The data of this study were obtained from the NHANES 1999–2004, which is a large, multilevel, cluster probability sampling survey that collects health examination data from the non-institutionalized population in the USA. Beginning in 1999, NHANES was conducted as annual surveys, with data being released every 2 years. A more detailed information can be found at https://wwwn.cdc.gov/nchs/nhanes.htm. In the present study, subjects with missing values of ABI, blood glucose, total cholesterol, and other interested covariates were excluded from the analyses. To minimize the potential confounding from prior morbidity, we excluded participants with low cardiovascular fitness level (11). Finally, three consecutive NHANES cycles (1999–2000, 2001–2002, and 2003–2004, N = 31,126) were analyzed, and a total of 1,370 participants older than 40 years with complete date were included in our analysis. The available data had been weighed demographically, and the NHANES study protocol was approved by the National Center for Health Statistics ethics review board with all participants providing written informed consent.

Briefly, the subjects lie supine on the exam table, and their systolic pressure are measured on the right arm (brachial artery) and both ankles (posterior tibial arteries) (12). If the participant has a rash or open wound on the right arm, dialysis shunt, right-sided radical mastectomy or any other condition that would interfere with accurate measurement or would cause discomfort to the participant, the left arm is used for the brachial pressure measurement. Systolic blood pressure is measured twice at each site for participants aged 40–59 years and once at each site for participants aged 60 years and older. If a health technician is unable to obtain a reading at a site, they may attempt another reading at the same site after a brief resting period. PAD was defined as ABI ≦0.9.

Each participant completed a PA questionnaire that was based on the Global Physical Activity Questionnaire (GPAQ) (13). The questionnaire collected the type, frequency (number of days per month), and duration (amount of time spent on a typical day) of PA in the past 30 days for a minimum of 10 min, including moderate- and vigorous-intensity recreational activities, moderate- and vigorous-intensity walking/bicycling for transportation and work. Vigorous-intensity activities were activities that cause significant increases in breathing or heart rate, while moderate-intensity activities were those that cause mild increase in breathing or heart rare. Work-related PA referred to paid or unpaid work, household chores, and yard work. Recreational PA referred to sports, fitness, and recreational activities. For the active transportation domain, information was collected on walking and bicycling activities.

Metabolic equivalent (MET) is a unit that describes the energy consumption while performing a specific activity and 1 MET is equal to 3.5 ml O₂ kg⁻¹ min⁻¹. NHANES suggested that MET for vigorous work-related activity was 8.0, moderate work-related activity 4.0, walking or bicycling for transportation 4.0, vigorous leisure-time PA 8.0 and moderate leisure-time PA 4.0, respectively (14). For each activity, PA was calculated by multiplying the number of days by the mean duration by the recommended MET and then summed the values to obtain a value of total PA.

The basic characteristics of the involved subjects, including age (categorized into 40–64, 65–75, 75–85, and ≧85 years), gender (male or female), race (Mexican American, non-Hispanic White, non-Hispanic Black, or other race), socioeconomic characteristics including education level (<9th grade, 9–11th grade, high school graduate, some college, or AA degree or above), marital status (married/living with partner, widowed/divorced/separated, or never married), family poverty-to-income ratio (PIR ≦ 100% or >100%, PIR ≦ 100% indicates poverty), body mass index (BMI <18.5, 18.5 to <25.0, 25.0 to <30.0, or <30 kg m⁻²), were all analyzed to control the bias. In addition, the risk factors for PAD such as smoking (yes or no), drinking alcohol (yes or no), blood total cholesterol, and plasma fasting glucose were also homogenized. Except for BMI (15), blood total cholesterol (16), and plasma fasting glucose (17), all covariates were self-reported. BMI data were measured by NHANES technicians. For plasma fasting glucose test, participants were examined in the morning after fasting at least 8 h or more but <24 h. Blood specimens are processed, stored, and shipped to the University of Missouri—Columbia for analysis. For total cholesterol test, vials were shipped to the Johns Hopkins University Lipoprotein Analytical Lab for testing. Detailed specimen collection and processing instructions are presented in the NHANES Laboratory/Medical Technologists Procedures Manual.

According to the NHANES analytic guidelines (18), sampling weight was incorporated to account for its complex design. Data are presented as mean with SD or median with interquartile range (IQR) according to their distributions. Categorical variables were presented as frequency. Student's t-test or Mann–Whitney test were applied to compare continuous variates between groups. Chi-square test was employed to detect differences in categorical variates between groups. Rank-sum test was utilized to compare ordinal categorical variates between groups. Multivariate logistic regressions were conducted to explore the independent relationship between PA and ABI values, while adjusting for potential confounders. A trend test was used to observe linear trends between PA and the incidence of PAD, which is described previously (19).

RCS were used to detect the dose–response relationship between PA and ABI values, using the 25, 50, and 75th percentiles of the distribution of PA as fixed knots. Non-linear relationship between PA and the incidence of PAD was examined using RCS models. Non-linearity was tested for using likelihood ratio test to compare two models: one including only a linear effect and the second also including cubic spline terms (20, 21). The RCS models were adjusted for age, sex, race, education level, marital status, family PIR, health insurance, smoking status, drinking-alcohol status, BMI, total cholesterol, and plasma fasting glucose. Analyses stratified by sex were conducted. R software version 3.53 (R Foundation for Statistical Computing, Vienna, Austria) and SAS 9.4 (Cary, NC) were used for the analyses, and p < 0.05 was considered statistically significant.

The prevalence of PAD was 6.2% of all individuals, 5.6% for male and 6.9% for female. The median MET were lower in participants with PAD than that without PAD (2,100.0 vs. 3,652.5 p < 0.05). The distribution of variables of interest is listed in Table 1. The incidence of PAD is significantly associated with age (p < 0.0001), education level (p < 0.05), smoking (p < 0.01), and plasma fasting glucose (p < 0.01). Overall, subjects with PAD were more likely to be older, widowed/separated/divorced, and to have a lower education level, a higher smoking proportion, and higher fasting plasma glucose level. However, there was no statistical differences in gender, race, family PIR, total cholesterol, and drinking alcohol between groups (all p > 0.05).

The distribution of MET quartiles is listed in Table 2, and the odds ratio (OR) and 95% CIs of the MET quartiles for PAD are listed in Table 3. The participants were divided into four groups according to the calculated MET. The first quartile (Q1) participants were used as a reference group (OR values = 1.0). The univariate analysis showed that compared with the Q1 of MET, with the increase in MET, the ORs of PAD for those with Q2, Q3, and Q4 of MET were 0.618 (95% CI = 0.614–0.621), 0.451 (95% CI = 0.449–0.454), and 0.597 (95% CI = 0.594–0.600), respectively (all p < 0.0001). The results indicated that, with the increase in PA, the likelihood of PAD decreased more than 50% for Q3 participants and 40% for Q4 participants.

Then we included age, sex, education level, marital status, family PIR, smoking, drinking alcohol, BMI, total cholesterol, and fasting plasma glucose into the model. After adjusting for these confounding factors, the odds ratio for those with Q2, Q3, and Q4 of MET compared with Q1 of MET were 0.688 (95% CI = 0.684–0.692), 0.463 (95% CI = 0.460–0.466), 0.816 (95% CI = 0.812–0.821), respectively (all p < 0.001).

The RCS model showed a non-linear negative dose–response correlation between PA (i.e., MET) and the OR for PAD (p for non-linearity = 0.0001) (Figure 1). The OR for PAD decreased as the PA increased, reaching the minimum at ~5,800 MET-min month⁻¹ (OR = 0.5012, 95% CI = 0.5008–0.5015). After exceeding 5,800, the OR for PAD increased with excessive PA. Furthermore, the gender-stratified dose–response analysis (Figure 2) showed that PA and the incidence of PAD presented non-linear trends for both males and females similar to the total population (all p for non-linearity < 0.0001). For males, the curve decreases more sharply than that of females, and there is no obvious plateau. However, for females, the curve decreases more slowly reaching the minimum at ~5,800 MET-min month⁻¹. After exceeding 5,800 MET-min month⁻¹, the OR starts to increase. These findings suggest that males can do more PA, which can benefit their ABI, but for females, proper amount of PA is important, and excessive exercise may reduce the protective effects of PA.

We further analyzed the PA patterns of each group, we found that a higher proportion (Table 2) of individuals had no vigorous-intensity PA in the PAD group compared with individuals without PAD (p < 0.001). Next, we divided all the participants into two groups by the ratio of MET of vigorous/moderate (V/M) intensity PA. The V/M ratio for participant without vigorous PA in the past 1 month was 0 and >0 for participant with vigorous PA. The ORs and 95% CIs are listed in Table 4. The univariate analysis showed that compared with the V/M = 0 group, the OR of PAD for the V/M > 0 group was 0.422 (95% CI = 0.420–0.424). After adjusting for confounding factors, the OR was 0.667 (95% CI = 0.663–0.670) (all p < 0.001). These results indicated that proper vigorous PA could benefit peripheral artery health.