This study utilised data from the seven-year KNHANES from 2017 to 2023, which was conducted by the Korean Centers for Disease Control and Prevention. The objectives of the KNHANES include monitoring trends in health risk factors and their prevalence, and the survey consisted of health screenings, health interviews, and nutritional surveys conducted by trained medical staff and interviewers. The Institutional Review Board of the Korean Centres for Disease Control and Prevention approved the survey and all participants signed an informed consent form. This study was a secondary analysis of de-identified data from the KNHANES and therefore did not require additional ethical approval. Detailed information and description of the database can be found on the KNHANES website (https://knhanes.kdca.go.kr). For the 2017-2023 survey, a total of 51,872 individuals were sampled, of which 45,742 completed the 48-hour dietary recall (48RC) survey. Of the 45742 participants, we excluded the following participants: those who did not fully answer questions about AD and ages <20 and >80 years (n=2073); those who did not answer whether they had ever smoked or drank alcohol (n=41); those with incomplete personal information (did not fully answer questions about level of education, household income (quartiles), occupation, and exercise), and those who did not have a measured body mass index (BMI) (n=8732); in addition, we excluded participants with abnormal Energy intake (EK) (EK <500Kcal or EK >6000Kcal) (n=1863). Therefore, a total of 32,763 participants were eligible for our study ( Figure 1 ).

We used the following variables as covariates: age (years), sex (male and female), area of residence (urban and rural), level of education (below university and university and above), household income (which will be categorised as low, medium, upper-middle, and upper based on quartiles), alcohol consumption (categorised as drinker and non-drinker based on current drinking or not), smoking (categorised as non-smoker, current smoker, and past smoker), exercise (categorised as no exercise, four and less, and five and more based on the number of number of days in a week that exercise (walking and strength) was performed was classified as no exercise, four or less, and five or more), and BMI calculated as weight in kilograms divided by the square of height in metres, based on the reference values for obesity in Asian populations (11), subjects were classified into three groups: underweight (BMI < 18.5), normal (18.5 ≤ BMI < 25), and obese (BMI ≥ 25).

In this study, the calculation of DII was based on an established standard model. The model first standardises the intake of the 45 dietary nutrients to the Z-score of the global average intake, then multiplies the standardised values with the corresponding inflammatory impact weights of each component, and finally adds all the weighted scores to give an individual’s total DII score. Higher DII scores indicate that the diet is pro-inflammatory, while lower scores reflect that the diet has anti-inflammatory potential (9).

Nutrient intake data in the study were obtained from KNHANES, where trained investigators obtained information on subjects’ food intake over the past 48 hours through a 1-hour dietary recall interview and calculated individual daily nutrient intakes by referring to the 10th edition of the Korean Food Composition Table by the Korea Rural Advancement Agency. Due to data availability constraints in KNHANES, we utilised a modified DII based on 25 nutrients (listed in Figure 2 ). Prior studies support the validity of this approach for population-level analyses (9).

The presence of AD was determined using a standardised approach validated for epidemiological studies. Participants were asked: “Have you ever been diagnosed with atopic dermatitis (eczema) by a physician?” Those responding “yes” were classified as AD cases. This approach of self-reported physician diagnosis is commonly used in large-scale population surveys and shares conceptual similarities with validated questionnaire-based assessments employed in major international studies such as the International Study of Asthma and Allergies in Childhood (ISAAC) (12, 13). Both methodologies have demonstrated reliability in population-based studies of allergic diseases (14).

We used the PROC SURVEY procedure in SAS software (version 9.4, SAS Institute Inc., Cary, NC, USA) and applied survey weights to account for the complex sampling design of KNHANES. The basic characteristics of the study population were described as weighted means (standard error, SE) for continuous variables and weighted percentages (weighted standard error, SE) for categorical variables. Significance tests for continuous variables were performed using analysis of variance (ANOVA), while categorical variables were assessed for differences between subgroups by the Rao-Scott chi-square test. Associations analysed by multinomial logistic regression models, two models were used for covariate control: (Model 1) a base-adjusted model (age, sex, education level, household income, occupation, place of residence); and (Model 2) a multivariate-adjusted model further controlling for the effects of lifestyle factors (alcohol consumption, smoking, exercise), BMI, and total energy intake; the present study analysed dietary inflammation by weighted multivariate logistic regression model to analyse the association between DII and disease risk, and stratified by sex (male/female) and age (≤54 vs >54 years), the aOR of DII quartiles for each subgroup was calculated, along with their 95% confidence intervals; the Wald test was used to assess the interaction of sex and age, and the interaction between the same DII level and the same DII level among subgroups was calculated by calculating the The Wald test was used to assess the interaction between sex and age, and the significance of the interaction was determined by calculating the log(OR) difference at the same DII level and its standard error among different subgroups; the nonlinear association between DII and disease risk was also explored using RCS with the nodes at the 10th, 50th, and 90th percentiles of the DII distribution, and all the models adjusted for socio-demographic covariates such as sex, age, and education. Results are presented as OR and 95% CI using two-sided tests with a significance threshold of p<0.05.

Table 1 presents the characteristics of the study participants according to DII quartiles. The data showed that higher DII scores were significantly correlated with increasing age, being male, having less than a college education, a non-smoking, non-alcoholic lifestyle, lower household income, more frequent strength training, less walking training, and lower BMI.

Table 2 demonstrates the cross-sectional association between DII score and AD. Without adjusting for covariates, higher DII scores were significantly associated to AD status, participants in the highest DII quartile had 73% greater odds of having AD versus the lowest quartile (aOR = 1.73, 95% CI = 1.45-2.07). This positive association remained significant after adjusting for the basic life variable model and the multivariate model (OR 1.23, 95% CI 1.02-1.49, p trend = 0.03041, OR 1.25, 95% CI 1.03-1.51, p trend = 0.0213), with the risk of AD prevalence elevated by approximately 15% for each unit of elevated DII.

Figure 2 shows the associations between the components of the DII and other nutrients included in the survey data and AD. After adjusting for covariates included based on model3, analyses of the components of the DII and other nutrients included in the survey data showed that carotenoid and dietary fibre intake were negatively associated with DII scores, whereas total energy and sugar intake were positively associated with the prevalence of AD, and these results suggest that specific dietary components may play a unique role in the mechanisms of AD.

Table 3 demonstrates the association between DII scores and covariates and AD through stratified analyses by sex and age. The analyses showed that among males, higher DII scores, living in the countryside, and current smoking were risk factors and higher household income was a protective factor, whereas among females, living in the countryside and engaging in mental labour significantly reduced the risk of AD, and alcohol consumption, higher levels of education, and performing weight and walking training also significantly increased the risk of AD. In the lower age group, living in the countryside, higher level of education, and engaging in mental labour significantly reduced the risk of AD, whereas higher DII scores, living in the countryside, drinking alcohol, higher level of education, and doing weight and walking training significantly increased the risk of AD, whereas no significant results were found in the higher age group. The pro-inflammatory effects of DII and smoking predominated in men, whereas in women socio-behavioural factors (level of education, type of occupation) may influence AD indirectly through stress or lifestyle. The lack of significant associations in the higher age groups may be related to the role of immune senescence in weakening dietary inflammation (15).

An exploratory analysis of the potential non-linear association between the DII and the risk of AD prevalence was carried out using the RCS model ( Figure 3 ). The results showed that the overall association between this variable and the outcome variable was statistically significant (p < 0.0001 for the overall test), and the non-linear component did not reach the level of significance (p = 0.18 for the non-linear test). This suggests a possible linear trend in the association between DII and risk of disease.