Abstract
Background Breastfeeding has been associated with maternal and infant health benefits but has been inversely associated with body mass index (BMI) prepartum. Breastfeeding and BMI are both linked to socioeconomic factors.
Methods Data from parous female participants with available breastfeeding information from the Million Veteran Program cohort was included. BMI at enrollment and earliest BMI available were extracted, and polygenic scores (PGS) for BMI were calculated. We modeled breastfeeding for one month or more as a function of BMI at enrollment; earliest BMI where available pre-pregnancy; and PGS for BMI. We conducted Mendelian randomization for breastfeeding initiation using PGS as an instrumental variable.
Results A higher BMI predicted a lower likelihood of breastfeeding for one month or more in all analyses. A +5 kg/m2 BMI pre-pregnancy was associated with a 24% reduced odds of breastfeeding, and a +5 kg/m2 genetically predicted BMI was associated with a 17% reduced odds of breastfeeding.
Conclusions BMI predicts a lower likelihood of breastfeeding for one month or longer. Given the high success of breastfeeding initiation regardless of BMI in supportive environments as well as potential health benefits, patients with elevated BMI may benefit from additional postpartum breastfeeding support.
Background
Breastfeeding is associated with maternal health benefits, such as lower incidence of type 2 diabetes, in addition to well established benefits for infants. Body mass may have a complex time-dependent relationship with breastfeeding. Some studies have found that breastfeeding is associated with greater postpartum weight loss, while others have found no effect1. In the other direction, body mass prepartum may impact breastfeeding. Prospective studies have shown that those who breastfeed, or breastfeed for a longer duration, already had a lower pre-pregnancy body mass index (BMI)2, 3.
Breastfeeding is associated with socioeconomic factors. Women who breastfeed are more likely to be married and highly educated; they have a higher average household income, while they have lower rates of smoking and alcohol use4-6. Therefore, breastfeeding may be in part a marker of overall health status7. BMI is socioeconomically linked, and the association between BMI and breastfeeding could merely reflect practicalities such as work obligations that interfere with breastfeeding. However, this association has held even in a population with substantial postpartum work leave8, suggesting plausible biologic factors. A delayed onset of lactogenesis observed among overweight and obese mothers could be a factor9.
Use of a genetic risk score can mitigate some of the effects of confounding given that genetic variation is randomly assigned at conception due to the law of independent assortment. In this study, we look at the association between BMI and breastfeeding initiation in the Veteran population using both observational and genetic risk score methods.
Methods
Data Source
Million Veteran Program (MVP) is an observational cohort study and biobank of participants enrolled from the Department of Veterans Affairs (VA) health care system from 2011-present10. Data obtained for this project included phenotype data from the Baseline Survey, the Lifestyle Survey (which contains reproductive health questions), and electronic health records (EHR), as well as genotype data from the blood specimen drawn after enrollment. Participants in our study had enrolled between January 2011 and September 2021.
Outcome and exposures
The Lifestyle Survey included sex-specific reproductive health questions, including the binary outcome variable for this study, “Did you breastfeed your child(ren) for a least 1 month?”, hereafter referred to as breastfeeding initiation. Participants eligible for inclusion were parous female participants who had completed the Lifestyle Survey with non-missing values for number of live births and breastfeeding initiation. BMI information was extracted from the CoreVitals table previously created by the MVP Data Analytics team. Weight and height in this table had been produced by harmonizing the EHR and Baseline Survey data.
The Baseline Survey value was used if the values agreed or if no measured value existed, but EHR data was used if the EHR and survey values were contradictory11. We also extracted directly from the EHR the earliest weight and height measured and the age at this measurement to obtain the earliest BMI. We used Python 3.9.1 for data extraction/transformation and R 4.0.3 for data analyses.
Covariates
We obtained covariates of age, race, ethnicity, education, income, smoking status, and marital/partner status from the Baseline Survey. We obtained number of children born, breastfeeding initiation status, and age at first childbirth from the Lifestyle Survey. Details of these covariates are listed in Supplementary File Table S1. Summary statistics were extracted for covariates and breastfeeding initiation as a function of BMI variables.
Statistical Analysis
Observational Associations
We modeled breastfeeding initiation as a function of BMI using logistic regression. We used the two measures of BMI as previously described – BMI at enrollment and earliest BMI. We stratified the subset of participants whose earliest BMI was prior to childbirth as a sensitivity analysis due to the potential for reverse causation. BMI was considered both as a continuous variable and by quartile. Each analysis was conducted with three sets of covariates: (1) as a univariate analysis; (2) adjusted for age (at the date of the respective BMI value), race/ethnicity, education, income, smoking; (3) additionally adjusted for married/partnered status and number of children born.
Genetic Associations
We used the PGS catalog entry 27 (https://www.pgscatalog.org/score/PGS000027/) to calculate a BMI polygenic score (PGS) in MVP participants. The entry included a set of weights for 2,100,302 genetic variants, or single nucleotide polymorphisms (SNPs) derived from a genome wide association study of 238,944 individuals and validated in an independent data set of 119,95112. Both the training and validation datasets were independent of MVP. Genetic data in MVP originated from a custom Axiom array based on the Axiom Genotyping Platform and was imputed using the TOPMed panel in GRCh38/hg3813. A PGS for each individual was calculated as the sum of the products of imputed genetic dosage at each SNP and the corresponding PGS catalog weight14. These risk scores represent genetic liability for a higher BMI. Major ancestry groups had previously been determined based on Genetically Inferred Ancestry15. We standardized the PGS within each stratified ancestry group.
For genetic analyses, women were stratified based on their genetically inferred ancestry rather than their self-identified race/ethnicity. Population membership to all genotyped participants was assigned using a reference dataset of unrelated individuals from the 1000 Genomes Project (1KGP). This assignment was performed centrally as part of a core MVP project and made available as a core resource for all MVP investigators15. We verified the association between the PGS and BMI. Within each genetically inferred major ancestry stratum, we evaluated the breastfeeding initiation outcome as a function of the PGS, both continuously and by quartile. We adjusted for age at enrollment and the first 10 genetic principal components.
We further used the PGS as an instrumental variable for a one-sample Mendelian Randomization analysis. We conducted a Wald ratio analysis within each major ancestry stratum, adjusting for the first 10 genetic principal components, using the R MendelianRandomization package.
Results
The data yielded 20,375 individuals eligible for inclusion in this study, with BMI at enrollment available for 20,293 and BMI at/prior to the first offspring birth available for 532 (Supplementary File Figure S2). A total of 12,315 (60.7%) of participants had initiated breastfeeding. BMI at enrollment was positively associated with younger age of enrollment, self-identified Black or African American race, education less than a Bachelor’s degree, low or medium income level, non-response to marital status and smoking survey questions, and formerly smoking, while it was inversely associated with currently smoking and currently married/partnered (Table 1a). BMI was strongly associated with the within-ancestry scaled PGS for BMI (p=6.51 × 10−142) and was inversely associated with having initiated breastfeeding (p=7.53 × 10−13). The mean BMI was 29.9 kg/m2 (standard deviation 6.6), with means by quartile of 22.4, 27.1, 31.2, and 38.7. A standard deviation increase in the PGS was responsible for an increase in BMI at enrollment of 0.8-1.5 kg/m2 and an increase in earliest pre-pregnancy BMI of 0.8-1.2 kg/m2, depending on the ancestry group (Table 1b). The PGS had better performance in participants of European ancestry, consistent with the background of the training dataset that had been used to generate the PGS catalog entry.
The observational analysis revealed that a higher BMI at enrollment was associated with a lower rate of breastfeeding initiation across all covariate models, whether BMI was considered continuously or by quartile (Table 2a; odds ratio (OR) = 0.93; 95% confidence interval [0.91, 0.95] per +5 kg/m2 BMI for the fully adjusted model). Quartile estimates decreased monotonically (Q1, OR=1 (reference); Q2, OR=0.95 [0.86, 1.04]; Q3, 0.81 [0.74, 0.89]; Q4, 0.77 [0.70, 0.84]). In the subset of participants with a recorded BMI at or before the year of birth of the first offspring, effect sizes were larger with OR=0.76 [0.58, 1.00] per +5 BMI for the fully adjusted model (Table 2b). The fourth quartile showed an especially pronounced inverse association with breastfeeding initiation (OR=0.34 [0.13, 0.87]).
Genetic analyses included 11568, 3589, and 1327 participants of European (EUR), African (AFR), and Admixed American (AMR) ancestry, respectively, with an insufficient count of East Asian ancestry participants (131) to generate stable estimates. A higher PGS (and thus higher BMI) was predictive of a lower chance of breastfeeding initiation (Table 2c; OR=0.94 [0.91, 0.98] EUR; 0.94 [0.87, 1.01] AFR; 0.84 [0.73, 0.98] AMR per standard deviation of PGS).
Finally, the instrumental variable analysis with the PGS via Mendelian randomization showed a decrease in breastfeeding initiation with increased BMI (Table 3; per +5 kg/m2 BMI, OR=0.83 [0.74, 0.93] EUR; 0.69 [0.44, 1.09] AFR; 0.62 [0.41, 0.95] AMR).
Discussion
We found that BMI was inversely associated with breastfeeding initiation in all analyses in this study. Our analyses have mitigated against the likely pitfall of reverse causation (from a bidirectional relationship between breastfeeding and body mass) in two different ways: (1) through an observational pre-childbirth measurement in a subset and (2) via use of genetic risk scores. In the subset of individuals with pre-childbirth BMI available, a +5 kg/m2 BMI was associated with a 24% reduced odds of breastfeeding, with a nonlinear trend indicating the highest BMIs led to the greatest risk of not initiating breastfeeding. The Mendelian randomization analysis found the genetic liability for +5 kg/m2 BMI associated with at least a 17% reduced odds of breastfeeding. These findings suggest the importance of targeted, early breastfeeding support to those with higher BMIs who are at greater risk of being unable to initiate breastfeeding. Given that a lack of breastfeeding and a higher BMI are each independently linked with type 2 diabetes2, 16, providing this targeted support could provide an important intervention for mitigating against risk of diabetes.
Multiple large scale studies have previously found a relationship between pre-pregnancy body mass and breastfeeding. The MIHA study at UCSF found that pre-pregnancy obesity predicted never breastfeeding17. Obesity and overweight status were also associated with lower breastfeeding initiation in the Norwegian MoBa data18 and the Australian National Health Survey19. A retrospective analysis of over 176,000 women in the North West Thames region showed antenatal overweight/obesity associated with a markedly decreased odds of breastfeeding at discharge20. Pre-pregnancy body mass has also been implicated in earlier cessation of breastfeeding. In Nurses Health Study II data, BMI at age 18 years was inversely associated with duration of breastfeeding2. Pre-pregnancy BMI was also a predictor of duration of breastfeeding in the Danish National Birth Cohort, despite a nearly universal initiation of breastfeeding21.
Evidence is not compelling that intention to breastfeed by BMI drives these differences. Intent to breastfeed did not differ between BMI groups in the Infant Feeding Practices Study II (n=2824)22, though obesity was associated with a shorter intended duration of breastfeeding in a smaller study from Cornell (n=114)23. Moreover, animal studies have also found that elevated body mass is associated with declines in lactation. Obese and overly lean pigs in late gestation have reduced mammary development24. Fat cow syndrome and high feeding before puberty are both linked to low milk yield in cows25, 26. Pups of obese rats gained significantly less weight in 3 weeks compared to controls27. Altogether this evidence suggests a biological mechanism for the impact of body weight on lactation.
Metabolic profile changes impairing lactogenesis II (the onset of copious milk production) could be driving the link between decreased breastfeeding and increased BMI. Delayed lactogenesis has previously been associated with cessation of breastfeeding by 4 weeks postpartum in the Infant Feeding Practices Study28. Lactogenesis II is driven by drop in progesterone with placental ejection and occurs regardless of attempting breastfeeding29, and it is not associated with intention to breastfeed30. A delayed onset of lactation may lead to formula supplementation, which in turn may decrease the removal of milk from the breast, interfering with the supply-and-demand mechanism that drives the steady-state lactogenesis III by day 10 postpartum.
Metabolic factors, including BMI, have been consistently linked to a delay in lactogenesis II. Risk of delayed onset of lactation has been greater with overweight and obese BMI compared to normal range BMI in multiple studies, including a general population from UC Davis Medical Center9 and a gestational diabetes mellitus specific population in SWIFT30. A study of 26 primiparas at a prenatal clinic found that 56% of the variation in the time to lactogenesis II was explained by a 1-hour post-glucose challenge serum insulin concentration31. Finally, a study of 40 women in the rural US found that the serum prolactin response to infant sucking was lower in those within an overweight/obese BMI range than in a normal BMI range32. Prolactin is a lactogenic hormone that enables milk production, but it also affects insulin sensitivity and lipid metabolism33. Serum prolactin – both during pregnancy34 and at 6-9 weeks postpartum35 – is inversely associated with later type 2 diabetes.
Structural changes in the breast could also be responsible for differences in lactogenesis. A study of diet-induced obese mice found delayed lactogenesis by one day compared to lean mice. In another subset of mice from this study, microscopy of mammary gland tissue cells from day 14 of pregnancy revealed abnormal alveolar development with reduced ductal branching in obese mice36.
Researchers have questioned the direction of effect between breastfeeding and metabolic syndrome – breastfeeding as a reset of metabolic disruptions from pregnancy vs. metabolic disease adversely affecting breastfeeding such that adverse lactation outcomes are a marker of metabolic dysregulation7, 37. Both may be true. We have shown in this paper that a genetically predicted higher BMI associates with reduced likelihood of breastfeeding initiation. In light of other evidence, we consider that an impaired metabolic profile, as most easily measured by BMI, is responsible for a decrease in breastfeeding initiation.
A strength of our study is the sample size. Although the Veteran population is mostly male, the large number of participants enrolled in MVP led to a substantial female population eligible for inclusion. Despite the relatively smaller number of participants with pre-partum measurements, a sample size of over 500 in this stratum was substantial and sufficient to detect an effect. Nevertheless, replication in additional populations would be warranted. Another strength is diversity, with over 40% of the genetic samples in subjects of non-European ancestry.
A weakness of our study was a limited measure of breastfeeding, with information only on whether breastfeeding occurred for one month and no further information on breastfeeding duration. We do not have information on whether breastfeeding was attempted but not sustained vs. never attempted. Moreover, the survey question does not differentiate between multiparas who breastfed all children vs. only one child for one month or more. The survey also did not include information on exclusivity of breastfeeding vs. combined breastfeeding and formula feeding. Further studies are needed in data sets with additional variables. Another weakness was that most measured BMIs occurred after pregnancy, so the observational analysis of BMI at enrollment may suffer from reverse causation.
In conclusion, we have shown that a higher BMI predicts a lower likelihood of initiation of breastfeeding. Importantly, this finding does not imply that patients with elevated BMI should not attempt breastfeeding; in fact, when adequately supported, breastfeeding initiation can be nearly universal21. Instead, additional support should be provided to those with higher BMI in the postpartum time to ensure successful breastfeeding initiation, which may represent an important intervention time with later maternal health benefits.
IRB statement
The study received ethical and study protocol approval from the U.S. Veterans Affairs Central Institutional Review Board.
Data Availability
Access to individual level data used in the present work requires approval by the U.S. Veterans Administration.
Funding
JL is supported by the Big Data Scientific Training Enhancement Program (BD-STEP) through the Veterans Affairs and National Cancer Institute. TLA is supported by a grant from the National Institute of Diabetes and Digestive and Kidney Diseases (R01DK114183). This research is based on data from the Million Veteran Program, Office of Research and Development, Veterans Health Administration, and was supported by Veterans Administration awards I01-01BX003362. The content of this manuscript does not represent the views of the Department of Veterans Affairs or the United States Government.
Consortia
The VA Million Veteran Program. Members of the consortium can be found in the Supplementary File Text.
References
