Development and validation of a nomogram for predicting calcification of arteriovenous access in hemodialysis patients

ABSTRACT:

Background

In patients with end-stage renal disease (ESRD), vascular calcification significantly impairs hemodialysis (HD) vascular access functionality, compromising both dialysis efficacy and long-term patency. Early risk prediction of vascular calcification facilitates timely clinical interventions to preserve vascular access integrity.

Methods

A cross-sectional analysis was performed. Risk factors for vascular calcification in CKD patients were identified from the literature and Kidney Disease: Improving Global Outcomes guidelines. All variable selection and model training procedures were conducted on the training set. Univariate logistic regression was performed for all candidate variables. A nomogram was then constructed based on the final multivariate logistic model to facilitate clinical interpretation.

Result

A total of 136 HD patients were included. The predictive model, relying on arteriovenous (AV) access usage time, hip circumference, and diabetes status, is reliable and clinically actionable tool for predicting AV access calcification. Its robust performance across validation and subgroup analyses supports its potential for integration into routine clinical practice.

Conclusion

This study developed a nomogram-based predictive model for calcification, providing a simple, cost-effective, and reliable tool for early risk assessment. Monitoring hip circumference may serve as a practical approach for identifying high-risk patients, allowing for timely intervention and improved vascular access outcomes.

1. Introduction

End-stage renal disease (ESRD) is a significant global public health challenge. According to the International Society of Nephrology’s 2019 Global Kidney Health Atlas cross-sectional survey, the average global incidence of ESRD is 144 cases per million individuals in the general population [1]. Approximately 78% of ESRD patients require dialysis as renal replacement therapy, with 89% undergoing hemodialysis (HD) and 11% receiving peritoneal dialysis [2]. For HD patients, vascular access is critical for effective treatment. The success of hemodialysis depends on reliable vascular access, which not only enhances dialysis quality but also improves patient survival [3]. Common types of dialysis vascular access include autogenous arteriovenous fistulas (AVFs), arteriovenous grafts (AVGs), and central venous catheters (CVCs). The 2019 update of the Kidney Disease Outcomes Quality Initiative (KDOQI) Clinical Practice Guidelines for Vascular Access recommends AV access (AVF or AVG) over CVC due to a lower risk of infection and associated complications [4].

Calcification of AV access in HD patients is a complex and multidimensional problem for which management often requires addressing multiple parameters [5]. Therefore, early prediction is crucial to prevent its onset and delay progression, ultimately prolonging AV access longevity. Recent studies have found that plasma desphosphorylated uncarboxylated matrix gla protein, serum sclerostin, and serum MicroRNA-125b can predict vascular calcification in ESRD, but primary hospitals visits and high costs limited the access for patients [6–9]. Compared to biological markers, anthropometric measurements for predicting vascular calcification are noninvasive, inexpensive, and easy to perform. Waist circumference and waist-to-hip ratio have been reported to be significantly associated with vascular calcification [10,11]. However, to the best of our knowledge, the association between anthropometric measurements and calcification of AV access in HD patients has not been reported.

A nomogram is an intuitive graphical tool based on regression analysis that integrates multiple variables to estimate the probability of a specific event. Widely regarded as a practical and accessible statistical method for risk assessment [12], nomograms enhance the visualization of complex risk factors in medical practice, facilitating interpretation and clinical decision-making [13].

Based on a cross-sectional study, we explored the association between anthropometric measurements and calcification and developed a nomogram for AV access calcification. The nomogram provides a reliable, evidence-based approach for predicting the calcification risk of AV access, supporting early intervention strategies.

2. Materials and methods

2.1. Study population

This cross-sectional study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Medical Ethics Committee. Written informed consent was obtained from all participants. The study was conducted between July and November 2024. The inclusion criteria include patients who were: (1) aged ≥ 18 years, (2) undergoing hemodialysis for ≥ 3 months, (3) with a life expectancy of more than 12 months. The exclusion criteria include patients who were: (1) aged ≥ 75 years, (2) using a central venous catheter (CVC) for hemodialysis, (3) AV access usage time ≤ 1 month, (4) had insufficient data, including past medical history, anthropometric measurements and data about AV access. A total of 136 patients were enrolled. The patient recruitment process and study design are illustrated in Figure 1.

Figure 1.

Flow chart.

2.2. Data collection and definition

AV access calcification refers to vascular calcification that occurs in AVF or AVG, which can be detected by X-ray, computed tomography (CT), and ultrasound. CT offers high accuracy but is limited by radiation exposure and cost, whereas dual-energy X-ray absorptiometry (DXA) is a versatile imaging modality with lower radiation risk but reduced precision compared to CT. Ultrasound, despite a lower positive detection rate than CT and DXA, was selected for this study due to its noninvasive nature, absence of radiation, and cost-effectiveness [14,15]. On ultrasound, AV access calcification was defined as hyperechoic lesions near the AV wall, with or without posterior shadowing. Examinations were performed by three experienced vascular sonographers using a Clover 60 portable color Doppler ultrasound system (Wisonic, China). Prior to the examination, an assistant recorded key AV access information, including whether it was the patient’s first access, AV access usage time, location, type of vascular access, and history of percutaneous transluminal angioplasty. Ultrasound parameters included AV access depth, brachial artery blood flow, arterial/venous diameters, and calcification presence. All measurements were obtained in triplicate, with mean values used for analysis.

Risk factors for vascular calcification in CKD patients were identified from literature [16–20], including age, sex, weight, duration of hemodialysis, hypertension, diabetes, waist/hip circumference, nutritional status, serum calcium/phosphorus, parathyroid hormone (PTH), albumin, hemoglobin, blood urea nitrogen, and total cholesterol. Additionally, indicators recommended by the Kidney Disease: Improving Global Outcomes (KDIGO) guidelines, including: glomerular filtration rate, urinary albumin, cystatin C, blood glucose, lactate dehydrogenase (LDH), serum ferritin, and a history of hypertension and diabetes [21], were included to enhance clinical applicability.

Demographic data (gender, age, body mass index (BMI), marital/employment status, educational level, income) and medical history (duration of hemodialysis, comorbidities) were extracted from electronic medical records and patient questionnaires. Anthropometric measurements included hip, waist, upper arm, thigh, and calf circumferences, obtained by trained researchers using standardized protocols with a single-vendor soft tape measure. Three independent investigators recorded measurements at different time points, with final values averaged. Laboratory data was collected within three months of ultrasound evaluation. Nutritional status and quality of life were collected via grip strength tests, the Kidney Disease Quality of Life Short Form (KDQOL-SF 1.3), and the Subjective Global Assessment (SGA). KDQOL-SF 1.3 and SGA data were collected via questionnaires distributed through the WeChat subscription account ‘Health Dialysis’ [22]. Grip strength was also measured in triplicate within one week, with mean values calculated.

2.3. Statistical analysis

Data preprocessing was performed prior to model development. Missing values were imputed using the MICE (Multivariate Imputation by Chained Equations) package in R. Continuous variables were tested for normality using the Shapiro–Wilk test. Descriptive statistics were used to compare baseline characteristics between patients with and without AV access calcification. Normally distributed variables were compared using independent t-tests (reported as mean ± standard deviation (SD)), while non-normally distributed variables were analyzed via Mann–Whitney U-tests (reported as median and interquartile range (IQR)). Categorical variables were described as counts and percentages, and group comparisons were conducted using chi-square tests or Fisher’s exact tests. Following initial data characterization, numerical features were standardized using z-score normalization. Additionally, Spearman correlation coefficients were calculated to assess pairwise associations between variables.

Then, we followed these steps for variable selection and model development: (1) The dataset was randomly split into a training set (70%) and a validation set (30%) to ensure balanced proportions of calcification cases between groups. All variable selection and model training procedures were conducted on the training set to minimize overfitting and ensure unbiased evaluation on the validation set. Elastic net regularization was additionally performed to address potential collinearity in high-dimensional data, with details described in the Supplementary Materials. (2) Univariate logistic regression was performed for all candidate variables, and those with p < 0.05 were considered for inclusion in the multivariate model. Additionally, several key comorbidities recommended by the 2024 KDIGO guidelines and demographic information were included regardless of their univariate significance. (3) A multivariate logistic regression model was refined using bidirectional stepwise selection (forward and backward) based on the Akaike Information Criterion (AIC). A nomogram was then constructed based on the final multivariate logistic model to facilitate clinical interpretation.

Model performance was evaluated in the validation set. The area under the receiver operating characteristic curve (AUC) was calculated, with 95% confidence intervals (CI) estimated using bootstrapping. Sensitivity, specificity, positive and negative predictive values, and Youden’s index were computed to evaluate classification accuracy. The Hosmer–Lemeshow test and calibration curves were used to assess agreement between predicted probabilities and observed outcomes. A Brier score was calculated to quantify overall prediction accuracy. Decision curve analysis (DCA) and clinical impact curves were used to evaluate net benefit across threshold probabilities. To ensure generalizability, subgroup analyses were conducted, including males, females, duration of hemodialysis (< 10 years), and a complete-case dataset without imputation. All statistical analyses were conducted using R software (version 4.4.1, R Foundation for Statistical Computing, Vienna, Austria), analyses were performed with two-tailed tests, and p-values < 0.05 were considered statistically significant.

3. Result

3.1. Clinical characteristics of participants

In the baseline characteristics (Table 1), a total of 136 patients were included, with 27 (19.9%) in the AV access calcification group and 109 (80.1%) in the non-calcification group. Gender differed significantly, with a higher female proportion in the calcification group (63.0% vs. 34.9%, p = 0.02). The AV access usage time (9.00 vs. 3.00 years, p < 0.01) and hemodialysis duration (9.00 vs. 3.00 years, p < 0.01) were longer in the calcification group. Anthropometrically, patients with calcification had lower weight (54.30 vs. 60.00 kg, p = 0.02) and shorter height (1.60 vs. 1.64 m, p < 0.01). No significant differences were noted in age, marital status, educational level, income, hypertension, diabetes, coronary heart disease, cerebrovascular disease, or vascular access location (p > 0.05).

Table 1.

Baseline characteristics.

Variables	Non-calcification	Calcification	P
Case, n (%)	109 (80.1)	27 (19.9%)
Gender, n (%)			0.02
Male	71 (65.1)	10 (37.0)
Female,	38 (34.9)	17 (63.0)
Age, n (%)			0.93
≤ 60,	61 (56.0)	16 (59.3)
≥ 60	48 (44.0)	11 (40.7)
Marital status, n (%)			0.27
Single	1 (0.9)	0 (0.0)
Married	78 (71.6)	24 (88.9)
Divorced	24 (22.0)	3 (11.1)
Widowed	6 (5.5)	0 (0.0)
Degree, n (%)			0.13
Junior High School or Below	52 (47.7)	14 (51.9)
High School/Vocational School	28 (25.7)	4 (14.8)
College/Associate Degree	20 (18.3)	3 (11.1)
Bachelor’s Degree	9 (8.3)	6 (22.2)
Unemployed, n (%)	97 (89.0)	25 (92.6)	0.84
Income (yuan), n (%)			0.55
<5000	94 (86.2)	23 (85.2)
5000–10000	14 (12.8)	3 (11.1)
>10000	1 (0.9)	1 (3.7)
Hypertension, n (%)	94 (86.2)	24 (88.9)	0.96
Diabetes, n (%)	44 (40.4)	9 (33.3)	0.65
Coronary heart disease, n (%)	13 (11.9)	2 (7.4)	0.74
Cerebrovascular disease, n (%)	10 (9.2)	3 (11.1)	1.00
Left Arm, n (%)	95 (87.2)	22 (81.5)	0.65
Forearm, n (%)	105 (96.3)	27 (100.0)	0.71
AVF, n (%)	104 (95.4)	27 (100.0)	0.57
AV access usage time (year), median [IQR]	3.00 [1.00, 5.00]	9.00 [7.50, 10.00]	<0.01
Duration of hemodialysis (year), median [IQR]	3.00 [1.00, 5.00]	9.00 [7.50, 10.00]	<0.01
Weight (kg), median [IQR]	60.00 [52.00, 69.00]	54.30 [47.75, 59.85]	0.02
Height (m), mean (SD)	1.64 (0.07)	1.60 (0.06)	<0.01

AV = arteriovenous; AVF = arteriovenous fistula.

3.2. Variable selection

We considered vascular access calcification as the dependent variable and evaluated its association with clinical indicators (Table 2). Univariate analysis identified AV access usage time (OR = 1.66, 95% CI 1.32 − 2.09, p < 0.01), duration of hemodialysis (OR = 1.63, 95% CI 1.31 − 2.04, p < 0.01), venous vessel diameter (OR = 1.36, 95% CI 1.17 − 1.58, p < 0.01), female (OR = 3.39, 95% CI 1.16 − 9.90, p = 0.03) as risk factors. Conversely, height (OR = 0.01, 95% CI 0.00 − 0.31, p = 0.03), grip strength (OR = 0.92, 95%CI 0.85 − 1.00, p = 0.05), and hip circumference (OR = 0.93, 95% CI 0.86 − 1.00, p = 0.04) acted as protective factors. In multivariate analysis, AV access usage time (OR = 1.85, 95% CI 1.37 − 2.49, p < 0.01), hip circumference (OR = 0.9, 95% CI 0.83 − 0.99, p = 0.03), and diabetes (OR = 3.55, 95% CI 0.73–17.34, p = 0.12) were retained. Although diabetes showed borderline significance (p = 0.12), it was included based on clinical guidelines (2024 KDIGO) and its contribution to model fit, serving as a critical covariate to enhance the model’s clinical relevance and predictive performance. Other univariate-significant variables were excluded due to collinearity (Supplementary Table 1).

Table 2.

Influencing factors of calcification.

Variable	Univariate Logistic Regression			Multivariate Logistic Regression
	OR	95% CI	P	OR	95% CI	P
AV access usage time	1.66	(1.32 − 2.09)	<0.01	1.85	(1.37 − 2.49)	<0.01
Duration of hemodialysis	1.63	(1.31 − 2.04)	<0.01
Venous vessel diameter	1.36	(1.17 − 1.58)	<0.01
Female	3.39	(1.16 − 9.90)	0.03
Height	0.01	(0.00 − 0.31)	0.03
Grip strength	0.92	(0.85 − 1.00)	0.05
Hip circumference	0.93	(0.86 − 1.00)	0.04	0.9	(0.83 − 0.99)	0.03
Age	1.02	(0.37 − 2.83)	0.97
Weight	0.97	(0.93 − 1.01)	0.19
Hypertension	0.89	(0.22 − 3.56)	0.87
Diabetes	0.81	(0.28 − 2.36)	0.7	3.55	(0.73 − 17.34)	0.12
Coronary heart disease	1.18	(0.22 − 6.18)	0.85
Cerebrovascular disease	2.22	(0.50 − 9.83)	0.29

3.3. Development and validation of the predictive model

The predictive model, incorporating three key variables, AV access usage time, hip circumference, and diabetes (Table 2 and Figure 2), demonstrated robust performance. In the validation set, the model achieved an Area Under the Curve (AUC) of 0.90 (95% CI: 0.79–0.99), indicating excellent discriminatory ability (Figure 3(A)). The optimal classification threshold, determined by Youden’s J statistic, was 0.076, balancing sensitivity (1.00) and specificity (0.72) effectively. The model’s overall accuracy was 0.78, with a positive predictive value (PPV) of 0.47 and a negative predictive value (NPV) of 1.00, highlighting its precision in identifying high-risk patients (Table 3).

Figure 2.

Calcification prediction nomogram.

Figure 3.

Evaluating predictive efficacy and clinical utility of model.

Table 3.

Performance of the prediction model.

Dataset	AUC (95% CI)	Accuracy	Sensitivity	Specificity	Positive predictive value	Negative predictive value
Validation set	0.90 (95% CI: 0.79 − 0.99)	0.78	1.00	0.72	0.47	1.00
Subgroup-Male	0.96 (95% CI: 0.91–1.00)	0.90	0.93	0.63	0.96	0.50
Subgroup-Female	0.78 (95% CI: 0.64 − 0.92)	0.78	0.81	0.69	0.89	0.53
Duration of hemodialysis < 10 years	0.88 (95% CI: 0.82 − 0.95)	0.88	0.92	0.65	0.93	0.59
Complete subset	0.89 (95% CI: 0.82–0.96)	0.88	0.91	0.75	0.93	0.67

The calibration curve (Figure 3(B)) showed close alignment between predicted probabilities and observed outcomes, with low prediction errors (mean absolute error = 0.032, mean squared error = 0.063). While a sharp segment in the curve suggested minor calibration discrepancies in intermediate probability ranges, the Hosmer–Lemeshow test (p = 0.67) confirmed overall good calibration. The Brier score of 0.11 further validated the model’s precision in risk estimation.

The clinical impact curve (Figure 3(C)) demonstrated high sensitivity at low threshold probabilities, effectively capturing true positives. As thresholds increased, specificity improved, reducing false positives. Decision curve analysis (DCA) (Figure 3(D)) revealed that the model outperformed "treat all" and "treat none" strategies across clinically relevant thresholds, underscoring its utility in optimizing patient management while minimizing overtreatment.

To assess generalizability, subgroup analyses were conducted across four distinct populations (Table 3). Subgroup analyses revealed varying model performances: in the male subgroup, the model achieved a high AUC of 0.96 (95% CI: 0.91–1.00), accompanied by superior accuracy (0.90), sensitivity (0.93), and positive predictive value (PPV, 0.96), outperforming other subgroups. In contrast, the female subgroup showed a moderate AUC of 0.78 (95% CI: 0.64–0.92), with acceptable accuracy (0.78) but a lower negative predictive value (NPV, 0.53), indicating potential gender-related variability. For patients with hemodialysis duration < 10 years, the model demonstrated strong discriminative ability (AUC = 0.88, 95% CI: 0.82–0.95) and high accuracy (0.88), with an NPV of 0.59, reflecting robust performance in this cohort. Meanwhile, the complete-case dataset (without imputation) yielded an AUC of 0.89 (95% CI: 0.82–0.96), with accuracy (0.88) and NPV (0.67) comparable to the validation set, confirming model stability across data imputation strategies. Overall, the model exhibited consistent efficacy in most subgroups, though the slightly reduced performance in females (lower AUC and NPV) merits further exploration of gender-specific risk factors or model refinements for this population.

In summary, the final model, relying on AV access usage time, hip circumference, and diabetes status, is reliable and clinically actionable tool for predicting AV access calcification. Its robust performance across validation and subgroup analyses supports its potential for integration into routine clinical practice.

4. Discussion

AV access is the lifeline for HD patients, and calcification is a critical factor contributing to AV access failure [23]. Identifying risk factors for AV access calcification and developing predictive models can help reduce calcification incidence, prolong fistula lifespan, and improve both hemodialysis efficiency and patient survival. However, research on AV access calcification remains limited.

In this study, univariate regression analysis identified AV access usage time, duration of hemodialysis, venous vessel diameter, gender, height, grip strength, and hip circumference as factors associated with AV access calcification. In multivariate analysis, AV access usage time, hip circumference, and diabetes were retained. The predictive model, relying on three key variables, is reliable and clinically actionable tool for predicting AV access calcification. And its robust performance across validation and subgroup analyses supports its potential for integration into routine clinical practice.

In HD patients, the incidence of vascular calcification is significantly higher in those with diabetes compared to non-diabetic individuals [24,25]. As an integral component of the systemic vasculature, diabetes undoubtedly influences the calcification of the AV access. The mechanisms by which diabetes contributes to AV access calcification in HD patients are complex and multifactorial. Notably, in patients with diabetes and ESRD, medial arterial calcification occurs, characterized by the deposition of calcium-phosphate crystals and hydroxyapatite in the middle layer of the arterial intima. This process is mediated by calcification inducers such as Bone Morphogenetic Protein (BMP) and Fetuin-A [26,27]. Furthermore, advanced glycation end-products (AGEs) are significantly elevated in individuals with diabetes and CKD, particularly in those undergoing dialysis. This increase is attributed to augmented production, impaired excretion, and inefficient removal of AGEs. AGEs interact with their receptors to induce oxidative stress, further promoting the differentiation of vascular smooth muscle cells into osteoblast-like cells, thereby facilitating vascular calcification [28,29]. Additionally, hyperglycemia enhances the production of reactive oxygen species (ROS) and activates the polyol pathway via protein kinase C (PKC), which in turn stimulates pro-inflammatory cytokine production, contributing to the development of calcification [30].

Adipose tissue distribution may also influence vascular metabolic risk. Previous studies have demonstrated that a higher waist-to-hip ratio (WHR) is associated with an increased risk of arterial calcification in non-dialysis CKD patients [31]. WHR is a widely used and validated indicator of fat distribution, and it can be modulated by increasing the denominator (hip circumference). Hip circumference serves as an indicator of lower body fat accumulation. Unlike upper body fat, which is primarily visceral, lower body fat accumulation is inversely associated with metabolic risk factors and is linked to a lower incidence of cardiovascular disease [32]. A cohort study in CKD patients found that peripheral fat, including hip fat, may offer protection against cardiovascular disease compared to visceral fat [11]. Moreover, increased hip and leg fat mass has been associated with a lower risk of aortic calcification and atherosclerosis, as well as a reduced progression of aortic calcification [33]. In our study, we observed a negative correlation between hip circumference and the occurrence of AV access calcification in HD patients, supporting the notion that lower body fat accumulation may have a protective effect on the vasculature. This protective role may be attributed to the capacity of lower body fat to function as a "metabolic reservoir," buffering dietary lipid influx and protecting other tissues from lipotoxicity associated with lipid spillover and ectopic fat deposition [34,35]. Additionally, gluteal fat has been shown to exhibit "resistance" to the pro-inflammatory and low metabolic profile characteristic of abdominal fat by secreting more beneficial adipokines and fewer pro-inflammatory molecules, thereby creating a protective adipokine profile [36,37].

Biologic sex is another important risk factor for AV access calcification. The 2025 Hemodialysis Vascular Access Core Curriculum identifies female sex as an independent risk factor for fistula failure in HD patients [38]. This effect may be mediated through sex hormones, as more than half of the women in our study were premenopausal and under the age of 50. Endogenous estrogen has been shown to inhibit the proliferation of vascular smooth muscle cells, prevent the development of atherosclerotic plaques, and inhibit vascular calcification via the vascular RANKL system [39,40]. Other studies have similarly confirmed the influence of gender on vascular health. For instance, a cohort study by Ellen Boakye observed a higher incidence of aortic valve calcification and stenosis in men compared to women, even in elderly individuals with a mean age of 80 years [41]. This suggests that factors beyond hormonal regulation may contribute to sex differences in vascular calcification. To assess the performance of our prediction model, we stratified the population by gender. In female patients, the negative predictive value was only 0.53. The model exhibited poor performance in negative predictions and was prone to false-positive results, particularly in the overall cohort and in the male subgroup. However, we believe that these limitations are outweighed by the clinical benefits. False-positive predictions can increase clinician and patient awareness, prompting proactive measures such as regular vascular color Doppler ultrasound screening, which may facilitate the early detection of pre-calcified lesions and micro-calcification.

The parameters of the nomogram included diabetes, hip circumference, and AV access usage time. These can be easily collected in clinical practice. Therefore, according to the patient’s previous medical history, it is recommended to increase the frequency of vascular ultrasound to detect early vascular calcification in high-risk patients. In addition, dynamic monitoring of hip circumference is recommended, especially in obese patients with small hip circumference and large waist circumference, and increased frequency of vascular calcification screening is also recommended.

Several limitations must be acknowledged in this study. First, the small sample size and single-center design not only limited generalizability but also posed a risk of imbalanced distribution of key variables (e.g., age, gender, comorbidity profiles), which may have affected the statistical power and internal validity of the results. Multi-center, large-sample studies are needed to validate its clinical utility and reliability. Second, the mechanism of vascular calcification in HD patients is complex, this study only investigated a subset of influencing factors. Confounding factors such as diet, drugs, and dialysis on calcium and vascular calcification were not fully controlled, which may affect the performance of the model. Furthermore, as a cross-sectional study, we cannot establish causal relationships between the identified risk factors and AV access calcification.

Funding Statement

Not applicable.

Author contributions statement

CRediT: Xueying Li: Methodology, Writing – review & editing; Xiaocui Wang: Investigation, Writing – original draft; Bonan Yan: Data curation; Yuanke Zhou: Data curation; Ling Li: Investigation; Xiaopeng Huang: Formal analysis, Investigation; Qiqi Wang: Data curation, Investigation; Enjie Tang: Formal analysis, Methodology, Visualization.

Consent to participate

All patients have given informed consent and agreed to participate in the study.

Disclosure statement

This study is compliance with STROBE guidelines. The authors report there are no competing interests to declare.