Abstract
Background
Breastfeeding resets insulin resistance caused by pregnancy however, studies on the association between breastfeeding and diabetes mellitus (DM) have reported inconsistent results. Therefore, we aimed to investigate the risk of DM according to breastfeeding duration in large-scale population-based retrospective study. In addition, machine-learning prediction models for DM and hemoglobin A1c (HbA1c) were developed to further evaluate this association.
Methods
We used the Korean National Health and Nutrition Examination Surveys database, a nationwide and population-based health survey from 2010 to 2020. We included 15,946 postmenopausal women with a history of delivery, whom we divided into three groups according to the average breastfeeding duration: (1) no breastfeeding, (2) < 12 months breastfeeding, and (3) ≥ 12 months breastfeeding. Prediction models for DM and HbA1c were developed using an artificial neural network, decision tree, logistic regression, Naïve Bayes, random forest, and support vector machine.
Results
In total, 2248 (14.1%) women had DM and 14,402 (90.3%) had a history of breastfeeding. The prevalence of DM was the lowest in the < 12 breastfeeding group (no breastfeeding vs. < 12 months breastfeeding vs. ≥ 12 months breastfeeding; 161 [10.4%] vs. 362 [9.0%] vs. 1,725 [16.7%], p < 0.001). HbA1c levels were also the lowest in the < 12 breastfeeding group (HbA1c: no breastfeeding vs. < 12 months breastfeeding vs. ≥ 12 months breastfeeding; 5.9% vs. 5.9% vs. 6.1%, respectively, p < 0.001). After adjustment for covariates, the risk of DM was significantly increased in both, the no breastfeeding (adjusted odds ratio [aOR] 1.29; 95% CI 1.29, 1.62]) and ≥ 12 months of breastfeeding groups (aOR 1.18; 95% CI 1.01, 1.37) compared to that in the < 12 months breastfeeding group. The accuracy and the area under the receiver-operating-characteristic curve of the DM prediction model were 0.93 and 0.95, respectively. The average breastfeeding duration was ranked among the top 15 determinants of DM, which supported the strong association between breastfeeding duration and DM. This association was also observed in a prediction model for HbA1c.
Conclusions
Women who did not breasted had a higher risk of developing DM than those who breastfed for up to 12 months.
Supplementary Information
The online version contains supplementary material available at 10.1186/s13006-024-00642-z.
Keywords: Diabetes, Breastfeeding, Prevention, Machine learning
Background
Breastfeeding is known to not only decrease the risk of neonatal/infant morbidity and mortality, but also have beneficial effects on maternal health [1–7]. Breastfeeding is reported to help the resetting process of metabolic changes caused by pregnancy which includes insulin resistance [8]. However, the results of previous studies on breastfeeding and maternal risk of diabetes mellitus (DM) are not consistent. Some studies which have demonstrated a possible risk reduction of maternal DM, hypertension, and metabolic diseases [9–13], while others have found no association between breastfeeding and DM [14–18]. There could be various reasons for such inconsistent results, but factors such as the duration of breastfeeding or parity could be the potential causes. In previous studies, the duration of breastfeeding varied, and some studies analyzed the association between breastfeeding and DM without considering the duration of breastfeeding [14, 16–19].
In this study, we aimed to investigate the risk of DM according to breastfeeding duration in a large-scale population-based retrospective design in South Korea. Furthermore, we used machine learning to optimize the analysis of large-scale data containing a large number of confounding variables for diabetes, including medical history; obstetric characteristics; and demographic, socioeconomic, and lifestyle characteristics. We developed a prediction model for DM and hemoglobin A1c (HbA1c) using machine learning and evaluated the association between breastfeeding duration and DM.
Methods
Participants
This retrospective cross-sectional study used data from the Korean National Health and Nutrition Examination Surveys (KNHANES) database. The KNHANES, a nationwide and population-based health survey, was performed by the Korean Ministry of Health and Welfare. The KNHANES included a health examination, health interview, and nutrition survey. The KNHANES measured anthropometric measures and biochemical profiles from the participants’ blood and urine samples. The information on socioeconomic status and health behaviors were also collected through the interview and surveys and the interviews and surveys were conducted by well-trained researchers. The KNHANES has been described in previous studies [20]. The current study used data from the fifth (2010–2012), sixth (2013–2015), seventh (2016–2018), or eighth (2019–2020) KNHANES. Among the KNHANES participants, postmenopausal women with a history of delivery were selected as the study population. Women with incomplete data on gravidity, breastfeeding, or biochemical profiles were excluded, and 15,946 postmenopausal/parous women were included in the final analysis (Fig. 1).
Fig. 1.
Flow diagram of the study population
Variables
This study considered various covariates, including sociodemographic characteristics, lifestyle characteristics, medical and obstetric history (including breasting duration), and biochemical profiles (Table 1). Data regarding sociodemographic characteristics, health-related variables, and medical and obstetric histories (including breasting duration) were obtained from the questionnaires. The presence of the following diseases was defined as an answer of “yes” to questions regarding whether the participants had ever been diagnosed by a physician: (1) hypertension, (2) DM, (3) myocardial infarction, (4) angina, (5) stroke, (6) hypercholesterolemia, (7) dyslipidemia, (8) osteoarthritis, (9) rheumatoid arthritis, (10) pulmonary tuberculosis, (11) asthma, (12) thyroid-related disease, (13) major depressive disorder, (14) kidney failure, (15) hepatitis B, (16) hepatitis C, (17) liver cirrhosis, (18) gastric cancer, (19) hepatic cancer, (20) colorectal cancer, (21) breast cancer, (22) cervical cancer, and (23) lung cancer. The total breastfeeding duration was defined based on a questionnaire in which female participants responded as the total duration of breastfeeding during their lifetime. The “average breastfeeding duration” was calculated using the following formula: average breastfeeding duration = (total breastfeeding duration) / (number of breastfed children).
Table 1.
Baseline characteristics of the study population
| Variables | DM group (n = 2,248) |
Control group (n = 13,691) |
p-value |
|---|---|---|---|
| Age at enrollment (years) | 67.9 (± 8.2) | 63.2 (± 9.0) | < 0.0001 |
| Body mass index (kg/m2) | 25.2 (± 3.5) | 24.1 (± 3.2) | < 0.0001 |
| Waist circumference (cm) | 86.5 (± 9.3) | 81.8 (± 90) | < 0.0001 |
| Systolic blood pressure (mmHg) | 129.3 (± 17.2) | 124.8 (± 17.8) | < 0.0001 |
| Diastolic blood pressure (mmHg) | 72.3 (± 9.6) | 75.8 (± 9.8) | < 0.0001 |
| Breastfeeding | |||
| Average breastfeeding duration (month) | 15.5 (± 9.7) | 12.81(± 9.2) | < 0.0001 |
| Breastfeeding duration group | < 0.0001 | ||
| No breastfeeding | 161 (7.2%) | 1,383 (10.1%) | |
| < 12 months | 362 (16.1%) | 3,676 (26.9%) | |
| ≥ 12 months | 1,725 (76.7%) | 8,632 (63.1%) | |
| Obstetric characteristics | |||
| Age at menarche (years) | 15.5 (± 2.03) | 15.3 (± 2.0) | < 0.0001 |
| Age at menopause (years) | 49.0 (± 5.5) | 49.2 (± 4.9) | 0.047 |
| Married | 2,248 (100.0%) | 13,680 (99.9%) | 0.603 |
| Gravidity | 5.1 (± 2.4) | 4.5 (± 2.1) | < 0.0001 |
| Age at first delivery (years) | 23.3 (± 3.5) | 24.2 (± 3.6) | < 0.0001 |
| Age at last delivery (years) | 29.8 (± 4.6) | 29.6 (± 4.4) | 0.014 |
| Prevalence of medical diseases | |||
| Hypertension | 1,524 (67.8%) | 4,894 (35.8%) | < 0.0001 |
| Dyslipidemia | 1,216 (54.1%) | 3,634 (26.5%) | < 0.0001 |
| Stroke | 145 (6.5%) | 347 (2.5%) | < 0.0001 |
| Myocardial infarction | 65 (2.9%) | 110 (0.8%) | < 0.0001 |
| Kidney failure | 30 (1.3%) | 74 (0.5%) | < 0.0001 |
| Liver cirrhosis | 15 (0.7%) | 42 (0.3%) | 0.008 |
| Biochemical profile at enrollment | |||
| Fasting glucose (mg/dL) | 134.7 (± 40.7) | 97.7 (± 14.4) | < 0.0001 |
| HbA1c | 7.3 (± 1.3) | 5.8 (± 0.5) | < 0.0001 |
| Total cholesterol (mg/dL) | 175.2 (± 38.8) | 202.2 (± 37.3) | < 0.0001 |
| HDL cholesterol (mg/dL) | 47.1 (± 11.2) | 52.2 (± 12.1) | < 0.0001 |
| Triglycerides (mg/dL) | 148.3 (± 85.1) | 128.6 (± 80.8) | < 0.0001 |
| LDL cholesterol (mg/dL) | 100.0 (± 33.7) | 124.6 (± 33.7) | < 0.0001 |
| Family history | |||
| Paternal family history of diabetes mellitus | 153 (8.5%) | 565 (4.8%) | < 0.0001 |
| Maternal family history of diabetes mellitus | 340 (17.8%) | 1,068 (8. 6%) | < 0.0001 |
| Socioeconomic and lifestyle characteristics | |||
| Education level | < 0.0001 | ||
| Elementary school and below | 1,469 (65.4%) | 6,475 (47.3%) | |
| Middle school | 335 (14.9%) | 2,297 (16.8%) | |
| High school | 334 (14.9%) | 3,366 (24.6%) | |
| College and above | 109 (4.9%) | 1,540 (11.3%) | |
| Household income | < 0.0001 | ||
| Low | 961 (43.0%) | 3,976 (29.2%) | |
| Medium-low | 601 (26.9%) | 3,565 (26.2%) | |
| Medium-high | 384 (17.2%) | 3004 (22.1%) | |
| High | 287 (12.9%) | 3,062 (22.5%) | |
| Economic activity | 709(31.5%) | 6,158(45.0%) | < 0.0001 |
| Residential areas | 0.267 | ||
| Urban | 1,684 (74.9%) | 10,404 (76.0%) | |
| Rural | 564 (25.1%) | 3,287 (24.0%) | |
| Frequency of drinking per year | < 0.0001 | ||
| Never | 832 (37.3%) | 3,649 (26.8%) | |
| Have not drunk in the last 1 year | 564 (25.3%) | 2,862 (21.0%) | |
| Less than once a month | 436 (19.5%) | 3,246 (23.8%) | |
| Once a month | 141 (6.3%) | 1,244 (9.1%) | |
| 2–4 times a month | 171 (7.7%) | 1,711 (12.6%) | |
| 2–3 times a week | 54 (2.4%) | 653 (4.8%) | |
| ≥ 4 times a week | 34 (1.5%) | 269 (2.0%) | |
| Smoking status | 0.020 | ||
| Smoker | 84 (3.8%) | 446 (3.3%) | |
| Ex-smoker | 92 (4.1%) | 425 (3.1%) | |
| Non-smoker | 2,056 (92.1%) | 12,767 (93.6%) | |
| Weekly weight training routines | < 0.0001 | ||
| 0 day | 2,019(89.8%) | 11,485 (83.9%) | |
| 1 day | 28 (1.3%) | 366 (2.7%) | |
| 2 days | 40 (1.8%) | 478 (3.5%) | |
| 3 days | 43 (1.9%) | 466 (3.4%) | |
| 4 days | 25 (1.1%) | 238 (1.7%) | |
| ≥5 days | 93 (4.1%) | 656 (4.8%) | |
| Stress awareness | < 0.0001 | ||
| Feel a great deal of stress | 126 (5.7%) | 624 (4.6%) | |
| Feel much stress | 439 (19.7%) | 2,616 (19.2%) | |
| Feel some stress | 1,085 (48.6%) | 7,618 (55.9%) | |
| Feel almost no stress | 582 (26.1%) | 2,776 (20.4%) | |
| Feeling depression in last 1 year | 265 (11.8%) | 1,523 (11.1%) | 0.340 |
DM Diabetes mellitus, HbA1c Hemoglobin A1c, LDL Low-density lipoprotein, HDL High-density lipoprotein
Biochemical profiles, including HbA1c, fasting glucose, high-density lipoprotein (HDL), total cholesterol, and triglyceride (TG), were measured at the time of the survey. The measured low-density lipoprotein (LDL)-C was used for this study for subjects whose LDL-C was measured directly, and when it was not measured and serum TG levels were < 400 mg/dL, LDL-C was calculated using the Friedewald formula.
The characteristics of women with DM were compared to those of women without. To evaluate the association between breastfeeding and DM, the study population was divided into three groups by average breastfeeding duration: no breastfeeding group, < 12 months breastfeeding group, and ≥ 12 months breastfeeding group. The risk of DM was compared among the three groups.
Statistical analyses
The variables were presented as mean values (standard deviation) or numbers (percentages). The chi-square test or Fisher’s exact test was used to compare categorical variables, and the t-test was used to compare continuous variables. Multivariable logistic regression analysis was performed to evaluate the risk of DM according to breastfeeding duration. The results from logistic regression are presented as odds ratios (OR), adjusted odds ratio (aOR), and 95% confidence intervals (95% CI). The < 12 months breastfeeding group was used as the reference group. For multivariate logistic regression analysis, covariates included sociodemographic characteristics (age at enrollment, body mass index (BMI), household income, areas of residence, occupation, and education level), obstetric characteristics (gravidity, age at first delivery, age at last delivery, age at menarche, and age at menopause), medical diseases (hypertension, myocardial infarction, angina, stroke, hypercholesterolemia, hyperlipidemia, osteoarthritis, rheumatoid arthritis, pulmonary tuberculosis, asthma, thyroid-related disease, major depressive disorder, kidney failure, hepatitis B, hepatitis C, liver cirrhosis, gastric cancer, hepatic cancer, colorectal cancer, breast cancer, cervical cancer, and lung cancer), family history, and socioeconomic and lifestyle characteristics (history of smoking, alcohol consumption, weekly weight training routines, stress awareness, and feeling depression within a year).
To evaluate the association between the average breastfeeding duration and DM, a machine-learning method was used. An artificial neural network, decision tree, logistic regression, naïve Bayes, random forest, and support vector machine were used to predict DM. Data on 15,939 observations with full information were divided into training and validation sets in a 70:30 ratio (11,157:4,782). The area under the receiver-operating-characteristic curve and accuracy (the ratio of correct predictions among the 4,782 observations in the validation set) were employed as the standards for model validation. The random-forest variable importance, the contribution of a certain variable to random-forest performance (accuracy), was used to examine the major predictors of DM. The importance of the random-forest variable was used to investigate the main factors of HbA1c. Variable importance provides the explanatory ability of the contribution of a specific variable to the outcome variable. Therefore, variable importance indicated important variables that had a strong association with the outcome, which was DM in the current study [21]. Then, random-forest Shapley Additive Explanations (SHAP) values were derived to examine the direction of association between DM and its major predictor. The SHAP value of a particular predictor for a particular observation measures the difference between what the model (random forest) predicts the probability of DM for the observation with and without the predictor (https://github.com/shap/shap). The inclusion of a predictor (average breastfeeding duration) into machine learning (random forest) will decrease or increase the probability of the dependent variable (DM). The SHAP values are skewed for the max value hence it can be concluded that there is a positive association between the average breastfeeding duration and DM.
The SHAP approach has a specific advantage compared to linear or logistic regression: the former incorporates all realistic scenarios unlike the latter. Let us make a simplistic assumption that there are three predictors of DM, i.e., BMI at enrollment, age at enrollment and the average breastfeeding duration. As explained above, the SHAP value of the average breastfeeding duration for DM for a particular participant is the gap between what machine learning predicts for the probability of DM with and without the average breastfeeding duration for the participant. Precisely speaking, the SHAP value for the participant is the average of the following four possible situations for the participant: (1) BMI at enrollment excluded, age at enrollment excluded; (2) BMI at enrollment included, age at enrollment excluded; (3) BMI at enrollment excluded, age at enrollment included; and (4) BMI at enrollment included, age at enrollment included [22]. That is, the SHAP value incorporates the outcomes of all possible sub-group analyses, which are missing in linear or logistic regression with an unrealistic presumption of ceteris paribus, i.e., “all the other variables staying constant”. Random split and analysis were repeated 50 times and averaged for external validation. R-Studio 1.3.959 (R-Studio Inc., Boston, United States) and Python 3.52 (CreateSpace: Scotts Valley, United States) were employed for the analysis between September 1, 2022-Ocober 30, 2022.
Ethics
This study was approved by the Institutional Review Board (IRB) of the Korea University Anam Hospital (K2023-1954-001). The IRB waived the requirement for informed consent due to the anonymization of participant information. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [23].
Results
Among the 15,946 participants, 2,248 (14.1%) women had DM and 14,402 (90.3%) had a history of breastfeeding. The study population was divided into two groups: the diabetes group consisted of women with a history of DM and the control group consisted of those without. The baseline characteristics of the study groups are presented in Table 1. The DM group included fewer women with a history of breastfeeding < 12 months (DM group vs. control group; 16.1% vs. 26.9%), but more women with a history of breastfeeding ≥ 12 months (DM group vs. control group; 76.7% vs. 63.1%) and with no history of breastfeeding (DM group vs. control group; 7.2% vs. 10.1%) than the control group. Women in the DM group delivered more children than those in the control group (DM group vs. control group; 5.1 vs. 4.5). More comorbidities were observed in the women in the DM group than in the control group.
The prevalence of DM was lowest in the < 12 breastfeeding group (no breastfeeding vs. < 12 months breastfeeding vs. ≥ 12 months breastfeeding: 161 (10.4%) vs. 362 (9.0%) vs. 1,725 (16.7%), p < 0.001) (Fig. 2). HbA1c and fasting glucose level measured at enrollment were also lowest in < 12 breastfeeding group (HbA1c: No breastfeeding vs. < 12 months breastfeeding vs. ≥ 12 months breastfeeding; 5.9% vs. 5.9% vs. 6.1%, p < 0.001, respectively; fasting glucose: No breastfeeding vs. < 12 months breastfeeding vs. ≥ 12 months breastfeeding; 101.1 mg/dL vs. 99.9 mg/dL vs. 104.4 mg/dL, respectively; p < 0.001). After adjustment for covariates, the risk of DM was increased in both, the no breastfeeding group (aOR 1.29; 95% CI 1.29, 1.62) and the ≥ 12 months of breastfeeding group (aOR 1.18; 95% CI 1.01, 1.37) compared to that in the < 12 months breastfeeding group. The protective effect of breastfeeding < 12 months on the risk of DM remained significant when analyzed the association using no breastfeeding group as a reference group (aOR 0.78; 95% CI 0.62, 0.98) (Supplementary Table 2) In addition, the association between the total breastfeeding duration and DM was evaluated. The risk of DM decreased in the < 12 months breastfeeding group compared to the no breastfeeding group, but a trend was observed where the protective effects diminished in the > 12 months breastfeeding group (aOR 2.61; 95% CI 1.00, 2.96), although the protective effect did not reach statistical significance in the < 12 months breastfeeding group (aOR 0.53; 95% CI 0.60, 1.07) (Supplementary Table 3).
Fig. 2.
Risk of diabetes mellitus according to the average breastfeeding duration. Adjusted for sociodemographic characteristics (age at enrollment, BMI, household income, areas of residence, occupations, and education level), obstetric characteristics (gravidity, age at first delivery, age at the last delivery, age at menarche, and age at menopause), medical diseases (hypertension, myocardial infarction, angina, stroke, hypercholesterolemia, hyperlipidemia, osteoarthritis, rheumatoid arthritis, pulmonary tuberculosis, asthma, thyroid-related disease, major depressive disorder, kidney failure, hepatitis B, hepatitis C, liver cirrhosis, gastric cancer, hepatic cancer, colorectal cancer, breast cancer, cervical cancer, and lung cancer), family histories, and socioeconomic and lifestyle characteristics (history of smoking, alcohol consumption, weekly weight training routines, stress awareness, and feeling depression within one year)
The performance measures of the six machine-learning models for DM are presented in Table 2. The accuracy and area under the receiver-operating-characteristic curve (AUC) were used as performance measures. The performance measures of the random forest and logistic regression were the best across the board. Their respective accuracy results were 0.93 and 0.91 respectively, and their corresponding AUC results were 0.95 and 0.93. Among the 59 variables, we extracted the top 15 variables that had the greatest association with the prediction of DM using variable importance (Table 3).
Table 2.
Model performance for diabetes mellitus: accuracy and AUC
| Model | Accuracy | AUC |
|---|---|---|
| Logistic Regression | 0.91 | 0.93 |
| Decision Tree | 0.89 | 0.78 |
| Naïve Bayes | 0.82 | 0.87 |
| Random Forest | 0.93 | 0.95 |
| Support Vector Machine | 0.90 | 0.91 |
| Artificial Neural Network | 0.88 | 0.66 |
Abbreviation: AUC Area under the receiver-operating characteristic curve
Table 3.
Random forest feature importance (top 15 variables)
| (a) Random-forest feature importance of prediction model for diabetes mellitus | ||
| Rank | Variables | Variable importance |
| 1 | HbA1c | 0.26 |
| 2 | Fasting glucose | 0.18 |
| 3 | LDL | 0.05 |
| 4 | Total cholesterol | 0.05 |
| 5 | HDL | 0.03 |
| 6 | Waist circumference | 0.03 |
| 7 | DBP | 0.03 |
| 8 | Triglyceride | 0.03 |
| 9 | BMI at enrollment | 0.03 |
| 10 | SBP | 0.02 |
| 11 | Age at enrollment | 0.02 |
| 12 | Age at menopause | 0.02 |
| 13 | Age at last delivery | 0.02 |
| 14 | Age at first delivery | 0.02 |
| 15 | Total breastfeeding duration | 0.02 |
| (b) Random-forest feature importance of prediction model for HbA1c | ||
| Rank | Variables | Variable importance |
| 1 | Diabetes mellitus | 2757 |
| 2 | BMI at enrollment | 543 |
| 3 | Age at enrollment | 350 |
| 4 | Age at menopause | 307 |
| 5 | Age at last delivery | 275 |
| 6 | Frequency of alcohol consumption | 269 |
| 7 | Age at menarche | 255 |
| 8 | Age at first delivery | 236 |
| 9 | Total breastfeeding duration | 229 |
| 10 | Occupation | 217 |
| 11 | Gravidity | 202 |
| 12 | Average breastfeeding duration | 179 |
| 13 | Household income | 169 |
| 14 | Stress awareness | 164 |
| 15 | Breastfeeding duration group | 150 |
DBP Diastolic blood pressure, HbA1c Hemoglobin A1c, LDL Low-density lipoprotein, HDL High-density lipoprotein, SBP Systolic blood pressure, BMI Body mass index, HbA1c Hemoglobin A1c
The top 15 predictors of DM were HbA1c, fasting glucose, LDL, total cholesterol, HDL, waist circumference, diastolic blood pressure, TG, BMI at enrollment, systolic blood pressure, age at enrollment, age at menopause, age at last delivery, age at first delivery, and total breastfeeding duration. The average breastfeeding duration was ranked the 17th. Here, the accuracy of the prediction model decreased by 1.5% if the values of the average breastfeeding duration were randomly permutated (or shuffled). The random-forest SHAP values are shown in Fig. 3. As explained above, the SHAP value of a particular predictor for a particular observation measures the gap between what the model (random forest) predicts the probability of DM for the observation with and without the predictor. The range of SHAP values for the average breastfeeding duration was from 0 to 0.05 (Supplementary Table 1). The SHAP values are skewed for the max value and there is a positive association between the average breastfeeding duration and DM (Fig. 3). In summary, the average breastfeeding duration was a significant predictor of DM, which has a strong association with DM, considering that the variable of average breastfeeding duration ranked in the upper half of the total variables in terms of both variable importance and SHAP value.
Fig. 3.
Random-forest SHAP summary plot
Furthermore, the prediction model for HbA1c was established by random forest, and the root mean square error over the mean was 0.11. The variables of both the average breastfeeding duration’ and breastfeeding duration groups were included in the top 15 variables of the prediction model (Table 3). Therefore, the average breastfeeding duration was associated with HbA1c level, similar to the prediction model of DM.
Discussion
This large population-based study identified an optimal breastfeeding duration to prevent DM in women. The risk of DM in parous women was the lowest in women who had breastfed for less than an average of 12 months compared to that in parous women who had never breastfed or had breastfed for more than 12 months. The protective effect of breastfeeding on maternal DM increased during the first 12 months and weakened after 12 months of breastfeeding.
This study confirmed the protective effect of breastfeeding on DM. A possible mechanism for this protective effect is well established [8, 24–27]. Breastfeeding recuperates insulin resistance and glucose intolerance, which increase during pregnancy [8]. Breastfeeding has been associated with the proliferation of pancreatic β-cells in animal studies [24]. Likewise, the insulin concentration and glucose production rate were both lower in lactating rats than in non-lactating rats [25]. Moreover, milk production and breastfeeding consume additional energy and glucose, approximately 478 calories and 50 g of glucose per day [26, 27]. We found a weakened protective effect of breastfeeding for an average breastfeeding duration of more than 12 months. This finding is consistent with the previous studies showing that the risk of DM is reduced only during a short period of breastfeeding per child [9]. We suggest that prolongation of amenorrhea due to breastfeeding is a possible explanation for this finding. Previous research indicated the cessation of ovulation and menstruation during breastfeeding because of elevated level of prolactin and suppressed pulsatile [28–30]. Excessive breastfeeding duration beyond the period sufficient to restore the metabolic derangement caused by pregnancy is assumed to have a similar effect to prolonging the period of menopause, such as early menopause. Studies have demonstrated an association between early menopause and DM. Women who experience early or premature menopause have an increased risk of DM [31–33]. Likewise, premature ovarian insufficiency is also associated with an increased risk of DM [33]. Increased insulin resistance and pancreatic β-cell apoptosis and decreased insulin signaling through estrogen receptor α (ER α) caused by decreased estrogen levels contribute to an increased risk of DM [34].
This large population-based study found that the heightened beneficial effect of breastfeeding on DM in breastfeeding duration under 12 months. Likewise, machine learning analysis indicated a positive association between the average breastfeeding duration shown in and DM implies that an increased duration of breastfeeding does not decrease the risk of DM. Many studies have reported the beneficial effect of breastfeeding in preventing DM; however, several studies have focused on breastfeeding duration of less than 12 months [35–38]. Although some studies have analyzed the protective effect of breastfeeding for more than 12 months compared to no breastfeeding, the change in the breastfeeding effect on DM according to breastfeeding duration has not been analyzed [10, 39]. Moreover, through detailed interviews and questionnaires obtained by trained researchers, we adjusted for various risk factors for diabetes, including educational level, household income, smoking status, drinking and exercise habits, family history, and comorbidities. In addition, we comprehensively analyzed the association between breastfeeding duration and diabetes using machine learning, considering these various confounding factors. Machine learning is optimized for analyzing associations and establishing prediction models for diseases by considering various variables. To our knowledge, the current study is the first to analyze the breastfeeding effect on DM by machine learning. Furthermore, we analyzed the association between breastfeeding and DM using laboratory data (HbA1c), while most previous studies analyzed the breastfeeding effect on diabetes based on self-reported questionnaires [15, 19, 32, 35–38].
The strong association between breastfeeding and HbA1c found using machine-learning analysis reinforced the association between breastfeeding and DM. The limitation of this study was that the breastfeeding duration was based on questionnaires, and we did not consider gestational diabetes, which is a risk factor for diabetes in parous women. In addition, we could not assess the nutritional factors or eating habit as the confounding factors. Although KNHANES provides the database of food intake frequency surveys, we considered it challenging to accurately convert this database into variables for carbohydrate, fat, and protein calories. Furthermore, we were unable to distinguish between type I and Type II DM. However, considering the low prevalence of type I DM in Korea ranging from 0.017 to 0.021%, this limitation is unlikely to significantly impact the overall results [40].
In conclusion, breastfeeding is associated with a lower risk of DM. This protective effect persists for 12 months, then weakens after one year of breastfeeding. Breastfeeding is recommended by obstetricians and pediatricians in various societies, but the rate of breastfeeding is disappointing [41–43]. In South Korea, breastfeeding is encouraged as part of the national health plan by the Ministry of Health and Welfare, but the rate of breastfeeding has decreased over the decades [44, 45]. In the United States, the rate of breastfed infants at 1 year of age was only 35.0% in 2008, and the rate was only 33.0% in 2016–2018 [42, 44]. The results of this study validated the association between breastfeeding and the risk of DM development in women while also reinforcing the positive effects of breastfeeding on maternal health. Further research is needed to determine the optimal duration of breastfeeding in terms of lowering the risk of hypertension or other metabolic diseases.
Supplementary Information
Acknowledgements
The authors appreciate all researchers and participants of KNHANES.
Authors’ relationships and activities
The authors declare that there are no relationships or activities that might bias, or be perceived to bias, their work.
Abbreviations
- aOR
Adjusted odds ratio
- CI
Confidence interval
- DM
Diabetes mellitus
- KNHANES
Korean National Health and Nutrition Examination Surveys
- OR
Odds ratio
- SHAP
Shapley additive explanations
Authors’ contributions
E-S. C., J.S. L., and K.H. A. conceptualized and designed the study. H. L., K-S. L. performed the statistical analysis. All authors reviewed and participated the interpretation of results. E-S. C., H.L, and K-S. L. wrote the draft of the manuscript. K-S. L. and K.H. A. made contribution to the integrity of the manuscript.
Funding
This work was supported by (1) the Korea University Medical Center grant (No. K1925051), (2) the Korea Health Industry Development Institute grant funded by the Ministry of Health & Welfare of South Korea (No. HI21C156001), and (3) the Korea Health Industry Development Institute grant (Korea Health Technology R&D Project) funded by the Ministry of Health & Welfare of South Korea (No. HI22C1302). The funders had no role in the design of the study, the collection, analysis, and interpretation of the data, and the writing of the manuscript.
Availability of data and materials
The data presented in this study are not publicly available. However, the data are available from the corresponding author upon reasonable request and under the permission of Korea National Health Insurance Service.
Declarations
Ethics approval and consent to participate
This study was approved by the Institutional Review Board (IRB) of the Korea University Anam Hospital (K2023-1954-001). The IRB waived the requirement for informed consent owing to anonymization of the participant information. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Eun-Saem Choi and Jue Seong Lee contributed equally to this work.
Contributor Information
Kwang-Sig Lee, Email: ecophy@hanmail.net.
Ki Hoon Ahn, Email: akh1220@hanmail.net.
References
- 1.Quigley MA, Carson C, Sacker A, Kelly Y. Exclusive breastfeeding duration and infant infection. Eur J Clin Nutr. 2016;70:1420–7. doi: 10.1038/ejcn.2016.135. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Ware JL, Chen A, Morrow AL, Kmet J. Associations between breastfeeding initiation and infant mortality in an urban population. Breastfeed Med. 2019;14:465–74. doi: 10.1089/bfm.2019.0067. [DOI] [PubMed] [Google Scholar]
- 3.Bowatte G, Tham R, Allen KJ, Tan DJ, Lau M, Dai X, et al. Breastfeeding and childhood acute otitis media: a systematic review and meta-analysis. Acta Paediatr. 2015;104:85–95. doi: 10.1111/apa.13151. [DOI] [PubMed] [Google Scholar]
- 4.Li R, Ware J, Chen A, Nelson JM, Kmet JM, Parks SE, et al. Breastfeeding and post-perinatal infant deaths in the United States, a national prospective cohort analysis. Lancet Reg Health Am. 2022;5:100094. doi: 10.1016/j.lana.2021.100094. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Xue M, Dehaas E, Chaudhary N, O’Byrne P, Satia I, Kurmi OP. Breastfeeding and risk of childhood asthma: a systematic review and meta-analysis. ERJ Open Res. 2021;7:00504–2021. doi: 10.1183/23120541.00504-2021. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Lund-Blix NA, Dydensborg Sander S, Stordal K, Nybo Andersen AM, Ronningen KS, Joner G, et al. Infant feeding and risk of type 1 diabetes in two large scandinavian birth cohorts. Diabetes Care. 2017;40:920–7. doi: 10.2337/dc17-0016. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Amitay EL, Keinan-Boker L. Breastfeeding and childhood leukemia incidence: a meta-analysis and systematic review. JAMA Pediatr. 2015;169:e151025. doi: 10.1001/jamapediatrics.2015.1025. [DOI] [PubMed] [Google Scholar]
- 8.Stuebe AM, Rich-Edwards JW. The reset hypothesis: Lactation and maternal metabolism. Am J Perinatol. 2009;26:81–8. doi: 10.1055/s-0028-1103034. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Aune D, Norat T, Romundstad P, Vatten LJ. Breastfeeding and the maternal risk of type 2 diabetes: a systematic review and dose-response meta-analysis of cohort studies. Nutr Metab Cardiovasc Dis. 2014;24:107–15. doi: 10.1016/j.numecd.2013.10.028. [DOI] [PubMed] [Google Scholar]
- 10.Rameez RM, Sadana D, Kaur S, Ahmed T, Patel J, Khan MS, et al. Association of maternal lactation with diabetes and hypertension: a systematic review and meta-analysis. JAMA Netw Open. 2019;2:e1913401. doi: 10.1001/jamanetworkopen.2019.13401. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Park S, Choi NK. Breastfeeding and maternal hypertension. Am J Hypertens. 2018;31:615–21. doi: 10.1093/ajh/hpx219. [DOI] [PubMed] [Google Scholar]
- 12.Suliga E, Ciesla E, Gluszek-Osuch M, Lysek-Gladysinska M, Wawrzycka I, Gluszek S. Breastfeeding and prevalence of metabolic syndrome among perimenopausal women. Nutrients. 2020;12:2691. doi: 10.3390/nu12092691. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Matsunaga T, Kadomatsu Y, Tsukamoto M, Kubo Y, Okada R, Nagayoshi M, et al. Associations of breastfeeding history with metabolic syndrome and cardiovascular risk factors in community-dwelling parous women: the Japan multi-institutional collaborative cohort study. PLoS ONE. 2022;17:e0262252. doi: 10.1371/journal.pone.0262252. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Gunderson EP, Lewis CE, Wei GS, Whitmer RA, Quesenberry CP, Sidney S. Lactation and changes in maternal metabolic risk factors. Obstet Gynecol. 2007;109:729–38. doi: 10.1097/01.AOG.0000252831.06695.03. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Villegas R, Gao YT, Yang G, Li HL, Elasy T, Zheng W, et al. Duration of breast-feeding and the incidence of type 2 diabetes mellitus in the Shanghai women’s health study. Diabetologia. 2008;51:258–66. doi: 10.1007/s00125-007-0885-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Capula C, Chiefari E, Vero A, Foti DP, Brunetti A, Vero R. Prevalence and predictors of postpartum glucose intolerance in Italian women with gestational diabetes mellitus. Diabetes Res Clin Pract. 2014;105:223–30. doi: 10.1016/j.diabres.2014.05.008. [DOI] [PubMed] [Google Scholar]
- 17.Nelson AL, Le MH, Musherraf Z, Vanberckelaer A. Intermediate-term glucose tolerance in women with a history of gestational diabetes: natural history and potential associations with breastfeeding and contraception. Am J Obstet Gynecol. 2008;198:699. doi: 10.1016/j.ajog.2008.03.029. [DOI] [PubMed] [Google Scholar]
- 18.Kim SH, Kim MY, Yang JH, Park SY, Yim CH, Han KO, et al. Nutritional risk factors of early development of postpartum prediabetes and diabetes in women with gestational diabetes mellitus. Nutrition. 2011;27:782–8. doi: 10.1016/j.nut.2010.08.019. [DOI] [PubMed] [Google Scholar]
- 19.Martens PJ, Shafer LA, Dean HJ, Sellers EAC, Yamamoto J, Ludwig S, et al. Breastfeeding initiation associated with reduced incidence of diabetes in mothers and offspring. Obstet Gynecol. 2016;128:1095–104. doi: 10.1097/AOG.0000000000001689. [DOI] [PubMed] [Google Scholar]
- 20.Kweon S, Kim Y, Jang MJ, Kim Y, Kim K, Choi S, et al. Data resource profile: the Korea national health and nutrition examination survey (knhanes) Int J Epidemiol. 2014;43:69–77. doi: 10.1093/ije/dyt228. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Ishwaran H. Variable importance in binary regression trees and forests. Electron J Stat. 2007;1:519–37. doi: 10.1214/07-EJS039. [DOI] [Google Scholar]
- 22.Scott Lundberg S-IL. A unified approach to interpreting model predictions. arXiv; 2017. [Google Scholar]
- 23.von Elm E, Altman DG, Egger M, Pocock SJ, Gotzsche PC, Vandenbroucke JP, et al. Strengthening the reporting of observational studies in epidemiology (strobe) statement: guidelines for reporting observational studies. BMJ. 2007;335:806–8. doi: 10.1136/bmj.39335.541782.AD. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Drynda R, Peters CJ, Jones PM, Bowe JE. The role of non-placental signals in the adaptation of islets to pregnancy. Horm Metab Res. 2015;47:64–71. doi: 10.1055/s-0034-1395691. [DOI] [PubMed] [Google Scholar]
- 25.Burnol AF, Leturque A, Ferre P, Kande J, Girard J. Increased insulin sensitivity and responsiveness during lactation in rats. Am J Physiol. 1986;251:E537–41. doi: 10.1152/ajpendo.1986.251.5.E537. [DOI] [PubMed] [Google Scholar]
- 26.Butte NF, Wong WW, Hopkinson JM. Energy requirements of lactating women derived from doubly labeled water and milk energy output. J Nutr. 2001;131:53–8. doi: 10.1093/jn/131.1.53. [DOI] [PubMed] [Google Scholar]
- 27.Butte NF, Hopkinson JM, Mehta N, Moon JK, Smith EO. Adjustments in energy expenditure and substrate utilization during late pregnancy and lactation. Am J Clin Nutr. 1999;69:299–307. doi: 10.1093/ajcn/69.2.299. [DOI] [PubMed] [Google Scholar]
- 28.Vekemans M. Postpartum contraception: the lactational amenorrhea method. Eur J Contracept Reprod Health Care. 1997;2:105–11. doi: 10.3109/13625189709167463. [DOI] [PubMed] [Google Scholar]
- 29.Chao S. The effect of lactation on ovulation and fertility. Clin Perinatol. 1987;14:39–50. doi: 10.1016/S0095-5108(18)30780-2. [DOI] [PubMed] [Google Scholar]
- 30.Gray RH, Campbell OM, Apelo R, Eslami SS, Zacur H, Ramos RM, et al. Risk of ovulation during lactation. Lancet. 1990;335:25–9. doi: 10.1016/0140-6736(90)90147-W. [DOI] [PubMed] [Google Scholar]
- 31.Muka T, Asllanaj E, Avazverdi N, Jaspers L, Stringa N, Milic J, et al. Age at natural menopause and risk of type 2 diabetes: a prospective cohort study. Diabetologia. 2017;60:1951–60. doi: 10.1007/s00125-017-4346-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Shen L, Song L, Li H, Liu B, Zheng X, Zhang L, et al. Association between earlier age at natural menopause and risk of diabetes in middle-aged and older Chinese women: the dongfeng-tongji cohort study. Diabetes Metab. 2017;43:345–50. doi: 10.1016/j.diabet.2016.12.011. [DOI] [PubMed] [Google Scholar]
- 33.Anagnostis P, Christou K, Artzouchaltzi AM, Gkekas NK, Kosmidou N, Siolos P, et al. Early menopause and premature ovarian insufficiency are associated with increased risk of type 2 diabetes: a systematic review and meta-analysis. Eur J Endocrinol. 2019;180:41–50. doi: 10.1530/EJE-18-0602. [DOI] [PubMed] [Google Scholar]
- 34.Lambrinoudaki I, Paschou SA, Armeni E, Goulis DG. The interplay between diabetes mellitus and menopause: clinical implications. Nat Rev Endocrinol. 2022;18:608–22. doi: 10.1038/s41574-022-00708-0. [DOI] [PubMed] [Google Scholar]
- 35.Tanase-Nakao K, Arata N, Kawasaki M, Yasuhi I, Sone H, Mori R, et al. Potential protective effect of lactation against incidence of type 2 diabetes mellitus in women with previous gestational diabetes mellitus: a systematic review and meta-analysis. Diabetes Metab Res Rev. 2017;33:e2875. doi: 10.1002/dmrr.2875. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Ziegler AG, Wallner M, Kaiser I, Rossbauer M, Harsunen MH, Lachmann L, et al. Long-term protective effect of lactation on the development of type 2 diabetes in women with recent gestational diabetes mellitus. Diabetes. 2012;61:3167–71. doi: 10.2337/db12-0393. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Schwarz EB, Brown JS, Creasman JM, Stuebe A, McClure CK, Van Den Eeden SK, et al. Lactation and maternal risk of type 2 diabetes: a population-based study. Am J Med. 2010;123:e8631–6. doi: 10.1016/j.amjmed.2010.03.016. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Liu B, Jorm L, Banks E. Parity, breastfeeding, and the subsequent risk of maternal type 2 diabetes. Diabetes Care. 2010;33:1239–41. doi: 10.2337/dc10-0347. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Schwarz EB, Ray RM, Stuebe AM, Allison MA, Ness RB, Freiberg MS, et al. Duration of lactation and risk factors for maternal cardiovascular disease. Obstet Gynecol. 2009;113:974–82. doi: 10.1097/01.AOG.0000346884.67796.ca. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Lee JW, Kang HT, Lim HJ, Park B. Trends in Diabetes prevalence among Korean adults based on Korean national health and nutrition examination surveys iii-vi. Diabetes Res Clin Pract. 2018;138:57–65. doi: 10.1016/j.diabres.2018.01.013. [DOI] [PubMed] [Google Scholar]
- 41.World Health Organization, UNICEF . Global strategy for infant and young child feeding. Geneva: World Health Organization; 2003. [Google Scholar]
- 42.Meek JY, Noble L. Section on Breastfeeding. Policy statement: breastfeeding and the use of human milk. Pediatrics. 2022;150:e2022057988. doi: 10.1542/peds.2022-057988. [DOI] [PubMed] [Google Scholar]
- 43.Committee on Obstetric Practice, Breastfeeding Expert Work Group Breastfeeding challenges: Acog Committee opinion, number 820. Obstet Gynecol. 2021;137:e42–53. doi: 10.1097/AOG.0000000000004253. [DOI] [PubMed] [Google Scholar]
- 44.Ministry of Health and Welfare, Korea Health Promotion Institute . The 5th national health plan (2021–2030) Seoul: Korea Health Promotion Institute; 2022. [Google Scholar]
- 45.Ministry of Health and Welfare, Korea Health Promotion Institute . The 4th health plan 2020 (2016 ~ 2020) Seoul: Korea Health Promotion Institute; 2015. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data presented in this study are not publicly available. However, the data are available from the corresponding author upon reasonable request and under the permission of Korea National Health Insurance Service.