This cross-sectional study extracted the data of 2,081 patients with diabetes from the NHANES database. Conducted by the Centers for Disease Control and Prevention’s National Center for Health Statistics, the NHANES performed a comprehensive monitoring of the nation’s nutrition and health status through direct physical examinations, clinical and laboratory tests, personal interviews, and related measurement procedures. The examinations are conducted in mobile examination centers that travel to various locations throughout the country, ensuring a standardized environment for the health examinations (14). In our study, diabetes was diagnosed by fasting glucose/HbA1c, physician diagnosis, or those who had anti-diabetic drug. The excluded criteria were 1) <18 years, 2) without measurement of the left or right ankle brachial pressure index (ABPI), and 3) without measurement of RDW or albumin. Finally, 1,125 participants were included.

Age (<65 years or ≥65 years), gender (male or female), race (non-Hispanic White, non-Hispanic Black, or others), education [less than 9th grade, 9th to 11th grade (includes 12th grade with no diploma), high school graduate/general equivalent diploma (GED), or equivalent or some college or Associate of Arts (AA) degree/college graduate or above], poverty-to-income ratio (≤1.0, 1.0–2.0, >2.0, or unknown), marriage (never married, married, or others), physical activity [<450 metabolic equivalent of task (MET) × min/week, ≥450 MET × min/week, or unknown], smoking (yes or no), drinking (<once/week, ≥once/week, or no), hypertension (yes or no), dyslipidemia (yes or no), CVD (yes or no), diabetic retinopathy (yes or no), CKD (yes or no), family history of CVD (yes or no), body mass index (BMI) (<25 kg/m², 25 kg/m² to 30 kg/m², or ≥30 kg/m²), waist circumference (cm), energy (kcal), hemoglobin (g/dL), C-reaction protein (mg/dL), anti-platelet drug (yes or no), anti-coagulants drug (yes or no), adrenal cortical steroids (yes or no), and diabetes drug (yes or no).

Physical activity was converted into energy consumption based on the questionnaire in the database. Energy consumption (MET × min) = recommended MET × exercise time of corresponding activity (min), which can be converted into weekly energy consumption, and divided into three categories including <450 MET × min/week, ≥450 MET × min/week, and unknown. BMI was grouped into normal (18.5 kg/m² to 24.9 kg/m²), underweight (<8.5 kg/m²), overweight (25 kg/m² to 29.9 kg/m²), and obese (≥30 kg/m²). Because there were only two people in the underweight group, the underweight group was combined with the normal group.

The RAR was the main variable that was calculated on the basis of RAR. The continuous variable and the categorical variable of RAR were used for analysis. As a categorical variable, RAR was divided into <2.80, 2.80–2.98, 2.98–3.25, and ≥3.25 according to quarters. RDW was divided into <12.18%, 12.18%–12.56%, 12.56%–13.09%, and ≥13.09%, whereas albumin was divided into <3.95 g/dL, 3.95 g/dL to 4.19 g/dL, 4.19 g/dL to 4.39 g/dL, and ≥4.39 g/dL according to the respective quarters. PAD was the outcome, which was diagnosed on the basis of the left or right ABPI <0.9 (15).

Kolmogorov–Smirnov normality test was used for quantitative data. Normally distributed measurement data were described as mean (standard error) [mean (SE)], independent sample t-test was used for comparison between two groups, and analysis of variance was used for comparison between multiple groups. Non-normally distributed measurement data were described as median and quartiles [M (Q₁, Q₃)], and Kruskal–Wallis test was used for comparison among groups. The enumeration data were described as the numbers and percentages of cases [n (%)], Chi-square test was used for comparison between groups, and rank sum test was used for rank data. A weighted univariable logistic regression model was used to explore variables associated with PAD. With PAD as the outcome variable, a weighted logistic backward regression model was established. In Model I, no variable was adjusted; in Model II, age and gender were adjusted; and in Model III, age, gender, education, poverty-to-income ratio, smoking status, CVD, CKD, and anti-tuberculosis drug were adjusted. Missing values were manipulated via random forest using python miceforest package for interpolation processing ( Supplementary Table 1 ). Poverty-to-income ratio and physical activity had missing values >5%, and the missing values were classified as the unknown group. Sensitivity analysis was performed to compare the missing data before and after interpolation ( Supplementary Table 2 ). Subgroup analysis was stratified on the basis of age, CVD, and smoking status. The odds ratio (OR) and 95% confidence interval (CI) were effect size. Receiver operator characteristic curves of RAR, RDW, and albumin for the risk of PAD in patients with diabetes were plotted, and the areas under the curve (AUCs) were calculated and compared via Delong test. All statistical tests were conducted by a two-sided test with the test level α = 0.05. Python 3.9 was used for missing value processing, and SAS 9.4 (SAS Institute Inc., Cary, NC, USA) was used for model statistical analysis.

In total, the data of 2,081 patients with diabetes were retrieved from the NHANES database. Among them, participants whose age <18 years (n = 43), patients without measurement of left or right ABPI (n = 860), and those without measurement of RDW (n = 25) or albumin (n = 28) were excluded. Finally, 1,125 participants were included. The screen process is displayed in Figure 1 .

The mean RAR in the PAD group was higher than that in the non-PAD group (3.32 vs. 3.04). The percentages of patients with different RAR levels in the PAD group were different compared with that in the non-PAD group. The mean RDW in the PAD group was higher than that in the non-PAD group (13.50% vs. 12.78%). The mean albumin in the PAD group was lower than that in the non-PAD group (4.11 g/dL vs. 4.23 g/dL). The percentages of subjects with CKD in the PAD group was higher than that in the non-PAD group (27.88% vs. 5.07%). More detailed information is presented in Table 1 .

The results of weighted univariable logistic regression model revealed that age (OR = 4.82; 95% CI: 2.91–7.97), race (OR = 0.32; 95% CI: 0.17–0.60), education, poverty-to-income ratio, physical activity (OR = 2.29; 95% CI: 1.38–3.79), smoking status (OR = 2.60; 95% CI: 1.54–4.41), hypertension (OR = 4.94; 95% CI: 2.20–11.07), CVD (OR = 3.59; 95% CI: 2.08–6.21), CKD (OR = 7.23, 95% CI: 4.00–13.09), hemoglobin (OR = 0.76; 95% CI: 0.66–0.88), C-reaction protein (OR = 1.17; 95% CI: 1.04–1.32), anti-platelet drug (OR = 5.85; 95% CI: 2.60–13.15), anti-coagulants drug (OR = 3.54; 95% CI: 1.28–9.79), and diabetes drug (OR = 1.82; 95% CI: 1.06–3.13) were potential covariates associated with the risk of PAD in patients with diabetes ( Supplementary Table 3 ). Backward stepwise regression data indicated that age, gender, education, poverty-to-income ratio, smoking status, CVD, CKD, and anti-tuberculosis drug were covariates. In the crude model, increased RAR might be associated with the elevated risk of PAD in patients with diabetes. Compared with the RAR <2.80 group, RAR ≥3.25 might increase the risk of PAD in patients with diabetes. After adjusting for covariates, the risk of PAD in patients with diabetes was observed in those with higher RAR (OR = 1.83; 95% CI: 1.06–3.15). In addition, RAR ≥3.25 was related to the increased risk of PAD in patients with diabetes (OR = 2.04; 95% CI: 1.05–3.95) ( Table 2 ). No significant association between RDW and the risk of PAD in patients with diabetes was observed (P > 0.005). Moreover, the association between albumin and the risk of PAD in patients with diabetes was not statistically different (P > 0.005) ( Table 2 ). The AUC of RAR (AUC = 0.631; 95% CI: 0.588–0.675) for the risk of PAD in patients with diabetes was higher than that of albumin (AUC = 0.572; 95% CI: 0.526–0.618) ( Figure 2 , Table 3 ). No significant difference was found between the AUC of RAR and that of RDW (AUC = 0.649; 95% CI: 0.608–0.690) for the risk of PAD in patients with diabetes (P > 0.05) ( Figure 2 , Table 3 ).

In people with diabetes aged ≥65, RAR was a risk factor for PAD with an OR value of 2.67 (95% CI: 1.30–5.46). RAR ≥3.25 was associated with the increased risk of PAD (OR = 3.06; 95% CI: 1.15–8.11) relative to RAR <2.80. No significant association between the RAR and the risk of PAD was found in patients with diabetes <65 years (P > 0.05). In people with diabetes who smoked, the risk of PAD was elevated in those with RAR ≥3.25 (OR = 2.85; 95% CI: 1.28–6.32). The association between the RAR and the risk of PAD was not statistically different in non-smoking patients with diabetes (P > 0.05). As for patients with CVD, the risk of PAD was elevated as the increase of RAR (OR = 2.31; 95% CI: 1.05–5.10). RAR ≥3.25 was correlated with the increased risk of PAD (OR = 3.75; 95% CI: 1.42–9.87). No association was identified in people with diabetes who are not complicated with CVD (P > 0.05) ( Table 4 ).