The NHANES is an ongoing survey that measures the health and nutrition of adults and children in the United States. It employs sophisticated multistage probability cluster designs for data collection and study methodology, ensuring the gathering of precise and comprehensive data. The NHANES project was approved by the Research Ethics Committee of the National Center for Health Statistics (NCHS), with informed consent obtained from all participants. Additional details are available on the NCHS website. The secondary analysis conducted for this study did not require further Institutional Review Board approval.

This study was a retrospective analysis using data from NHANES (1999–2018), which included 55,081 participants aged 20 years and older. Exclusion criteria for individuals included pregnant (n = 1,547), absence of DI-GM components (n = 6,200), missing COPD survey data (n = 6), or missing covariates data (n = 7,310), such as marital status, poverty income ratio (PIR), body mass index (BMI), educational level, drinking and smoking status, cardiovascular disease (CVD), hypertension, diabetes, and hyperlipidemia (Figure 1).

COPD was defined based on responses to the clinical questions “Has a doctor ever said you had COPD”, “Have you ever been told you had chronic bronchitis” or “Have you ever been told you had emphysema” from the medical conditions questionnaire, as validated in a previous study (10).

According to the scoring criteria established by Kase et al., the DI-GM was composed of 14 specific food items or nutrients (Supplementary Table 1). These included beneficial items such as avocado, broccoli, soybean, chickpeas, cranberries, fermented dairy, coffee, green tea (data unavailable in NHANES due to the lack of specific tea information), whole grains, and fiber. In contrast, refined grains, red meat, processed meat, and diets with ≥40% of energy from fat were categorized as adverse components (9). The DI-GM was calculated using dietary recall data from the NHANES dataset (1999–2018). Individuals whose consumption was above the sex-specific median for beneficial components or below the median for unfavorable components were assigned a score of 1. In contrast, a score of 0 was given to individuals whose consumption was below the sex-specific median for beneficial components or above the median for unfavorable components. The individual component scores were then summed to derive the overall DI-GM score, which ranged from 0 to 13 (with scores from 0 to 9 for beneficial to gut microbiota and 0 to 4 for unfavorable to gut microbiota). A higher DI-GM score indicates a more favorable gut microbiota. In this study, the DI-GM was categorized into groups according to 0–3, 4, 5, and ≥6 (11).

All data were gathered and documented by investigators who received uniform training. Informed by existing research and clinical expertise, several potential confounding factors were considered, such as age, sex, race/ethnicity, education, PIR, marital status, smoking status, drinking status, BMI, hypertension, diabetes, CVD, and hyperlipidemia (11, 12).

Sex was categorized as female or male. Race/ethnicity was grouped as non-Hispanic White, non-Hispanic Black, other Hispanic, Mexican American, or other. Marital status was categorized into two groups: married or living with a partner, and living alone. Education levels were classified into three groups: above high school, high school or equivalent, and less than high school. PIR was grouped into three levels based on the values 1.30 and 3.50. Smoking status was classified into three categories: never smoked, current smoker, or former smoker. Drinking status was categorized into three groups: never, former, or current. BMI was calculated using the standard method based on weight and height.

CVD diagnosis was based on self-reported physician diagnoses obtained through a standardized questionnaire on medical conditions. The participants were asked, “Has a doctor or other health expert ever informed you that you have congestive heart failure/coronary heart disease/angina pectoris/myocardial Infarction/stroke?” individuals who responded with “yes” to any of these questions were considered to have CVD (13). Hypertension was diagnosed if the mean systolic blood pressure was ≥140 mmHg, or the mean diastolic blood pressure was ≥90 mmHg, or if the individual had ever used antihypertensive drugs or had been previously diagnosed with hypertension by a doctor or healthcare professional (14). Diabetes was diagnosed if any of the following conditions were met: a physician diagnosed the individual with diabetes; fasting glucose was ≥7.0 mmol/l; random blood glucose was ≥11.1 mmol/l; glycohemoglobin was ≥6.5%; 2-h oral glucose tolerance test blood glucose was ≥11.1 mmol/l; or the individual was using diabetes medication or insulin (15). Hyperlipidemia was defined as having any of the following: triglycerides ≥ 150 mg/dl, total cholesterol ≥ 200 mg/dl, low-density lipoprotein ≥ 130 mg/dl, or high-density lipoprotein ≤ 50 mg/dl in females and ≤ 40 mg/dl in males. Furthermore, individuals who indicated the use of lipid-lowering medications were classified as having hyperlipidemia (14).

The characteristics of all individuals were summarized based on the presence or absence of COPD. Categorical variables were described by proportions (%) whereas continuous variables were presented as mean with standard deviation (SD) or medians with interquartile range (IQR), depending on the data distribution. Depending on the normality of the distribution, either the independent samples Student's t-test or Mann-Whitney U-test was used to compare the continuous data among groups. Chi-square or Fisher's exact test was used to compare categorical data, as appropriate.

Following the NHANES analysis guidelines, this analysis accounted for complex sampling designs and sampling weights (16). The analysis included the following variables: masked variance pseudo-cluster (SDMVPSU), masked variance pseudo-stratum (SDMVSTRA), and dietary weights. Specifically, dietary day one 4-year sample weight (WTDR4YR) was utilized, as the data from NHANES 1999–2000 and 2001–2002 were included. For the NHANES 2003–2018 data, dietary day one 2-year sample weight (WTDRD1) was applied. The sampling weights for the 1999–2018 period were calculated as follows: for the 1999–2002 data, the weights were 1/5 × WTDR4YR, while for other years, the weights were 1/10 × WTDRD1.

Multivariable weighted logistic regression models were employed to estimate the odds ratios (OR) and 95 percent confidence intervals (95% CIs) for the association between DI-GM and COPD. Four progressively adjusted models were used. Model 1 is severed as an unadjusted or crude model. Model 2 was adjusted for age, sex, and race/ethnicity. Model 3 was further adjusted for marital status, PIR, educational level, smoking status, drinking status, and BMI in addition to the variables in model 2. Model 4 included all variables from model 3 and additionally accounted for comorbidities, including CVD, hyperlipidemia, hypertension, and diabetes.

Furthermore, interaction and subgroup analyses were performed to evaluate the consistency of the relationship between DI-GM and COPD across various populations. These analyses were stratified by sex (male vs. female), age (20–60 vs. ≥60 year), BMI (<25, 25–30, or ≥30 kg/m²), CVD (yes vs. no), diabetes (yes vs. no), hyperlipidemia (yes vs. no), hypertension (yes vs. no), constipation status (yes vs. no), PIR groups (≤1.3, 1.3–3.5 or >3.5), and education levels (“above high school”, “high school or equivalent”, or “less than high school”).

Multiple sensitivity analyses were performed to examine the robustness of the results: (1) To address missing data, multiple imputations by chained equations were applied. Five imputed datasets were generated using the variables included in the final statistical model (17); (2) Multivariable unweighted logistic regression models were employed; (3) Based on model 4, we separately adjusted energy, protein, saturated fat, and dietary fiber intake to assess the association between the DI-GM and COPD; (4) we compared DI-GM and established indices such as alternative Mediterranean Diet Score (aMED) (18), Healthy Eating Index-2015 (HEI-2015) (19), and the Dietary Inflammatory Index (DII) (20) to assess their association with COPD. Receiver operating characteristic (ROC) curves were generated to assess the predictive performance for COPD.

All analyses were performed using R Statistical Software (Version 4.2.2, http://www.R-project.org, The R Foundation) and Free Statistics analysis platform (Version 2.0, Beijing, China). A P-value of < 0.05 in a two-sided test was regarded as statistically significant.

Table 1 presents the baseline characteristics of a sample representing 185.88 million U.S. adults, with an average age of 47.33 (16.89) years. Of these, 14.85 million were diagnosed with COPD. Individuals with COPD tended to be older, female, non-Hispanic White, and more likely to be smokers and drinkers. They also had higher BMI, and a greater incidence of CVD, hypertension, diabetes, and hyperlipidemia, as well as lower DI-GM (Table 1).

In multivariable logistic regression analyses, DI-GM expressed as a continuous variable was negatively correlated with an increased prevalence of COPD (OR, 0.95; 95% CI, 0.92–0.99; P = 0.006; Table 2, model 1). After adjusting for the potential confounders, each one-point increase in DI-GM was linked to a 4% reduction in the risk of COPD (OR, 0.96; 95% CI, 0.92–0.99; P = 0.016; Table 2, model 4). The association remained consistent when the DI-GM was converted into a categorical variable. In the fully adjusted model, individuals with DI-GM ≥ 6 showed a significant negative association with the risk of COPD (OR = 0.84; 95% CI, 0.73–0.98; P = 0.023; Table 2, model 4). Additionally, as the beneficial to gut microbiota increased, the prevalence of COPD significantly decreased (OR = 0.95; 95% CI, 0.91–1.00; P = 0.030; Table 2, model 4), whereas no significant association was observed between the unfavorable to gut microbiota and COPD.

Stratified analysis was conducted in several subgroups to evaluate potential effect modifications on the association between DI-GM and COPD (Figure 2 and Supplementary Table 2). No significant interactions were observed in any subgroups after stratification by sex, age (20–60 vs. ≥60 year), BMI (< 25 vs. 25–30 or ≥30 kg/m²), CVD, diabetes, hyperlipidemia, and hypertension (all P for interaction > 0.05) (Figure 2).

In the sensitivity analysis, unweighted logistic regression analysis was performed. The DI-GM was found to be significantly associated with COPD after adjusting for all confounders (OR = 0.96; 95% CI, 0.94–0.99; P = 0.004; Table 2). Additionally, multiple interpolation was used to address missing data. Among the 47,328 participants included, 3,677 (7.7%) had COPD. The association between DI-GM and COPD remained statistically significant. After adjusting for confounders, an elevation in DI-GM was linked to a 4% reduction in the incidence of COPD (95% CI, 0.94–0.99; P = 0.004; Table 3). The adjusted OR for individuals with DI-GM ≥ 6 was 0.86 (95% CI, 0.77–0.97, P = 0.011), compared to those with lower DI-GM (Table 3).

Based on model 4, we separately adjusted energy, protein, saturated fat, and dietary fiber intake. The result showed that the association between DI-GM and COPD remained stable (Supplementary Table 3). Furthermore, we evaluated the associations of HEI-2015, aMED, and DII with COPD prevalence. While all indices showed statistically significant associations with COPD in adjusted models (Supplementary Table 4), their area under the curve (AUC) from ROC analyses were low, indicating limited discriminative ability (Supplementary Figure 1).