Association between non-high-density lipoprotein cholesterol to high-density lipoprotein cholesterol ratio (NHHR) and high-sensitivity C-reactive protein in US adults: Results from NHANES 2015–2018

Abstract

The non-high-density lipoprotein to high-density lipoprotein cholesterol ratio (NHHR) is an emerging lipid index linked to cardiovascular and metabolic risk. High-sensitivity C-reactive protein (hs-CRP) serves as a well-established marker of systemic inflammation. However, the association between NHHR and hs-CRP in the general population remains unclear. This study aimed to investigate their relationship using nationally representative US data. We analyzed data from 5994 adults aged ≥20 years from the 2015 to 2018 National Health and Nutrition Examination Survey (NHANES). NHHR was calculated as (total cholesterol – high-density lipoprotein cholesterol [HDL-C])/high-density lipoprotein cholesterol. Survey-weighted multivariable linear regression models were used to evaluate the association between NHHR (both continuous and quartiles) and hs-CRP levels. Restricted cubic spline analysis assessed nonlinear patterns. Subgroup and interaction analyses were conducted by age, sex, body mass index, diabetes, and hypertension. After full adjustment, NHHR was positively associated with hs-CRP (β = 0.91, 95% CI: 0.42–1.40, P = .002). Participants in Q3 and Q4 had significantly higher hs-CRP levels than those in Q1. Restricted cubic spline analysis revealed a significant nonlinear (inverted U-shaped) relationship (P for nonlinearity < .001). Subgroup analyses showed stronger associations in women and individuals with hypertension (P for interaction < .05). In this exploratory, cross-sectional analysis, our findings suggest that NHHR may be independently associated with hs-CRP and may exhibit a nonlinear relationship in US adults, suggesting its potential utility as an accessible marker of low-grade systemic inflammation; however, longitudinal cohort studies are needed to confirm these associations and to evaluate its potential predictive value and clinical applicability.

1. Introduction

Prospective studies in adults have shown that chronic low-grade systemic inflammation may emerge as a pivotal contributor to the development and progression of numerous chronic diseases, including cardiovascular disease (CVD),^[1] diabetes,^[2,3] neurodegenerative disorders,^[4] cancer^[5] and autoimmune diseases.^[6] Among various inflammatory biomarkers, high-sensitivity C-reactive protein (hs-CRP) is one of the most extensively validated and clinically relevant indicators.^[7] Synthesized by hepatocytes in response to pro-inflammatory cytokines such as IL-6 and TNF-α, hs-CRP serves as a marker of systemic inflammation and has been widely employed in both clinical and epidemiological settings to evaluate future risk of cardiovascular and metabolic diseases.^[8,9]

The non-high-density lipoprotein cholesterol to high-density lipoprotein cholesterol ratio (NHHR), computed by subtracting HDL-C from total cholesterol (TC) and dividing by HDL-C, serves as an integrated lipid marker representing the equilibrium between atherogenic and protective lipoprotein fractions. Unlike isolated lipid parameters such as LDL-C or triglycerides, NHHR incorporates a broader spectrum of lipid risk and is recognized as a powerful predictor for cardiometabolic disorders,^[10] including atherosclerosis,^[11] insulin resistance,^[12] and metabolic syndrome.^[13] Dyslipidemia and chronic low-grade inflammation are both critical factors involved in the initiation and progression of atherosclerosis, which underlies most CVDs.^[14] HDL-C has long been regarded as beneficial for cardiovascular protection, such as promoting reverse cholesterol transport, reducing oxidative stress, exerting anti-inflammatory effects, and supporting endothelial integrity.^[15] Several population-based studies have shown that HDL-C levels are predictive of all-cause and cardiovascular-related mortality among diverse cohorts.^[16,17] Low-grade systemic inflammation, characterized by a persistent yet mild elevation in inflammatory biomarkers, has become an area of increasing research interest.^[18] hs-CRP serves as a key marker of such inflammation. It is produced primarily by liver cells, aortic endothelium, and vascular smooth muscle cells in response to inflammatory stimuli such as IL-6 and TNF-α.^[19,20] A growing body of literature has confirmed that elevated hs-CRP concentrations are strongly associated with both the onset and recurrence of CVD events.^[21-24] Nonetheless, variations across populations and the presence of residual or unaccounted confounders limit the ability of hs-CRP or lipid markers alone to serve as dependable predictors of long-term cardiovascular risk.^[25-27]

Given the growing burden of chronic inflammation-related diseases and the widespread availability of lipid profiles in clinical practice, clarifying the relationship between NHHR and hs-CRP has important implications. NHHR is easy to calculate, cost-effective, and already embedded in routine lipid testing, making it a potentially valuable biomarker for early risk stratification. However, whether NHHR is independently associated with systemic inflammation – as indicated by hs-CRP – remains unclear. To address this gap, we conducted a hypothesis-generating, exploratory cross-sectional analysis using NHANES 2015–2018 data to evaluate the association between NHHR and hs-CRP, assess its linear and nonlinear patterns, and explore effect modification across demographic and clinical subgroups.

2. Materials and methods

2.1. Data source

The data utilized in this study were extracted from the National Health and Nutrition Examination Survey (NHANES), a nationally representative, cross-sectional survey program conducted by the National Center for Health Statistics (NCHS), part of the centers for disease control (CDC) and Prevention. NHANES employs a stratified, multistage probability sampling design to assess the health and nutritional status of the civilian, non-institutionalized US population. It integrates interviews, physical examinations, and laboratory tests. Data from NHANES are publicly accessible and can be obtained from the CDC website (https://www.cdc.gov/nchs/nhanes/). For the present analysis, data were derived from 2 continuous cycles: 2015 to 2016 and 2017 to 2018. The NHANES protocol received ethical approval from NCHS Ethics Review Board at the CDC, in accordance with established ethical standards. Written informed consent was obtained from all participants prior to their involvement in the survey.

2.2. Study population

An initial sample of 19,225 participants was drawn from the 2015 to 2018 NHANES waves. We applied sequential exclusion criteria to ensure data completeness and analytical validity. First, individuals younger than 20 years were removed (n = 7937), leaving 11,288 adults aged ≥20 years. We then excluded participants with missing covariate information, including sociodemographic, lifestyle, body mass index (BMI), diabetes, and hypertension variables (n = 4916), resulting in 6372 eligible individuals. Subsequently, participants without TC or HDL-C measurements were excluded (n = 344), yielding 6028 participants. Finally, individuals missing hs-CRP data were removed (n = 34). Overall, 5294 participants (46.9% [5294/11,288]) were excluded due to missing data on covariates and key exposure/outcome-related variables. After applying these exclusion criteria, 5994 eligible adult participants remained for final analysis, as depicted in Figure 1.

Figure 1.

Flowchart of participants screening. HDL-C = high-density lipoprotein cholesterol, hs-CRP = high-sensitivity C-reactive protein, NHANES = National Health and Nutrition Examination Survey.

2.3. Definition of NHHR

NHHR was calculated as the ratio of non-HDL cholesterol (non-HDL-C) to HDL-C, with non-HDL-C derived by subtracting HDL-C from TC:

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{N}\mathrm{H}\mathrm{H}\mathrm{R}=(\mathrm{T}\mathrm{C}-\mathrm{H}\mathrm{D}\mathrm{L}-\mathrm{C})/\mathrm{H}\mathrm{D}\mathrm{L}-\mathrm{C}}\end{array}}$$

TC and HDL-C concentrations were obtained from standard laboratory measurements provided by NHANES. In the analysis, NHHR was evaluated as both a continuous measure and a categorical variable divided into quartiles.^[28] For categorical analysis, NHHR was stratified into quartiles (Q1–Q4) according to its weighted distribution within the study sample, using Q1 as the referent.

2.4. Definition of hs-CRP

hs-CRP levels were measured as part of NHANES laboratory procedures. Serum samples were collected at mobile examination centers using standardized venipuncture protocols. Following specimen processing and storage, hs-CRP was quantified using the Beckman UniCel^® DxC 600/660i Synchron Clinical System (immunoturbidimetric method), which detects antigen-antibody complexes formed between serum CRP and antibodies, resulting in light scattering proportional to CRP concentration. Quality assurance and quality control protocols included daily testing of bio-rad liquid unassayed multiqual controls, with results verified via the laboratory information system.^[29] The lower limit of detection for hs-CRP was 0.11 mg/L. For statistical analysis, hs-CRP was treated as a continuous variable.

2.5. Covariates

The covariates included in this study were selected based on prior literature and clinical relevance, and were grouped into 4 categories: demographic, lifestyle, anthropometric, and comorbidity-related variables.^[28,30,31] Demographic variables included age, sex, racial and ethnic background, and educational level. Lifestyle factors comprised alcohol consumption and smoking status. BMI was considered an anthropometric indicator. Comorbidity-related variables included self-reported diagnoses of diabetes and hypertension. All covariate information was obtained using standardized questionnaires administered during the in-home interview phase of the NHANES survey.

Age was classified into 3 groups: 20 to 59 years, 60 to 79 years, and ≥80 years. Participants’ self-reported race and ethnicity were categorized as non-Hispanic White, non-Hispanic Black, Mexican American, other Hispanic, and other race including multiracial. Educational level was grouped into 3 levels: less than high school, high school or equivalent, and college or above. Lifestyle factors included smoking and alcohol drinking status. Smoking status was divided into 3 categories: current smoker (individuals who had smoked at least 100 cigarettes in their lifetime and currently smoke), former smoker (those who had smoked ≥100 cigarettes but no longer smoke), and never smoker. Alcohol consumption was classified into 3 categories: “Never drinker,” “Former drinker,” and “Current drinker.” Due to inconsistencies in alcohol-related questionnaire items across NHANES cycles, data from 2015 to 2016 and 2017 to 2018 were harmonized using available variables. Participants who reported never having consumed 12 or more alcoholic beverages in their lifetime were classified as “Never drinkers.” Those who had previously consumed alcohol but reported no recent consumption (indicated by either a frequency of 0 or missing recent drinking frequency) were categorized as “Former drinkers.” Participants with a reported non-zero drinking frequency were considered “Current drinkers.” Anthropometric indicator included BMI, which was calculated as weight in kilograms divided by height in meters squared (kg/m²). BMI was categorized into 4 groups: underweight (<18.5 kg/m²), normal weight (18.5–24.9 kg/m²), overweight (25–29.9 kg/m²), and obesity (≥30 kg/m²), with obesity defined as BMI ≥ 30 kg/m² in accordance with current CDC and WHO criteria. Measures of central adiposity (e.g., waist circumference) and additional lipid fractions (such as triglycerides), as well as formal definitions of metabolic syndrome, were available only in subsets of participants and were not included as covariates in the primary models to preserve sample size and model stability; this limitation is explicitly acknowledged in the Discussion.

Comorbid conditions included diabetes and hypertension. Diabetes was defined as having self-reported physician diagnosis of diabetes, use of antidiabetic medications, or a fasting plasma glucose ≥ 126 mg/dL or HbA1c ≥ 6.5%. Hypertension was defined as a self-reported diagnosis, use of antihypertensive medication, or measured systolic blood pressure ≥ 140 mm Hg and/or diastolic blood pressure ≥ 90 mm Hg. All covariates were harmonized across NHANES survey cycles and included as categorical variables in the regression models.

2.6. Statistical analysis

All statistical procedures were performed in R (version 4.3.3). To ensure national representativeness and accommodate NHANES’s complex multistage sampling design, all analyses utilized the “survey” package, incorporating sampling weights, strata, and primary sampling units (PSUs). The combined 4-year mobile examination center weights were calculated by dividing the 2-year weights (WTMEC2YR) by 2, as recommended by the NCHS guidelines for merged NHANES cycles.

To properly account for the NHANES complex survey structure, mobile examination center examination weights, design strata, and (PSUs) were incorporated exactly as provided by NHANES. The design variables SDMVSTRA (strata) and SDMVPSU (PSUs) were already uniquely coded across cycles; therefore, no recoding was necessary. All survey analyses were specified with `nest = TRUE` and `survey.lonely.psu = “adjust”` to ensure valid variance estimation under the Taylor series linearization method.

Given the cross-sectional nature of NHANES, all analyses were explicitly exploratory and focused on characterizing associations between NHHR and hs-CRP rather than making causal or predictive inferences.

Descriptive statistics were first used to summarize baseline characteristics of participants across quartiles of the NHHR. For descriptive analysis, continuous variables were presented as weighted means with standard errors, and categorical variables as weighted percentages. Between-group comparisons were conducted using the Rao–Scott χ² test for categorical variables and weighted linear regression models for continuous variables.

Due to the skewed distribution of hs-CRP, the outcome variable was analyzed in its original scale, as hs-CRP is a clinically interpretable biomarker typically evaluated in absolute concentrations. Survey-weighted residual diagnostics also indicated acceptable model fit without transformation. In contrast, NHHR was log-transformed to reduce skewness and improve model performance. A sensitivity analysis using log-transformed hs-CRP as the outcome yielded similar effect estimates and significance across all models (Table S1, Supplemental Digital Content 1, https://links.lww.com/MD/R508).

Missing data were handled using listwise deletion. Participants with missing values for NHHR, hs-CRP, lipid measurements, or covariates were excluded from the analytic sample. Because the proportion of missing data was low and not indicative of systematic patterns, multiple imputation was not performed, consistent with standard practice in NHANES-based analyses.

Covariates were selected a priori based on established evidence and conceptual causal frameworks linking demographic factors, lifestyle behaviors, and metabolic conditions to both lipid profiles and systemic inflammation.^[32,33] Age, sex, and race/ethnicity were considered potential confounders given their known associations with lipid profiles in prior NHANES-based analyses.^[34] Lifestyle and socioeconomic variables^[33] – including smoking status,^[35] alcohol consumption,^[36] BMI,^[37] and educational attainment^[38] – are known to influence inflammatory biomarkers and lipid abnormalities. Diabetes^[39] and hypertension^[40] were included as key metabolic comorbidities associated with dyslipidemia and systemic inflammation. This covariate selection strategy is consistent with prior NHANES-based studies evaluating lipid-related inflammatory or metabolic outcomes.

Multivariable linear regression models were employed to examine the association between NHHR and hs-CRP. In addition to treating NHHR as a continuous variable, we also categorized it into quartiles, with the lowest quartile (Q1) as the reference. Three models were successively constructed. No covariates were adjusted in model 1. Model 2 was adjusted for demographic factors, including age group, sex, race/ethnicity, and education. Model 3 was refined by incorporating lifestyle and health factors, including BMI category, smoking status, alcohol use, diabetes, and hypertension.

The “rms” package was used to fit RCS models, allowing for the evaluation of nonlinear associations between NHHR and hs-CRP. Models with 3 to 7 knots were evaluated, and the optimal number was determined using the lowest Akaike information criterion (AIC). A final model with 4 knots was used for restricted cubic spline (RCS) analysis. The median NHHR value was used as the reference point, and sampling weights were scaled to account for the survey’s complex design. Both the overall and nonlinear trends were tested using Wald statistics.

To assess effect heterogeneity, subgroup analyses and interaction tests were conducted. Stratified linear regression models were fitted by sex, age group, BMI group, smoking status, alcohol consumption, diabetes status, and hypertension. Interaction terms (e.g., NHHR × sex) were included in the models, and Wald tests were used to evaluate interaction significance. Due to the small sample size in the underweight subgroup (n = 84), the multivariable regression model for this group was simplified by including only key confounders (age, sex, diabetes, and hypertension) to avoid overfitting and ensure model convergence. Results were visually presented using forest plots, and P values for interaction were reported.

Several sensitivity analyses were conducted to assess the robustness of the findings. First, to minimize the influence of acute inflammation, we repeated the main multivariable models after excluding participants with hs-CRP levels >10 mg/L. Second, we fitted additional models using log-transformed hs-CRP as the outcome to examine whether the choice of scale for hs-CRP affected the results. These sensitivity analyses yielded effect estimates that were similar in direction and magnitude to those of the primary models. Model assumptions were evaluated by inspecting residual-versus-fitted plots and normal Q–Q plots of residuals, and by calculating variance inflation factors for all covariates to assess multicollinearity.

Because NHANES is based on a nationally representative, fixed-probability sampling design, a priori sample size calculation is not applicable. Instead, the statistical power of observational analyses relies on the achieved sample size and the precision of survey-weighted estimates. The final analytic sample of 5994 adults is comparable to or larger than sample sizes used in previous NHANES studies examining lipid biomarkers and inflammatory markers, which have been shown to reliably detect small-to-moderate effect sizes within a complex survey framework. Therefore, the sample size of the present study is sufficient to detect the modest associations observed.

All statistical tests were 2-sided, and P < .05 was considered statistically significant.

3. Results

3.1. Characteristics of the included participants

The final analytic sample comprised 5994 individuals aged 20 years or above. Quartile classification of participants was conducted using their NHHR distributions (see Table 1). Significant differences were observed across NHHR quartiles for most baseline characteristics (all P < .05), except for alcohol intake (P = .30). The majority of participants were aged 20 to 59 years (69.4%), female (55%), and non-Hispanic White (62%). Higher NHHR quartiles were associated with a greater proportion of males, younger participants, and those with obesity. Specifically, the proportion of male participants increased from 30% in Q1 to 63% in Q4, while the proportion of participants aged ≥80 years decreased from 8.8% to 2.6%. In Q4, 58% of participants were obese compared to 26% in Q1. A declining trend in educational attainment was observed across increasing NHHR quartiles, with the proportion of college-educated participants falling from 65% in Q1 to 56% in Q4.

Table 1.

Baseline characteristics of participants according to NHHR quartiles in NHANES 2015–2018.

Characteristic	Overall, N = 5994 (100%)	NHHR				P value
		Q1, N = 1499 (25%)	Q2, N = 1498 (25%)	Q3, N = 1499 (25%)	Q4, N = 1498 (25%)
Age (yr)						<.001
20–59	3668 (69.4%)	877 (67.0%)	885 (68.5%)	896 (67.1%)	1010 (75.2%)
60–79	1849 (24.9%)	442 (24.3%)	480 (25.0%)	504 (28.1%)	423 (22.2%)
80+	477 (5.7%)	180 (8.8%)	133 (6.6%)	99 (4.8%)	65 (2.6%)
Sex						<.001
Male	2629 (45%)	479 (30%)	562 (38%)	700 (49%)	888 (63%)
Female	3365 (55%)	1020 (70%)	936 (62%)	799 (51%)	610 (37%)
Race						<.001
Non-Hispanic White	1992 (62%)	487 (62%)	516 (62%)	498 (62%)	491 (60%)
Non-Hispanic Black	1330 (11%)	445 (15%)	356 (12%)	300 (9.9%)	229 (7.8%)
Mexican American	922 (9.3%)	181 (7.4%)	204 (8.4%)	258 (10%)	279 (12%)
Other Hispanic	635 (6.9%)	113 (5.7%)	155 (6.7%)	169 (6.7%)	198 (8.7%)
Other/multiracial	1115 (11%)	273 (9.4%)	267 (11%)	274 (11%)	301 (12%)
Education level						.02
Less than high school	1302 (13%)	294 (11%)	283 (12%)	346 (12%)	379 (16%)
High school or equivalent	1394 (27%)	347 (25%)	348 (25%)	351 (29%)	348 (28%)
College or above	3298 (61%)	858 (65%)	867 (64%)	802 (59%)	771 (56%)
BMI						<.001
Underweight (<18.5)	84 (1.5%)	52 (3.2%)	14 (1.0%)	10 (1.1%)	8 (0.7%)
Normal (18.5 to <25)	1447 (25%)	597 (43%)	403 (29%)	265 (16%)	182 (11%)
Overweight (25 to <30)	1883 (31%)	425 (28%)	466 (30%)	495 (34%)	497 (31%)
Obese (30 or greater)	2580 (43%)	425 (26%)	615 (40%)	729 (50%)	811 (58%)
Alcohol intake						.30
Current drinker	3471 (67%)	901 (70%)	852 (68%)	855 (65%)	863 (66%)
Former drinker	1278 (18%)	287 (15%)	330 (17%)	328 (20%)	333 (19%)
Never drinker	1245 (15%)	311 (16%)	316 (15%)	316 (15%)	302 (15%)
Smoking status						<.001
Current smoker	942 (16%)	212 (13%)	205 (14%)	241 (16%)	284 (20%)
Former smoker	1310 (23%)	287 (20%)	334 (24%)	338 (22%)	351 (27%)
Never smoker	3742 (61%)	1000 (67%)	959 (61%)	920 (62%)	863 (53%)
HDL	51 (42, 62)	68 (60, 78)	56 (48, 63)	48 (42, 54)	41 (35, 45)	<.001
Total cholesterol	186 (161, 214)	165 (143, 187)	177 (155, 200)	189 (170, 214)	216 (192, 243)	<.001
Diabetes						.02
Yes	1326 (16%)	299 (13%)	302 (15%)	348 (17%)	377 (20%)
No	4668 (84%)	1200 (87%)	1196 (85%)	1151 (83%)	1121 (80%)
Hypertension						.003
Yes	2781 (40%)	653 (36%)	679 (38%)	722 (43%)	727 (46%)
No	3213 (60%)	846 (64%)	819 (62%)	777 (57%)	771 (54%)
hs-CRP	1.9 (0.9, 4.5)	1.0 (0.6, 2.8)	1.7 (0.8, 4.0)	2.2 (1.1, 5.1)	2.8 (1.3, 5.6)	<.001

Continuous variables: values are expressed as mean ± standard error.

Categorical variables: values are expressed as numbers (%).

Bold values indicate statistical significance (P < .05).

BMI = body mass index, HDL-C = high-density lipoprotein cholesterol, hs-CRP = high-sensitivity C-reactive protein, NHHR = non-high-density lipoprotein cholesterol to high-density lipoprotein cholesterol ratio.

Regarding lifestyle factors, the prevalence of current smokers increased across quartiles (from 13% to 20%), while never smokers decreased. No significant difference was found in alcohol consumption across NHHR quartiles. Median HDL levels progressively decreased (from 68 to 41 mg/dL), whereas TC levels increased (from 165 to 216 mg/dL). As NHHR increased, the proportion of participants with diabetes and hypertension rose, accompanied by an elevation in median hs-CRP levels from 1.0 mg/L in Q1 to 2.8 mg/L in Q4.

3.2. Univariate linear regression analyses

Univariate linear regression analyses were conducted to assess the independent associations between each covariate and hs-CRP levels (Table S2, Supplemental Digital Content 2, https://links.lww.com/MD/R508). Female participants had significantly higher hs-CRP levels than males (β = 1.34, 95% CI: 0.85–1.82, P < .001). Compared to non-Hispanic Whites, non-Hispanic Blacks (β = 0.83, 95% CI: 0.25–1.41, P = .006) showed significantly higher hs-CRP levels, while other racial groups showed no significant differences. Higher education, particularly college level or above, correlates with reduced hs-CRP levels compared to individuals with less than a high school education (β = −1.08, 95% CI: −1.82 to −0.35, P = .005). Regarding lifestyle factors, former drinkers had higher hs-CRP levels than current drinkers (β = 1.40, 95% CI: 0.68–2.12, P < .001). There was no significant association between smoking status and hs-CRP levels. Among lipid parameters, HDL showed a negative association with hs-CRP (β = −0.049, 95% CI: −0.060 to −0.038, P < .001), while TC was not significantly related to hs-CRP levels. NHHR was positively associated with hs-CRP (β = 0.344, 95% CI: 0.229–0.459, P < .001). Participants without diabetes or hypertension had significantly lower hs-CRP levels compared to those with these conditions (diabetes: β = −2.98, P < .001; hypertension: β = −1.46, P < .001).

3.3. Multivariable linear regression analysis

Table 2 presents the findings from weighted multivariable linear regression analyses examining the relationship between NHHR and hs-CRP levels. Log-transformed NHHR, when treated as a continuous variable, was significantly positively associated with hs-CRP across all models. In the unadjusted model (model 1), a 1-unit increase in log(NHHR) was associated with an increase of 1.4 units in hs-CRP (β = 1.4, 95% CI: 1.1–1.8, P < .001). After adjusting for age, sex, race, and education in model 2, the association persisted strong (β = 1.9, 95% CI: 1.4–2.4, P < .001). In the fully adjusted model (model 3), which accounted for comorbid conditions and lifestyle variables, the observed association was weakened but still reached statistical significance (β = 0.91, 95% CI: 0.42–1.4, P = .002). In the fully adjusted model, each 1-unit increase in log(NHHR) was associated with an average 0.91 mg/L higher hs-CRP level. Given that the median hs-CRP concentration in this population was approximately 1.6 mg/L, this corresponds to a roughly 50% to 60% relative increase. Although statistically significant, this coefficient indicates a relatively modest change in hs-CRP, with predicted values remaining largely within the range of low-grade systemic inflammation. After exclusion of participants with hs-CRP > 10 mg/L, the positive association between log(NHHR) and hs-CRP remained statistically significant, although the effect size was attenuated (fully adjusted β changed from 0.91 [95% CI: 0.42–1.40] to 0.40 [95% CI: 0.34–0.47]; Table S3, Supplemental Digital Content 3, https://links.lww.com/MD/R508). This suggests that the main findings are robust and are not fully driven by individuals with possible acute inflammation. Diagnostic plots did not reveal major violations of linear regression assumptions, and VIF values indicated no serious multicollinearity among covariates.

Table 2.

Association between NHHR and hs-CRP in the multiple linear regression model.

	Model 1		Model 2		Model 3
Characteristic	β (95% CI)	P-value	β (95% CI)	P-value	β (95% CI)	P-value
log(NHHR)	1.4 (1.1, 1.8)	<.001	1.9 (1.4, 2.4)	<.001	0.91 (0.42, 1.4)	.002
Quartile
Q1	Reference		Reference		Reference
Q2	0.93 (0.30, 1.55)	.005	1.09 (0.46, 1.72)	.002	0.65 (−0.07, 1.36)	.07
Q3	1.76 (0.93, 2.59)	<.001	2.11 (1.22, 3.00)	<.001	1.31 (0.26, 2.35)	.02
Q4	1.61 (0.99, 2.24)	<.001	2.19 (1.48, 2.91)	<.001	1.02 (0.26, 1.78)	.01
P for trend		<.001		<.001		.007

Model 1 = unadjusted.

Model 2 = adjusted for age, sex, race, and education level.

Model 3 = adjusted for age, sex, race, education level, smoking status, alcohol consumption, BMI, diabetes, and hypertension.

Bold values indicate statistical significance (P < .05).

CI = confidence interval, hs-CRP = high-sensitivity C-reactive protein, NHHR = non–high-density lipoprotein cholesterol to high-density lipoprotein cholesterol ratio.

When NHHR was categorized into quartiles, hs-CRP levels were significantly higher in Q3 and Q4 compared to Q1 in the model with full covariate adjustment. Specifically, those in the higher NHHR categories (Q3 and Q4) had higher hs-CRP levels by 1.31 units (95% CI: 0.26–2.35, P = .02) and 1.02 units (95% CI: 0.26–1.78, P = .01), respectively. Although Q2 did not show a statistically meaningful association (β = 0.65, 95% CI: −0.07 to 1.36, P = .07), the overall trend remained significant (P = .007), indicating a positive dose–response relationship between NHHR and hs-CRP.

3.4. Nonlinear association between NHHR and hs-CRP

To further characterize the dose–response relationship between NHHR and hs-CRP, a RCS analysis was performed based on a weighted multivariable linear regression model. Using the AIC for model selection, a 4-knot model was identified as the optimal fit. Sensitivity analyses comparing models with 3 to 7 knots showed very similar goodness-of-fit, with the 4-knot model yielding the lowest AIC and comparable R² values (Table S4, Supplemental Digital Content 4, https://links.lww.com/MD/R508). RCS modeling revealed both a statistically significant overall relationship (P-overall < .001) and a meaningful nonlinear component. As illustrated in Figure 2, hs-CRP levels initially increased with NHHR, reaching a peak at approximately log(NHHR) = 1.17, followed by a modest decline at higher values of NHHR. The reference point was set at the median log(NHHR) = 0.79. The widening confidence intervals at extreme NHHR values indicated increased uncertainty in the model predictions. Overall, the RCS analysis provided further evidence of a nonlinear, inverted U-shaped association between NHHR and hs-CRP levels.

Figure 2.

Restricted cubic spline analysis depicting the nonlinear association between log(NHHR) and hs-CRP. Adjusted for all covariates (age, sex, race, education level, smoking status, alcohol consumption, BMI, diabetes, and hypertension). The blue line represents the estimated β coefficients; the shaded area is the 95% confidence interval. The reference log(NHHR) value is 0.79. The peak hs-CRP level appeared at log(NHHR) = 1.17. Both overall and nonlinear associations were significant (P < .001). AIC = Akaike information criterion, hs-CRP = high-sensitivity C-reactive protein, NHHR = non-high-density lipoprotein cholesterol to high-density lipoprotein cholesterol ratio, RCS = restricted cubic spline.

3.5. Subgroup analysis

To assess the stability of the NHHR–hs-CRP association and explore potential subgroup differences, we performed stratified analyses by age, sex, ethnoracial group, education, BMI, alcohol use, smoking, diabetes, and hypertension (Fig. 3). After adjusting for relevant covariates within each subgroup, we found that the positive association between NHHR and hs-CRP was significantly stronger in females (β = 0.51, 95% CI: 0.31–0.71, P < .001) compared to males (β = −0.003, 95% CI: −0.12 to 0.12, P = .96), with a statistically significant interaction (P for interaction < .001). Similarly, the association was more pronounced in participants with hypertension (β = 0.49, 95% CI: 0.23−0.75, P = .001) than those without (P for interaction = .008).

Figure 3.

Subgroup analyses of the association between NHHR and hs-CRP. All models were adjusted for age, sex, race, education level, smoking status, alcohol consumption, BMI, diabetes, and hypertension, except the stratification variable. Due to limited sample size in the underweight group (n = 84), only key covariates were included in its model. BMI = body mass index, hs-CRP = high-sensitivity C-reactive protein, NHHR = non-high-density lipoprotein cholesterol to high-density lipoprotein cholesterol ratio.

Interactions were not statistically significant across other subgroup variables, including age, race, education, BMI, alcohol use, smoking status and diabetes, with no significant interaction detected (all P > .05). Of note, due to the limited sample size in the underweight group (n = 84), the regression model for this group included only age, sex, diabetes, and hypertension to ensure model stability.

4. Discussion

In this large, nationally representative exploratory cross-sectional study using NHANES 2015–2018 data, we found that higher NHHR was positively associated with hs-CRP, an established marker of systemic inflammation, even after adjusting for a comprehensive set of demographic, behavioral, and metabolic covariates including age, sex, ethnicity, educational attainment, smoking and alcohol behavior, BMI, and the presence of diabetes and hypertension. These findings reflect statistical associations in cross-sectional data and should not be interpreted as evidence of a causal or predictive relationship. Although this association was statistically robust across multiple models, the magnitude of the effect was modest, suggesting that NHHR is likely one of several contributors to low-grade systemic inflammation rather than a dominant causal determinant.

In fully adjusted models, both the continuous and quartile-based analyses supported a positive association between NHHR and hs-CRP, with higher quartiles of NHHR quartiles corresponding to higher hs-CRP levels. The RCS analysis further indicated a positive but nonlinear, inverted U-shaped association, with estimated hs-CRP values increasing up to log(NHHR) = 1.17 and then plateauing or slightly declining. This suggests a threshold effect, wherein moderately elevated NHHR levels may be associated with the greatest systemic inflammatory burden. These cross-sectional associations were generally consistent across subgroups defined by age, sex, and BMI strata, and are consistent with previous studies linking higher NHHR to inflammatory and metabolic disorders such as type 2 diabetes, sarcopenia, and hyperuricemia.

The observed association between NHHR and hs‑CRP likely reflects a complex interplay among lipid metabolism, inflammation, and metabolic regulation. Based on previous experimental and clinical studies, NHHR, defined as non‑HDL-C/HDL-C, offers a holistic measure of lipid asymmetry – capturing both atherogenic (e.g., LDL, VLDL, oxidized LDL) and anti-atherogenic (HDL) lipid particles.^[41,42] Non-HDL particles-especially oxidized LDL, can trigger endothelial cell activation by engaging scavenger and Toll-like receptors, thereby inducing oxidative stress and stimulating pro-inflammatory cytokine release via pathways such as NF-κB.^[43,44] These signals induce hepatic CRP synthesis, raising systemic hs‑CRP levels.^[45] Conversely, HDL exhibits anti-inflammatory and antioxidant functions, including inhibiting monocyte adhesion, reducing reactive oxygen species, and suppressing pro-inflammatory cytokine release.^[46,47] A lower HDL component in NHHR thus eliminates this protective buffer, tilting the balance toward systemic inflammation.^[48] Moreover, CRP can amplify inflammatory signaling by binding to endothelial Fc gamma receptors, which activates downstream NF-κB pathways.^[49] This activation leads to the degradation of IκB-α and increased expression of interleukin-8,^[50] thereby promoting further inflammation and creating a self-reinforcing cycle of CRP production and vascular activation. The inverted U-shaped pattern observed suggests a potential threshold effect: moderate increases in NHHR may maximize inflammatory activation, but further rises may activate compensatory mechanisms or reflect lipid function saturation. Similar inflection points have been reported in studies linking NHHR with metabolic disorders and COPD.^[51]

Subgroup analyses confirmed the overall robustness of the association between NHHR and hs‑CRP, while also revealing meaningful variations across population groups. The association appeared particularly stronger among women, which is consistent with extensive evidence suggesting that females tend to exhibit more pronounced lipid-related inflammatory responses. These differences may arise from variations in body fat distribution and the modulatory effects of sex hormones on inflammatory pathways.^[52] Furthermore, one study reported that, compared to traditional lipid biomarkers, indicators such as non-HDL-C, the non-HDL-C to HDL-C ratio, and the TC to HDL-C ratio showed greater discriminatory power for predicting 10-year CVD risk – especially in women – further supporting the potential utility of NHHR in sex-specific risk stratification.^[42] Among hypertensive individuals, the relationship also appeared substantially strengthened. This observation mirrors previous findings that dyslipidemia and elevated CRP exert synergistic effects in the presence of hypertension, thereby amplifying vascular inflammation and cardiovascular risk.^[53] Although interactions by age, BMI, diabetes, or smoking status were not statistically significant, slight trends in former smokers and current drinkers suggest that lifestyle factors may modulate the NHHR–inflammation axis and warrant further research. Clinically, these subgroup nuances reinforce the importance of interpreting NHHR within demographic and clinical contexts, particularly for tailoring risk stratification strategies.

Beyond our main findings, evidence from US and other Western cohorts provides additional support for the relationship between dyslipidemia, adiposity, and systemic inflammation. Large NHANES-based analyses have consistently shown that elevated CRP levels are strongly linked to metabolic abnormalities such as central obesity, dyslipidemia, insulin resistance, and hypertension.^[54,55] Likewise, major Western population cohorts have demonstrated that atherogenic lipid patterns and impaired HDL function are closely linked to systemic inflammation and increased cardiometabolic risk, as shown by the strong associations between inflammatory biomarkers and incident coronary heart disease reported by Pai et al,^[56] and by the reduced cholesterol efflux capacity associated with greater atherosclerotic burden described by Khera et al^[57] These findings align with our observations and reinforce the biological plausibility of lipid-driven low-grade inflammation in US adults. At the same time, BMI categories may not fully capture central obesity or the broader metabolic syndrome phenotype, so part of the observed NHHR–hs-CRP association could reflect unmeasured variation in waist circumference, triglyceride levels, or related metabolic abnormalities. In line with contemporary views of obesity as a heterogeneous condition that depends not only on total body mass but also on adipose tissue distribution and quality, different adiposity phenotypes may exhibit distinct patterns of so-called low-grade inflammation even at similar BMI levels.

Collectively, these results from this exploratory cross-sectional analysis suggest that NHHR is associated with systemic inflammatory status and may have potential as a feasible and integrative marker of low-grade inflammation. Given its availability from standard lipid panels and its association with hs-CRP, NHHR might help to identify individuals who tend to have higher hs-CRP levels. The observed sex- and hypertension-specific differences further suggest that the clinical interpretation of NHHR should consider demographic and clinical context. As a noninvasive and cost-effective indicator, NHHR warrants further evaluation in longitudinal studies to determine whether it can meaningfully contribute to risk stratification and prevention strategies targeting low-grade inflammation and lipid-related disorders.

5. Study strengths and limitations

This study has several notable strengths. First, it utilized data from NHANES, a large, nationally representative survey with rigorous sampling design and standardized data collection, enhancing the generalizability of our findings to the US adult population. Second, the use of complex survey-weighted regression models allowed for appropriate estimation accounting for sampling design and population structure. Third, we incorporated both categorical and continuous forms of NHHR and employed restricted examined nonlinear relationships using RCSs, which provided a more nuanced characterization of the NHHR–hs-CRP association. Furthermore, subgroup and interaction analyses provided additional insight into population heterogeneity.

Despite these strengths, several limitations merit consideration. Due to the cross-sectional design, we cannot establish temporality or infer causality between NHHR and hs-CRP, and the observed cross-sectional associations may be influenced by reverse causation and unmeasured confounding. hs-CRP was measured at a single time point, which may not fully reflect long-term inflammatory status. Furthermore, although we adjusted for a wide range of demographic, lifestyle, and metabolic covariates, the possibility of residual confounding cannot be fully excluded. In particular, obesity and metabolic syndrome are only partially captured by BMI categories. Key indicators of central adiposity and metabolic status, such as waist circumference, triglyceride levels, and formal definitions of metabolic syndrome, were not incorporated into the final models due to limited availability or substantial missingness in the NHANES cycles used. These unmeasured factors may lead to residual confounding by central obesity and metabolic syndrome components and could partially influence the observed cross-sectional association, which should be explored in future longitudinal studies with more comprehensive metabolic profiling. Finally, our mechanistic interpretations are hypothesis-generating and require confirmation in prospective cohort studies and experimental research. Although residual Q–Q plots indicated some right-skewness and slightly heavier tails at higher hs-CRP values, the large sample size, use of survey weighting, and consistent results across multiple sensitivity analyses (including exclusion of hs-CRP > 10 mg/L, log-transformed hs-CRP, and alternative spline specifications) suggest that any minor departures from model assumptions are unlikely to have materially biased our estimates.

6. Conclusion

Using nationally representative NHANES 2015–2018 data, our findings suggest that NHHR may be modestly and non-linearly associated with hs-CRP, a marker of systemic inflammation. The inverted U-shaped dose–response pattern indicates that individuals with moderately elevated NHHR may experience higher levels of low-grade inflammation. However, the effect sizes were relatively small, and predicted hs-CRP values largely remained within the range of low-grade systemic inflammation. In interpreting these findings, it is important to note that obesity in our analyses was operationalized using conventional BMI categories, whereas contemporary concepts of obesity increasingly emphasize adiposity indices, body fat distribution, and visceral fat rather than BMI alone. Indices of central or visceral adiposity may therefore shift the hs-CRP profile and modify what is considered “low-grade” inflammation across different obesity phenotypes. Taken together, these findings suggest that NHHR may provide incremental information on systemic inflammatory status rather than serving as a stand-alone clinical indicator. Given that NHHR can be readily derived from routine lipid panels, it may still be a useful, low-cost adjunct for identifying individuals at potentially higher cardiometabolic risk, but its clinical significance should be interpreted cautiously, and the associations observed in this exploratory cross-sectional analysis require confirmation in future longitudinal cohort studies and should be regarded as hypothesis-generating rather than causal or predictive.

Author contributions

Conceptualization: Hao Wang, Yu Peng.

Data curation: Hao Wang, Jia Li.

Formal analysis: Hao Wang.

Methodology: Hui Chen.

Project administration: Yu Peng.

Resources: Jia Li.

Software: Hui Chen.

Supervision: Yu Peng.

Validation: Hui Chen.

Visualization: Hui Chen.

Writing – original draft: Hao Wang, Hui Chen.

Writing – review & editing: Jia Li, Yu Peng.