Introduction

Diabetic foot ulcers (DFUs) are among the most disabling complications of diabetes.1 With the rising global prevalence of diabetes, DFUs affect 19% to 34% of patients with long-standing disease (>10 years).2 DFUs are the leading cause of non-traumatic lower-extremity amputation (LEA), accounting for around 80% of such procedures and conferring a substantially increased risk of amputation.3 The 5-year mortality after major amputation is reported to be 45–55%, comparable to that seen in several solid malignancies.4 In the French CODIA cohort of 3102 hospitalized DFU patients, the 12-month cumulative incidence of death and LEA was 21.6% and 24.2%, respectively, driven by age, comorbidity burden, and systemic inflammation.5 A nationwide German analysis of 3.3 million insured individuals showed that infected DFUs carry a 5-year mortality of 64.0%, compared with 51.3% in non-infected DFUs, alongside markedly higher rates of minor (25.4% vs 3.0%) and major amputation (6.7% vs 1.2%).6

Despite standard care—glycemic control, debridement, and off-loading—clinical outcomes remain disappointing.7 Ulcer recurrence within one year exceeds 40%, and for severe ischemic cases (Wagner ≥3), the risk of amputation is markedly elevated.2,7 Early identification of patients at high risk of amputation is therefore essential to enable timely escalation of care, multidisciplinary referral, and individualized intervention.

Nomograms convert multivariable regression models into easy-to-use bedside tools and have been increasingly applied to DFU risk prediction.8 Yet several issues limit the use of existing models in everyday practice. Many rely on variables that are not consistently available in primary-care or resource-limited settings, such as the ankle-brachial index, transcutaneous oxygen pressure, or advanced vascular imaging. Others use only a single round of variable selection (typically univariate filtering or plain stepwise regression), and the variables retained therefore depend heavily on the particular sample at hand.9 Reporting practices are also uneven, with many studies predating the TRIPOD+AI 2024 guidance.10 A simpler, more reproducible tool built on routinely available indicators is therefore needed to support early risk stratification at the point of hospital care. Because the pathophysiology and case mix of type 1 and type 2 diabetes differ, and type 2 diabetes accounts for the large majority of DFU admissions at our center, we restricted the present analysis to patients with type 2 diabetes.

In this study we set out to build such a tool. Candidate predictors were limited to inexpensive, routinely available clinical and laboratory variables, and variable selection was stabilized using LASSO together with bootstrap resampling. The resulting four-variable nomogram was developed and internally validated in a single-center cohort of hospitalized patients with type 2 diabetes-related DFUs, with in-hospital LEA as the outcome, and is reported in line with both the TRIPOD+AI 2024 statement and the RECORD statement for observational studies using routinely collected health data.11

Methods Study Participants

This study is a retrospective cohort study, collecting patients with diabetic foot ulcers who visited the Hand and Foot Surgery Department of Linyi Central Hospital from January 2019 to June 2025. This study has been approved by the Ethics Committee of Linyi Central Hospital (No.: LCH-LW-2024102). The inclusion and exclusion criteria are shown in Table 1. The overall study workflow, including patient selection, cohort partitioning, variable selection, and model development, is illustrated in Figure 1.

Table 1 Inclusion and Exclusion Criteria

Figure 1 Flow diagram of the study. The training cohort was used for variable selection and model construction; the validation cohort was held out and applied only to the final nomogram for performance evaluation. λ.1se, the largest LASSO penalty value (λ) within one standard error of the value that minimizes the 10-fold cross-validation error, selected to yield a more parsimonious model.

Abbreviations: ALB, serum albumin; PLT, platelet count; SH, smoking history; HH, hypertension; CV, cross-validation; ROC, receiver operating characteristic; DCA, decision curve analysis.

Monitoring Indicators

Baseline clinical data and laboratory parameters were retrospectively collected. Demographic and clinical history variables included age, sex, body mass index (BMI), smoking history (SH), alcohol use history (AUH), and comorbidities such as hypertension (HH), coronary heart disease (CHD), and cerebrovascular disease (CVD). Laboratory indicators comprised routine blood parameters [white blood cell count (WBC), hemoglobin (Hb), platelet count (PLT), lymphocyte count (LYM), neutrophil percentage (NEUT%), neutrophil count (NEUT), platelet distribution width (PDW)], inflammatory markers [erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), D-dimer, plasma fibrinogen (FIB)], and biochemical markers [serum albumin (ALB), high-density lipoprotein (HDL), low-density lipoprotein (LDL), fasting plasma glucose (FPG), triglyceride (TG)]. All laboratory tests were performed using standardized hospital protocols on blood samples collected after at least 8 hours of fasting to minimize interference from dietary intake. All laboratory and clinical variables were those recorded at admission, before any decision on amputation or major surgical intervention, to avoid overlap between predictors and outcome. Complete blood counts were measured on an automated hematology analyzer, biochemical markers on an automated biochemistry analyzer, C-reactive protein by immunoturbidimetry, and the erythrocyte sedimentation rate by the Westergren method.

Statistical Analysis

The primary outcome was lower-extremity amputation (LEA) at any anatomical level during the index hospitalization, including both minor (below-ankle) and major (at or above the ankle) amputation. Of the 508 included patients, 119 (23.4%) underwent amputation. Model development, including variable selection, was performed in the training cohort, which contained 84 amputation events. The events-per-variable (EPV) is the number of outcome events divided by the number of predictors considered. Because 24 candidate variables were screened, the EPV at the variable-selection stage was low (approximately 3.5); for the four-variable final model the EPV was approximately 21 in the training set (84/4) and approximately 30 in the full cohort (119/4). The low selection-stage EPV is a regime in which any single selection method is unstable; the dual-method (LASSO plus bootstrap) intersection described below was adopted specifically to mitigate—though not fully eliminate—this instability, and internal validity was further assessed by bootstrap optimism correction. Continuous variables were summarized as mean ± SD or median (interquartile range) according to distribution and compared between groups using the t test or Mann–Whitney U-test; categorical variables were summarized as n (%) and compared using the χ2-test or Fisher’s exact test as appropriate. Variables with more than 20% missing values were excluded; those with ≤20% missing were imputed using multiple imputation by chained equations (m = 5 imputations) implemented in the mice R package, and pooled estimates were obtained according to Rubin’s rules. Missing data were infrequent: only three variables had any missing values (BMI, 62/508 [12.2%]; fasting plasma glucose, 35/508 [6.9%]; C-reactive protein, 24/508 [4.7%]); all remaining 21 variables, including the four retained in the final model (serum albumin, platelet count, hypertension, and smoking history), were complete in all 508 patients. Because overall missingness was low and the final predictors had no missing values, any risk of information leakage from pre-split imputation is expected to be negligible. The cohort was randomly partitioned into training (70%) and validation (30%) sets, stratified on the outcome using the caret::createDataPartition function. All analyses were performed in R version 4.4.3, using the glmnet (LASSO), MASS (stepwise), boot, mice (imputation), rms (nomogram, calibration), pROC (ROC analysis), ResourceSelection (Hosmer–Lemeshow), and rmda (decision curve analysis) packages. Two-sided P < 0.05 was considered statistically significant.

Variable selection was performed in two independent steps and the results were then compared. First, least absolute shrinkage and selection operator (LASSO) regression was applied to the 24 candidate predictors in the training set using 10-fold cross-validation. We used λ.1se rather than λ.min to favor a sparser model, and any variable with a non-zero coefficient at this value was retained. Second, 1000 bootstrap samples (with replacement) were drawn from the training set; within each sample, bidirectional AIC-based stepwise logistic regression was applied to the full set of 24 candidates, and the inclusion frequency of each variable was recorded across the 1000 runs. Variables selected in more than 80% of the bootstrap samples were considered stable. Variables retained by both procedures—LASSO at λ.1se and bootstrap inclusion >80%—were entered into the final multivariable logistic regression model used to build the nomogram. Discrimination was assessed using the receiver operating characteristic (ROC) curve.12 Calibration was evaluated graphically and quantitatively using the Hosmer–Lemeshow test with 10 groups. Clinical utility was assessed using decision curve analysis (DCA), which estimates the net benefit of model-guided decisions across threshold probabilities compared with the “treat-all” and “treat-none” strategies.13

Results Demographics and Clinical Characteristics

Baseline characteristics of the study population are summarized in Table 2. The training cohort consisted of 357 patients (273 non-amputation, 84 amputation), and the validation cohort included 151 patients (116 non-amputation, 35 amputation). No significant differences were observed between the training and validation cohorts for any of the variables (all Pb> 0.05); the distributions of baseline variables were generally similar between the two cohorts. In the training cohort, compared with non-amputation patients, those requiring amputation had significantly higher proportions of smoking (34.5% vs 21.2%, Pa = 0.020) and hypertension (51.2% vs 38.1%, Pa = 0.045). Additionally, the amputation group showed markedly elevated levels of white blood cell count, neutrophil count and percentage, platelet count, erythrocyte sedimentation rate, C-reactive protein, and fibrinogen (all Pa < 0.001), whereas levels of albumin, high-density lipoprotein, and low-density lipoprotein were significantly lower (all Pa < 0.01). In the validation cohort, similar patterns were observed for most inflammatory and nutritional markers (all Pa < 0.05 except for low-density lipoprotein, which showed a trend but did not reach significance, Pa = 0.053).

Table 2 Comparison of Variable Characteristics Between Training Cohort and Validation Cohort

Variable Selection

Of the 24 candidate predictors, LASSO regression with 10-fold cross-validation retained 15 variables with non-zero coefficients at λ.1se (λ.1se = 0.018; λ.min = 0.001; Figure 2A and B): serum albumin (ALB), platelet count (PLT), smoking history (SH), hypertension (HH), C-reactive protein (CRP), fasting plasma glucose (FPG), sex, body mass index (BMI), hemoglobin (Hb), neutrophil percentage (NEUT%), platelet distribution width (PDW), erythrocyte sedimentation rate (ESR), D-dimer, high-density lipoprotein (HDL-C), and low-density lipoprotein (LDL-C). The 1000 bootstrap stepwise regression replicates identified four variables exceeding the 80% selection-frequency threshold (Figure 2C): ALB and PLT, both selected in 100% of bootstrap replicates, together with HH (90%) and SH (90%). CRP, Hb, and neutrophil-related indicators showed moderate selection frequencies (57–70%), whereas age, sex, and HDL-C were rarely selected (≤20%). LDL-C, although retained by LASSO at λ.1se, did not exceed the 80% bootstrap selection threshold and was therefore not included in the final model. All four bootstrap-stable variables (ALB, PLT, HH, SH) were also among the LASSO-retained set at λ.1se. These four variables, supported by both selection procedures, were entered into the final multivariable logistic regression model used to construct the nomogram.

Figure 2 Dual-method variable selection. (A) LASSO coefficient path against log(λ), where λ is the LASSO penalty (regularization) parameter; (B) 10-fold cross-validation curve for LASSO with vertical dashed lines indicating λ.min and λ.1se (λ.min is the value of the penalty parameter λ that minimizes the 10-fold cross-validation error, and λ.1se is the largest λ within one standard error of λ.min, which yields a more parsimonious model); (C) bootstrap selection-frequency bar chart from 1000 stepwise logistic regression replicates, with the red dashed line marking the 80% stability threshold.

These four high--stability variables (ALB, PLT, HH, SH) were entered into a binary logistic regression model to construct a predictive model for amputation risk in patients with diabetic foot ulcers (Table 3). The results showed that smoking history (sh: OR = 2.416, 95% CI: 1.317–4.435, P = 0.004), hypertension (hh: OR = 2.106, 95% CI: 1.206–3.678, P = 0.009), and platelet count (plt: OR = 1.004 per 1×109/L, 95% CI: 1.001–1.006; equivalently OR = 1.041 per 10×109/L, P = 0.008) were independent predictors of amputation. In contrast, albumin (alb: OR = 0.869, 95% CI: 0.824–0.916, P = 0.001) emerged as a protective factor.

Table 3 Multivariate Logistic Regression Analysis of Factors Associated with Amputation in Patients with Type 2 Diabetes and Diabetic Foot Ulcers

Construction and Validation of the Nomogram

A nomogram incorporating the four independent predictors (ALB, HH, SH, PLT) was constructed to predict amputation risk (Figure 3).

Figure 3 Nomogram for predicting amputation risk. Points are assigned to each of the four variables (ALB, PLT, HH, SH) based on the patient’s clinical measurements. The sum of the points yields a total score, which directly reflects the predicted probability of amputation. Higher total scores indicate higher risk.

Model performance was evaluated in both the training and validation cohorts. In the training cohort, the nomogram achieved an AUC of 0.788 (95% CI: 0.735–0.840). In the validation cohort, the AUC was 0.755 (95% CI: 0.671–0.839); the wider validation interval reflects the limited number of validation events (n = 35). On bootstrap internal validation (1000 resamples on the full cohort of 508 patients), the optimism-corrected AUC was 0.776, with a bootstrap-corrected calibration slope of 0.944 (Figure 4A–C). Calibration curves showed acceptable agreement in both cohorts (training: Hosmer–Lemeshow P = 0.794, calibration slope 1.000, intercept 0.000, Brier score 0.146; validation: Hosmer–Lemeshow P = 0.238, calibration slope 0.805, intercept −0.177, Brier score 0.160) (Figure 4D and E). Decision curve analysis showed that the nomogram provided positive net benefit relative to the “treat-all” and “treat-none” strategies across threshold probabilities of approximately 10–60% (Figure 4F–H). Taken together, these results indicate that the nomogram demonstrated moderate discrimination, acceptable calibration, and potential clinical utility for estimating in-hospital amputation risk in patients with type 2 diabetes-related DFUs.

Figure 4 Performance evaluation of the nomogram. (A) ROC curve in the training cohort (AUC = 0.788); (B) ROC curve in the validation cohort (AUC = 0.755). (C) ROC curves overlaid for training and validation cohorts. (D and E) Calibration curves with bias correction in the training cohort (D Hosmer–Lemeshow P = 0.794) and validation cohort (E Hosmer–Lemeshow P = 0.238). (F–H) Decision curve analyses in the training (F), validation (G), and both cohorts (H), demonstrating positive net benefit over a wide range of threshold probabilities. The validation calibration curve deviates from the ideal line in the medium-to-high probability range owing to the small validation sample.

Discussion

In this single-center retrospective cohort of 508 hospitalized patients with type 2 diabetes-related DFUs, we built and internally validated a four-variable nomogram for LEA based on serum albumin, platelet count, smoking history, and hypertension. The AUC was 0.788 in the training set and 0.755 in the validation set; calibration was acceptable and decision curve analysis suggested potential clinical utility. Importantly, all four predictors were retained by both LASSO at λ.1se and by >80% of bootstrap stepwise samples; albumin and platelet count were selected in almost every bootstrap replicate. The agreement between two structurally different selection procedures gives us more confidence in the choice of variables than either method alone would.

Serum albumin was the strongest protective factor in our model (OR = 0.869), retained in nearly every bootstrap replicate. Low serum albumin is a familiar marker of poor nutritional and catabolic states.14 Patients with DFUs often combine chronic infection, hypermetabolism, and poor intake, so hypoalbuminemia in this setting reflects more than diet alone—it tracks with impaired wound healing, immune compromise, and poor granulation.15 Our estimate is consistent with a multicentre retrospective analysis of 526 DFU patients (OR = 0.850) and with a meta-analysis pooling 5689 patients, both of which reported a similar protective effect.16,17 Albumin remained in the model after both LASSO and bootstrap screening, suggesting that the signal is not merely an artifact of correlated inflammation. Albumin is also potentially modifiable. Nutritional assessment, oral or enteral protein supplementation in patients with protein-energy malnutrition, and aggressive control of underlying infection are reasonable components of routine DFU care.

Platelet count, the other variable retained at near-100% bootstrap frequency, was an independent predictor in our model. Patients in the amputation group had higher platelet counts, with each 10×109/L increase corresponding to roughly a 4% rise in odds (OR = 1.041 per 10×109/L). This is in line with earlier work in DFU populations, where elevated platelet count has been associated with major amputation, plausibly through platelet-mediated impairment of microcirculation.18 Several mechanisms may contribute. Systemic inflammation in DFU drives reactive thrombocytosis through IL-6 and TNF-α.19 Activated platelets release CD40L and PF4, sustaining the inflammatory loop, and they form aggregates with leukocytes that promote microthrombosis and worsen tissue perfusion—an effect particularly damaging in diabetic microvascular disease.20 Because it is a cheap and universally available test, platelet count is well suited to risk stratification in everyday practice; whether antiplatelet or platelet-modulating strategies can alter outcomes in this subgroup remains an open question.

Smoking history was an independent risk factor in our cohort (OR = 2.416). The biological basis is multilayered. Nicotine drives sympathetic activation and norepinephrine release, leading to sustained vasoconstriction and reduced cutaneous perfusion.21 Nicotine also has direct cellular effects on wound healing, inhibiting fibroblast proliferation and collagen synthesis.22,23 These mechanisms are reflected at the population level: a meta-analysis of 14 studies reported a large negative effect of smoking on endothelial function (SMD = −3.15, 95% CI −3.84 to −2.46),23 and research indicates that smoking-induced sympathetic overactivity further aggravates peripheral ischemia.24

Hypertension was the fourth independent predictor in our model (OR = 2.106). Long-standing high pressure drives vascular remodeling, with intimal thickening, medial hypertrophy, arterial stiffness, and luminal narrowing.25,26 In patients with diabetes, this process is compounded by hyperglycemia-driven accumulation of advanced glycation end-products, and the combined mechanical and metabolic insult further compromises distal blood flow.27,28 The association between hypertension and amputation in our model persisted after adjustment for other clinical variables. Routine blood pressure monitoring and control may therefore form part of the multidisciplinary management of DFUs, alongside glycemic control, wound care, and infection management.

Several nomograms for DFU amputation have been reported in the past five years. Peng et al developed a single-center five-variable model in 125 patients incorporating course of diabetes, PAD, HbA1c, WBC, and FIB (AUC 0.876).29 Che et al reported a multi-centre model based on 526 patients from five Anhui hospitals, with six independent predictors: blood type-B, history of amputation, positive bacterial culture, Wagner grade (3 vs 4–5), PAD, and serum albumin (AUC 0.85).17 Xu et al constructed a single-center nomogram in 144 patients from an initial pool of 20 candidate variables, retaining 12 predictors including Wagner grade, PAD, CKD, CRP, and age.30 Our model has a somewhat lower apparent AUC (0.755 on validation) than these earlier models, but it differs in three practical ways. First, all four predictors are available from routine clinical history (smoking and hypertension) and standard laboratory testing (albumin and platelet count), without imaging, vascular assessment, or microbiological culture, which is advantageous where such specialized testing is not consistently available. Second, variable selection was stabilized by combining LASSO with 1000 bootstrap stepwise replicates and retaining only the intersection; this addresses the well-documented instability of single-pass stepwise selection that has limited the reproducibility of earlier DFU nomograms. Third, restricting the model to four variables reduces the overfitting risk that is otherwise hard to avoid in single-center DFU cohorts of this size. A direct head-to-head comparison against SINBAD, WIfI, Wagner, and the published nomograms above, in an external multi-centre cohort, is needed and is planned as a follow-up study.

This study has several limitations. First, it was a single-center, retrospective analysis with internal validation only: the 7:3 split was random rather than temporal, and the validation set contained only 35 amputations, so discrimination may be overestimated, the validation AUC is imprecise, and generalizability—including across the 2019–2025 study period, which spanned the COVID-19 pandemic—remains untested. Second, several established DFU predictors (Wagner/SINBAD/WIfI classification, ankle-brachial index, HbA1c, diabetes duration, prior amputation, and wound depth or area) were unavailable in our database, which probably understates the achievable predictive performance. Third, the primary outcome combined minor and major amputation, which differ in prognosis, and was restricted to the index hospitalization; a major-amputation-only model was not feasible given the few major events, and performance at first presentation or in outpatient settings is unknown. Fourth, variable selection was performed at a low selection-stage EPV, so—despite the dual-method intersection and bootstrap optimism correction—some overfitting cannot be excluded; consistent with this, smoking (P = 0.329) and hypertension (P = 1.000) were not significant in the validation cohort, most likely reflecting the small number of validation events. External, prospective, multi-centre validation—incorporating the variables above and a head-to-head comparison with SINBAD and WIfI—is therefore required before clinical use.

Conclusion

We built and internally validated a parsimonious four-variable nomogram—serum albumin, platelet count, smoking history, and hypertension—for estimating in-hospital LEA risk in hospitalized patients with type 2 diabetes-related DFUs. Discrimination was moderate and calibration was acceptable, and decision curve analysis suggested potential clinical utility. Because every predictor is available from routine clinical history and standard laboratory testing, the model may support early in-hospital risk stratification; as none of the four variables is fully under clinician control, the nomogram is intended as a stratification tool rather than a treatment-decision guide. As this was a single-center study with internal validation only, external multi-centre validation is required before routine clinical use.

Use of AI Tools

During the preparation of this work, the authors used ChatGPT (GPT-5, OpenAI) for language editing only. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

Data Sharing Statement

The de-identified datasets and R analysis code used in this study are available from the corresponding author on reasonable request, subject to approval by the Ethics Committee of Linyi Central Hospital.

Ethics Approval and Consent to Participate

This study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Linyi Central Hospital (approval No. LCH-LW-2024102). Because the study was retrospective and used only de-identified data extracted from existing medical records, posed no more than minimal risk, and did not affect patient care, the Ethics Committee waived the requirement for individual informed consent. All patient records were de-identified before analysis; data were stored on password-protected institutional servers accessible only to the research team, and no individually identifiable information is reported.

Acknowledgments

The authors thank the Department of Hand and Foot Surgery, Linyi Central Hospital, for providing access to the clinical records used in this study.

Author Contributions

Haipeng Zhang: Conceptualization, Methodology, Formal analysis, Investigation, Writing – original draft. Jian Guo: Conceptualization, Methodology, Formal analysis, Data curation, Writing – original draft. Haipeng Zhang and Jian Guo contributed equally and share first authorship. Yongfang Ma: Investigation, Data curation, Writing – review & editing. Chuang Ma: Conceptualization, Supervision, Project administration, Funding acquisition, Writing – review & editing. All authors gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Funding

This project was funded by: Research Project of Xinjiang Uygur Autonomous Region Health Commission (No. 2025001CGZHYDXM650030090); National Natural Science Foundation of China (No. 82260425); The “Tianshan Talent” Youth Science and Technology Top Talent Program in the Xinjiang Uygur Autonomous Region (No. 2022TSYCCX0113); The Xinjiang Uygur Autonomous Region Graduate Innovation Program (No. XJ2025G189).

Disclosure

Dr Haipeng Zhang reports Support for the manuscript from Research Project of Xinjiang Uygur Autonomous Region Health Commission, National Natural Science Foundation of China, The “Tianshan Talent” Youth Science and Technology Top Talent Program in the Xinjiang Uygur Autonomous Region, The Xinjiang Uygur Autonomous Region Graduate Innovation Program, during the conduct of the study. The other authors declare that they have no competing interests.

References

1. Monteiro-Soares M, Hamilton EJ, Russell DA, et al. Guidelines on the classification of foot ulcers in people with diabetes (IWGDF 2023 update). Diabetes Metab Res Rev. 2024;40(3):e3648. doi:10.1002/dmrr.3648

2. Armstrong DG, Boulton AJM, Bus SA. Diabetic foot ulcers and their recurrence. N Engl J Med. 2017;376(24):2367–12. doi:10.1056/NEJMra1615439

3. Chen L, Sun S, Gao Y, et al. Global mortality of diabetic foot ulcer: a systematic review and meta-analysis of observational studies. Diabetes Obes Metab. 2023;25(1):36–45. doi:10.1111/dom.14840

4. McDermott K, Fang M, Boulton AJM, et al. Etiology, epidemiology, and disparities in the burden of diabetic foot ulcers. Diabetes Care. 2023;46(1):209–221. doi:10.2337/dci22-0043

5. Julla JB, Jolivet T, Estellat C, et al. Incidence of death and amputation in patients with a first diabetic foot ulcer: results from the CODIA cohort. Diabetes Metab. 2025;51(6):101700. doi:10.1016/j.diabet.2025.101700

6. Wernecke J, Risse F, Möller M, et al. Epidemiology and the medical burden of diabetic foot ulcers especially in patients with infection — a population-based analysis from Germany. Int Wound J. 2025;22(3):e70157. doi:10.1111/iwj.70157

7. Mo R, Ding Y, Hua Q. Research advances in limb salvage treatment of diabetic foot using tibial transverse transport. Zhongguo Xiu Fu Chong Jian Wai Ke Za Zhi. 2025;39(8):942–949. doi:10.7507/1002-1892.202505026

8. Guo P, Tu Y, Liu R, et al. Performance of risk prediction models for diabetic foot ulcer: a meta-analysis. PeerJ. 2024;12:e17770. doi:10.7717/peerj.17770

9. Silva MA, Hamilton EJ, Russell DA, et al. Diabetic foot ulcer classification models using artificial intelligence and machine learning techniques: systematic review. J Med Internet Res. 2025;27:e69408. doi:10.2196/69408

10. Collins GS, Moons KGM, Dhiman P, et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 2024;385:e078378. doi:10.1136/bmj-2023-078378

11. Benchimol EI, Smeeth L, Guttmann A, et al. The REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) statement. PLoS Med. 2015;12(10):e1001885. doi:10.1371/journal.pmed.1001885

12. Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology. 1982;143(1):29–36. doi:10.1148/radiology.143.1.7063747

13. Vickers AJ, Elkin EB. Decision curve analysis: a novel method for evaluating prediction models. Med Decis Making. 2006;26(6):565–574. doi:10.1177/0272989X06295361

14. Aragón-Sánchez J, Víquez-Molina G, López-Valverde ME, et al. Clinical, microbiological and inflammatory markers of severe diabetic foot infections. Diabet Med. 2021;38(10):e14648. doi:10.1111/dme.14648

15. Ran Q, Xu W, Zhao X, et al. Risk factors for malnutrition in patients with diabetic foot ulcer and its association with prolonged length of hospitalization. Nutr Diabetes. 2024;14(1):26. doi:10.1038/s41387-024-00290-6

16. Luo Z, Fan W, Xie X, Yang Y, Xu Z, Zhang J. The incidence of lower extremity amputation and its associated risk factors in patients with diabetic foot ulcers: a meta-analysis. Int Wound J. 2024;21(7):e14931. doi:10.1111/iwj.14931

17. Che D, Jiang Z, Xiang X, et al. Predictors of amputation in patients with diabetic foot ulcers: a multi-centre retrospective cohort study. Endocrine. 2024;85(1):181–189. doi:10.1007/s12020-024-03704-8

18. Demirkol D, Aktaş Ş, Özcan T, et al. Analysis of risk factors for amputation in patients with diabetic foot ulcers: a cohort study from a tertiary center. Acta Orthop Traumatol Turc. 2022;56(5):333–339. doi:10.5152/j.aott.2022.22052

19. Aydın MS, Eren MA, Uyar N, et al. Relationship between systemic immune inflammation index and amputation in patients with diabetic foot ulcer. J Orthop Sci. 2024;29(4):1060–1063. doi:10.1016/j.jos.2023.07.015

20. Elalamy I, Chakroun T, Gerotziafas GT, et al. Circulating platelet-leukocyte aggregates: a marker of microvascular injury and thrombotic risk in inflammatory states. Thromb Res. 2008;121(6):843–848. doi:10.1016/j.thromres.2007.07.019

21. Whitehead AK, Erwin AP, Xu X. Nicotine and vascular dysfunction. Acta Physiol. 2021;231(1):e13631. doi:10.1111/apha.13631

22. Chen AL, Shepherd RK, Bhagwat SP, et al. Nicotine impairs wound healing by inhibiting fibroblast proliferation and collagen synthesis via the α7-nAChR pathway. Lab Invest. 2003;83(3):321–327. doi:10.1097/01.LAB.0000059928.80656.FC

23. Yin X, Sun Y, Ding H, et al. The association between smoking exposure and endothelial function evaluated using flow-mediated dilation values: a meta-analysis. BMC Cardiovasc Disord. 2024;24(1):188. doi:10.1186/s12872-024-03915-x

24. Dimitriadis K, Narkiewicz K, Leontsinis I, et al. Acute effects of electronic and tobacco cigarette smoking on sympathetic nerve activity and blood pressure in humans. Int J Environ Res Public Health. 2022;19(6):3237. doi:10.3390/ijerph19063237

25. Renna NF, de Las Heras N, Miatello RM. Pathophysiology of vascular remodeling in hypertension. Int J Hypertens. 2013;2013:808353. doi:10.1155/2013/808353

26. Sun X, Wu J, Zhang X, et al. Atlas of cell repertoire within neointimal lesions is metabolically altered in hypertensive rats. Hypertension. 2024;81(4):787–800. doi:10.1161/HYPERTENSIONAHA.123.22057

27. Lin H, Yang Y, Wang X, et al. Targeting the AGEs-RAGE axis: pathogenic mechanisms and therapeutic interventions in diabetic wound healing. Front Med. 2025;12:1667620. doi:10.3389/fmed.2025.1667620

28. Wang B, Jiang T, Qi Y, et al. AGE-RAGE axis and cardiovascular diseases: pathophysiologic mechanisms and prospects for clinical applications. Cardiovasc Drugs Ther. 2025;39(6):1489–1506. doi:10.1007/s10557-024-07639-0

29. Peng B, Min R, Liao Y, et al. Development of predictive nomograms for clinical use to quantify the risk of amputation in patients with diabetic foot ulcer. J Diabetes Res. 2021;2021:6621035. doi:10.1155/2021/6621035

30. Xu T, Hu L, Xie B, et al. Analysis of clinical characteristics in patients with diabetic foot ulcers undergoing amputation and establishment of a nomogram prediction model. Sci Rep. 2024;14(1):27934. doi:10.1038/s41598-024-78215-7