This study retrospectively analyzed 590 individuals diagnosed with T2DM who received treatment at Hefei First People’s Hospital between January 2022 and August 2024 (Fig. 1). The inclusion criteria were as follows: (1) a verified diagnosis of T2DM according to the 2025 American Diabetes Association (ADA) guidelines, (2) age of 18 years or older, and (3) availability of complete clinical data, including Lp(a) measurements and PAD status. Exclusion criteria included (1) history of cardiovascular events not related to PAD (such as myocardial infarction, stroke), (2) presence of severe hepatic or renal dysfunction (such as decompensated liver cirrhosis and uremia), (3) active infection or malignancy, and (4) presence of serious diabetic complications, including diabetic ketoacidosis, diabetic foot ulcers, or advanced diabetic retinopathy. Eligible participants were identified using T2DM diagnostic codes extracted from the hospital’s electronic medical record system. After initial electronic screening, all cases underwent manual verification to ensure conformity with the defined inclusion and exclusion parameters. The research was carried out in compliance with the ethical principles outlined in the Declaration of Helsinki. The study protocol received approval from the Ethics Committee of Hefei First People’s Hospital [Approval No. 2022 (74)], and all subjects provided informed consent before participation.

Fig. 1.

Flowchart of patient selection and grouping process. T2DM, type 2 diabetes mellitus; PAD, peripheral arterial disease; Lp(a), lipoprotein(a).

In this study, essential patient information—including age, gender, smoking habits, clinical characteristics, and laboratory test results—was obtained. Smoking status was determined through the patients’ electronic medical records and classified as smoker (yes) or non-smoker (no). Hypertension was identified either by a prior diagnosis or by elevated blood pressure measured during hospitalization—defined as systolic blood pressure (SBP) $\ge$140 mmHg or diastolic blood pressure (DBP) $\ge$90 mmHg [22]. Hyperlipidemia was determined based on either current lipid panel results or previously documented clinical history. Medication history, including the use of lipid-lowering drugs and antihypertensive drugs, was extracted from the patients’ past medical records and recorded as yes or no. Chronic kidney disease (CKD) was determined by either a sustained estimated glomerular filtration rate (eGFR) below 60 mL/min/1.73 m² or the presence of kidney damage markers lasting over three months [23]. Blood pressure (SBP and DBP), heart rate, and body mass index (BMI) were obtained using established protocols and documented. Fasting glucose (mmol/L) and HbA1c (%) were used to assess the patients’ glycemic control. Blood lipid analysis included evaluation of triglycerides, total cholesterol, and both low-density lipoprotein cholesterol (LDL-C) and high-density lipoprotein cholesterol (HDL-C), with results expressed in mmol/L. Kidney function was assessed via the eGFR, calculated using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) formula [24]. Serum uric acid was measured in $\mu$mol/L and included as a continuous variable in the analysis.

PAD was diagnosed based on a combination of clinical symptoms, physical examination, and objective diagnostic tests, with the ankle-brachial index (ABI) being the primary assessment tool [25]. ABI was calculated by dividing the SBP at the ankle by the SBP at the brachial artery. Following a minimum 10-minute rest in the supine position, SBP was assessed bilaterally at the upper limbs and at the posterior tibial and dorsalis pedis arteries in the lower limbs, using a handheld Doppler ultrasound device (Bidop ES-100V3, Hadeco Inc., Kawasaki, Japan). An ABI value of less than 0.90 was diagnostic of PAD [25].

To assess Lp(a), an immunoturbidimetric assay was employed, utilizing antigen–antibody interactions that produce a turbidity signal proportional to Lp(a) content [26]. Participants provided fasting blood samples following an overnight fast of $\ge$8 hours. Absorbance at 340 nm was recorded using an automated analyzer (Hitachi 7600, Hitachi High-Technologies Corporation, Tokyo, Japan), with measurements processed via a commercial kit (KAI-017, Sekisui Medical Co., Ltd., Tokyo, Japan). Based on measured values, individuals were categorized into tertiles: T1 ($\le$132 mg/L), T2 (132–340 mg/L), and T3 ($>$340 mg/L), ensuring comparable group sizes. This stratification enabled exploration of Lp(a)-related variation in PAD risk, supporting detection of potential dose-dependent relationships.

Data analyses were executed by R (version 4.1.4, R Foundation for Statistical Computing, Vienna, Austria) and SPSS (version 26.0, IBM Corp., Armonk, NY, USA). The Shapiro-Wilk procedure was utilized to assess the normality assumption for continuous variables. For variables that followed a normal distribution, results were described using means and standard deviations (SDs), and group differences were evaluated through one-way analysis of variance (ANOVA). Levene’s test was employed to examine equality of variances. In contrast, skewed data were summarized as medians and interquartile ranges (IQR), and non-parametric comparisons were conducted via the Kruskal-Wallis method. Categorical variables were expressed as counts (n) and percentages (%), and group differences were assessed using the chi-square ($\chi$²) test.

The relation between Lp(a) and PAD was assessed through a four-step hierarchical logistic-regression strategy that produced odds ratios (ORs) and 95% confidence intervals (CIs): Model 1 (baseline): no covariate adjustment; Model 2 (+demographics): Model 1 plus age and sex; Model 3 (+lifestyle/clinical): Model 2 plus smoking status, hypertension, antihypertensive treatment, and CKD; Model 4 (full adjustment): Model 3 plus heart rate, BMI, DBP, fasting glucose, HbA1c, triglycerides, eGFR, and uric acid. Covariates were chosen from univariable analyses in which p 0.05 signified potential confounding. Multicollinearity was checked with variance-inflation factors (all 2.5), confirming variable independence. Robustness was probed by repeating the analysis within strata defined by age ($\le$60 vs. $>$60 years), sex and the presence of hypertension, hyperlipidemia or CKD. Predictive performance of Lp(a) for PAD was quantified with receiver operating characteristic (ROC) curves and the corresponding area under the curve (AUC). Possible non-linear dose-response patterns were investigated with a restricted cubic spline (RCS). Statistical significance was set at a two-tailed p-value 0.05.

Table 1 summarized baseline data: participants were separated into Lp(a) tertiles (T1–T3), and their baseline profiles were contrasted across these categories. First, the incidence of PAD differed significantly between the groups with varying Lp(a) levels (p = 0.001). The prevalence of PAD rose with higher Lp(a) concentrations. Specifically, in the T1 group (Lp(a) $\le$132 mg/L), the PAD incidence was 13.6%, while in the T3 group (Lp(a) $>$340 mg/L), the incidence increased to 28.6%. Second, age likewise varied by tertile (p = 0.020): participants in T1 averaged 61.92 years, whereas those in T2 and T3 were, on average, older than 64 years. Additionally, hypertension and the use of antihypertensive drugs differed significantly between the groups (p = 0.037 and p = 0.017, respectively), with the T1 group showing a higher proportion of hypertension (78.3%) compared to the other groups. Notably, CKD also showed a clear step-wise rise with higher Lp(a) (p 0.001), climbing from 14.1% in T1 to 35.2% in T3. Moreover, LDL-C and total cholesterol levels significantly increased with higher Lp(a) levels, while eGFR showed a significant decline as Lp(a) levels increased (p 0.05). No inter-group variation was observed in sex distribution, smoking prevalence, BMI, heart rate, blood pressure, plasma glucose, or HbA1c (p $>$ 0.05).

Univariate logistic regression (Table 2) revealed that age, sex, smoking, hypertension, use of antihypertensive drugs, CKD, BMI, heart rate, DBP, fasting glucose, HbA1c, triglycerides, eGFR, and uric acid were all significantly associated with PAD risk (p 0.05).

As presented in Table 3, higher Lp(a) concentrations were independently correlated with an increased likelihood of PAD. Without covariate adjustment (Model 1), individuals in the highest tertile (T3) exhibited a markedly greater risk of PAD relative to those in the lowest group (T1), with an OR of 2.533 (95% CI: 1.520–4.222; p 0.001). This association persisted even after comprehensive adjustment in Model 4, which accounted for demographic factors, clinical conditions, and biochemical markers—including age, sex, smoking, hypertension, antihypertensive drugs, CKD, heart rate, BMI, DBP, glucose, HbA1c, triglycerides, eGFR, and uric acid—yielding an adjusted OR of 1.961 (95% CI: 1.071–3.588; p = 0.029).

Treating Lp(a) as a continuous metric—whether assessed per unit, on a logarithmic scale, or standardized by SD—consistently revealed a robust link with PAD risk across all regression models. In the crude analysis (Model 1), each 1 mg/L increase in Lp(a) corresponded to an OR of 1.001 (95% CI: 1.001–1.002, p 0.001), with this effect persisting after successive adjustments in Models 2 through 4 (p 0.05). Log-transformation of Lp(a) further strengthened the association, yielding an OR of 2.585 (95% CI: 1.582–4.225, p 0.001) in Model 1. Although attenuation was noted in the fully adjusted model, statistical significance was maintained (OR = 1.803, 95% CI: 1.013–3.209, p = 0.045). Similarly, when analyzed per SD, Lp(a) remained a significant predictor of PAD, with Model 1 showing an OR of 1.423 (95% CI: 1.186–1.707, p 0.001), and Model 4 yielding a slightly lower but still meaningful OR of 1.257 (95% CI: 1.016–1.555, p = 0.035).

As presented in the subgroup analysis (Table 4), elevated Lp(a) levels were significantly linked to increased PAD risk within multiple population strata. In individuals older than 60, those in the highest tertile (T3) exhibited a markedly greater likelihood of developing PAD relative to the lowest tertile (T1), with an OR of 1.869 (95% CI: 1.006–3.469, p = 0.048). In male patients, the PAD risk in the T3 group also significantly increased (OR = 2.131, 95% CI: 1.087–4.177, p = 0.028). Similarly, in patients with hypertension, the PAD risk in the T3 group was significantly elevated (OR = 1.996, 95% CI: 1.080–3.690, p = 0.028). For patients with hyperlipidemia, the OR for PAD in the highest

Lp(a) tertile (T3) was 2.364 (95% CI: 1.115–5.013, p = 0.025). Likewise, among those without CKD, the T3 group showed a similarly elevated PAD risk, with an OR of 2.479 (95% CI: 1.189–5.170, p = 0.015).

As shown in Fig. 2, the ROC curve demonstrated the discriminative capacity of Lp(a) for identifying PAD. The AUC reached 0.622 (95% CI: 0.568–0.677; p 0.001), suggesting a statistically significant yet modest predictive performance. An optimal threshold of 279 mg/L was identified, corresponding to 59.8% sensitivity and 62.4% specificity.

Fig. 2.

The ROC curve of Lp(a) for predicting PAD. The ROC curve, depicted in blue, reflected the diagnostic performance of Lp(a), whereas the red diagonal served as the reference line (AUC = 0.5). With an AUC of 0.622 (95% CI: 0.568–0.677; p 0.001), Lp(a) demonstrated limited but statistically significant capacity to distinguish individuals with PAD. ROC, receiver operating characteristic; PAD, peripheral arterial disease; Lp(a), lipoprotein(a); AUC, area under the curve; CI, confidence interval.

As illustrated in Fig. 3, a strong linear relationship between Log₁₀Lp(a) and PAD risk was observed in Model 1 (unadjusted), with statistical significance (p 0.001) and no indication of nonlinearity (p for nonlinearity = 0.437). This pattern remained evident after controlling for age and sex in Model 2 (p = 0.004; p for nonlinearity = 0.386). Even with further adjustments in Model 3—including smoking status, hypertension, antihypertensive therapy, and CKD—the association persisted (p = 0.029), and the linear trend remained (p for nonlinearity = 0.367). Finally, in Model 4, which included full adjustments for heart rate, BMI, DBP, fasting glucose, HbA1c, triglycerides, eGFR, and uric acid, the significant association continued (p = 0.031) without nonlinearity (p for nonlinearity = 0.266).

Fig. 3.

The restricted cubic spline plots of Lp(a) with PAD in four models. The x-axis represented Log₁₀-transformed Lp(a) concentrations, while the y-axis showed the OR with 95% CI for PAD. The red solid lines indicated the fitted spline curves, shaded areas represented the 95% CI, and the grey bars displayed the distribution of participants. A significant association between Lp(a) and PAD risk was found in all models, with no indication of nonlinearity (p for nonlinearity $>$ 0.05). (A) Model 1 (baseline): no covariate adjustment. (B) Model 2 (+demographics): Model 1 plus age and sex. (C) Model 3 (+lifestyle/clinical): Model 2 plus smoking status, hypertension, antihypertensive treatment, and CKD. (D) Model 4 (full adjustment): Model 3 plus heart rate, body mass index, diastolic blood pressure, fasting glucose, hemoglobin A1c, triglycerides, estimated glomerular filtration rate, and uric acid. Lp(a), lipoprotein(a); PAD, peripheral arterial disease; OR, odds ratio; CI, confidence interval.