NHANES data has been an ongoing program since the early 1960s under the U.S. National Center for Health Statistics (NCHS). It employs a stratified multistage probability sampling technique to collect health and nutrition data (26). A key objective of NHANES is to extensively gather examination data from the U.S. elderly population to enhance our understanding of this demographic. The NCHS Ethics Review Board approved the NHANES survey protocol, and all participants provided written informed consent before their inclusion.

This study initially included cross-sectional data from 82,298 participants from the NHANES (2009–2020). The exclusion criteria were set as follows: (1) participants without stroke data (n = 82,298); (2) participants without albumin data (n = 5,240); (3) participants without globulin data (n = 62); (4) participants without BMI data (n = 60). A total of 52,119 participants were included in the analysis. A flowchart of participant enrollment is presented in Figure 1.

The method used by the DxC800 to measure albumin concentration is a two-color digital endpoint method. In the reaction, albumin binds to bromocresol purple (BCP) reagent to form a complex. The system monitors the change in absorbance at 600 nm. The change in absorbance is proportional to the concentration of albumin in the sample. Serum A/G = Albumin (g/L)/Globulin (g/L).

Participants were considered to have a history of stroke if they reported being previously informed by a doctor or other healthcare professional that they had a stroke (27, 28).

The baseline variables in this study included age, gender, ethnicity, diabetes, smoking status, body mass index (BMI), hypertension, income, education level, coronary heart disease, alcohol consumption habits, stroke history, and physical exercise. Ethnicity is divided into five categories Mexican American, Other Hispanic, Non-Hispanic White, Non-Hispanic Black and Other race. Being diagnosed with diabetes means that a doctor or other health professional has informed you that you have diabetes. Smoking is defined as subjects who smoked at least 100 cigarettes in their entire life. BMI stands for Body Mass Index, calculated as weight in kilograms divided by height in meters squared (kg/m^2). Normal weight is considered less than 25 kg/m^2, overweight is classified as having a BMI between 25 and 30 kg/m^2, and obesity is defined as having a BMI of 30 kg/m^2 or higher. Hypertension is defined as a condition characterized by a systolic blood pressure (SBP) over 140 mmHg and/or a diastolic blood pressure (DBP) over 90 mmHg. If a doctor or other health care professional has told you that you have high blood pressure and that you are recommended to take prescription medication, answering yes to both questions is a diagnosis of high blood pressure. Education level is divided into six categories: less than 9th grade, 9th-11th grade, high school graduation/GED or equivalent, partial college or AA degree, college graduation or higher, and others. Coronary heart disease is defined as having been told by a doctor or other health professional that you have congestive heart failure. A person is considered to be a drinker if he or she has had at least 12 alcoholic beverages of any type in his or her lifetime. Conditions such as hypertension, diabetes, coronary heart disease, and stroke were determined either through self-report or a doctor’s diagnosis. Details of all study variable measurement procedures are available at https://www.cdc.gov/nchs/nhanes/.

Sample sizes were determined based on existing NHANES data. All statistical analyses were performed according to CDC guidelines using appropriate NHANES sampling weights and considered the complexity of the multistage cluster survey design. Continuous variables are summarized as means with standard errors (SE), and categorical variables are presented as percentages. Multivariate regression analysis was conducted based on tertiles of baseline characteristics or selected past studies. Model 1 was unadjusted. Model 2 adjusted for sex, age, and race. Model 3 adjusted for age, sex, ethnicity, diabetes, smoking habits, BMI, hypertension, income, education level, coronary heart disease, alcohol consumption, and physical exercise. Subgroup analyses were conducted according to sex, hypertension, diabetes, smoking status, race, and BMI categories, which were also considered potential effect modifiers, to examine whether there was a significant interaction between these covariates and A/G and stroke. In addition to this, we performed smoothed curve fitting and threshold effects to analyze and predict the non-linear relationship between A/G and stroke. All analyses were conducted using empowerstats 2.0.¹ The significance level was set at p < 0.05.

This study involved 52,119 participants, comprising 25,216 males (48.38%) and 26,903 females (51.62%). Table 1 summarizes the baseline characteristics of participants stratified by tertiles of baseline serum A/G levels. Among all participants, 1,264 (2.43%) had a history of stroke, with a decreasing incidence correlating with higher A/G ratios. Participants in the highest A/G tertile were more likely to be male, Non-Hispanic White, non-smokers, have a normal BMI, be free of hypertension, have a higher income, and hold a college degree or higher. Higher A/G ratios were also associated with lower incidences of moderate exercise, diabetes, and coronary heart disease. Overall, the baseline data were stable.

Table 2 lists results from multivariable logistic regression analyses exploring the relationship between stroke and variables such as A/G, diabetes, hypertension, BMI, income, physical exercise, and education level. In Models 1 and 2, significant positive correlations were observed between stroke, A/G, and diabetes (p < 0.05). In Model 3, this positive correlation remained statistically significant within the highest tertile of A/G (OR = 0.01; 95% CI, 0.00–0.02; p = 0.0302), with a 1% increase in stroke incidence corresponding to an increase in the highest A/G tertile. Across all three models, stroke was positively correlated with higher tertiles of hypertension, income, and education, showing significant differences.

To assess the consistency of the relationship between A/G ratios and stroke across the general population and to identify potential differences across demographic subgroups, we conducted subgroup analyses and interaction tests stratified by gender, hypertension, diabetes, smoking, ethnicity, and BMI. The results indicated an inconsistent relationship between the two. As shown in Table 3, significant interactions were found between A/G score and gender, and smoking (P for interaction <0.05), but no interaction effects were observed for hypertension, diabetes, ethnicity, or BMI. The relationship between A/G and stroke showed significant positive correlations across the stratified subgroups of gender, hypertension, diabetes, smoking, and BMI (p < 0.05). However, in the ethnicity-stratified subgroup, a statistically significant association was observed only among non-Mexican, non-American populations (β = 0.65; 95% CI, 0.43–1.00; p = 0.0477). Our findings suggest that the association between A/G ratios and stroke depends on gender, hypertension, diabetes, smoking, and BMI conditions.

Specifically, interaction effects were noted for hypertension (p = 0.0048) and BMI (p = 0.0447) in this relationship. Within the hypertension-stratified subgroup, each unit increase in A/G was associated with a 3% increase in stroke incidence among participants without hypertension (OR = 0.03; 95% CI, 0.00–0.55; p = 0.0172). No significant differences were observed in the interaction tests for gender, diabetes, smoking, and ethnicity (P for interaction >0.05).

The effect of A/G ratio on the incidence of stroke was stratified by sex, race, history of diabetes, history of hypertension, and body mass index, and risk ratio analyses and interaction tests were performed. Our results indicate that these associations were not consistent, as shown in Figure 2, where we detected a significant interaction for sex (p < 0.05) and no significant interaction for the rest. The association between A/G ratio and stroke was negatively associated with risk ratios for sex, other Hispanic origin, history of diabetes and hypertension, and BMI subgroups. In the association between A/G ratio and stroke, patients with diabetes and hypertension had a higher risk than non-diabetics and non-hypertensives, while the risk ratio also increased with increasing BMI.

Through curve-fitting threshold analysis, we identified a nonlinear relationship between A/G ratios and stroke incidence. The threshold (K) for the A/G ratio was determined to be 1.88 g/L. To the left of this threshold, no statistically significant relationship was found between A/G and stroke (β = 0.01; 95% CI, −0.00-0.03; p = 0.1309). However, to the right of the threshold, a positive correlation was observed between A/G and stroke (β = 0.06; 95% CI, 0.03–0.09; p = 0.0002). The log-likelihood ratio test, before and after adjustment for covariates, yielded a p-value of 0.017 (see Figure 3).

We also explored whether there was a potential sex differences between the A/G ratio and stroke prevalence. In men, the association between A/G and stroke was nonlinear and positive. The threshold (K) for the A/G ratio was determined to be 1.55 g/L. To the left of this threshold, there was a statistically significant relationship between A/G and stroke (β = 0.21; 95% CI, 0.06–0.35; p = 0.0048). However, to the right of the threshold, a positive association between A/G and stroke was observed (β = 0.12; 95% CI, 0.05–0.18; p = 0.0003). The log-likelihood ratio test yielded a p value of 0.059 before and after adjusting for covariates. In women, the association between A/G and stroke was also nonlinear and positive. The threshold (K) for the A/G ratio was determined to be 1.48 g/L. To the left of this threshold, there was a statistically significant relationship between A/G and stroke (β = 0.26; 95% CI, 0.06–0.46; p = 0.0098). However, to the right of the threshold, a positive association was observed between A/G and stroke (β = 0.13; 95% CI, 0.03–0.23; p = 0.0112). The log-likelihood ratio test yielded a p value of 0.024 before and after adjusting for covariates (see Figure 4).