for normal weight and 1.06 (95% CI 0.92; 1.21) for overweight pre-pregnancy BMI], and more evidence of mediation via other pathways [ORs for natural direct effect=2.34 (95% CI 1.24; 4.40) for normal weight and 1.97 (95% CI 1.08; 3.60) for overweight pre-pregnancy BMI].

Conclusions:

While twin pregnancies with normal weight or overweight pre-pregnancy BMI experience higher odds of gestational diabetes versus singletons, most of this effect is explained by pathways not involving mid-pregnancy weight gain.

Keywords: Gestational Diabetes, Gestational Weight Gain, Pregnancy, Twin Pregnancy, Mediation Analysis

Introduction

Gestational diabetes describes glucose intolerance caused by insufficient insulin and/or excessive blood glucose first diagnosed during pregnancy.1 Studies estimate a two- to three-fold increased risk of gestational diabetes among twin versus singleton pregnancies;2–7 however, differences in gestational diabetes by plurality are not always observed.8,9 Gestational diabetes is associated with adverse perinatal outcomes in both twin and singleton pregnancies, including preeclampsia, large for gestational age, and neonatal intensive care unit admissions, although evidence for adverse effects of gestational diabetes in singletons is more conclusive than that in twins.10–12

In singletons, gestational weight gain is a potentially modifiable risk factor for gestational diabetes. Observational studies have found that excessive weight gain before glucose screening is correlated with higher risk for gestational diabetes among singletons.13–15 Systematic reviews of randomized controlled trials for the prevention of gestational diabetes found that diet- and exercise-based interventions decreased average weight gain by 0.89 kg and risk of gestational diabetes by 15% to 44%.16,17 In mothers with overweight/obese pre-pregnancy body mass index (BMI), who are at higher risk for gestational diabetes, physical activity interventions decreased average weight gain by 1.14 kg, and risk of gestational diabetes by 29%.18 Taken together, these studies may support a causal relationship between weight gain and risk of gestational diabetes, although it is plausible that exercise- and diet-based interventions themselves also decrease risk for gestational diabetes.

Current guidelines recommend that twin gestations gain more weight during pregnancy than singletons,19 and total weight gain is observed to be higher in twin versus singleton pregnancies.3 It is possible that this higher weight gain contributes to the increased risk for gestational diabetes in twins compared to singletons; however, the role of weight gain in conferring higher risk of gestational diabetes in twins pregnancies is unclear. Few studies have described the relationship between weight gain and gestational diabetes in twin pregnancies.20–23 Moreover, many previous studies evaluated total weight gain, which may be impacted by interventions that limit weight gain after gestational diabetes is diagnosed mid-pregnancy.24 The role of weight gain in conferring additional risk of gestational diabetes in twin pregnancies, and how the relationship between weight gain and gestational diabetes relates to that in singletons, is poorly understood. Since current weight gain guidelines for twins are provisional, and were primarily established using limited data on neonatal outcomes,19 it is important to determine if higher weight gain in twins contributes to higher risk of gestational diabetes compared to singletons.

We investigated the extent to which increased risk of gestational diabetes in twin compared to singleton pregnancies is explained by higher mid-pregnancy weight gain among women with normal weight and overweight pre-pregnancy BMI.

Methods

Study Population

We analyzed previously-collected data from two parent studies of pregnancies delivered at Magee Women’s Hospital in Pittsburgh, Pennsylvania.25,26 The first study was a cohort of all twin pregnancies delivered from 1998–2013 (i.e. twin cohort).25 The second study was a case–cohort of singleton pregnancies delivered from 1998–2011; specifically, in each of six pre-pregnancy BMI groups (i.e. underweight or <18.5 kg/m2, normal weight or 18.5–24.9 kg/m2, overweight or 25–29.9 kg/m2, obese I or 30–34.9 kg/m2, obese II or 25–29.9 kg/m2, and obese III or ≥40.0 kg/m2), the parent study selected a random sample of all singleton pregnancies (i.e. singleton subcohort) and a random sample of singleton pregnancies with physician diagnosis of gestational diabetes (i.e. singleton cases).26

Data Sources

The parent study collected maternal and pregnancy characteristics, including self-reported pre-pregnancy weight, from the hospital-maintained Magee Obstetric Maternal and Infant database supplemented by linked vital statistics information from the state of Pennsylvania. The parent study abstracted serial maternal weights and glucose screening and tolerance test information from medical charts.

Inclusion/Exclusion Criteria

We included only twin and singleton pregnancies with either normal or overweight pre-pregnancy body mass index (BMI) in the current study to limit effect measure modification of the relationship between weight gain and gestational diabetes by pre-pregnancy BMI. We excluded the second twin pregnancy per mother and twin pregnancies with monochorionic placentation. We additionally excluded twin and singleton pregnancies resulting in delivery or fetal death before 24 weeks’ gestational age, as well as twin and singleton pregnancies with missing pre-pregnancy weight or height, missing or implausible gestational age at birth, positive or missing diagnosis of pre-existing diabetes, and missing values for covariates. In primary analyses, we excluded pregnancies with no glucose screening or tolerance tests conducted between 20 and 32 weeks’ gestational age and/or no available weight measurements before these tests.

Gestational Diabetes

In this setting, clinicians generally determined gestational diabetes by applying Carpenter and Coustan criteria to 50g glucose screening tests conducted between 24–28 weeks’ gestational age and subsequent 100g glucose tolerance/diagnostic tests (eAppendix 1).27,28 Gestational diabetes was occasionally diagnosed at glucose screening when blood glucose was ≥200 mg/dL. Additionally, some pregnancies had inconclusive, incomplete, or missing glucose screening and tolerance test data. For this reason, and to leverage all available pregnancies, we determined gestational diabetes classification from glucose tolerance test, glucose screening test, and International Classification of Disease (ICD-9) code for physician diagnosis in decreasing priority (eAppendix 2).

Gestational Weight Gain

We calculated mid-pregnancy weight gain in kilograms by subtracting self-reported pre-pregnancy weight from a prenatal weight measurement. We used the maternal weight measured at or most recently before the first glucose screening/tolerance test conducted from 20–32 weeks’ gestational age. We did not consider glucose screening or tolerance tests conducted before 20 weeks’ gestational age in weight measurement selection.

Statistical Analysis

We first examined crude relationships between weight gain and gestational diabetes separately in twins and singletons and within normal and overweight pre-pregnancy BMI strata using density plots and logistic regression. We then analyzed twin and singleton pregnancies together and investigated the extent to which increased risk of gestational diabetes in twin compared to singleton pregnancies was explained by higher weight gain using causal mediation analysis.29 Briefly, causal mediation analyses assess the extent to which an observed relationship between an exposure (i.e. twin versus singleton pregnancy) and outcome (i.e. gestational diabetes) can be attributed to mechanisms that include the mediator of interest (i.e. mid-pregnancy weight gain). A causal diagram of our hypothesized relationships and corresponding assumptions are presented in Figure 1.

Figure 1.

Figure 1.

Open in a new tab

Causal diagram for the relationship between plurality, gestational weight gain, and gestational diabetes. C and its solid arrows indicate measured confounders (gestational age at weight gain measurement, parity, maternal age, delivery year, marital status, ever smoker, insurance, and maternal race), U3 and U4 and its dashed arrows indicate unmeasured confounders, the presence of which violate identifiability conditions in the presence of exposure–mediator interaction only, and U1 and U2 and its dotted arrows indicate unmeasured confounders, the presence of which violate identifiability conditions regardless of exposure–mediator interaction.

Causal mediation analysis estimates controlled direct, natural direct, natural indirect, and total effects. In our study, the natural indirect effect quantifies relative change in odds of gestational diabetes if weight gain was set to the value it would take if each pregnancy was twin versus singleton, and plurality was set to singleton for all pregnancies. The natural direct effect quantifies the relative change in odds of gestational diabetes if all pregnancies were twin versus singleton, where mid-pregnancy weight gain is set to the value it would have if plurality were singleton for each individual pregnancy. The controlled direct effect estimates the relative change in odds of gestational diabetes if all pregnancies were twin versus singleton and mid-pregnancy weight gain is set to a constant value. The total effect quantifies the relative change in odds of gestational diabetes if all pregnancies were twin versus singleton and weight gain is set to the value it would have if plurality were twin versus singleton (eAppendix 3).

Causal mediation analyses estimate models for the mediator and outcome. We used linear regression to estimate the mediator model, or the effect of plurality on weight gain, and logistic regression to estimate outcome model, or the effect of plurality on gestational diabetes while holding mid-pregnancy weight gain constant. We estimated natural direct, natural indirect, controlled direct, and total effects on the odds ratio (OR) scale, which approximated the risk ratio (RR) scale, since gestational diabetes is rare. If no effect modification of the relationship between weight gain and gestational diabetes by plurality is assumed, the natural and controlled direct effects are equivalent and can be expressed as a single value. However, we allowed for statistical interaction between plurality and weight gain in all models, regardless of statistical significance, due to biologic plausibility of effect modification by plurality. Specifically, weight gain of dichorionic twin pregnancies includes an additional fetus and placenta; thus, equivalent weight gain in singletons and twins may have different implications for risk of gestational diabetes. In the absence of exposure–mediator interaction, the natural and controlled direct effects are the same and can be expressed as a single value. In the presence of exposure–mediator interaction, the controlled direct effect could be expressed as several values at which mid-pregnancy weight gain could be set for all pregnancies. We plotted controlled direct effects by the value at which the mediator was set across a range of plausible values for weight gain.

The parent singleton case-cohort study oversampled cases of gestational diabetes. Since our exposure of interest was twin versus singleton pregnancy, it was important to maintain the relationship between plurality and gestational diabetes observed in the study population. For this reason, we conducted two separate analyses: (1) we pooled twin cohort and singleton subcohort pregnancies and disregarded sampling weights (i.e. unenriched models); (2) we pooled twin cohort, singleton subcohort, and singleton case pregnancies, removed duplicates (i.e. singleton subcohort pregnancies that were also selected as cases), and weighted all models by the inverse probability of selection into the parent and current studies (i.e. enriched models; sampling weights in eAppendix 4).30–32 eAppendix 5 illustrates pregnancies that were included in each of the unenriched and enriched models.

Our unadjusted models controlled only for gestational age in days at which weight gain was measured, which is consistent with previous research.33 Our adjusted models additionally controlled for covariates that may confound relationships between plurality, weight gain, and/or gestational diabetes, including parity, maternal age, delivery year, marital status, ever smoker, insurance, and maternal race (details in eAppendix 6). It was not possible to control for pre-existing hypertension because there were no cases of gestational diabetes among twin cohort or singleton subcohort pregnancies with normal weight pre-pregnancy BMI and pre-existing hypertension. We calculated confidence intervals for effects using the delta method (for unenriched models) and bootstrap method sampling pregnancies within strata of parent study dataset, namely twin cohort, singleton sub-cohort, and singleton case datasets, with 200 iterations (for enriched models).

Sensitivity Analyses

In primary analyses, we considered only glucose screening or glucose tolerance test(s) conducted from 20–32 weeks’ gestational age; we included in primary analysis pregnancies with glucose screening or glucose tolerance test(s) conducted before 20 weeks’ gestational only if an additional test(s) was conducted from 20–32 weeks’ gestational age. In sensitivity analyses, we both broadened and restricted these criteria; for example, we considered pregnancies and glucose screening/tolerance test(s) regardless of gestational age at which they were conducted, which incorporated pregnancies with less or more accurate or precise outcome ascertainment. Details of these and additional sensitivity analyses are available alongside results in eAppendices 15 and 16. We additionally calculated marginal effects on the OR, RR, and risk difference (RD) scale for enriched and unenriched models using contrasts of the marginal probabilities obtained from the mediator and outcome models (formulae available in eAppendix 3). Confidence intervals for enriched and unenriched marginal effects were bootstrapped as described above.

We conducted statistical analyses using Stata 14.2 (StataCorp 2015, Texas), and fit causal mediation models primarily using the paramed command34 or derivatives thereof. Ethics approvals for the parent studies were obtained from the University of Pittsburgh, and additionally from the McGill University Faculty of Medicine Institutional Review Board for the current study.

Results

Overall, we included 1397 twin cohort, 2622 singleton subcohort, and 495 singleton case pregnancies that met criteria for either primary or sensitivity analyses in the current study; eAppendices 7–9 enumerate all exclusions for the twin cohort, singleton subcohort, and singleton case pregnancy samples, respectively. For our primary exposure definition, which necessitated a glucose screening or tolerance test conducted from 20–32 weeks’ gestational age and a weight measurement either on or before the glucose screening/tolerance test, we excluded an additional 442 twin cohort, 845 singleton subcohort, and 113 singleton case pregnancies, or approximately one third of each sample.

Pregnant women in the twin cohort were older and more frequently classified as college graduate, white, married, privately-insured, and never-smoker, and more likely to be diagnosed with preeclampsia than pregnant women in the singleton subcohort (eAppendix 10). Twin pregnancies were more likely to have gestational diabetes diagnosis confirmed with glucose screening and/or tolerance test, yet also more likely to have inconclusive or missing tests compared to singleton subcohort pregnancies (Table 1). Twin pregnancies also had slightly lower gestational age at glucose screening or tolerance test compared to the singletons. eAppendices 11 and 12 display maternal and pregnancy characteristics by plurality and plurality-specific weight gain quartile for women with normal weight and overweight pre-pregnancy BMI, respectively.

Table 1.

Selected study characteristics by plurality and pre-pregnancy BMI (complete table available in eAppendix 10).

Normal Weight Pre-Pregnancy BMI Overweight Pre-Pregnancy BMI
Twin Cohort Singleton Subcohort Singleton Cases Twin Cohort Singleton Subcohort Singleton Cases
Number 914 1360 259 483 1262 236
Glucose Tolerance/Screening Test Results [n(%)]
   Confirmed No 719 (78.7) 1178 (86.6) 2 (0.8) 357 (73.9) 1051 (83.3) 3 (1.3)
   Confirmed Yes 43 (4.7) 23 (1.7) 253 (97.7) 28 (5.8) 42 (3.3) 228 (96.6)
   Inconclusive 46 (5.0) 45 (3.3) 4 (1.5) 31 (6.4) 51 (4.0) 5 (2.1)
   missing 106 (11.6) 114 (8.4) 0 (0.0) 67 (13.9) 118 (9.4) 0 (0.0)
Gestational Diabetes [n(%)]
   Yes 48 (5.3) 26 (1.9) 257 (99.2) 35 (7.2) 55 (4.4) 232 (98.3)
   No 866 (94.7) 1334 (98.1) 2 (0.8) 448 (92.8) 1207 (95.6) 4 (1.7)
Gestational Age of Glucose Screening (weeks) [median (25th; 75th percentile)] 26.4 (25.1; 27.7) 26.9 (25.7; 28.0) 27.3 (26.1; 28.4) 26.4 (25.3; 27.7) 27.0 (25.7; 28.3) 27.6 (26.4; 28.7)
   missing 280 (30.6) 429 (31.5) 59 (22.8) 162 (33.5) 416 (33.0) 54 (22.9)
Gestational Age of Weight Gain Measurement (weeks) [median (25th; 75th percentile)] 25.6 (24.1; 27.1) 26.0 (24.3; 27.6) 26.4 (24.5; 27.7) 25.7 (23.9; 27.1) 26.1 (24.6; 27.7) 26.4 (24.6; 27.9)
   missing 280 (30.6) 429 (31.5) 59 (22.8) 162 (33.5) 416 (33.0) 54 (22.9)
Mid-Pregnancy Weight Gain (kg) [mean (standard error)] 11.9 (4.4) 9.2 (4.0) 10.9 (4.7) 10.3 (4.9) 9.0 (5.1) 10.1 (5.4)
   missing 280 (30.6) 429 (31.5) 59 (22.8) 162 (33.5) 416 (33.0) 54 (22.9)
Open in a new tab

Average mid-pregnancy weight gain was 11.6 kg among normal weight twin cohort, 10.2 kg among overweight twin cohort, and 8.8 kg among both normal and overweight singleton subcohort pregnancies (eAppendices 13 and 14). When we evaluated the relationship between weight gain and gestational diabetes separately in twins and singletons, relative odds of gestational diabetes increased by factors of 1.05 (95% CI 0.99; 1.11) among twins and 1.03 (95% CI 0.98; 1.10) among singletons per 1-kg increase in weight gain within models adjusted for gestational age of weight gain measurement.

The total effect compares the odds of gestational diabetes if all pregnancies were twin versus the odds of gestational diabetes if all pregnancies were singleton via pathways mediated and not mediated by weight gain. Among normal weight pregnancies, the OR for the total effect was 2.83 (95% CI 1.54; 5.19) in the unenriched analysis and 2.14 (95% CI 1.44; 3.07) in the enriched analysis (Table 2). Among overweight pregnancies, the OR for the total effect of twin versus singleton pregnancy on gestational diabetes was 2.09 (95% CI 1.16; 3.75) in the unenriched and 1.66 (1.00; 2.66) in the enriched analysis.

Table 2.

Natural direct, natural indirect, and total effects estimated by causal mediation analyses by pre-pregnancy BMI.

Normal Weight Pre-Pregnancy BMI Overweight Pre-Pregnancy BMI
N ORNDE ORNIE ORTE N ORNDE ORNIE ORTE
Unadjusteda
   Unenrichedb 1565 2.67
(1.46; 4.88)		 1.18
(0.94; 1.48)		 3.14
(1.77; 5.59)		 1167 2.26
(1.32; 3.88)		 1.06
(0.92; 1.21)		 2.40
(1.42; 4.06)
   Enrichedc 1750 2.11
(1.34; 2.89)		 1.10
(0.87; 1.34)		 2.31
(1.63; 3.15)		 1321 1.94
(1.12; 2.94)		 1.05
(0.90; 1.21)		 2.02
(1.23; 2.87)
Adjustedd
   Unenrichedb 1565 2.34
(1.24; 4.40)		 1.21
(0.95; 1.54)		 2.83
(1.54; 5.19)		 1167 1.97
(1.08; 3.60)		 1.06
(0.90; 1.24)		 2.09
(1.16; 3.75)
   Enrichedc 1750 1.91
(1.18; 2.91)		 1.12
(0.88; 1.41)		 2.14
(1.44; 3.07)		 1321 1.58
(0.93; 2.54)		 1.05
(0.90; 1.29)		 1.66
(1.00; 2.66)
Open in a new tab
a

Adjusted only for gestational age at gestational weight gain measurement

b

Unenriched analyses included twin cohort and singleton subcohort pregnancies, did not incorporate sampling weights, and were conducted using the paramed command

c

Enriched analyses included twin cohort, singleton subcohort, and singleton case pregnancies, incorporated sampling weights, and were conducted using extensions of the paramed command

d

Adjusted for gestational age at gestational weight gain measurement, parity, maternal age, delivery year, marital status, ever smoker, insurance, and maternal race

The natural indirect effect compares the odds of gestational diabetes if all pregnancies were singleton, but each pregnancy gained the amount of weight it would have gained if it were twin, versus the odds of gestational diabetes if all pregnancies were singleton, and each pregnancy gained the amount it would have gained if it were singleton; the OR for the natural indirect effect was 1.21 (95% CI 0.95; 1.54) in the unenriched and 1.12 (95% CI 0.88; 1.41) in the enriched analysis among women with normal weight pre-pregnancy BMI; and 1.06 (95% CI 0.90; 1.24) in the unenriched and 1.05 (95% CI 0.90; 1.29) in the enriched analysis among women with overweight pre-pregnancy BMI (Table 2). The natural direct effect compares the odds of gestational diabetes if each pregnancy were twin but gained the amount of weight it would have gained if it were singleton versus the odds of gestational diabetes if each pregnancy were singleton and gained the amount of weight it would have gained if it were singleton; the OR for the natural direct effect was 2.34 (95% CI 1.24; 4.40) in the unenriched and 1.19 (95% CI 1.18; 2.91) in the enriched analysis among women with normal weight pre-pregnancy BMI; and 1.97 (95% CI 1.08; 3.60) in the unenriched and 1.58 (95% CI 0.93; 2.54) in the enriched analysis among women with overweight pre-pregnancy BMI (Table 2).

Since we allowed statistical interaction between plurality and weight gain, the controlled direct effect cannot be expressed as a single value across all values of weight gain, and instead must be calculated for each specific value of mid-pregnancy weight gain. Figures 2 and 3 display controlled direct effects by values at which weight gain was set for their calculation. Since magnitudes of coefficients for statistical interaction between plurality and weight gain were small in both normal weight and overweight pre-pregnancy BMI groups, controlled direct effects were similar across a range of weight gain, and therefore could be approximated by the natural direct effect when expressed on the relative scale.

Figure 2.

Figure 2.

Open in a new tab

Controlled direct effect on odds ratio (OR) scale by value at which gestational weight gain is set for unenriched multivariate model among normal weight pre-pregnancy BMI.

Figure 3.

Figure 3.

Open in a new tab

Controlled direct effect on odds ratio (OR) scale by value at which gestational weight gain is set for unenriched multivariate model among overweight pre-pregnancy BMI.

Inferences regarding natural indirect effect were similar throughout sensitivity analyses, yet we observed substantial variation in the total and natural direct effects. The total effect of plurality on gestational diabetes ranged from 1.75 (95% CI 1.19; 2.28) to 3.24 (95% CI 1.53; 6.48) among normal weight and 1.19 (0.79; 1.64) to 2.40 (1.42; 4.06) among overweight pre-pregnancy BMI groups. Generally, the total effect moved closer to the null as exposure and outcome definitions (and analytical samples, by extension) were broadened [i.e. incorporated glucose screening/tolerance test(s) regardless of gestational age of administration and pregnancies without glucose screening/tolerance tests(s)], and away from the null as exposure and outcome definitions were restricted [i.e. included pregnancies with glucose screening test(s) and excluded pregnancies with only glucose tolerance test(s) conducted 20–32 weeks’ gestational age]. Effect estimates were also attenuated throughout enriched analysis that incorporated singleton cases (eAppendices 15 and 16). When expressed as risk differences, controlled direct effects increased as weight gain increased among women with normal weight and overweight pre-pregnancy BMI (eAppendices 17 and 18).

Discussion

Our findings suggest a limited role of mid-pregnancy weight gain in mediating the relationship between plurality and gestational diabetes among pregnancies with normal and overweight pre-pregnancy BMI. Odds of gestational diabetes were up to three times higher in twins versus singletons pregnancies, yet the natural indirect effect remained relatively small in magnitude. This suggests that the relationship between twin pregnancy and gestational diabetes is primarily mediated by causal pathways that do not involve mid-pregnancy weight gain. Guidelines for gestational weight gain recommend higher weight gain in twin versus singleton pregnancies within pre-pregnancy BMI groups, yet the former are currently based on limited evidence.19 Since excessive weight gain is considered a risk factor for developing gestational diabetes13–15, these provisional guidelines may cause concern among clinicians and patients that higher weight gain among twin pregnancies may increase risk for gestational diabetes. Our findings provide reassurance that higher weight gain guidelines for twin pregnancies are not unintentionally increasing risk of gestational diabetes.

Our total effects (and natural direct effects, by extension) were highly sensitive to analyses that included or excluded groups of pregnancies, which suggests potential for selection bias in this and similar studies of gestational diabetes. Gestational diabetes is diagnosed using glucose screening or tolerance tests during pregnancy, and administration of these tests may be determined by perceived risk of gestational diabetes in clinical settings. At Magee-Women’s Hospital, gestational diabetes is diagnosed using Carpenter and Coustan criteria27,28, and universal gestational diabetes screening during pregnancy has been recommended since 1998. In the United States (US) more broadly, the US Preventive Services Task Force updated recommendations for gestational diabetes screening and diagnosis in 1996, 2003, and 2008.35–37 In these updates, the Task Force cited insufficient evidence for recommending routine gestational diabetes screening for all pregnant women, recommended clinicians and/or patients considered associated risks and benefits, and listed obesity, age, race, family history of diabetes, and history of gestational diabetes or associated sequalae as characteristics that may increase risk for gestational diabetes.35–37 It was not until 2014 that the Task Force recommended routine gestational diabetes screening for all pregnant women.38 Twin pregnancies and singleton pregnancies in overweight or obese pre-pregnancy BMI groups may have been at higher perceived risk of gestational diabetes than singleton pregnancies in the normal weight pre-pregnancy BMI group, and may have been more likely to be assessed from 1998 to 2013. Notably, we observed that pregnancies in the lowest weight gain tertiles consistently had more missing glucose screening/tolerance test information, which suggests that perceived risk of gestational diabetes may play a role in its assessment. Sensitivity analyses that relied on physician diagnosis when glucose screening or tolerance tests were not available may have reduced the potential for this selection bias by including more pregnancies. However, incorporation of less precise outcome measurement may have also biased effect estimates towards the null. Analyses that incorporated singleton case pregnancies, which were identified using physician diagnosis of gestational diabetes (i.e. instead of glucose screening or tolerance test information), yielded similarly attenuated but more precise effect estimates.

Our observation that the total and natural direct effects varied with more or less restrictive exclusion criteria may explain discrepant findings with respect to previous studies of the risk of gestational diabetes in twin versus singleton pregnancies. Studies that have investigated this relationship in various settings of pregnancies delivered from 1971 to 2016 have observed crude effect estimates in the range of 1.0 to 2.6 and adjusted effect estimates in the range of 1.0 to 2.2.2–9 Many of these studies used existing clinical or administrative data to ascertain diagnosis of gestational diabetes2–7, while only a subset indicated that screening for gestational diabetes was routinely performed among all pregnancies in their settings.3,8 It follows from these findings as well as the current study that the relationship between twin versus singleton pregnancy and gestational diabetes, which we present as the total effect, may be sensitive to bias due to lack of confounding adjustment, sample selection, and source population of pregnancies. Nonetheless, mid-pregnancy weight gain consistently explained a small proportion of the observed increased risk of gestational diabetes among twins in our study.

We restricted to normal weight and overweight pre-pregnancy BMI for preliminary assessment of the plausibility of this biologic mechanism, and did not explore this relationship in women with obese pre-pregnancy BMI, who generally experience higher risk of gestational diabetes.39 However, we observed higher baseline risk for gestational diabetes among overweight versus normal weight women and our point estimates for total and natural direct effects were greater in magnitude among normal weight versus overweight women. This suggests that the relative importance of plurality with respect to gestational diabetes may decrease as pre-pregnancy BMI increases. To further examine this hypothesis, we calculated the incidence of gestational diabetes by pre-pregnancy BMI strata and plurality among obese twin and singleton subcohort pregnancies that would have been included in the current study per our other inclusion and exclusion criteria, and observed a consistent gradient in the crude OR comparing twin versus singleton pregnancies across normal weight, overweight, obese I, obese II, and obese III pre-pregnancy BMI strata (eAppendix 19). Nonetheless, we acknowledge that the inferences of our mechanistic study may be limited to women within normal weight and overweight pre-pregnancy BMI strata.

Strengths of our study include concurrent analysis of population-based samples of twin and singleton pregnancies. In the parent studies, twin cohort and singleton subcohort pregnancies were selected to represent all twin and singleton pregnancies delivered at Magee Women’s Hospital, which enabled us to estimate the effect of plurality on gestational diabetes. Parent studies collected detailed data not commonly available in pregnancy databases, including serial weight measurements and glucose screening/tolerance test data abstracted from medical charts, for large samples of both twin and singleton pregnancies. We used serial maternal weight measurements to calculate mid-pregnancy weight gain at or prior to glucose screening and tolerance tests; this is an important improvement on prior research, which largely examines the relationship between total weight gain and gestational diabetes. We extended current causal mediation methods to estimate both absolute and relative effects, as well as effects that incorporated oversampled gestational diabetes cases and sampling weights from the parent singleton case–cohort study.

Our statistical model assumed no unmeasured confounding between plurality and gestational diabetes, mid-pregnancy weight gain and gestational diabetes, or plurality and mid-pregnancy weight gain, and no measured or unmeasured confounding between mid-pregnancy weight gain and gestational diabetes that is caused by plurality (U1 to U4 in Figure 1). We were unable to adjust for use of assisted reproductive technology, which was unavailable or incomplete for singletons, and increases risk of multiple birth40 as well as gestational diabetes among twin and singleton pregnancies;41,42 thus, the assumption of no unmeasured exposure-outcome confounding (i.e. U1 in Figure 1) may not be met, and our total effect estimate may be biased away from the null. Additionally, there are conceivably many unmeasured common causes of weight gain and gestational diabetes, including genetic,43 socioeconomic, and other pre-pregnancy characteristics. Unmeasured common causes of weight gain and gestational diabetes may bias the natural indirect and/or direct effects either towards or away from the null but have no implications for the observed relationship between plurality and gestational diabetes. Notably, crude relationships between mid-pregnancy weight gain and gestational diabetes were similar by plurality, despite the diverse sociodemographic profiles observed in twin and singleton pregnancies; thus, we suggest that any violation of no mediator-outcome confounding (i.e. U2 in Figure 1) may be limited. Gestational ages at which glucose screening/tolerance test was performed and weight was measured were less than one week lower in twins; since twins are at higher risk for gestational diabetes, this may constitute a measured confounder of mid-pregnancy weight gain and gestational diabetes that is caused by plurality (i.e. U4 in Figure 1). However, minimal interaction between plurality and mid-pregnancy weight gain was observed, and violation of this assumption as well as that of no unmeasured exposure–mediator confounding (i.e. U3 in Figure 1) may be acceptable in the current study.44 Previous studies indicate that self-reported pre-pregnancy weight may be underestimated by up to 3 kilograms. 45 We expect this measurement bias to be non-differential with respect to pregnancy plurality and thus not affect the interpretation of the total effect, while the natural indirect effect may be underestimated and natural/controlled direct effects may be overestimated due to corresponding mediator imprecision.46

In the context of our causal model, plurality is only one non-modifiable cause of mid-pregnancy weight gain. Weight gain is a summary measure of changes in maternal physiology, such as changes in fat/protein mass, total body water, or fetal/placental size.47 Several different unmodifiable (plurality/genetics) and modifiable (diet/exercise) causes of weight gain may exist, and changes in weight gain caused by these mechanisms may result in different increases/decreases in risk for perinatal outcomes. Since weight gain can be intervened upon in different ways, and different interventions may have different effects on gestational diabetes even given equivalent changes in weight gain, the consistency assumption needed for causal inference may be violated.48 A Cochrane review of randomized clinical trials aimed at preventing gestational diabetes in singletons reported an approximately 39% risk reduction in gestational diabetes when scaled to the mean difference in mid-pregnancy weight gain by plurality observed in our study;16 this is greater than our estimated natural indirect effects. Further studies are needed to discern whether interventions that equivalently change weight gain observe similar effects on health among both twins and singletons. To our knowledge, our study is the first to examine mid-pregnancy weight gain and its relationship with gestational diabetes concurrently in twin and singleton pregnancies.

We estimate that higher mid-pregnancy weight gain in twin pregnancies does not substantially contribute to increased risk for gestational diabetes among women with normal weight and overweight pre-pregnancy BMI. We recommend that future research further investigate the relationship between plurality and gestational diabetes, and examine the role of other physiologic mechanisms that occur only among or to a greater extent in twin pregnancies in mediating this relationship. More broadly, we encourage research comparing relationships between weight gain and maternal morbidity in twin and singleton pregnancies to inform future weight gain guidelines.

Supplementary Material

Supplemental Digital Content

Acknowledgements:

We thank Sara Parisi for her technical contributions. JAH proposed the current study; JAH and LMB lead parent cohorts; MCD designed the current study, conducted analysis, and wrote manuscript; JSK, LMB, RWP, KPH, JAH provided feedback on manuscript. All authors read and approved the final study.

Sources of Funding:

The results reported herein correspond to specific aims of grants R01 NR014245 and R01 HD094777 to investigators LMB and JAH from the National Institutes of Health. MCD was supported by Graduate Award (Institute for Health and Social Policy, 2016), Alexander McFee and Graduate Excellence Fellowships (McGill University Faculty of Medicine, 2016–2017), Ferring Fellowship (McGill University Faculty of Medicine, 2017–2018), and Fonds de Recherche Santé Doctoral Award (Government of Quebec, 2018–2020).

Footnotes

Conflicts of Interest:

None declared.

Meetings and Related Work/Preprint Servers:

Poster presentation at the Canadian National Perinatal Research Meeting 2020. Online poster presentation at the Society for Pediatric and Perinatal Epidemiologic Research Meeting 2020. Online oral presentation at the Society for Epidemiologic Research Annual Meeting 2020. Doctoral thesis available on McGill University Institutional Digital Repository: https://escholarship.mcgill.ca/concern/theses/dz010v32x?locale=en

Code/Data Availability:

Code for mediation analyses may be obtained from MCD on reasonable request. Data may be obtained from LMB and JAH conditional on appropriate ethics and data steward approvals.

References

  • 1.Reece EA, Leguizamón G, Wiznitzer A. Gestational diabetes: the need for a common ground. Lancet 2009;373(9677):1789–1797. doi: 10.1016/S0140-6736(09)60515-8 [DOI] [PubMed] [Google Scholar]
  • 2.Wein P, Warwick MM, Beischer NA. Gestational Diabetes in Twin Pregnancy: Prevalence and Long-term Implications. Aust New Zeal J Obstet Gynaecol 1992;32(4):325–327. doi: 10.1111/j.1479-828X.1992.tb02843.x [DOI] [PubMed] [Google Scholar]
  • 3.Schwartz DB, Daoud Y, Zazula P, et al. Gestational diabetes mellitus: metabolic and blood glucose parameters in singleton versus twin pregnancies. Am J Obstet Gynecol 1999;181(4):912–914. [DOI] [PubMed] [Google Scholar]
  • 4.Rauh-Hain JA, Rana S, Tamez H, et al. Risk for developing gestational diabetes in women with twin pregnancies. J Matern Fetal Neonatal Med 2009;22(4):293–299. doi: 10.1080/14767050802663194 [DOI] [PubMed] [Google Scholar]
  • 5.Retnakaran R, Shah BR. Impact of Twin Gestation and Fetal Sex on Maternal Risk of Diabetes During and After Pregnancy. Diabetes Care 2016;39(8):e110–e111. doi: 10.2337/dc16-0825 [DOI] [PubMed] [Google Scholar]
  • 6.Weissman A, Drugan A. Glucose tolerance in singleton, twin and triplet pregnancies. J Perinat Med 2016;44(8). doi: 10.1515/jpm-2016-0186 [DOI] [PubMed] [Google Scholar]
  • 7.Hiersch L, Berger H, Okby R, et al. Incidence and risk factors for gestational diabetes mellitus in twin versus singleton pregnancies. Arch Gynecol Obstet 2018;298(3):579–587. doi: 10.1007/s00404-018-4847-9 [DOI] [PubMed] [Google Scholar]
  • 8.Henderson CE, Scarpelli S, LaRosa D, Divon MY. Assessing the risk of gestational diabetes in twin gestation. J Natl Med Assoc 1995;87(10):757–758. [PMC free article] [PubMed] [Google Scholar]
  • 9.Buhling KJ, Henrich W, Starr E, et al. Risk for gestational diabetes and hypertension for women with twin pregnancy compared to singleton pregnancy. Arch Gynecol Obstet 2003;269(1):33–36. doi: 10.1007/s00404-003-0483-z [DOI] [PubMed] [Google Scholar]
  • 10.Hiersch L, Berger H, Okby R, et al. Gestational diabetes mellitus is associated with adverse outcomes in twin pregnancies. Am J Obstet Gynecol 2019;220(1):102.e1–102.e8. doi: 10.1016/j.ajog.2018.10.027 [DOI] [PubMed] [Google Scholar]
  • 11.McGrath RT, Hocking SL, Scott ES, Seeho SK, Fulcher GR, Glastras SJ. Outcomes of twin pregnancies complicated by gestational diabetes: a meta-analysis of observational studies. J Perinatol 2017;37(4):360–368. doi: 10.1038/jp.2016.254 [DOI] [PubMed] [Google Scholar]
  • 12.Vani K, Scifres C, Himes K. The Impact of Gestational Diabetes in Twin Pregnancies on Maternal and Neonatal Outcomes. Am J Obstet Gynecol 2018;218(1):S590. [Google Scholar]
  • 13.Brunner S, Stecher L, Ziebarth S, et al. Excessive gestational weight gain prior to glucose screening and the risk of gestational diabetes: a meta-analysis. Diabetologia 2015;58(10):2229–2237. doi: 10.1007/s00125-015-3686-5 [DOI] [PubMed] [Google Scholar]
  • 14.Hedderson MM, Gunderson EP, Ferrara A. Gestational Weight Gain and Risk of Gestational Diabetes Mellitus. Obstet Gynecol 2010;115(3):597–604. doi: 10.1097/AOG.0b013e3181cfce4f [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Zhong C, Li X, Chen R, et al. Greater early and mid-pregnancy gestational weight gain are associated with increased risk of gestational diabetes mellitus: A prospective cohort study. Clin Nutr ESPEN 2017;22:48–53. doi: 10.1016/j.clnesp.2017.08.013 [DOI] [PubMed] [Google Scholar]
  • 16.Shepherd E, Gomersall JC, Tieu J, Han S, Crowther CA, Middleton P. Combined diet and exercise interventions for preventing gestational diabetes mellitus. Cochrane Database Syst Rev November 2017. doi: 10.1002/14651858.CD010443.pub3 [DOI] [PMC free article] [PubMed]
  • 17.Bennett CJ, Walker RE, Blumfield ML, et al. Interventions designed to reduce excessive gestational weight gain can reduce the incidence of gestational diabetes mellitus: A systematic review and meta-analysis of randomised controlled trials. Diabetes Res Clin Pract 2018;141:69–79. doi: 10.1016/j.diabres.2018.04.010 [DOI] [PubMed] [Google Scholar]
  • 18.Du M, Ouyang Y, Nie X, Huang Y, Redding SR. Effects of physical exercise during pregnancy on maternal and infant outcomes in overweight and obese pregnant women: A meta‐analysis. Birth 2019;46(2):211–221. doi: 10.1111/birt.12396 [DOI] [PubMed] [Google Scholar]
  • 19.Institute of Medicine (US) and National Research Council (US) Committee to Reexamine IOM Pregnancy Weight Guidelines, Rasmussen K, Yaktine A, Editors. Weight Gain During Pregnancy: Reexamining the Guidelines Vol 33. Washington, DC: The National Academies Press; 2009. [PubMed] [Google Scholar]
  • 20.Shamshirsaz AA, Haeri S, Ravangard SF, et al. Perinatal outcomes based on the institute of medicine guidelines for weight gain in twin pregnancies. J Matern Neonatal Med 2014;27(6):552–556. doi: 10.3109/14767058.2013.836177 [DOI] [PubMed] [Google Scholar]
  • 21.Pettit KE, Lacoursiere DY, Schrimmer DB, Alblewi H, Moore TR, Ramos GA. The association of inadequate mid-pregnancy weight gain and preterm birth in twin pregnancies. J Perinatol 2015;35(2):85–89. doi: 10.1038/jp.2014.160 [DOI] [PubMed] [Google Scholar]
  • 22.Kosinska-Kaczynska K, Szymusik I, Kaczynski B, Wielgos M. Observational study of associations between gestational weight gain and perinatal outcomes in dichorionic twin pregnancies. Int J Gynecol Obstet 2017;138(1):94–99. doi: 10.1002/ijgo.12171 [DOI] [PubMed] [Google Scholar]
  • 23.Pécheux O, Garabedian C, Drumez E, et al. Maternal and neonatal outcomes according to gestational weight gain in twin pregnancies: Are the Institute of Medicine guidelines associated with better outcomes? Eur J Obstet Gynecol Reprod Biol 2019;234:190–194. doi: 10.1016/j.ejogrb.2019.01.010 [DOI] [PubMed] [Google Scholar]
  • 24.Brown J, Alwan NA, West J, et al. Lifestyle interventions for the treatment of women with gestational diabetes. Cochrane Database Syst Rev May 2017. doi: 10.1002/14651858.CD011970.pub2 [DOI] [PMC free article] [PubMed]
  • 25.Hutcheon JA, Platt RW, Abrams B, et al. Pregnancy Weight Gain by Gestational Age in Women with Uncomplicated Dichorionic Twin Pregnancies. Paediatr Perinat Epidemiol 2018;32(2):172–180. doi: 10.1111/ppe.12446 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Bodnar LM, Himes KP, Abrams B, Parisi SM, Hutcheon JA. Early-pregnancy weight gain and the risk of preeclampsia: A case-cohort study. Pregnancy Hypertens 2018;14:205–212. doi: 10.1016/j.preghy.2018.10.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Carpenter M, Coustan D. Criteria for screening tests for gestational diabetes. Am J Obstet Gynecol 1982;144(7):768–773. [DOI] [PubMed] [Google Scholar]
  • 28.Metzger BE, Coustan DR. Summary and recommendations of the Fourth International Workshop-Conference on Gestational Diabetes Mellitus. The Organizing Committee. Diabetes Care 1998;21 Suppl 2:B161–7. [PubMed] [Google Scholar]
  • 29.VanderWeele TJ. Explanation in Causal Inference: Methods for Mediation and Interaction Oxford University Press; 2015. [Google Scholar]
  • 30.Johansson ALV The Case-Cohort design: What it is and how it can be used in register-based research https://www.stata.com/meeting/nordic-and-baltic16/slides/norway16_johansson.pdf. Accessed July 28, 2020.
  • 31.Borgan O, Langholz B, Samuelsen SO, Goldstein L, Pogoda J. Exposure Stratified Case-Cohort Designs. Lifetime Data Anal 2000;6(1):39–58. doi: 10.1023/A:1009661900674 [DOI] [PubMed] [Google Scholar]
  • 32.Kulathinal S, Karvanen J, Saarela O, Kuulasmaa K. Case-cohort design in practice – experiences from the MORGAM Project. Epidemiol Perspect Innov 2007;4(1):15. doi: 10.1186/1742-5573-4-15 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.MacDonald SC, Bodnar LM, Himes KP, Hutcheon JA. Patterns of Gestational Weight Gain in Early Pregnancy and Risk of Gestational Diabetes Mellitus. Epidemiology 2017;28(3):419–427. doi: 10.1097/EDE.0000000000000629 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Emsley R, Liu H. PARAMED: Stata module to perform causal mediation analysis using parametric regression models 2013.
  • 35.US Preventive Services Task Force. Screening for Diabetes Mellitus. In: Guide to Clinical Preventive Services: Report of the U.S. Preventive Services Task Force 2nd ed. Baltimore, MD: Williams & Wilkins; 1996. [Google Scholar]
  • 36.US Preventive Services Task Force. Screening for Gestational Diabetes Mellitus: Recommendation and Rationale. Am Fam Physician 2003;68(2):331–335. [PubMed] [Google Scholar]
  • 37.Screening for Gestational Diabetes Mellitus: U.S. Preventive Services Task Force Recommendation Statement. Ann Intern Med 2008;148(10):759. doi: 10.7326/0003-4819-148-10-200805200-00008 [DOI] [PubMed] [Google Scholar]
  • 38.Moyer VA. Screening for Gestational Diabetes Mellitus: U.S. Preventive Services Task Force Recommendation Statement. Ann Intern Med 2014;160(6):414–420. doi: 10.7326/M13-2905 [DOI] [PubMed] [Google Scholar]
  • 39.Kim SY, England L, Wilson HG, Bish C, Satten GA, Dietz P. Percentage of gestational diabetes mellitus attributable to overweight and obesity. Am J Public Health 2010;100(6):1047–1052. doi: 10.2105/AJPH.2009.172890 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Lynch A Assisted reproductive interventions and multiple birth. Obstet Gynecol 2001;97(2):195–200. doi: 10.1016/S0029-7844(00)01145-5 [DOI] [PubMed] [Google Scholar]
  • 41.Qin J, Wang H, Sheng X, Liang D, Tan H, Xia J. Pregnancy-related complications and adverse pregnancy outcomes in multiple pregnancies resulting from assisted reproductive technology: a meta-analysis of cohort studies. Fertil Steril 2015;103(6):1492–1508.e7. doi: 10.1016/j.fertnstert.2015.03.018 [DOI] [PubMed] [Google Scholar]
  • 42.Qin J, Liu X, Sheng X, Wang H, Gao S. Assisted reproductive technology and the risk of pregnancy-related complications and adverse pregnancy outcomes in singleton pregnancies: a meta-analysis of cohort studies. Fertil Steril 2016;105(1):73–85.e6. doi: 10.1016/j.fertnstert.2015.09.007 [DOI] [PubMed] [Google Scholar]
  • 43.Andersson ES, Silventoinen K, Tynelius P, Nohr EA, Sørensen TIA, Rasmussen F. Heritability of Gestational Weight Gain – A Swedish Register-Based Twin Study. Twin Res Hum Genet 2015;18(4):410–418. doi: 10.1017/thg.2015.38 [DOI] [PubMed] [Google Scholar]
  • 44.Valeri L, VanderWeele TJ. Mediation analysis allowing for exposure–mediator interactions and causal interpretation: Theoretical assumptions and implementation with SAS and SPSS macros. Psychol Methods 2013;18(2):137–150. doi: 10.1037/a0031034 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.Headen I, Cohen AK, Mujahid M, Abrams B. The accuracy of self-reported pregnancy-related weight: a systematic review. Obes Rev 2017;18(3):350–369. doi: 10.1111/obr.12486 [DOI] [PubMed] [Google Scholar]
  • 46.VanderWeele TJ, Valeri L, Ogburn EL. The role of measurement error and misclassification in mediation analysis: mediation and measurement error. Epidemiology 2012;23(4):561–564. doi: 10.1097/EDE.0b013e318258f5e4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Gandhi M, Gandhi R, Mack LM, et al. Impact of changes in maternal body composition on birth weight and neonatal fat mass in dichorionic twin pregnancies. Am J Clin Nutr 2018;108(4):716–721. doi: 10.1093/ajcn/nqy180 [DOI] [PubMed] [Google Scholar]
  • 48.Hernán MA, Robins J. Causal Inference Boca Raton, FL: Chapman & Hall/CRC, forthcoming.; 2015. [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplemental Digital Content
NIHMS1762111-supplement-Supplemental_Digital_Content.pdf (953.5KB, pdf)

Data Availability Statement

Code for mediation analyses may be obtained from MCD on reasonable request. Data may be obtained from LMB and JAH conditional on appropriate ethics and data steward approvals.

RESOURCES