Gestational Weight Gain and Adverse Birth Outcomes in Twin Pregnancies

Abstract

Objective:

To evaluate the association between gestational weight gain in twin pregnancies and small- and large-for-gestational-age birth (SGA, LGA), preterm birth <32 weeks, cesarean delivery, and infant death within each prepregnancy body mass index (BMI) category.

Methods:

Data in this population-based study came from Pennsylvania linked infant birth and death records (2003–2013). We studied 54,836 twins born alive at <39 weeks gestation. Total pregnancy weight gain (kg) was converted to gestational age-standardized z scores. Multivariable modified Poisson regression models stratified by prepregnancy BMI were used to estimate associations between z scores and outcomes. A probabilistic bias analysis, informed by an internal validation study, evaluated the impact of BMI and weight gain misclassification.

Results:

Gestational weight gain z score was negatively associated with SGA and positively associated with LGA and cesarean delivery in all BMI groups. The relation between weight gain and preterm birth was U-shaped in non-obese women. An increased risk of infant death was observed for very low weight gain among normal weight women and for high weight gain among women without obesity. Most excess risks of these outcomes were observed at weight gains 37-week gestation equivalent to <14 kg or >27 kg in underweight or normal weight women, <11 kg or >28 kg in overweight women, and <6.4 kg or >26 kg in women with obesity. The bias analysis supported the validity of the conventional analysis.

Conclusions:

Very low or very high weight gains were associated with the adverse outcomes we studied. If the associations we observed are even partially reflective of causality, then targeted modification of pregnancy weight gain in women carrying twins might improve pregnancy outcomes.

Precis

Most of the excess risk of poor birth outcomes in twin pregnancies occurs at very high or very low gestational weight gains relative to moderate gains.

Introduction

Over the past 40 years, the number of twins born in the U.S. doubled (1,2). This unprecedented rise is a major public health concern because twin gestations disproportionately experience poor maternal and infant outcomes. Compared with singletons, twins are 2.5 times as likely to die in utero (3) and are more than four times as likely to die in the first year of life (4). The dramatic elevation in the risk of perinatal death is due to the higher rates of pregnancy complications and poor birth outcomes, most notably preterm birth. While twins make up 3.3% of all births, they account for over 20% of the burden of preterm births (2) Twin pregnancies are also more likely to experience diabetes, preeclampsia, and cesarean deliveries (5,6).

Gestational weight gain is a potentially modifiable factor that has been linked with many poor pregnancy and birth outcomes that are common to twin pregnancies (7,8). In 1990 (9) and again in 2009 (8), the Institute of Medicine (IOM) (now called the National Academy of Medicine) and the National Research Council recognized the importance of understanding optimal pregnancy weight gain ranges for women with twin gestations. Yet, the current evidence base to inform gestational weight gain guidelines for twin pregnancies is methodologically limited. As a result of the limited evidence base, the 2009 IOM Committee could only provide provisional guidelines that were based on the interquartile range of weight gain observed in a single study of 2234 self-selected mothers who delivered twins weighing at least 2500 g at ≥36 weeks’ gestation (8,11). This evidence base and the resulting provisional guidelines are insufficient for optimizing maternal and child health.

Our objective was to evaluate the association between gestational weight gain z scores, a measure standardized for gestational age, and SGA, LGA, preterm birth <32 weeks, cesarean delivery, and infant death separately by prepregnancy BMI category in a large, population-based cohort of twin births.

Methods

Penn MOMS is a population-based study of maternal obesity, gestational weight gain, and adverse birth outcomes (12,13) that was approved by the University of Pittsburgh institutional review board. Data came from Pennsylvania linked infant birth and death vital statistics records from 2003 to 2013. We cleaned the data to eliminate duplicate records, confirm plurality and match twin siblings, create unique maternal and pregnancy identifiers, and confirm all birth-death matches. Of the 55,471 live-born twins, we excluded 635 (1.1%) infants born at 39 weeks gestation or greater because the American College of Obstetricians and Gynecologists recommend the delivery of twin pregnancies by 39 weeks (14), and gestational weight gain z scores (15) and the fetal growth percentiles (16) for twins are available up to 38 weeks of gestation. Our final analytic sample included 54,836 live-born twins, contributed by 27,723 twin pregnancies. Missing values were filled first using information from the record of the twin sibling, as appropriate, and if missing in both twins, by multiple imputation, described below.

Birth registry staff interviewed the mother before hospital discharge to obtain prepregnancy height and weight, which we used to calculate prepregnancy BMI (weight (kg)/height (m)²). We classified women as underweight (<18.5 kg/m²), normal weight (18.5–24.9 kg/m²), overweight (25–29.9 kg/m²), grade 1 obese (30–34.9 kg/m²), grade 2 obese (35–39.9 kg/m²), or grade 3 obese (≥40 kg/m²) (17). Maternal weight at delivery was ascertained from the medical record, either from the last recorded antenatal weight or from measured or self-reported weight at delivery. We calculated total gestational weight gain by subtracting prepregnancy weight from maternal weight at delivery. We standardized total gestational weight gain (kg) for gestational age at delivery using prepregnancy BMI category-specific z score charts for twin pregnancies that were produced for Magee-Womens Hospital in Pittsburgh, Pennsylvania (15). Converting women’s total weight gain to gestational age-standardized z scores produces a measure that is independent of gestational duration (15,18,19). Z-scores were calculated by comparing a woman’s weight gain to the gestational age-specific mean and standard deviation of weight gain in the population using the formula: z score=(observed weight- gestational age-specific population mean weight)/gestational age-specific population standard deviation (15). We present the total weight gain (kg) equivalent to each z score, assuming a 37-week gestation.

The Pennsylvania Bureau of Health Statistics and Research uses birth certificate record number, county of birth, date of birth, and mother and infant names to link infant death certificates to the birth certificates with a match rate of 98.5%. We defined infant mortality as the death of a live-born infant before 365 days old, and neonatal death as death before 28 days old. We evaluated preterm birth before 32 weeks’ gestation because the risk of neonatal morbidity is highest at <32 weeks (20). Gestational age was based on the best obstetric estimate at delivery (21). We validated gestational age from the birth certificate against the customary gestational age derivation based on last menstrual period and ultrasound dating in the medical record (22) from 2251 twin pregnancies at Magee-Womens Hospital (Spearman’s rank correlation: 0.97; preterm birth classification: 99% agreement). Small- and large-for-gestational age births were classified using twin-specific ultrasound-based intrauterine fetal weight standards at <10th percentile or >90th percentile, respectively (16).

Maternal interview before discharge provided birth certificate data on maternal race or ethnicity, maternal education, marital status, and smoking during pregnancy. Medical records were used to ascertain birth certificate data on parity, source of payment at delivery, preexisting conditions, and address of primary residence at delivery. A birth registry staff member uses medical record data to ascertain whether the “pregnancy resulted from infertility treatment” (yes/no). The Pennsylvania Bureau of Health Statistics and Research geocodes addresses of primary residence and provides census tracts and block groups, which we merged with U.S. Department of Agriculture Economic Research Service Urban-Rural Continuum Codes (23) to define urban residence. Using the 2000 United States Census data, we also created a census-tract level measure of the percentage of Black residents (24). We classified the neonatal care of the birth facility as level 1, 2, or 3 (25).

Statistical Analysis

Records were missing data on birth weight (0.7%), gestational age (0.4%), delivery weight (9.4%), prepregnancy weight (5.8%), height (0.9%), or one of the confounders in the final model (11%). Missing data were imputed using the multivariate normal imputation model, which uses an iterative Markov Chain Monte Carlo method (26,27). We included in the imputation model all variables in the dataset because this approach produces more accurate and stable estimates (28,29). We imputed 25 datasets, with initial values derived from 500 iterations. Data were imputed at the infant level, but the pregnancy identifier was included in the imputation model to inform the estimates. We performed imputation diagnostics comparing the imputed and observed datasets to inform the number of imputations and to ensure acceptability of the completed imputations (30).

To estimate the independent association between gestational weight gain z score and each adverse outcome, we used multivariable modified Poisson models with a log link and robust standard errors to account for clustering between twin pairs (31). Modified Poisson regression with robust standard errors is an acceptable alternative to log-binomial models, even when applied to binomially distributed data (31,32). We used a “twins-as-individuals” modeling approach because we were not interested in studying the twin pair itself (33).

We decided a priori to stratify models by prepregnancy BMI category because weight gain guidelines are BMI-specific (8). To accommodate flexible, nonlinear relations, we modeled gestational weight gain z scores as restricted cubic splines with four knots at the 5^th, 35^th, 65^th, and 95^th percentiles of the distribution (normal weight and overweight women) or three knots at the 10^th, 50^th, and 90^th percentiles (all other BMI categories) (34). We selected these specifications for each BMI group based on Akaike information criterion (34). We used theory-based causal diagrams (35) to identify confounders that we included in all models: maternal race or ethnicity, education, neonatal care level of birth facility (a proxy for antenatal care facility), parity, payment at delivery, smoking during pregnancy, marital status, year of birth, height, maternal age, pre-existing diabetes or hypertension, infertility treatment, infant sex, and racial composition of neighborhood (as a proxy of neighborhood-level socioeconomic status). All models included prepregnancy weight to account for potential confounding within BMI categories, and a statistical interaction term between maternal race or ethnicity and the racial composition of the neighborhood. We specified categorical confounders as indicator variables and continuous confounders as restricted cubic splines.

After model estimation, we calculated risk differences and 95% confidence intervals (CI) for selected z score values from −2 SD to +2 SD (approximately 4.5^th to 97.5^th percentiles of weight gain z scores observed in our cohort) compared with the z score of 0. We calculated the excess or prevented number of cases per 100 births by multiplying estimates by 100. We plotted the average adjusted risks and 95% CI for each outcome by BMI category. We compared these results with those generated using twins known to have a sex discordant pair because they were dichorionic diamniotic twins (n=19,406, 35% of twins). We also performed sensitivity analysis among twins without congenital anomalies (n= 54,048) or twins with complete data (n=43,668).

We previously conducted a validation study of self-reported prepregnancy weight and height, and the resulting estimates of BMI and gestational weight gain categories, derived from the birth certificate. Self-reported weight and height were validated with information abstracted from medical records for 1204 singleton and 186 twin Pennsylvania births at a single institution (36,37). We used the predictive values from the validation study to adjust our results for measurement error in BMI and gestational weight gain using probabilistic bias analysis methods as a supplemental analysis (38,39). Our approach has been described in detail previously (38). We randomly selected one multiply imputed dataset for the probabilistic bias analysis due to the computational intensity of joining probabilistic bias analysis with multiple imputation (38).

Results

Mothers of twins tended to be non-Hispanic white, college graduates, married, multiparous, 30 years or older, and had private insurance and no preexisting conditions (Table 1). Most mothers lived in metropolitan areas and delivered in facilities with Level 3 neonatal care. Approximately 16% of mothers received infertility treatment. Women who were underweight, normal weight, overweight, grade 1 obese, grade 2 obese, and grade 3 obese made up 3.4%, 48%, 24%, 13%, 6.6%, and 5.0% of the population, respectively, and their mean total weight gain (SD) was 18 (6.4), 19 (7.2), 18 (8.5), 16 (9.2), 13 (9.7), and 11 (11) kg.

Table 1.

Characteristics of mothers delivering live-born twins, Pennsylvania twin birth certificates, 2003–2013.

	Total Mothers N = 27,723 n (%)
Race or ethnicity
Non-Hispanic White	20,515 (74)
Non-Hispanic Black	4,436 (16)
Hispanic	1,830 (6.6)
Non-Hispanic Other	942 (3.4)
Prepregnancy body mass index
Underweight (<18.5 kg/m2)	915 (3.3)
Normal weight (18.5–24.9 kg/m2)	13,307 (48)
Overweight (25–29.9 kg/m2)	6,654 (24)
Grade 1 obese (30–34.9 kg/m2)	3,604 (13)
Grade 2 obese (35–39.9 kg/m2)	1,857 (6.7)
Grade 3 obese (>40 kg/m2)	1,386 (5.0)
Age
< 20 years	1,109 (4.0)
20-29 years	11,366 (41)
≥ 30 years	15,248 (55)
Education
High school or equivalent	9,149 (33)
Some college or associates degree	7,208 (26)
College graduate	11,366 (41)
Marital status
Unmarried	8,871 (32)
Married	18,852 (68)
Parity at index birth
1	6,099 (22)
2 or more	21,624 (78)
Received infertility treatment
No	23,287 (84)
Yes	4,436 (16)
Smoking during pregnancy
No	24,119 (87)
Yes	3,604 (13)
Preexisting diabetes or hypertension
No	26,891 (97)
Yes	832 (3.0)
Insurance
Private	19,129 (69)
Medicaid, self-pay, or other	8,594 (31)
Neonatal care level of birth facility
Level 1 or 2	6,654 (24)
Level 3	21,069 (76)
Urban residence
Metropolitan, ≥1 million	16,079 (58)
Metropolitan, 250,000–<1 million	7,208 (26)
Metropolitan, <250,000	2,190 (7.9)
Non-metropolitan urban/rural	2,246 (8.1)
Census tract racial distribution % Black residents (median (interquartile range))	2.7 (0.8, 11)
Gestational weight gain (mean (SD)), kg	17 (8.5)
Gestational weight gain z score (mean (SD)), SD	−0.048 (1.5)

SD, standard deviation

The unadjusted incidences of SGA, LGA, preterm birth <32 weeks, cesarean delivery, and infant death were 12%, 8%, 13%, 68%, and 3.1%, respectively. Approximately 84% of infant deaths occurred in the neonatal period. SGA and preterm birth <32 weeks were most common among underweight women, and did not vary greatly among the other BMI groups (Table 2). LGA and cesarean delivery were more common as prepregnancy BMI category increased. The incidence of infant death was highest at the extremes of BMI.

Table 2.

Unadjusted incidence of select adverse twin birth outcomes stratified by maternal prepregnancy body mass index, Pennsylvania twin birth certificates, 2003–2013.

Prepregnancy body mass index*	Births	Small-for-gestational-age birth		Large-for-gestational-age birth		Preterm birth <32 weeks		Cesarean delivery		Infant death
	Total, n	Cases, n	%	Cases, n	%	Cases, n	%	Cases, n	%	Cases, n	%
Underweight	1858	370	20	88	4.7	291	16	1093	59	75	4.0
Normal weight	26,011	3323	13	1756	6.8	3183	12	16,979	65	675	2.6
Overweight	13,373	1531	11	1146	8.6	1667	12	9237	69	409	3.1
Grade 1 obese	7153	864	12	627	8.8	865	12	5120	72	262	3.7
Grade 2 obese	3684	403	11	403	11	485	13	2668	72	136	3.7
Grade 3 obese	2757	285	10	317	12	376	14	2164	78	117	4.2

^*Underweight (<18.5 kg/m²), normal weight (18.5–24.9 kg/m²), overweight (25–29.9 kg/m²), grade 1 obese (30–34.9 kg/m²), grade 2 obese (35–39.9 kg/m²), or grade 3 obese (≥40 kg/m²)

As gestational weight gain z score category increased across all BMI groups, the unadjusted risk of SGA declined, while LGA and cesarean delivery increased (Table 3). The risk of preterm birth <32 weeks and infant death tended to be highest at the extremes of weight gain.

Table 3.

Unadjusted incidence of select adverse twin birth outcomes stratified by gestational weight gain z score and maternal prepregnancy body mass index, Pennsylvania twin birth certificates, 2003–2013.

Maternal prepregancy BMI	Gestational weight gain z score *	Total gestational weight gain equivalent at 37 weeks, kg		Small-for-gestational-age birth		Large-for-gestational-age birth		Preterm birth <32 weeks		Cesarean Delivery		Infant death
			Births, n	Cases, n	%	Cases, n	%	Cases, n	%	Cases, n	%	Cases, n	%
Underweight †	<−1 SD	<14		314	104	33	6	1.7	53	17	159	51	8	2.6
	−1 to +1 SD	14-27	1317	234	18	63	4.8	176	13	790	60	48	3.6
	>+1 SD	>27	227	32	14	19	8.6	62	27	144	64	19	8.5
Normal weight	<−1 SD	<14	4194	872	21	169	4.0	673	16	2549	61	144	3.4
	−1 to +1 SD	14-27	17,514	2049	12	1104	6.3	1941	11	11,427	65	394	2.3
	>+1 SD	>27	4303	402	9.3	483	11	569	13	3003	70	137	3.2
Overweight	<−1 SD	<11	1856	335	18	104	5.6	305	17	1226	66	68	3.6
	−1 to +1 SD	11–28	9220	971	11	730	7.9	1041	11	6324	69	253	2.7
	>+1 SD	>28	2297	225	9.8	312	14	321	14	1687	73	88	3.8
Grade 1 obese	<−1 SD	<6.4	769	133	17	31	4.0	116	15	492	64	35	4.5
	−1 to +1 SD	6.4–26	5088	608	12	422	8.3	560	11	3641	72	162	3.2
	>+1 SD	>26	1297	123	9.5	174	13	189	15	987	76	65	5.1
Grade 2 obese	<−1 SD	<6.4	670	99	15	43	6.5	110	16	465	69	32	4.9
	−1 to +1 SD	6.4–26	2486	253	10	280	11	302	12	1788	72	84	3.4
	>+1 SD	>26	528	51	9.7	80	15	73	14	415	79	20	3.8
Grade3 obese	<−1 SD	<6.4	744	113	15	55	7.4	114	15	566	76	35	4.8
	−1 to +1 SD	6.4-26	1684	149	8.8	195	12	218	13	1322	79	65	3.9
	>+1 SD	>26	329	23	7.0	67	20	44	13	276	84	17	5.2

SD, standard deviation

^*Gestational weight gain z scores <−1, −1 to +1, and >+1 SD correspond to total gestational weight gains at 37 weeks gestation. Total weight gain ranges are the same for all grades of obesity because there is one z score chart for obese women (BMI ≥30 kg/m²)

^†Underweight (<18.5 kg/m²), normal weight (18.5–24.9 kg/m²), overweight (25–29.9 kg/m²), grade 1 obese (30–34.9 kg/m²), grade 2 obese (35–39.9 kg/m²), or grade 3 obese (≥40 kg/m²)

Figure 1 shows the confounder-adjusted curvilinear associations between gestational weight gain and adverse outcomes for each prepregnancy BMI group (CI have been removed for readability). Table 4 shows adjusted risk differences (95% CI) for these relations for 5 different weight gain values. Pregnancy weight gain was negatively associated with SGA and positively associated with LGA and cesarean delivery in all BMI groups. For example, among normal weight women, compared with a pregnancy weight gain equivalent to 20 kg at 37 weeks (z score of 0 SD), a weight gain of 27 kg at 37 weeks (z score of +1 SD) was associated with 2.2 fewer cases of SGA but 2.9 more cases of LGA and 3.7 more cases of cesarean delivery.

Figure 1.

Association between gestational weight gain z score and the adjusted predicted risks of five adverse perinatal outcomes, shown separately for women who are underweight (n=1,858, light orange lines), normal weight (n=26,011, purple lines), overweight (n=13,373, blue lines), grade 1 obese (n=7,153, green lines), grade 2 obese (n=3,684, yellow lines), or grade 3 obese (n=2,757, dark orange lines), Pennsylvania twin birth certificates, 2003–2013. The solid lines represent the point estimate. The 95% confidence bands have been removed for readability. Effect estimates that correspond to this figure appear in Table 4. Gestational weight gain z scores −2, −1, 0, 1, and 2 standard deviations correspond to the following total gestational weight gains at 37 weeks of gestation: 9.4, 14, 20, 27, and 37 kg for underweight and normal weight women; 5.5, 11, 18, 28, and 41 kg for overweight women; 0.9, 6.4, 14, 26, and 43 kg for women with grade 1, 2, or 3 obesity. Risks were set at the population average for maternal race or ethnicity, maternal education, maternal age, marital status, parity, insurance source, smoking during pregnancy, preexisting diabetes or hypertension, height, prepregnancy weight, infertility treatment, infant sex, neonatal care level of birth facility, year of birth, urban residence, and racial composition of neighborhood. Small for gestational age (A), large for gestational age (B), cesarean delivery (C), early preterm birth (<32 weeks of gestation) (D), and infant death (E).

Table 4.

Associations between gestational age z score and select adverse twin birth outcomes stratified by gestational weight gain z score and maternal prepregnancy body mass index, Pennsylvania twin birth certificates, 2003–2013.

Gestational weight gain z score	Total gestational weight gain equivalent at 37 weeks, kg	Adjusted* number of excess cases† per 100 births (95% CI)
		Small-for-gestational-age birth	Large-for-gestational-age birth	Preterm birth <32 weeks	Cesarean delivery	Infant death	Neonatal death
Underweight‡
−2 SD§	9.4	18 (9.3–26)	−3.2 (−5.8–−0.6)	6.4 (−0.3–13)	−11 (−21–−0.5)	0.5 (−2.3–3.2)	0.8 (−1.6–3.2)
−1 SD	14	7.2 (4.8–9.5)	−1.9 (−3.7–−0.1)	1.3 (−0.8–3.3)	−4.8 (−9.2–−0.3)	−0.2 (−1.2–0.8)	−0.04 (−0.9–0.8)
0	20	0	0	0	0	0	0
1 SD	27	−4.8 (−7.8–−1.8)	2.1 (0.1–4.0)	4.2 (2.0–6.5)	1.6 (−2.7–5.9)	2.1 (0.8–3.5)	1.9 (0.7–3.2)
2 SD	37	−8.1 (−14–−2.3)	4.1 (−2.6–11)	16 (6.6–26)	1.0 (−10–12)	9.0 (1.1–17)	8.7 (0.7–17)
Normal weight
−2 SD	9.4	7.1 (5.8–8.3)	−1.1 (−1.8–−0.3)	3.9 (2.7–5.2)	−3.4 (−5.5–−1.2)	1.0 (0.5–1.6)	0.9 (0.3–1.4)
−1 SD	14	4.3 (3.4–5.2)	−1.3 (−1.7–−0.9)	2.2 (1.3–3.1)	−2.4 (−3.6–−1.1)	0.6 (0.2–1.0)	0.5 (0.1–0.8)
0	20	0	0	0	0	0	0
1 SD	27	−2.2 (−2.8–−1.7)	2.9 (2.3–3.6)	0.4 (−0.3–1.1)	3.7 (2.5–4.9)	0.1 (−0.2–0.4)	0.1 (−0.2–0.4)
2 SD	37	−1.9 (−3.4–−0.5)	6.3 (4.9–7.7)	5.0 (3.2–6.9)	7.8 (5.1–10)	1.3 (0.5–2.2)	1.2 (0.4–2.0)
Overweight
−2 SD	5.5	3.8 (2.6–5.0)	−1.6 (−2.4–−0.8)	2.7 (1.3–4.0)	−0.7 (−2.6–1.1)	0.6 (−0.01–1.2)	0.4 (−0.2–0.9)
−1 SD	11	2.8 (1.8–3.8)	−1.6 (−2.1–−1.0)	1.7 (0.6–2.7)	−1.1 (−2.6–0.4)	0.3 (−0.20–0.8)	0.1 (−0.3–0.5)
0	18	0	0	0	0	0	0
1 SD	28	−2.1 (−2.9–−1.3)	4.3 (3.1–5.5)	0.3 (−0.7–1.4)	3.3 (1.5–5.0)	0.5 (−0.09–1.0)	0.6 (0.1–1.1)
2 SD	41	−0.5 (−2.7–1.8)	7.7 (4.9–10)	6.2 (2.9–9)	6.3 (1.9–11)	2.7 (0.9–4.6)	2.6 (0.9–4.4)
Grade 1 Obese
−2 SD	0.9	2.1 ( 1.5–2.7)	−3.3 (−5.8–−0.8)	1.1 (0.2–2.0)	−1.6 (−4.3–1.2)	0.2 (−0.4–0.8)	0.1 (−0.5–0.7)
−1 SD	6.4	1.1 (0.7–1.4)	−1.9 (−3.4–−0.5)	0.4 (−0.07–0.8)	−1.1 (−2.3–0.2)	0.03 (−0.2–0.3)	−0.01 (−0.3–0.3)
0	14	0	0	0	0	0	0
1 SD	26	−1.5 (−2.8–−0.2)	3.3 (2.3–4.3)	1.0 (−0.4–2.5)	3.6 (1.7–5.6)	0.6 (−0.8–1.3)	0.6 (−0.1–1.2)
2 SD	43	−3.1 (−5.8–−0.4)	8.7 (4.7–13)	3.1 (−1.0–7.1)	8.8 (3.6–14)	1.8 (−0.4–4.1)	1.7 (−0.4–3.7)
Grade 2 Obese
−2 SD	0.9	1.3 (0.2–2.5)	−1.5 (−3.0–−0.1)	0.9 (−0.02–1.7)	−2.4 (−4.8–0.01)	0.4 (−0.1–0.9)	0.3 (−0.1–0.8)
−1 SD	6.4	1.0 (0.2–1.9)	−1.2 (−1.9–−0.6)	0.2 (−0.5–0.8)	−1.7 (−2.9–−0.6)	0.1 (−0.3–0.5)	0.1 (−0.3–0.4)
0	14	0	0	0	0	0	0
1 SD	26	−2.4 (−4.3–−0.5)	4.3 (2.6–6.0)	0.9 (−1.2–3.1)	4.2 (1.3–7.2)	0.3 (−0.9–1.5)	0.3 (−0.8–1.4)
2 SD	43	−4.6 (−7.9–−1.4)	12 (6.1–18)	2.3 (−2.9–7.6)	9.5 (2.5–17)	0.7 (−2.2–3.7)	0.7 (−2.1–3.5)
Grade 3 Obese
−2 SD	0.9	3.2 (1.4–4.9)	−2.3 (−3.1–−1.5)	1.0 (−0.7–2.6)	−2.1 (−4.1–−0.2)	0.04 (−0.7–0.7)	0.3 (−0.4–1.0)
−1 SD	6.4	2.3 (0.9–3.7)	−2.0 (−2.6–−1.4)	0.4 (−0.9–1.8)	−1.2 (−2.7–0.3)	−0.2 (−0.7–0.4)	0.04 (−0.5–0.6)
0	14	0	0	0	0	0	0
1 SD	26	−1.5 (−2.2–−0.8)	1.8 (1.1–2.5)	−0.5 (−1.4–0.5)	1.1 (−0.04–2.3)	0.05 (−0.4–0.5)	−0.1 (−0.5–0.3)
2 SD	43	−5.2 (−7.4–−3.0)	14 (6.6–20)	−0.7 (−5.7–4.3)	3.3 (−3.1–9.7)	1.4 (−1.7–4.6)	0.4 (−2.1–3.0)
Grade 2 or Grade 3 Obese
−2 SD	0.9	2.6 (1.2–3.9)	−1.8 (−2.5–−1.0)	1.0 (−0.3–2.2)	−2.0 (−3.8–−0.3)	0.2 (−0.4–0.7)	0.3 (−0.2–0.9)
−1 SD	6.4	1.9 (0.7–3.1)	−1.6 (−2.1–−1.1)	0.4 (−0.6–1.5)	−1.2 (−2.4–0.1)	−0.1 (−0.5–0.4)	0.1 (−0.4–0.6)
0	14	0	0	0	0	0	0
1 SD	26	−3.2 (−4.9–−1.4)	5.2 (3.0–7.3)	−0.3 (−2.9–2.3)	1.7 (−1.5–4.9)	0.6 (−0.7–2.0)	0.2 (−1.2–1.4)
2 SD	43	−5.6 (−8.1–−3.1)	14 (6.8–22)	−0.5 (−6.1–5.0)	3.6 (−3.4–11)	1.6 (−1.9–5.1)	0.5 (−2.3–3.2)

SD, standard deviation

^*Adjusted for maternal race or ethnicity, maternal education, maternal age, marital status, parity, insurance source, smoking during pregnancy, pre-existing diabetes or hypertension, height, prepregnancy weight, infertility treatment, infant sex, neonatal care level of birth facility, year of birth, urban residence, and racial composition of neighborhood.

^†When the estimate is a negative value, it is the adjusted number of prevented cases per 100 births.

^‡Underweight (<18.5 kg/m²), normal weight (18.5–24.9 kg/m²), overweight (25–29.9 kg/m²), grade 1 obese (30–34.9 kg/m²), grade 2 obese (35–39.9 kg/m²), or grade 3 obese (≥40 kg/m²)

^§Total weight gain ranges are the same for all grades of obesity because there is one z score chart for obese women (BMI ≥30 kg/m2)

Adjusted associations between gestational weight gain and preterm birth <32 weeks were U-shaped for all women except women with grade 2 or 3 obesity, among whom there was no association. An increased risk of infant death was observed for low weight gain among normal weight women (≤14 kg at 37 weeks [z score <−1 SD]) and for high weight gain among underweight (≥27 kg [z score +1 SD]), normal weight (≥37 kg [z score +2 SD]), and overweight women (≥41 kg [z score +2 SD]). Results were similar when we studied neonatal death (data available upon request).

Figure 2 shows associations between pregnancy weight gain and risk of outcomes simultaneously for women in each BMI category (grade 2 and 3 obesity are combined in Appendix 1). We overlaid the 2009 IOM provisional recommended total weight gain ranges for twins with vertical lines (underweight – no recommendation; normal weight 16.8–24.5 kg; overweight 14.1–22.7 kg; obese 11.3–19.1 kg). Within the IOM-recommended weight gain ranges for normal weight, overweight, and grade 1 obese women, there were meaningful declines in SGA risk and increases in LGA risk. Below the guidelines, risks of SGA, infant death (for normal weight women only), and preterm birth <32 weeks were elevated. At the upper limit of the IOM recommendations for these 3 BMI groups, the risk of SGA and LGA were approximately 10%, but weight gains greater than the upper cut-point were associated with increasing risk of preterm birth, infant death, and cesarean delivery. For women with severe obesity, weight gain below the provisional guidelines was associated with an elevated risk of SGA (adjusted predicted probabilities >12%), but not preterm birth or infant death. At gains higher than recommended, risk of LGA increased sharply. In sum, most excess risk was observed at weight gains more than 1 SD away from the mean: <14 kg or >27 kg in underweight or normal weight women, <11 kg or >28 kg in overweight women, and <6.4 kg or >26 kg in women with obesity.

Figure 2.

Adjusted predicted risks of four adverse perinatal outcomes by gestational weight gain z score among women with twin gestations who are underweight (n=1,858, A), normal weight (n=26,011, B), overweight (n=13,373, C), grade 1 obese (n=7,153, D), grade 2 obese (n=3,684, E), or grade 3 obese (n=2,757, F), Pennsylvania twin birth certificates, 2003–2013. The solid lines represent the point estimate and dashed lines represent its 95% confidence bands. Vertical dotted lines represent the current Institute of Medicine provisional gestational weight gain guidelines for twin pregnancies.^8,11 Gestational weight gain z scores have been converted to equivalent total weight gain at 37 weeks of gestation on the y-axis. Tick marks and values of total weight gain are given for z scores of −2, −1, 0, 1, and 2 standard deviations. Risks were set at the population average for maternal race or ethnicity, maternal education, maternal age, marital status, parity, insurance source, smoking during pregnancy, preexisting diabetes or hypertension, height, prepregnancy weight, infertility treatment, infant sex, neonatal care level of birth facility, year of birth, urban residence, and racial composition of neighborhood.

There were no meaningful differences when we limited to sex-discordant twins, twins without congenital malformations, or twins with complete data (Appendix 2–4).

Our probabilistic bias analysis showed that measurement error in BMI and gestational weight gain on the birth certificate led to biases both away from the null and towards the null, but was unlikely to explain the aforementioned associations (Appendix 5), conditional on the accuracy of the bias model.

Discussion

In this population-based cohort of Pennsylvania twins, we found that very low weight gain was related to excess risk of SGA in all women and preterm birth <32 weeks and infant death in women without obesity, while very high weight gain is related to excess risks of LGA and cesarean delivery in all women and preterm birth <32 weeks and infant death in women without obesity. These associations remained after adjustment for many important confounding factors including race or ethnicity, smoking, socioeconomic status, and infertility treatment and were supported after a quantitative evaluation of bias due to measurement error in BMI and gestational weight gain.

It is difficult to compare our findings to existing literature on gestational weight gain in twins because published studies have suffered from serious limitations, including the use of small, highly-selected samples, lack of stratification by prepregnancy BMI or control for confounders (10). Studies of fetal growth in twins tended to use absolute measures of birth weight, which are problematic because this measure cannot separate infants who are small due to preterm delivery from infants who are small due to intrauterine growth restriction (40). Nevertheless, our findings of a negative association between gestational weight gain and SGA in all BMI groups is consistent with several (41–44) but not all (45,46) large studies in twins.

Importantly, published work on weight gain in twin pregnancies has used measures of weight gain that failed to account for the impact of the duration of pregnancy on weight gain. Our use of BMI-specific z score charts to classify gestational weight gain (15) untangles gestational duration from weight gain, thereby removing this bias (18).

In our study, gestational weight gain had a U-shaped association with preterm birth <32 weeks among women without severe obesity. Infant death was elevated with high weight gain among women without obesity and low weight gain only among normal weight women. In singleton pregnancies, U-shaped relations have been observed between gestational weight gain z score and preterm birth and infant death in non-obese women, but associations have been more variable among women with obesity (12,13,48–50). Our observation of elevated risks of LGA birth and cesarean delivery and reduced risk of SGA birth as gestational weight gain z score increased in each BMI group is consistent with a large body of evidence in singletons (8,51).

We found that very low weight gain (0.9 to 6.4 kg at 37 weeks) among women with severe obesity increased the risk of SGA birth, but was not associated with preterm birth or infant death (outcomes related to low weight gain in non-obese samples (8,13)). Only a single study from 1990 has reported associations between gestational weight gain and adverse outcomes in twin pregnancies with severe obesity, but inferences were limited by the sparse data from only 179 women (45). Research evaluating a wider range of outcomes is needed to understand the safety of low gestational weight gain among pregnant women with severe obesity and carrying twins (8,11). Poor perinatal outcomes rise as obesity becomes more severe (52,53), and optimal gestational weight gain may attenuate these risks. Further, prepregnancy BMI and gestational weight gain contribute to an intergenerational cycle of obesity (54).

We lacked data to evaluate chorionicity as a confounder or effect modifier. Although chorionicity affects twin perinatal outcomes, it is unknown whether it influences gestational weight gain. When we limited our sample to sex-discordant pairs, which are known to be dichorionic, inferences were similar. Our measure of infertility treatment was based on a single question in the birth record that underestimates actual use (55,56). There is likely incomplete control for confounding by assisted reproductive technologies.

We did not include gestational diabetes or preeclampsia as outcomes because these conditions may alter total gestational weight gain, and data on early-pregnancy weight gain were not available to avoid biases due to reverse causality. We also did not adjust for these pregnancy complications because they likely lie on the causal pathway, and therefore do not meet the definition of a confounder (57). Strong system level factors shape the management of twin gestations. For example, some hospitals preclude twins from delivering vaginally. Relations with cesarean delivery may not be relevant for all twin pregnancies. The birth certificate does not collect clinical data that would have allowed us to distinguish spontaneous from induced preterm births or to classify twins based on achievement of their growth potential. We have documented measurement error in prepregnancy weight, height, and weight at delivery on Pennsylvania birth records (36,37), but our bias analysis found little evidence to support misclassification as importantly influencing the results or inferences from our study. In the study period, twin birth rates and the distribution of maternal age and race or ethnicity in twin births were similar between Pennsylvania and the U.S.(58), which lends support to the generalizability of our results.

Lifestyle interventions can modify gestational weight gain and improve some outcomes in singletons (59–62). While some evidence suggests that women with twin gestations who participate in a dietary intervention program are more likely than non-participants to meet weight gain goals (63,64), randomized trials have not tested whether interventions in twin pregnancies can modify weight gain and adverse outcomes. As a result, the causality and the biological basis of associations between pregnancy weight gain and outcomes in twin gestations are unknown.

We found that weight gains well above or well below the IOM provisional guidelines (<14 kg or >27 kg in underweight or normal weight women, <11 kg or >28 kg in overweight women, and <6.4 kg or >26 kg in women with obesity) were associated with the highest risk of adverse outcomes and that nearly 1 in 3 women had such weight gains. If the associations we observed are even partially reflective of causality, then targeted modification of pregnancy weight gain in women carrying twins might improve pregnancy outcomes. Data on a wide range of short- and long-term outcomes and information on the relative seriousness of these outcomes is needed to determine optimal gestational weight gain ranges for twin pregnancies.