Study design and the participants

The NHANES, conducted by the National Center for Health Statistics (NCHS) of the United States, is a cross-sectional survey aimed at gathering health and nutrition information regarding American adults and children. All NHANES data are publicly accessible and can be downloaded freely from NHANES website. Participants in NHANES provide written informed consent, and the entire process is approved by the Institutional Review Board of the Centers for Disease Control and Prevention. Data on gestational diabetes mellitus (GDM) history in NHANES were first collected from 2007 to 2008, with subsequent research focusing primarily on the period from 2007 to 2010. To explore this issue further, we incorporated cross-sectional data from two consecutive NHANES cycles (2011–2014), involving 10,072 female participants. We compared Dietary Inflammatory Index (DII) scores between reproductive-age females with and without a history of GDM. Additionally, we conducted a comprehensive investigation to study the association between DII and prediabetes in female participants with a history of GDM, and developed a nomogram model specifically to predict prediabetes for this subgroup. The Institutional Review Board of Fujian Maternity and Child Health Hospital determined that this study does not involve oversight of human subject research.

Exclusion criteria were applied as follows: (1) female participants aged outside the range of 20 to 44 years (n = 7,555); (2) female participants who had never given birth to a live infant (n = 1,197); (3) female participants reporting a doctor-diagnosed diabetes (n = 58) or borderline diabetes managed with oral hypoglycemic agents (n = 3), and those reporting previous critical GDM (n = 20), totaling 81 participants excluded; (4) female participants with missing DII data (n = 197) or GDM information (n = 11), resulting in 207 exclusions; and (5) female participants with missing covariate and other variable data (n = 61). After manual data filtering, a final cohort of 971 female participants without diabetes mellitus was selected for subsequent analysis. The detailed recruitment process of study participants is illustrated in Fig. 1.

Assessment of the dietary inflammation index (DII)

In this study, we utilized data from 28 food parameters sourced from 11 countries worldwide, including alcohol, caffeine, carbohydrates, fats, proteins, and vitamins A, B1, B2, B6, B12, C, D, E, folate, β-carotene, cholesterol, energy (kcal), fiber, iron, magnesium, selenium, zinc, saturated fatty acids, monounsaturated fatty acids, polyunsaturated fatty acids, N3 and N6 fatty acids, and niacin, to calculate the Dietary Inflammatory Index (DII). All data underwent standardized processing.

From numerous inflammation markers, we selected six representative markers to evaluate the impact of food parameters: pro-inflammatory markers (IL-1β, IL-6, TNF-α, C-reactive protein (CRP)) and anti-inflammatory markers (IL-4 and IL-10). Each food parameter was scored based on its weighted association with these six inflammation markers and its role in either promoting or inhibiting inflammation. A food parameter received a score of + 1 if it significantly increased IL-1β, IL-6, TNF-α, or CRP levels, or decreased IL-4 and IL-10 levels; conversely, it received a score of -1 otherwise. Z-scores were calculated based on participants’ exposure levels to each food parameter. Standardized estimates of dietary intake were converted to centiles for each DII component [18]. These centiles were then multiplied by the respective component-specific inflammation effect scores to derive the total DII score for each individual. In this study, DII was treated as a continuous variable, and participants were categorized into three groups: Low, Medium, and High.

History of gestational diabetes mellitus

Female participants reported their history of gestational diabetes mellitus (GDM) in the reproductive health questionnaire. They were asked, “During your pregnancy, did a doctor or other health professional ever tell you that you had diabetes, sugar diabetes, or gestational diabetes?” Female participants who answered “yes” to this question were categorized as having a history of GDM.

Prediabetes

Female participants who reported being informed by a doctor or other health professional that they had any of the following conditions—prediabetes, impaired fasting glucose, impaired glucose tolerance, borderline diabetes, or that their blood sugar was higher than normal but not high enough to be called diabetes or sugar diabetes—were classified as having prediabetes.

Fig. 1

Flowchart of the study participants

Covariates

In this study, we identified the following variables as potential confounding factors: age, race, BMI, birth history of macrosomia, age of menarche, marital status, education levels, smoking, drinking, hypertension, minutes of sedentary activity, thyroid problems, and mean platelet volume. Age, race/ethnicity, and education levels were gathered from demographic questionnaires. Age of menarche, birth history of macrosomia, prediabetes, health risk for diabetes, alcohol consumption, and smoking status were obtained from questionnaire data. The time point for age and BMI data in this study was independent of pregnancy status and based on cross-sectional attributes collected from the NHANES database. Race/ethnicity was categorized as non-Hispanic white, non-Hispanic black, Mexican American, other Hispanic, and others. Education level was classified into three categories: less than high school, high school, and more than high school. Additionally, information on arthritis, gout, heart failure, coronary heart disease, angina pectoris, heart attack, stroke, emphysema, thyroid problems, chronic bronchitis, and liver disease was obtained from responses to the medical conditions questionnaire provided by female participants. This section provided self- and proxy-reported personal interview data covering a wide range of health conditions and medical histories for all participants. Furthermore, participants’ complete blood count (CBC) and red blood cell (RBC) folate information were obtained from Laboratory Data. The CBC parameters were derived using Beckman Coulter methodology, which includes counting and sizing with an automatic diluting and mixing device for sample processing, and hemoglobinometry using a single-beam photometer. The Beckman Coulter DxH 800 instrument at the NHANES mobile examination center (MEC) was used to analyze blood specimens and provide a distribution of blood cells for all participants. CBC information collected included neutrophil count, platelet count, basophil count, lymphocyte count, eosinophil count, monocyte count, and white blood cell (WBC) count. RBC folate measurements encompassed Whole Blood Folate and Serum Total Folate. Whole-blood folate was assessed using a microbiologic assay, while serum folate forms were measured using isotope-dilution high-performance liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS).

Statistical analysis

Due to the complex sampling design utilized in NHANES surveys, we incorporated sample weights corresponding to different study periods in our analytical methods to generate accurate estimates of health-related statistical data. We compared baseline characteristics, including CBC components and DII scores among female participants across different groups based on independent t tests, chi-square test, and Mann-Whitney U test.

For analyzing the association between DII and prediabetes in female participants with a history of GDM, we used a multivariable logistic regression model. Specifically, we employed three models: an unadjusted model and two adjusted models (Model 1 and Model 2). Model 1 adjusted for age, race, and BMI, while Model 2 further adjusted for additional factors such as birth history of macrosomia, age of menarche, marital status, education level, smoking, drinking, hypertension, minutes of sedentary activity, thyroid problems, and mean platelet volume.

To explore nonlinear relationships, DII was categorized into three groups—Low, Medium, and High—using equal intervals: <0.89 for Low, [0.89–2.38) for Medium, and ≥ 2.38 for High. Additionally, we employed restricted cubic spline (RCS) regression with 3 knots (at the 10th, 50th, and 90th percentiles) to further investigate the nonlinear relationship between DII and prediabetes in female participants with a history of GDM.

Subgroup analyses were conducted based on various demographic and health-related factors to assess significant interactions with the association between DII and prediabetes in this population.

Given the significantly increased risk of type 2 diabetes in females with a history of GDM, our objective was to develop a predictive nomogram model for predicting prediabetes in these participants. To identify the most critical dietary factors associated with prediabetes and mitigate collinearity among variables, we employed the Least Absolute Shrinkage and Selection Operator (LASSO) regression model. In the LASSO model, we utilized cross-validation to evaluate model performance and select optimal parameters. This involved dividing the dataset into 10 subsets for training and testing the model multiple times, thereby assessing its performance across different settings. The lambda value, which minimizes bias during cross-validation, was selected with consideration of model stability and generalizability to new data. Additionally, a prediction nomogram model was developed based on key variables, and its discriminatory ability in predicting prediabetes in female participants with a history of GDM was validated using receiver operating characteristic (ROC) curves.

All statistical analyses were performed using R software version 4.1.6 (http://www.R-project.org, The R Foundation, Vienna, Austria), and statistical significance was determined by a two-tailed P-value < 0.05.

Baseline characteristics of study participants

This study included a total of 971 female participants from NHANES (2011–2014), with a weighted average age of 33.96 years. Among all participants, the weighted median DII score was 1.96 (0.88, 2.93), and 12.98% reported a history of GDM. Compared to females without a history of GDM, those with GDM tended to be older, from other racial backgrounds, and had higher rates of prediabetes. They also had an earlier age at menarche, a history of macrosomia at birth, and increased health risks related to diabetes (all P < 0.05). Detailed baseline characteristics of all female participants categorized by history of GDM are provided in Table 1.

Specifically, females with a history of GDM were on average 2 years older (36.00 (31.00, 41.00) vs. 34.00 (29.00, 40.00)) and experienced menarche earlier (12.00 (11.00, 13.00) vs. 12.00 (12.00, 14.00)) compared to those without such history. Furthermore, females with a history of GDM were more likely to have had macrosomic births (19.8% vs. 12.5%), report prediabetes (12.7% vs. 3.3%), and indicate health risks associated with diabetes (40.5% vs. 16.4%) (Table 1).

It is noteworthy that female participants with a history of GDM had a significantly lower DII score compared to those without (1.62 (0.58, 2.93) vs. 2.05 (0.91, 2.93)) (Table 1). Further comparisons of individual components of the DII between the two groups revealed that females with a history of GDM had lower inflammatory scores in Fiber (0.27 (-0.18, 0.59) vs. 0.38 (-0.00, 0.59)), Magnesium (0.11 (-0.09, 0.27) vs. 0.16 (-0.00, 0.29)), N3 fatty acids (-0.15 (-0.23, -0.03) vs. -0.11 (-0.23, 0.01)), and Selenium (-0.12 (-0.17, -0.06) vs. -0.11 (-0.17, -0.02)) (all P < 0.05, Table S1). Finally, we compared all components of the complete blood cell count between female participants with and without a history of GDM, finding that those with a history of GDM had a lower Hematocrit (38.35 (36.40, 40.20) vs. 38.60 (36.50, 40.70)) (P < 0.05, Table S2).

Table 1 Baseline characteristics of participants grouped by with or without a history of GDMCharacteristics of female participants with a history of GDM grouped by Prediabetes Status

Based on the findings presented in Table 1, female participants with a history of GDM exhibited a higher prevalence of prediabetes. To further explore this association, we analyzed the characteristics of these females stratified by prediabetes status. Out of 126 female participants with a history of GDM, 16 were diagnosed with prediabetes. Compared to those without prediabetes, females with prediabetes were more likely to have hypertension (37.5% vs. 13.6%) and thyroid disorders (25.0% vs. 5.5%). Additionally, they reported shorter sedentary activity times (240.00 (150.00, 360.00) vs. 390.00 (240.00, 540.00) minutes) (all P < 0.05, Table S3).

It is noteworthy that females with prediabetes had higher Dietary Inflammatory Index (DII) scores compared to those without prediabetes (2.75 (1.27, 3.42) vs. 1.59 (0.44, 2.81)). Specifically, females with prediabetes exhibited higher scores in Vitamin B6 (0.03 ± 0.15 vs. -0.09 ± 0.16) and Vitamin E (0.38 (0.05, 0.42) vs. 0.23 (-0.05, 0.38)) (all P < 0.05, Table 2).

Furthermore, we conducted a comparative analysis of complete blood count (CBC) components among female participants with a history of GDM stratified by prediabetes status. The results revealed that females with prediabetes had higher mean platelet volume (8.40 (8.00, 9.00) vs. 8.05 (7.15, 8.55)) (P < 0.05, Table 3).

Table 2 Comparison of dietary intake of each DII component between female participants with a history of GDM grouped by prediabetes statusTable 3 Comparison of each component of complete blood cell count (CBC) between female participants with a history of GDM grouped by prediabetes statusAssociation between DII and prediabetes in female participants with a history of GDM

We conducted a multivariable logistic regression analysis to explore the association between DII and prediabetes in female participants with a history of GDM. We observed that higher DII scores were correlated with prediabetes in this cohort. When analyzed as a continuous variable, DII showed a positive correlation with prediabetes. In the unadjusted logistic regression model, the odds ratio (OR) was 1.68 (95% confidence interval [CI]: 1.09–2.59). After adjusting for confounding factors, including age, race, BMI, history of macrosomia, age at menarche, marital status, education level, smoking, drinking, hypertension, sedentary activity time, thyroid issues, and mean platelet volume, the fully adjusted Model 2 indicated that DII remained significantly associated with prediabetes (OR: 1.97; 95% CI: 1.03–3.77) (Table 4). When considered as categorical variables, individuals with Medium DII (OR: 28.67; 95% CI: 1.34-614.39) and High DII (OR: 36.40; 95% CI: 1.71-772.83) showed a significantly higher risk of prediabetes compared to those with Low DII (Table 4).

Furthermore, RCS analysis was employed to explore the nonlinear relationship between DII and prediabetes among female participants with a history of GDM. The RCS curve indicated no significant nonlinear negative association between DII and GDM history (p for nonlinear = 0.617) (Fig. 2). We also conducted stratified analyses to assess whether the association between DII and prediabetes remained consistent across different subgroups among female participants with a history of GDM. However, no statistically significant differences were observed across subgroups stratified by macrosomia history (Yes, No), age (Below 35, Above 35), race (Mexican American, Other Hispanic, Non-Hispanic White, Non-Hispanic Black, Other Race), education level (Less than high school, High school, More than high school), marital status (Married, Single, Living with partner), smoking (No, Yes), drinking (No, Yes), and hypertension (No, Yes) (all p for interaction > 0.05) (Fig. 3).

Table 4 Multivariable weighted logistic regression model revealed the relationship between DII and prediabetes in female participants with a history of GDMFig. 2

The RCS analysis on the association between DII and prediabetes in female participants with a history of GDM. RCS, restricted cubic spline; DII, Dietary Inflammatory Index; GDM, gestational diabetes mellitus; OR, odds ratio; CI, confidence interval

Fig. 3

Subgroup analysis of the association between DII and prediabetes in female participants with a history of GDM. Each stratification was adjusted for age, race, BMI, birth history of macrosomia, age of menarche, marital status, education levels, smoking, drinking, hypertension, minutes sedentary activity, and mean platelet volume. OR, odds ratio; ; CI, confidence interval; BMI, body mass index; DII, Dietary Inflammation Index; Inf: Infinity

Identification of Key Prediabetes-related dietary factors in female participants with a history of GDM

We aimed to develop a predictive nomogram model using 10-fold cross-validation of LASSO regression, incorporating all 28 dietary components and four covariates (hypertension, minutes of sedentary activity, thyroid issues, and mean platelet volume) to pinpoint dietary factors most closely associated with prediabetes in female participants with a history of gestational diabetes mellitus (GDM) (Fig. 4). Through LASSO regression analysis, we identified hypertension, sedentary activity duration, thyroid issues, mean platelet volume, vitamin B6, β-Carotene, polyunsaturated fatty acids, saturated fat, and vitamin C as the nine predictive variables in the nomogram model. Subsequently, we validated the robust predictive performance of this model using ROC curve analysis, yielding an area under the curve (AUC) of 88.6% (79.9–97.4%) (Fig. 5).

Fig. 4

The LASSO penalized regression analysis for identifying key related factors. (A) The coefficient shrinkage process of all 28 dietary components and four covariates (hypertension, minutes sedentary activity, thyroid problem, and mean platelet volume), we represent the changes in coefficients of different features under various levels of shrinkage by drawing lines of different colors. (B) A 10-fold cross-validation of the LASSO regression model. LASSO, least absolute shrinkage and selection operator

Fig. 5

Establishment and validation of a risk prediction model for prediabetes in female participants with a history of GDM. (A) A nomogram model based on hypertension, minutes sedentary activity, thyroid problem, and mean platelet volume, and 5 key related dietary factors (vitamin B6, β-Carotene, polyunsaturated fatty acids, saturated fat, and vitamin C) identified by LASSO regression analysis. (B) ROC curve for evaluating the predictive power for predicting prediabetes in female participants with a history of GDM of the nomogram model. LASSO, least absolute shrinkage and selection operator; ROC, receiver operating characteristic. * P value < 0.05