The NHANES is a cross-sectional health survey in the U.S. that employs a stratified, multistage probability sampling design to obtain a representative sample of the non-institutionalized civilian population (16, 17). All protocols were approved by the National Center for Health Statistics Research Ethics Review Board, and participants provided written informed consent. Detailed information on NHANES is available at https://www.cdc.gov/nchs/nhanes/.

In this study, we used a sample from the 2017–2020 NHANES cycles. Participants aged 20 years or older with complete data on hepatic vibration-controlled transient elastography (VCTE), NHHR, and covariates were included in the analysis. Exclusion criteria were: (1) ineligible, not performed, or partial elastography examination status (N = 6,537); (2); hepatitis B (positive for hepatitis B surface antigen) or hepatitis C (positive for hepatitis C antibodies or hepatitis C RNA) infection (N = 205); (3) significant alcohol consumption (more than 2 drinks/day for females or 3 drinks/day for males) (N = 2,495); (4) use of steatogenic medications (e.g., prednisone, amiodarone, tamoxifen, methotrexate) for at least 3 months (N = 15); (5) incomplete NHHR data (N = 519); (6) younger than 20 years old (N = 1,300); (7) missing covariate data, including education (N = 9), marital status (N = 4), poverty income ratio (N = 616), smoking status (N = 1), physical activity (N = 5), and the Healthy Eating Index-2015 (N = 298). The final study population consisted of 3,529 participants (Supplementary Figure 1).

The NHHR was calculated using participants’ lipid profiles. Non-high-density lipoprotein cholesterol (non-HDL-C) was calculated by subtracting high-density lipoprotein cholesterol (HDL-C) from total cholesterol (TC). NHHR was computed by dividing non-HDL-C by HDL-C. The NHHR formula is: NHHR = Non-HDL-C/HDL-C, where Non-HDL-C = TC - HDL-C.

NAFLD diagnosis traditionally relies on imaging or histological assessment to detect the presence of hepatic steatosis (18). The NHANES staff assessed participants using VCTE with the FibroScan® model 502 V2 Touch. The device measures ultrasound attenuation, recording the controlled attenuation parameter (CAP) as an indicator of hepatic steatosis. We applied a CAP cutoff value of 285 dB/m as the marker for NAFLD status. This cutoff value, validated in the U.S. population, demonstrates 80% sensitivity and 77% specificity for detecting hepatic steatosis (19). Liver fibrosis was assessed using FibroScan®, which employs ultrasound and VCTE to measure liver stiffness. Liver stiffness measurement (LSM) was used to assess the degree of hepatic fibrosis.

To investigate the independent association between NAFLD and NHHR, adjustments were made for potential confounders, including sociodemographic, lifestyle, and health status variables. Sociodemographic factors, such as age, gender, race (Mexican American, Other Hispanic, Non-Hispanic White, Non-Hispanic Black, and Other Races), education level (less than high school, high school, and more than high school), marital status (married/living with partner, widowed/divorced/separated, and never married), poverty income ratio (<1.3, 1.3 to <3.5, and ≥ 3.5) were obtained through interviews. Lifestyle related covariates included smoking status, physical activity, and dietary metric. Smoking status was classified as never smoker former smoker, and current smoker. Physical activity was categorized as inactive, moderate, and vigorous, based on the interview data. Dietary metric was evaluated by the Healthy Eating Index-2015 (HEI-2015), which is a measure for assessing whether a set of foods aligns with the Dietary Guidelines for Americans (20). The components and scoring standards of HEI-2015 were detailed in Supplementary method 1. Body mass index (BMI), and history of hypertension or diabetes were all considered variables of health status. BMI was calculated as weight divided by height squared and categorized as normal weight (<25), overweight (≥25 and < 30), and obesity (≥30) (21). History of hypertension was defined by a prior diagnosis, current use of antihypertensive medications, or a systolic blood pressure of ≥130 mmHg and/or diastolic blood pressure of ≥80 mmHg (22). History of diabetes was defined by a prior diagnosis, current use of insulin or diabetic medications, fasting plasma glucose ≥126 mg/dL, or hemoglobin A1c ≥6.5% (23).

The analysis accounted for the complex multistage cluster survey design of NHANES by using appropriate sample weights. Participants were stratified into tertiles according to NHHR levels. Baseline characteristics were summarized using weighted means with 95% confidence intervals (CIs) for continuous variables and weighted percentages (95% CIs) for categorical variables. To compare baseline characteristics across NHHR tertiles, weighted linear regression was used for continuous variables, and chi-squared tests were used for categorical variables. Multivariable logistic regression models assessed the relationship between NAFLD and NHHR, while multivariate linear regression models examined the associations between NHHR and both CAP and LSM (Supplementary method 2). NHHR was modeled as both a continuous variable and a categorical variable (tertiles). Three models were assessed: Model 1 was unadjusted; Model 2 adjusted for gender, age, and race; Model 3 further adjusted for education, marital status, poverty income ratio, BMI, smoking status, physical activity, HEI-2015, hypertension, and diabetes. Results are presented as odds ratios (ORs) or β coefficients with 95% CIs. Linear trend tests were used to assess the consistency of relationships. Generalized additive models and smooth curve fittings were employed to explore potential nonlinear associations, adjusting for potential confounders. The inflection point was identified using a recursive algorithm, followed by the application of two-piecewise linear regression models. Likelihood ratio tests were used to compare the one-line linear regression model with the two-piecewise linear regression model.

Subgroup analyses examined the association between NHHR and NAFLD across different age groups, genders, BMI categories, smoking statuses, physical activity levels, and histories of hypertension and diabetes. Interaction tests were employed to assess the consistency of these associations across subgroups.

To evaluate NHHR as a diagnostic marker for NAFLD and compare its diagnostic performance with BMI, TC, and HDL-C, receiver operating characteristic (ROC) curves were constructed, and the area under the curve (AUC) was calculated for each metric. Sensitivity, specificity, and accuracy of the diagnostics were determined. The optimal cutoff value was identified using the Youden index (sensitivity + specificity - 1). AUC values for NHHR and other parameters were compared using the Z test, with Bonferroni correction applied to adjust p values for multiple comparisons.

All statistical analyses were performed using R¹ and EmpowerStats.² Statistical significance was set as two-sided p < 0.05.

The baseline characteristics of the included participants, categorized by tertiles of NHHR, are presented in Table 1. Among the 3,529 participants, the mean age was 51.34 years (95% CI: 49.97, 52.72), with 53.53% being male and 66.25% identified as Non-Hispanic White. The mean NHHR value was 2.83 (95% CI: 2.73, 2.94), and 38.46% of the participants were diagnosed with NAFLD. Compared to those in the lowest tertile of NHHR, participants in the highest tertile were more likely to be male, Non-Hispanic White, less likely to be coupled, have a lower HEI-2015 score, and higher rates of obesity and hypertension. Additionally, they had significantly higher prevalence of NAFLD, as well as higher CAP and LSM values.

The multivariate logistic regression analysis revealed a significant association between NAFLD and NHHR across all models (Table 2). NHHR was significantly associated with NAFLD in the non-adjusted model (OR: 1.56, 95% CI: 1.47, 1.66), the partially adjusted model (OR: 1.54, 95% CI: 1.45, 1.64), and the fully adjusted model (OR: 1.33, 95% CI: 1.24, 1.42). NHHR was further categorized into tertiles for sensitivity analysis. In the fully adjusted model, compared to the lowest tertile of NHHR, the probability of NAFLD was increased in both the second tertile (OR: 1.42,95% CI: 1.16, 1.75) and the third tertile (OR = 2.35, 95% CI: 1.91, 2.90), with a significant linear trend (p for trend <0.001). To assess potential nonlinearity, we employed smooth curve fitting, which revealed a segmented relationship between NHHR and NAFLD probability (Figure 1). An inflection point was identified at an NHHR of 4. Below this threshold, NHHR showed a significant positive association with NAFLD (OR: 1.52, 95% CI: 1.38, 1.68), whereas above the threshold, the association was not significant (OR: 1.03, 95% CI: 0.91, 1.16) (Table 3).

NHHR showed a significant positive association with CAP across all models, including the fully adjusted model (β: 7.41, 95% CI: 6.07, 8.75) (Supplementary Table 1). When categorized into tertiles, participants in the highest NHHR tertile had significantly higher CAP values compared to those in the lowest tertile (β: 25.83, 95% CI: 21.51, 30.16), with a significant linear trend (p for trend <0.001). To investigate potential nonlinear patterns, we applied smooth curve fitting, which identified a segmented relationship between NHHR and CAP (Supplementary Figure 2). The inflection point was observed at an NHHR of 4. Below this value, NHHR was significantly associated with CAP (β: 12.66, 95% CI: 10.65, 14.67), whereas above this value, the association was not significant (β: -0.18, 95% CI: −2.73, 2.38) (Supplementary Table 3).

In contrast to the findings for CAP, NHHR was not significantly associated with LSM in the fully adjusted model (β: 0.00, 95% CI: −0.13, 0.12) (Supplementary Table 2). Comparison across NHHR tertiles revealed no significant differences between the highest and lowest tertiles (β: -0.03, 95% CI: −0.44, 0.37), and no significant trend (p for trend = 0.866). To explore potential nonlinear relationships, we utilized smooth curve fitting (Supplementary Figure 3). The log-likelihood ratio test indicated that a two-piecewise linear regression model did not provide a better fit than a one-line linear regression model (p = 0.128) (Supplementary Table 4).

Subgroup analyses, as illustrated in Figure 2, indicated a positive association between NHHR and NAFLD across all subgroups. Significant interactions were identified between NHHR and both gender and BMI categories (p for interaction <0.05). The positive association between NHHR and NAFLD was stronger in females (OR: 1.45, 95% CI: 1.30, 1.63) than in males (OR: 1.25, 95% CI: 1.15, 1.37). The NHHR and BMI interaction revealed the strongest association in individuals with normal weight (OR: 1.73, 95% CI: 1.44, 2.09), followed by those who were overweight (OR: 1.31, 95% CI: 1.18, 1.46), and those with obesity (OR: 1.24, 95% CI: 1.12, 1.37). Other factors, including age, smoking status, physical activity, and history of hypertension and diabetes, did not significantly influence the association between NHHR and NAFLD (p for interaction >0.05).

ROC curve analyses were conducted to evaluate the diagnostic value of NHHR for NAFLD compared with other measures (Figure 3). As shown in Table 4, with a cutoff value of 2.40, NHHR achieved a sensitivity of 0.71 and a specificity of 0.55, resulting in an AUC of 0.66 (95% CI: 0.64, 0.68). NHHR outperformed TC, which had an AUC of 0.51 (p < 0.001). However, the diagnostic accuracy of NHHR was lower than that of HDL-C, which had an AUC of 0.68 (p = 0.015), and BMI, which showed the highest diagnostic efficacy with an AUC of 0.77 (p < 0.001).