Association of lactation with maternal risk of type 2 diabetes: A systematic review and meta‐analysis of observational studies

To investigate the association between lactation and maternal risk of type 2 diabetes, including a potential graded association according to lactation duration.


| INTRODUCTION
The World Health Organization (WHO) and the United Nations International Children's Emergency Fund (UNICEF) recommend exclusive breastfeeding for the first 6 months of life, followed by the introduction of complementary foods along with continued breastfeeding, up to at least the age of 2 years. 1 There is compelling evidence of the short-and long-term benefits of breastfeeding for children, such as protection against childhood conditions, particularly infections and obesity, as well as chronic diseases, including type 2 diabetes and cardiovascular disease later in life. [2][3][4] Comparably less attention has been given to the benefits of lactation (i.e. breastfeeding a child) for mothers, although previous studies have suggested that lifetime duration of lactation is associated with a reduced risk of breast and thyroid cancer, metabolic syndrome, atherosclerosis and hypertension. [5][6][7][8][9] However, evidence of the effects of lactation on women's risk of developing type 2 diabetes is conflicting, with some studies suggesting a protective effect, whilst others report no association. [10][11][12][13][14] Almost all studies are limited by their inability to account for metabolic risk profiles before lactation, and previous history of gestational diabetes (GD), a strong risk factor for type 2 diabetes in women. Furthermore, previous meta-analyses have only included subsets of available studies and have not investigated the possibility of a graded association between duration of lactation and type 2 diabetes in women. [15][16][17] Therefore, the aims of this study were to investigate the effects of lactation on maternal risk of type 2 diabetes overall and according to previous history of GD, and to investigate the existence of a doseresponse relationship between lactation and maternal risk of type 2 diabetes.

| Study design
A systematic review and meta-analysis of observational studies was conducted according to the Meta-analyses Of Observational Studies in Epidemiology (MOOSE) statement and the Cochrane Collaboration. 18,19 The protocol was registered with the PROS-PERO database of systematic reviews (CRD42020221183).

| Literature search
The bibliographic databases Medline (via PubMed) and Embase (via Ovid) were searched, from inception to February 2021, with MESH terms for lactation and breastfeeding and for type 2 diabetes (the search criteria are provided in the supporting information). No restrictions were applied to type of study or language. This was complemented with hand-searching of reference lists of eligible studies and related systematic reviews.

| Inclusion and exclusion criteria
Cohort, case control or cross-sectional studies that investigated the effects of lactation on risk of type 2 diabetes were included, irrespective of the duration and intensity (e.g. exclusive, predominant or intermittent) of lactation. No restriction on the number of participants or duration of follow-up was applied.
Studies that only reported outcomes for offspring, that compared the effects of different strategies to increase lactation, that only measured the effects of other perinatal exposures, or that only reported maternal outcomes other than type 2 diabetes, were excluded.

| Definition of exposure
The exposure to lactation was measured according to the definitions adopted by the included studies, which included average lactation per child and cumulative lifetime lactation and lactation for the index pregnancy (for studies that followed up women after an index pregnancy).

| Definition of outcome
Type 2 diabetes was defined according to the included studies. Prospective studies measured incidence of type 2 diabetes according to international guidelines based on oral glucose tolerance tests. Retrospective and cross-sectional studies used a combination of selfreported diagnoses, fasting plasma glucose testing or linked medical records.

| Study selection
Two investigators (ACPG and GM) independently screened titles and abstracts for all eligible studies according to the inclusion and exclusion criteria. Full-text articles were retrieved and reviewed in duplicate, with disagreements resolved by a third reviewer (AJ). EndNote X8 software was used to manage references and record screening and selection operations. Details of the study selection are documented in Figure S1.

| Data extraction
Data extraction was performed by one reviewer (ACPG) using a structured form. Data were collected for study and population characteristics, including demographics, history of GD, lactation mode and duration and risk of type 2 diabetes.

| Risk of bias and quality assessment
The methodological quality of eligible studies was assessed using the Newcastle-Ottawa Scale for non-randomized studies. 20 Studies were given stars depending on specific criteria in three fields: selection, comparability and outcomes.

| Data analysis
The association between lactation and maternal risk of type 2 diabetes was explored in two complementary analyses: (a) ever versus never lactation and (b) according to duration of lactation. All studies that reported adjusted risk estimates were included in the first meta-analysis, 14,[21][22][23][24][25][26][27][28][29][30][31] and those that reported effect sizes for intervals of duration of lactation were included in the second metaanalysis. 13,26,29,[31][32][33][34] Three studies contributed to both analyses. 26,29,31 Reported hazard ratios for type 2 diabetes with 95% confidence intervals (CIs), with adjustment for confounders, were extracted for cohort studies, when available. Reported risk or odds ratios and 95% CIs, with adjustment for confounders, were extracted if hazard ratios were not available for cohort studies. Reported odds ratios, with adjustment for confounders, with 95% CIs were extracted for crosssectional studies. When numbers of cases but not adjusted effect sizes were reported, studies were excluded from the meta-analysis.
Unadjusted effect sizes were not calculated for studies that only reported the number of cases in exposed versus unexposed individuals because of the high risk of confounding in observational studies.
Those studies were only included in the qualitative synthesis. [10][11][12]35,36 All studies were adjusted at least for age and body mass index (BMI) in multivariable models, with most adjusting for additional variables, such as ethnicity, education, smoking and parity. Whenever results for more than one model were reported, we selected the model with the highest number of covariates. The specific covariates used in multivariable models for each study are provided in Table S2.
Inverse variance weighted random effects meta-analyses, based on restricted maximum likelihood estimators (REML), were performed.
Pooled estimates were reported as relative risks (RRs) with 95% CIs and are presented using forest plots. Heterogeneity was quantified using the I 2 statistic and was tested using Cochran's Q test.
For the dose-response meta-analysis, the mean point of the interval for which the hazard ratio was available in each study was used (e.g. for 3 to 6 months, 4.5 months was used). The linearity of the association, on the log scale, was verified by drawing a bubble plot ( Figure S2). Next, a dose-response meta-analysis was conducted accounting for the clustering of subjects within each study. This was achieved using a one-stage random effects model, which took into account study-specific deviations from the population average doseresponse coefficients. 37 This model used weighted least squares regression with cluster robust error variances. The regression line with respective 95% CI was plotted to illustrate the fitted dose-response association. For the ever versus never meta-analysis, sensitivity analyses were conducted to explore the impact of including the study that compared lactation for under versus over 12 months, and excluding studies with a sample size of less than 500 participants [21][22][23]25,30 For the doseresponse meta-analysis, sensitivity analyses were conducted (a) for studies that considered total lifetime duration versus average duration per child or duration for the index pregnancy, (b) excluding crosssectional studies, and (c) for studies that included only women with a history of GD (i.e. excluding studies that included women without GD).
Publication bias was investigated using funnel plots and Egger's tests. All P values were calculated from two-tailed tests. Analyses were performed using STATA version 16. 38

| RESULTS
A total of 22 studies were eligible for inclusion in this systematic review, but only 16 of them contributed to the meta-analysis, as the remaining did not provide adjusted risk estimates. [10][11][12]35,36 One of the studies contributed with data for two different cohorts of women, and hence in reality there were 17 studies. 34 The characteristics of the included studies are summarized in Table 1 24,26,28,33 (c) total duration for a single pregnancy in studies that followed up women after a specific pregnancy, 10,12,21,22,25,27,[29][30][31]35 and (d) lactation at hospital discharge. 14,23 Follow-up varied from 4 to 12 weeks for studies that evaluated the short-term association between lactation and type 2 diabetes after birth, and up to 25 years.
The overall quality of the studies was modest (Table S1), with the main caveat being that lactation was self-reported and often many years after birth. There were some good quality cohort studies with large sample sizes and a low risk of bias. 13,31,32 However, there were also several small studies with a high risk of selection or information bias. [10][11][12]23,25,27 Overall, studies that investigated the association between lactation and the risk of type 2 diabetes in the first months after birth in women with GD reported conflicting results. Some studies suggested that lactation had a positive impact on glucose metabolism, 27,35 while others reported that lactation had no effect on glucose tolerance. [10][11][12]22 Studies with a longer follow-up showed a protective association between lactation and the RR of type 2 diabetes, with a stronger association in women with GD than in those without GD. 13,[31][32][33] The extent of the benefit seemed to increase in parallel with duration of lactation and to be independent of well-established risk factors for type 2 diabetes, such as obesity, ethnicity, diet, exercise, weight gain and antecedent metabolic risk profiles. 24,28,34 Meta-analysis of studies that compared ever with never lactation 14,21-31 suggested that lactation was associated with a 27% lower risk of type 2 diabetes (RR 0.73, 95% CI [0.65, 0.83]) ( Figure 1). Sensitivity analyses, including the study that compared lactation for under with over 12 months, 21 and studies with a sample size of less than 500 participants, 21-23,25,30 did not have a substantial impact on the effect size estimates (Figures S3 and S4). There was significant heterogeneity in the meta-analysis (I 2 = 69%), which was completely explained by high heterogeneity among cross-sectional studies (I 2 = 72%), as there was no heterogeneity among cohort studies (I 2 = 0%). The protective association between lactation and type 2 diabetes was more marked in cohort than in cross-sectional studies (0.73    Figure S10).

| DISCUSSION
This systematic review and aggregate data meta-analysis suggested that lactation is associated with a reduced risk of maternal type 2 diabetes in comparison with no lactation (i.e. use of any breastmilk substitutes), particularly in women with GD. It also suggested that lactation has incremental benefits, with each additional month of lactation being associated with a 1% lower risk of type 2 diabetes, with no evidence that the graded association differs between women with GD and without GD. However, the overall quality of the observational studies, particularly in women without GD, was modest.
A recent meta-analysis in women with prior GD, which included 13 cohort studies, showed that, compared with no lactation, lactation was significantly associated with a lower risk of type 2 diabetes (RR 0.66, 95% CI [0.48, 0.90]). 15 Although the authors reported that there was no significant relationship between long-term lactation and type 2 diabetes, this may be explained by the fact that long term was defined as over 3 months. This is at odds with a previous Heterogeneity: τ 2 = 0.00, I 2 = .%, H 2 = .

Weight
Random−effects REML model F I G U R E 2 Forest plot for meta-analysis of the association between lactation and maternal risk of type 2 diabetes stratified by history of gestational diabetes. Forest plot displays hazard ratios with 95% confidence intervals for each study and for the pooled estimate, as well as measures of heterogeneity. "GD" refers to women with gestational diabetes, "NGD" refers to no gestational diabetes, and "All" refers to a mixed population of women with and without gestational diabetes. Studies with short-term follow-up (up to 3 months after delivery) were excluded from this meta-analysis. Martens, 2016 (1) refers to women without gestational diabetes and Martens, 2016 (2) refers to women with gestational diabetes as an overall estimate was not reported. REML, restricted maximum likelihood estimators meta-analysis, including 14 reports of nine studies, which had shown that, in women with GD, lactation longer than 4 to 12 weeks was associated with a reduced risk of type 2 diabetes compared with shorter lactation. 17 Furthermore, the protective effect seemed to increase over time, as the risk reduction appeared larger when type 2 diabetes was evaluated more than 5 years after delivery. More recently, another meta-analysis including four studies showed that lactation for more than 12 months was associated with a RR reduction of 30% for type 2 diabetes (odds ratio 0.70, 95% CI [0.62, 0.78]). 16 This meta-analysis had important limitations as it only considered lactation longer than 12 months, thus it did not take into account variability in effects according to duration of lactation, nor did it perform subgroup analyses according to history of GD, type of study or duration of follow-up.
In this context, our systematic review and aggregate data meta- Albeit acknowledging that the association reported in observational studies may not be causal, the potential mechanisms underlying the beneficial effects of lactation on maternal risk of type 2 diabetes are worth discussing. 39 First, lactation has been associated with less weight retention after pregnancy as well as a lower risk of obesity in later life, which is the main risk factor for diabetes. 40 However, the protective effect of lactation appears to be independent of obesity or BMI. Second, the complex hormonal changes elicited by pregnancy, and the influence of lactation on this hormonal milieu, may influence glucose and insulin metabolism and beta-cell function, 41-44 thus reducing the risk of developing diabetes in later life. [45][46][47] The graded association observed in this meta-analysis would fit with a potentially stronger or long-lasting effect associated with longer or more intensive lactation. Third, it is possible that the observed association is partially confounded, as lactation has been associated with other healthy behaviours. 48 Further research is thus warranted to better clarify the mechanisms underpinning the association between lactation and the reduced risk of type 2 diabetes, including how this association may be modulated by exposure to other risk factors, such as obesity. 49 This study has important implications for healthcare professionals and policymakers worldwide. Although the WHO recommends exclusive breastfeeding (i.e. no other food or drink, not even water, except breastmilk) for the first 6 months followed by partial breastfeeding for as long as possible, in 2018 only 41% of women were exclusively breastfeeding for 6 months worldwide, and this dropped to 24% in upper-middle income countries. 50 Furthermore, lactation rates appear to be lower in women who develop GD, which may be related to delayed onset of lactogenesis, 51 despite our study showing that they probably afford a larger risk reduction in type 2 diabetes than in women without GD. 52 Reasons for suboptimal lactation rates across the globe, and especially in high-income countries, are multifactorial in that they are influenced by a variety of individual and structural factors. The WHO has acknowledged that poverty, malnutrition and a lack of education and training on infant feeding decrease the prevalence of lactation. 53 Other individual factors include cultural and religious beliefs, lack of support systems and mental health barriers, as women with depression in pregnancy or postpartum are less probable to initiate and continue lactation. 54 Broader structural factors also influence lactation rates, such as more mothers entering the labour force and negative attitudes towards lactation in public. 54 The unavailability of paid maternity leave is a key contributor, 55 as is the lack of employment policies that promote lactation. 54 Finally, aggressive marketing of breastmilk substitutes is a powerful disincentive for lactation, as it shapes perceived social norms and misinforms the public about scientifically recommended infant nutrition. 56 Considering the stark increase in the global burden of disease attributable to type 2 diabetes, 57 this study lends further support to the importance of implementing lactation-friendly policies that enable women and their children to accrue the benefits of lactation for their health and well-being. Ultimately, this will also be positive for healthcare and welfare systems by saving resources and reducing the disproportionate burden of type 2 diabetes on women. 58 On the other hand, equipping healthcare professionals with the skills and knowledge to provide individually tailored advice and support is paramount to improve lactation, 59 particularly for women who experience socioeconomic and cultural barriers. 60 Although this systematic review and aggregate data meta-analysis suggests that lactation may reduce the risk of type 2 diabetes in parous women, several questions have yet to be answered. First, despite the graded association between duration of lactation and the risk of type 2 diabetes, it remains unclear whether there is a threshold beyond which no further protection is achieved by prolonging lactation. Second, there is uncertainty on how the effects of lactation change over time, that is, whether benefits are evident shortly after birth or if they only accrue as women age. Third, evidence is lacking on the effects of different modalities of lactation. Although there is evidence that exclusive lactation could be more protective against type 2 diabetes than partial lactation, 31 the intensity of lactation that is required to observe a meaningful difference in risk reduction remains unknown. It has also yet to be understood whether there is an interaction between intensity and duration in a way that partial lactation could be compensated by prolonging the duration of lactation.
Fourth, as large cohort studies were mainly conducted in women with GD, no definite conclusions can be drawn about potential differences between the effects of lactation in women with and without GD. As lactation is an exposure that cannot be randomized, large, prospective cohort studies, including a representative sample of the population (e.g. diverse ethnicities and socioeconomic backgrounds), with detailed and objective assessment of lactation and type 2 diabetes, are warranted to understand the effects of lactation on maternal risk of developing type 2 diabetes. Importantly, these cohort studies should be powered to conduct subgroup analyses according to gestational diabetes status and obesity.
This systematic review summarizes the best evidence currently available on the association between lactation and maternal risk of diabetes. Nonetheless, there are some limitations to acknowledge.
First, it is affected by the limitations of the observational studies upon which it is based, such as selection bias, information bias and residual confounding. However, the nature of the exposure means that randomized controlled trials are not ethically feasible, and hence observational studies are the only source of evidence. Second, many studies were of low or modest quality, particularly small studies based at single hospitals and with substantial loss to follow-up. In addition, lactation was self-reported in all studies, and in many of them a long time after birth. However, the nature of this exposure makes it difficult to ascertain accurately. Third, there was significant heterogeneity across studies in some analyses, mainly because of methodological differences among studies. Although we performed subgroup analyses to investigate potential sources of heterogeneity, individual participant data are required to adequately explore heterogeneity and avoid ecological bias that may affect aggregate data meta-analysis. Fourth, the possibility of publication bias cannot be excluded. However, those limitations are unlikely to have a material impact on the key findings of this systematic review, which were robust to comprehensive sensitivity analyses.
In conclusion, lactation is associated with a significantly reduced risk of maternal type 2 diabetes over the life course, particularly in women with GD. The protective effect seems to increase with longer duration of lactation, but it remains uncertain whether there is a threshold above which the benefits become significant, or a ceiling effect above which no further benefits are accrued. Further research is also warranted to understand whether this association is modified by exposure to other risk factors.
F I G U R E 3 Dose-response meta-analysis for duration of lactation and maternal risk of type 2 diabetes. Plot displays dose-response relationship assuming a linear association (solid line) with 95% confidence intervals (dashed lines). Reference is zero, which means no lactation. Maternal risk of type 2 diabetes is plotted against the duration of lactation in months