Introduction

Diabetic peripheral neuropathy (DPN) is one of the most prevalent chronic complications of type 2 diabetes mellitus (T2DM), characterized by progressive dysfunction of sensory and motor nerves. Clinically, it typically manifests as limb numbness, pain, and sensory abnormalities and represents the major antecedent of diabetic foot ulcers and lower-extremity amputation, leading to considerable disability and healthcare burden.1 The prevalence of DPN among individuals with T2DM was estimated to be approximately 46% to 67.6%.2,3 The burden of DPN is likely higher in older adults with T2DM, as advanced age was a well-established risk factor that can accelerate diabetes-related peripheral nerve degeneration.4,5 Moreover, DPN often presents insidiously.6 In older adults, early symptoms are frequently obscured by age-related sensory loss, cognitive impairment, and physical frailty, delaying recognition and increasing the risk of severe complications.7 These characteristics collectively underscore the need for age-sensitive screening tools and early risk stratification strategies tailored to elderly patients with T2DM.

Despite its clinical importance, early detection of DPN remains challenging. Although nerve conduction studies represent the diagnostic gold standard, these methods are technically complex, costly, and impractical for widespread routine screening. Commonly used clinical screening tools, including the Michigan Neuropathy Screening Test and the Toronto Clinical Neuropathy Score, are simple and convenient; however, they heavily depend on patients’ subjective reports and clinicians’ judgment, limiting their objectivity and sensitivity for early stages. Therefore, early screening and risk management are essential for older adults with T2DM. Several risk prediction models for DPN have been developed. Traditional models for DPN have typically used moderate sample sizes, conventional clinical and laboratory variables–such as age, diabetes duration, glycemic control, and comorbidities, but have demonstrated only modest predictive performance.8–11 In contrast, Machine learning (ML) models offer substantial advantages for modeling complex, high-dimensional clinical data and has demonstrated improved predictive performance compared with conventional statistical approaches, particularly when combined with interpretable frameworks such as SHapley Additive exPlanations (SHAP).12 While studies in larger cohorts using ML have reported superior predictive accuracy,13–16 these studies did not specifically focus on older adults,8–11,13–16 a population at elevated risk. Moreover, these studies rarely incorporated lifestyle factors, such as sleep, physical activity, and dietary habits,17–19 which are recognized as important determinants of DPN risk. Incorporating these variables into predictive models could improve early risk stratification19 and guide personalized preventive strategies. Therefore, this study aims to develop and validate a machine learning–based predictive model for DPN in older adults with T2DM using a large real–world cohort by incorporating a comprehensive set of clinical, laboratory, and lifestyle risk factors. We will compare multiple ML algorithms to identify the optimal predictive model and apply SHAP analysis to enhance clinical interpretability, providing a clinically applicable tool for early risk stratification and targeted intervention in this vulnerable population.

Materials and MethodsStudy Population

A total of 1450 elderly patients with T2DM were retrospectively enrolled in this study using a convenience sampling method. These patients were recruited from the National Standardized Metabolism Management Center (MMC) of a tertiary Grade A hospital in Shanghai and received follow-up care between March 2022 and March 2025. The inclusion criteria were as follows: 1) age≥65 years, and 2) a confirmed diagnosis of T2DM meeting the established diagnostic criteria for elderly patients.20 The exclusion criteria included: 1) peripheral neuropathy attributable to other etiologies, including such as cervical/lumbar spondylosis, stroke, or the use of neurotoxic medications; and 2) incomplete medical records. The sample size was estimated based on a standard formula for clinical prediction model development:21 n = exp, whereφrepresents the incidence rate of the outcome event (reported in the literature as 39.24–49.4%), P is the number of candidate predictors (assumed to be 10), and MAPE (Mean Absolute Percentage Error) was set at 0.05. This calculation yielded a required sample size of 525 to 584 participants for model development. Subsequently, the total sample was to be randomly divided into a training set and a test set at a ratio of 7:3, necessitating a final total sample size of 750 to 834 participants. To enhance the robustness of the predictive model, this study ultimately included 1450 patients. The research was approved by the hospital ethics committee (Approval No.: GLYYls2024-066). Patient consent for medical record review was waived, and all data were de-identified and coded to ensure confidentiality.

Research Variables

A self-administered questionnaire titled “Risk Factors for Peripheral Neuropathy in Elderly Patients with T2DM” was developed based on clinical expertise and previous research evidence, This questionnaire comprised 40 items across three sections: (1) General Demographic and Lifestyle Data: including gender, age, education, income, body mass index (BMI), subcutaneous and abdominal fat area, physical activity level, sedentary behavior, sugar-sweetened beverage(SSB) intake, dietary habits, smoking and drinking, sleep patterns, and family history of diabetes; (2) Disease-Related Information: including diabetes duration, hypoglycemic treatment regimen, history of hypertension, peripheral arterial disease (PAD), diabetic complications, Charlson Comorbidity Index (CCI), and abdominal fat diagnosis; (3) Laboratory and Auxiliary Examination Data: including glycated hemoglobin (HbA1c), C-peptide, lipid profiles (total cholesterol [TC], triglycerides [TG], high-density lipoprotein cholesterol [HDL-C], low-density lipoprotein cholesterol [LDL-C]), free triiodothyronine (FT3), urinary creatinine, urinary microalbumin, 24-hour urinary protein quantification, urine albumin, white blood cell count (WBC), and hemoglobin (HB).

Operational Definitions of Key Variables:

Sedentary behavior was defined as a daily sitting time of ≥ 6 hours.22 Sugar-sweetened beverage (SSB) intake was defined as consuming SSBs (eg, carbonated soft drinks, fruit juice drinks, sweetened tea beverages) at least once per week or with a cumulative weekly volume of ≥ 500 mL. PAD was diagnosed by ankle-brachial index < 0.7.23 Poor sleep quality was defined as meeting any of the following criteria: ≥3 weekly episodes of difficulty falling asleep (onset latency >30 min) or <6 hours of daily sleep.24

The diagnostic criteria for peripheral neuropathy were established in accordance with relevant clinical guidelines,25 with diagnosis meeting any of the following:

① The presence of clinical symptoms of neuropathy (eg, pain, numbness, paresthesia) and at least one abnormal finding in five specific tests (ankle reflex, vibration sensation, pressure sensation, temperature sensation, pinprick sensation);

② For asymptomatic individuals, at least two abnormal findings among the five tests;

③ Reduced nerve conduction velocity or decreased wave amplitude.

In this study, the final determination of DPN was based on a clearly documented clinical diagnosis in the electronic medical records.

Statistical Analysis

All statistical analyses were conducted using Python software (version 3.13) with relevant packages including pandas, NumPy etc. Continuous variables were reported as mean±standard deviation or median (interquartile range), depending on their distribution. Categorical variables were summarized as frequency (percentage).

Data Preprocessing

To minimize bias from missing data, variables with a missing rate ≥30% were excluded. For variables with <30% missingness, continuous variables were handled using multiple imputation by MICE method, with five imputations and predictive mean matching for continuous variables. Whereas categorical variables were imputed using the mode. The dataset was randomly divided into training and testing sets in a 7:3 ratio with a fixed random seed (random_state = 42) for reproducibility. Within the training set, continuous variables were further standardized using Z-score transformation, and categorical variables were one-hot encoded.26 A comprehensive preprocessing pipeline was established to ensure that the transformation rules derived from the training set were consistently utilized for the testing set, thereby averting any potential data leakage during model evaluation.

Feature Selection

In the training set, variables associated with DPN were initially screened through univariable analysis. Inter-group comparisons were conducted using the Student’s t-test, Chi-square test, or Mann–Whitney U-test, as appropriate. Variables with statistical significance (P < 0.05) underwent further multicollinearity diagnosis (using variance inflation factor VIF < 5 as the criterion). Ultimately, eligible variables were incorporated into a multivariate logistic regression analysis to identify independent risk factors for DPN in elderly patients with T2DM.

Model Development and Evaluation

Prediction models were developed using four machine learning algorithms: Random Forest (RF), extreme gradient boosting (XGBoost), Logistic Regression (LR), and Naïve Bayes (NB). This selection encompassed both traditional (LR, NB) and advanced ensemble methods (RF, XGBoost) to capture linear, probabilistic, and complex nonlinear relationships, enabling comprehensive comparison across modeling paradigms. Hyperparameters were optimized via grid search with stratified 5-fold cross-validation using ROC-AUC as the selection metric (Python 3.13, scikit-learn 1.3.0, XGBoost 1.7.0, random_state=42). The optimal configurations were as follows: RF (n_estimators=50, max_depth=None, min_samples_split=2, min_samples_leaf=4); XGBoost (objective“binary: logisti”, colsample_bytree=0.8, learning_rate=0.2, max_depth=3, n_estimators=50, subsample=0.9); LR (penalty“l”, C=10, solver“sag”, max_iter=1000); and NB (var_smoothing=1e-5). Model performance was internally validated using the testing set. Discrimination was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, precision, recall, and F1 score; Calibration was measured with the Brier score and clinical utility was examined through decision curve analysis (DCA).

Model Explanation

To improve the interpretability of the predictive model, this study utilized the SHAP method to interpret and visualize the optimal model. The SHAP values for each feature variable were calculated. Subsequently, a feature importance summary plot and SHAP beeswarm plots were created to quantitatively and intuitively illustrate the contribution magnitude and directional influence of each feature on the model’s predictions.

ResultPatient Characteristics

The incidence of DPN among the included subjects was 42.9% (623/1450), The prevalence rates in the training set and test set were 42.95% (436/1015) and 42.98% (187/435), respectively. No statistically significant differences were observed between the training and test cohorts for any variables, as shown in Table 1.

Table 1 Characteristics Between Training and Test Sets of Elderly Type 2 Diabetes Patients

Feature Selection

The training set was categorized into a DPN group (n = 436) and a non-DPN group (n = 579). Univariable analysis identified 19 variables with statistically significant differences (P<0.05), as shown in Table 2. These variables were subsequently incorporated into a multivariable logistic regression analysis, which identified 9 independent predictive feature: CCI, diabetes duration, HbA1c, PAD, history of hypertension, smoking, sedentary behavior, sugar-sweetened beverage intake, and sleep status. The detailed results are shown in Table 3.

Table 2 Univariable Factors Analysis of DPN Selection Variables (Training Set, n=1015)

Table 3 Multivariable Factors Analysis of DPN Selection Variables (Training Set, n=1015)

Model Performance

Using the occurrence of DPN as the dependent variable and the nine previously screened variables as independent variables, we constructed four prediction models based on RF, XGBoost, LR, and NB. The performance evaluation demonstrated that the XGBoost model achieved the highest scores across all metrics except Accuracy, as shown in Table 4. The ROC analysis on the training set (Figure 1A) showed that all models achieved high AUC values, indicating excellent discriminative ability during training. In the test set (Figure 1B), the XGBoost model achieved the highest AUC (0.951) among all models. To assess calibration performance, the Hosmer-Lemeshow goodness-of-fit test was conducted, the XGBoost model demonstrated superior performance in the test set, yielding a χ2value of 4.935 with a P-value of 0.668, indicating no significant deviation between the predicted and observed outcomes. Furthermore, the calibration curves closely aligned with the ideal reference line (Figure 1C and D), suggesting strong agreement between predicted probabilities and actual event probabilities. DCA revealed that in the training set, the RF model outperformed all other models across a broad threshold probability range of 0–0.8 (Figure 1E). However, in the test set, the XGBoost model provided the highest net benefit across nearly the entire threshold range (0–0.9, Figure 1F). Based on this comprehensive assessment of discrimination, calibration, and clinical applicability, the XGBoost model was identified as the optimal predictive model.

Table 4 Performance Evaluation of the Four Prediction Models

Figure 1 Performance and comparison of four predictive models.(A) ROC curves for the training set. (B) ROC curves for the test set.(C) Calibration curves for the training set. (D) Calibration curves for the test set.(E) Decision curve analysis (DCA) for the training set. (F) Decision curve analysis (DCA) for the test set.

Model Explanation

We utilized the SHAP method to interpret the XGBoost model by calculating the contribution of each variable to the predictions. Figure 2 displays the average SHAP values for each feature, with the relative contributions of the feature variables ranked in descending order as follows: diabetes duration, HbA1c, sleep status, CCI, sugar-sweetened beverage intake, PDA, sedentary behavior, smoking, and hypertension, underscoring their pivotal role in predicting DPN. SHAP swarm Plot visualizes feature importance, with each point corresponds to a sample, and a color gradient from blue (low values) to red (high values) reflects the magnitude of the feature value. The vertical axis displays the ranked features, showing the correlation and distribution of feature values with their corresponding SHAP values.

Figure 2 SHAP analysis for feature interpretability. (A) Feature importance ranking for the XGBoost model. (B) SHAP Beeswarm Plot of features for the XGBoost model.

Discussion

The findings of this study demonstrate that disease duration (OR=1.127, P< 0.001) and HbA1c level (OR=1.448, P<0.001) are significant risk factors for DPN in elderly patients with T2DM. Previous studies have also showed that patients with T2DM for more than 10 years have a markedly increased risk of neuropathy, with each additional year of disease duration raising the risk of DPN by approximately 7%.27–29 This underlying pathological mechanism may be linked to cumulative microvascular and neural damage, which is mediated by the “metabolic memory effect”. In terms of glycemic control, HbA1c level showed a significant positive correlation with DPN risk. Supporting the findings of Wu and Darivemula.9,30 It was further quantified that individuals with HbA1c >7% had a 4.81-fold increased risk of DPN. This mechanism is primarily attributed to persistent hyperglycemia, which promotes the excessive formation of advanced glycation end products(AGEs), leading to endometrial ischemia and disruption of the blood-nerve barrier.31 Notably, unlike previous studies, this research specifically focuses on elderly T2DM patients. This population may exhibit greater susceptibility to long-term hyperglycemia and the metabolic memory effect due to age-related declines in neural repair capacity and vascular function. Therefore, the following targeted strategies are recommended for clinical practice: first, establish disease duration-stratified management pathway. For elderly patients with T2DM and a disease duration ≥ 10years, standardized examinations for peripheral neurological signs should be conducted semi-annually, and personal health records should be established to systematically monitor disease progression. In contrast, for those with a duration <10 years, annual screening should be prioritized, combined with preventive education, specifically guiding them in recognizing early symptoms such as limb numbness and pain. Furthermore, individualized glycemic management must be implemented. HbA1c targets should be personalized, considering the patient’s overall health status. For elderly patients with multiple chronic conditions or frailty, these targets may be appropriately relaxed to 8.0–8.5% to balance potential risks and benefits.32 Simultaneously, a multidimensional support network should be developed, incorporating family members, peers, and digital tools. By engaging family in daily management, organizing peer support groups for diabetes,33 and utilizing health platforms with data-sharing capabilities,34 continuous glucose monitoring and timely feedback can be achieved. These measures can effectively delay the onset and progression of DPN and improve the quality of life for elderly diabetic patients.

The findings of this study also indicate that the CCI (OR = 2.000, P < 0.001), PAD (OR = 9.911, P < 0.001), and history of hypertension (OR = 3.306, P < 0.001) are independent risk factors for DPN in this population. A higher CCI score signifies multiple chronic conditions, which may interact with diabetic metabolic dysregulation through chronic inflammatory pathways, thereby accelerating nerve damage progression, consistent with the findings of Luo.11 PAD, as a manifestation of systemic atherosclerosis, indicates significant macrovascular pathology in the lower limbs, directly leading to hypoperfusion in the neural microcirculation and placing nerve fibers in a persistent ischemic and hypoxic environment.35 Hypertension impairs vascular endothelial function and exacerbates microcirculatory disorders, disrupting blood-nerve barrier integrity in conjunction with hyperglycemia.36 Clinical evidence demonstrates a significant delay in DPN onset and improvement in nerve conduction function among patients using Angiotensin-Converting Enzyme Inhibitors (ACEIs),37 validating the critical role of blood pressure management in neuroprotection. Therefore, a comprehensive management strategy is recommended for elderly T2DM patients with these comorbidities: establish a comorbidity assessment system incorporating the CCI as a mandatory component of regular evaluations and initiate multidisciplinary collaborative management for patients with multiple chronic conditions; Enhance PAD screening by regularly assessing vascular status using objective measures such as the ankle-brachial index. For patients with hypertension, it is crucial to prioritize antihypertensive agents with potential neuroprotective properties, such as ACEIs, while emphasizing that blood pressure control is as critical as glycemic management. Multifactorial integrated interventions can effectively reduce the DPN risk in elderly patients with T2DM.

The results further demonstrate that poor sleep quality (OR = 6.258, P < 0.001), sugar-sweetened beverage intake (OR = 5.487, P < 0.001), sedentary behavior (OR = 2.693, P < 0.001), and smoking (OR = 1.711, P < 0.001) are independent, modifiable behavioral risk factors for DPN. These factors collectively promote DPN onset and progression through multiple interacting pathophysiological pathways. A bidirectional vicious cycle exists between sleep disorders and DPN: sleep disorders activate the hypothalamic-pituitary-adrenal axis and sympathetic nervous system, triggering systemic inflammation and oxidative stress responses, exacerbating insulin resistance and metabolic disturbances, thereby promoting neural damage.38,39 Conversely, neuropathic pain and paresthesia caused by DPN disrupt sleep architecture and quality.40 Frequent consumption of sugar-sweetened beverages not only impairs pancreatic β-cell function through glucotoxicity but also induces massive reactive oxygen species production, leading to oxidative stress damage. These mechanisms collectively compromise the integrity of the blood-nerve barrier and impair neuronal structure and function.41 A clear dose-response relationship exists between sedentary behavior and DPN risk. Research by Loprinzi et al indicated that both T2DM and DPN are independently associated with lower levels of physical activity and higher sedentary time.42 This association likely arises because prolonged sitting impedes venous return in the lower limbs, suppresses the muscle pump function, and causes inadequate perfusion and hypoxia in neural tissues, ultimately promoting nerve fiber degeneration.43 Conversely, regular exercise not only improves local neural blood flow and promotes the release of neurotrophic factors and nerve regeneration, but also exerts protective effects through systemic metabolic regulation.44,45 Furthermore, studies have identified a significant positive correlation between sleep deprivation and sedentary behavior, with both factors producing synergistic detrimental effects through weight gain and insulin resistance.46 Smoking increases DPN risk through multiple mechanisms, including inducing insulin resistance, promoting atherosclerosis, exacerbating systemic inflammatory responses, and impairing vascular endothelial function.47,48 Based on these findings, comprehensive interventions targeting these modifiable behavioral factors are recommended. Such as establishing regular sleep schedules, providing sleep education, and implementing cognitive behavioral therapy to improve sleep quality; developing individualized sedentary behavior intervention plans addressing psychological and behavioral aspects,49 incorporating daily moderate aerobic and resistance training; delivering nutritional guidance that explicitly restricts sugar-sweetened beverage consumption; and providing systematic smoking cessation support and follow-up. Establishing a prevention and control system centered on behavior management for DPN will effectively delay disease progression and improve patients’ quality of life.

This study developed a risk prediction model for DPN among older adults with T2DM. Systematic evaluations indicated that the XGBoost model achieved the best overall performance. In the study by Sun et al, although ten-fold cross-validation was used to improve model stability, the AUC dropped substantially from 0.933 in internal validation to 0.811 in an independent test set, reflecting notable performance variation across datasets.14 Our model maintained high and comparable performance in both the training and test sets (AUCs of 0.973 and 0.951, respectively), with decision curve analysis demonstrating stable net benefits across a wide threshold range (0–0.9). Beyond predictive performance, the clinical feasibility of a model depends on its feature selection. Liu et al incorporated inflammatory biomarkers such as C-reactive protein and total bile acid to improve predictive accuracy; however, these laboratory-dependent features are more feasible in resource-intensive inpatient settings, thereby limiting their generalizability to outpatient elderly populations.15 While Jiang et al employed traditional Chinese medicine symptom features that rely heavily on patient-reported information and physician interpretation, which may restrict standardization and cross-institutional applicability.13 In contrast, our model leverages the MMC platform’s standardized electronic follow-up system, integrating objective clinical indicators, such as HbA1c, as well as lifestyle factors derived from structured questionnaires, including sleep duration, exercise intensity, and sugar-sweetened beverage intake, ensuring both accessibility and reproducibility. Overall, the proposed model combines high predictive performance with standardized, easily obtainable features, enhancing its applicability and scalability for managing chronic disease in elderly T2DM patients.

To enhance interpretability, SHAP analysis was employed to visualize the contribution of each variable. The SHAP summary and dependence plots indicated that diabetes duration and HbA1c were the strongest contributors to a higher predicted risk of DPN, aligning with the well-established roles of long-term hyperglycemia, metabolic memory, and microvascular injury in neuropathy progression. Lifestyle behaviors, including poor sleep quality, sedentary time, and sugar-sweetened beverage intake, also demonstrated substantial positive SHAP values, indicating their meaningful influence on individual risk profiles. In addition, clinical comorbidities such as PAD, higher CCI scores, hypertension, and smoking showed pronounced SHAP contributions, suggesting cumulative risk effects in elderly patients. Consistent with the approach of Sun et al, who used SHAP to enhance transparency,14 this study further identifies potential targets for lifestyle intervention, providing actionable guidance for the personalized health management of elderly patients and supporting its potential integration of this predictive model into structured management platforms such as the MMC system to facilitate early screening and targeted intervention for DPN in elderly T2DM patients.

Limitations

In this study, several limitations should be acknowledged. First, this study used a retrospective design, which may introduce information bias. Second, the data were derived from a single center, limiting representativeness and potentially restricting the model’s applicability to other populations. Third, although the model achieved strong performance in internal validation, external validation using independent multi-center datasets was not conducted, which limits generalizability. Future studies should therefore include prospective, multi-center cohorts with external validation to further refine the model and enhance its clinical utility.

Conclusion

This study developed a risk prediction model for DPN in elderly patients with T2DM using four machine learning algorithms. Through comprehensive evaluation and comparison, the XGBoost model exhibited optimal performance in terms of discrimination, calibration, and clinical utility, indicating its potential value as an early screening tool in clinical practice. The model enables rapid identification of high-risk individuals based on easily obtainable clinical and lifestyle indicators, which may assist clinicians in implementing timely risk stratification and targeted interventions. Through variable importance analysis, we identified diabetes duration, HbA1c, poor sleep, CCI, sugar-sweetened beverage intake, PAD, sedentary behavior, smoking and hypertension as key predictors. These findings provide a valuable reference for identifying elderly T2DM patients at high risk of developing DPN. However, the model was developed using retrospective, single-center data and without external validation in independent cohorts, which may limit its generalizability. Future research should focus on multi-center, large-sample prospective studies with external validation to continuously optimize model performance and enhance its clinical applicability and generalizability.

Data Sharing Statement

The data used to support the findings of this study are available from the corresponding author upon request.

Ethics Approval and Consent to Participate

This study followed the principles of the Declaration of Helsinki and was approved by the ethics committee of Gongli hospital of Shanghai Pudong New Area (Approval No: GLYYls2024-066). The requirement for informed consent was waived by the same ethics committee due to the retrospective nature of the study.

Acknowledgments

We extend our heartfelt gratitude to the Endocrinology Department of Gongli Hospital in Pudong New Area, Shanghai, for their tremendous support and assistance. We also sincerely thank the patients who participated in the study for their cooperation and contributions.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; 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

The project was supported by the Shanghai Pudong New Area Health Talents Development Initiative (2024PDWSYCBJ-04).

Disclosure

The authors declare no conflicts of interest in this work.

References

1. Selvarajah D, Kar D, Khunti K, et al. Diabetic peripheral neuropathy: advances in diagnosis and strategies for screening and early intervention. Lancet Diabetes Endocrinol. 2019;7(12):938–948. doi:10.1016/S2213-8587(19)30081-6

2. Shiferaw WS, Akalu TY, Work Y, Aynalem YA. Prevalence of diabetic peripheral neuropathy in Africa: a systematic review and meta-analysis. BMC Endocr Disord. 2020;20(1):49. doi:10.1186/s12902-020-0534-5

3. Wang W, Ji Q, Ran X, et al. Prevalence and risk factors of diabetic peripheral neuropathy: a population-based cross-sectional study in China. Diabetes Metab Res Rev. 2023;39(8):e3702. doi:10.1002/dmrr.3702

4. Kimura K, Endo Y. Age-related changes in the tactile pressure sensitivity threshold in the sole of the foot in diabetic patients. J Phys Ther Sci. 2018;30(7):917–920. doi:10.1589/jpts.30.917

5. Andersen ST, Witte DR, Dalsgaard EM, et al. Risk factors for incident diabetic polyneuropathy in a cohort with screen-detected type 2 diabetes followed for 13 years: ADDITION-Denmark. Diabetes Care. 2018;41(5):1068–1075. doi:10.2337/dc17-2062

6. Savelieff MG, Elafros MA, Viswanathan V, Jensen TS, Bennett DL, Feldman EL. The global and regional burden of diabetic peripheral neuropathy. Nat Rev Neurol. 2025;21(1):17–31. doi:10.1038/s41582-024-01041-y

7. Ruan MH. Current status and influencing factors of delayed medical care among 236 elderly patients with diabetic peripheral neuropathy. J Nurs. 2025;32(16):64–68. doi:10.16460/j.issn2097-6569.2025.16.064

8. Zhang W, Chen L. A nomogram for predicting the possibility of peripheral neuropathy in patients with type 2 diabetes mellitus. Brain Sci. 2022;12(10):1328. doi:10.3390/brainsci12101328

9. Wu B, Niu Z, Hu F. Study on risk factors of peripheral neuropathy in type 2 diabetes mellitus and establishment of prediction model. Diabetes Metab J. 2021;45(4):526–538. doi:10.4093/dmj.2020.0100

10. Liu H, Liu Q, Chen M, Lu C, Feng P. Construction and validation of a nomogram model for predicting diabetic peripheral neuropathy. Front Endocrinol. 2024;15:1419115. doi:10.3389/fendo.2024.1419115

11. Luo L, Long X, Cheng C, Xu Q, Li J. Development and validation of a risk nomogram model for predicting peripheral neuropathy in patients with type 2 diabetes mellitus. Front Endocrinol. 2024;15:1338167. doi:10.3389/fendo.2024.1338167

12. Ning Y, Ong MEH, Chakraborty B, et al. Shapley variable importance cloud for interpretable machine learning. Patterns. 2022;3(4):100452. doi:10.1016/j.patter.2022.100452

13. Jiang A, Li J, Wang L, et al. Multi-feature, Chinese-western medicine-integrated prediction model for diabetic peripheral neuropathy based on machine learning and SHAP. Diabetes Metab Res Rev. 2024;40(4):e3801. doi:10.1002/dmrr.3801

14. Sun M, Sun X, Wang F, Liu L. Machine learning-based prediction of diabetic peripheral neuropathy: model development and clinical validation. Front Endocrinol. 2025;16:1614657. doi:10.3389/fendo.2025.1614657

15. Liu L, Bi B, Gui M, et al. Development and internal validation of an interpretable risk prediction model for diabetic peripheral neuropathy in type 2 diabetes: a single-centre retrospective cohort study in China. BMJ Open. 2025;15(4):e092463. doi:10.1136/bmjopen-2024-092463

16. Lian X, Qi J, Yuan M, et al. Study on risk factors of diabetic peripheral neuropathy and establishment of a prediction model by machine learning. BMC Med Inf Decis Making. 2023;23(1):146. doi:10.1186/s12911-023-02232-1

17. Davies MJ, Aroda VR, Collins BS, et al. Management of hyperglycemia in type 2 diabetes, 2022. A consensus report by the American diabetes association (ADA) and the european association for the study of diabetes (EASD). Diabetes Care. 2022;45(11):2753–2786. doi:10.2337/dci22-0034

18. Lee SWH, Ng KY, Chin WK. The impact of sleep amount and sleep quality on glycemic control in type 2 diabetes: a systematic review and meta-analysis. Sleep Med Rev. 2017;31:91–101. doi:10.1016/j.smrv.2016.02.001

19. American Diabetes Association Professional Practice Committee. 5. Facilitating positive health behaviors and well-being to improve health outcomes: standards of care in diabetes-2025. Diabetes Care. 2025;48(1 Suppl 1):S86–S127. doi:10.2337/dc25-S005

20. Chen W, Jiang H. Chinese guidelines for medical nutrition therapy for patients with diabetes (2022 edition). Asia Pac J Clin Nutr. 2024;33(2):118–152. doi:10.6133/apjcn.202406_33(2).0001

21. Riley RD, Ensor J, Snell KIE, et al. Calculating the sample size required for developing a clinical prediction model. BMJ. 2020;368:m441. doi:10.1136/bmj.m441

22. Luo Y, Li Y, Xie J, et al. Symptoms of depression are related to sedentary behavior and sleep duration in elderly individuals: a cross-sectional study of 49,317 older Chinese adults. J Affect Disord. 2022;308:407–412. doi:10.1016/j.jad.2022.04.102

23. Tehan PE, Mills J, Leask S, et al. Toe‐brachial index and toe systolic blood pressure for the diagnosis of peripheral arterial disease. Cochrane Database Syst Rev. 2024;2024(10):CD013783. doi:10.1002/14651858.CD013783.pub2

24. Meng LL, Liu Y, Geng RN, Tang YZ, Li DQ. Association of diabetic vascular complications with poor sleep complaints. Diabetol Metab Syndr. 2016;8:80. doi:10.1186/s13098-016-0195-8

25. Chinese Elderly Type 2 Diabetes Prevention and Treatment of Clinical Guidelines Writing Group, Geriatric Endocrinology and Metabolism Branch of Chinese Geriatric Society, Geriatric Endocrinology and Metabolism Branch of Chinese Geriatric Health Care Society, Geriatric Professional Committee of Beijing Medical Award Foundation, National Clinical Medical Research Center for Geriatric Diseases (PLA General Hospital). [clinical guidelines for prevention and treatment of type 2 diabetes mellitus in the elderly in China (2022 edition)]. Zhonghua Nei Ke Za Zhi. 2022;61(1):12–50. Polish. doi:10.3760/cma.j.cn112138-20211027-00751

26. Efthimiou O, Seo M, Chalkou K, Debray T, Egger M, Salanti G. Developing clinical prediction models: a step-by-step guide. BMJ. 2024;386:e078276. doi:10.1136/bmj-2023-078276

27. Pfannkuche A, Alhajjar A, Ming A, Walter I, Piehler C, Mertens PR. Prevalence and risk factors of diabetic peripheral neuropathy in a diabetics cohort: register initiative “diabetes and nerves. Endocr Metab Sci. 2020;1(1):100053. doi:10.1016/j.endmts.2020.100053

28. Tao Y, Zhang HY, MacGilchrist C, Kirwan E, McIntosh C. Prevalence and risk factors of painful diabetic neuropathy: a systematic review and meta-analysis. Diabet Res Clin Pract. 2025;222:112099. doi:10.1016/j.diabres.2025.112099

29. Khawaja N, Abu-Shennar J, Saleh M, Dahbour SS, Khader YS, Ajlouni KM. The prevalence and risk factors of peripheral neuropathy among patients with type 2 diabetes mellitus; the case of Jordan. Diabetol Metab Syndr. 2018;10:8. doi:10.1186/s13098-018-0309-6

30. Park HT, Kim YH, Lee KE, Kim JK. Behind the pathology of macrophage-associated demyelination in inflammatory neuropathies: demyelinating schwann cells. Cell Mol Life Sci. 2020;77(13):2497–2506. doi:10.1007/s00018-019-03431-8

31. Takeshita Y, Sato R, Kanda T. Blood-nerve barrier (BNB) pathology in diabetic peripheral neuropathy and In vitro human BNB model. Int J Mol Sci. 2020;22(1):62. doi:10.3390/ijms22010062

32. American Diabetes Association. Standards of medical care in diabetes-2021 abridged for primary care providers. Clin Diabetes. 2021;39(1):14–43. doi:10.2337/cd21-as01

33. Wang Y, Pei L, Wang M, Sun S, Wang S, Zhang Q. Effect of peer support on adults with diabetes-related peripheral neuropathy. Patient Educ Couns. 2022;105(4):828–834. doi:10.1016/j.pec.2021.07.047

34. Fan K, Zhao Y. Mobile health technology: a novel tool in chronic disease management. Intell Med. 2022;2(1):41–47. doi:10.1016/j.imed.2021.06.003

35. Hashemi-Madani N, Azadi M, Khajavi A, Emami Z, Khamseh ME. Peripheral arterial disease is associated with sensorimotor peripheral neuropathy in people with type 2 diabetes. Med J Islam Repub Iran. 2024;38:143. doi:10.47176/mjiri.38.143

36. Salih MH, Tsega SS, Alemayehu BT, Ferede AJ. Peripheral neuropathy and associated factors among type 2 diabetic patients attending referral hospitals in the amhara region, a multi-center cross-sectional study in Ethiopia. Sci Rep. 2024;14(1):19662. doi:10.1038/s41598-024-70314-9

37. Iwane S, Nemoto W, Miyamoto T, et al. Clinical and preclinical evidence that angiotensin-converting enzyme inhibitors and angiotensin receptor blockers prevent diabetic peripheral neuropathy. Sci Rep. 2024;14(1):1039. doi:10.1038/s41598-024-51572-z

38. Schipper SBJ, Van Veen MM, Elders PJM, et al. Sleep disorders in people with type 2 diabetes and associated health outcomes: a review of the literature. Diabetologia. 2021;64(11):2367–2377. doi:10.1007/s00125-021-05541-0

39. Galicia-Garcia U, Benito-Vicente A, Jebari S, et al. Pathophysiology of type 2 diabetes mellitus. Int J Mol Sci. 2020;21(17):6275. doi:10.3390/ijms21176275

40. Tan X, Chapman CD, Cedernaes J, Benedict C. Association between long sleep duration and increased risk of obesity and type 2 diabetes: a review of possible mechanisms. Sleep Med Rev. 2018;40:127–134. doi:10.1016/j.smrv.2017.11.001

41. Stumvoll M, Goldstein BJ, van Haeften TW. Type 2 diabetes: principles of pathogenesis and therapy. Lancet. 2005;365(9467):1333–1346. doi:10.1016/S0140-6736(05)61032-X

42. Ahaouari T, Berendsen B, Schaper N, et al. Combined associations of type 2 diabetes and peripheral neuropathy with device-measured physical activity and sedentary behaviour—the maastricht study. Diabetes Metab Res Rev. 2025;41(5):e70069. doi:10.1002/dmrr.70069

43. Heidari M, Zolaktaf V, Ghasemi G, Nejadian SL. Integrated exercise and glycemic and peripheral sensation control in diabetic neuropathy: a single-blind, randomized controlled trial. Int J Prev Med. 2021;12:169. doi:10.4103/ijpvm.IJPVM_306_20

44. Enders J, Elliott D, Wright DE. Emerging nonpharmacologic interventions to treat diabetic peripheral neuropathy. Antioxid Redox Signal. 2023;38(13–15):989–1000. doi:10.1089/ars.2022.0158

45. Elafros MA, Andersen H, Bennett DL, et al. Towards prevention of diabetic peripheral neuropathy: clinical presentation, pathogenesis, and new treatments. Lancet Neurol. 2022;21(10):922–936. doi:10.1016/S1474-4422(22)00188-0

46. Chen Y, You Y, Wei M, et al. Exploration of physical activity, sedentary behavior and insulin level among short sleepers. Front Endocrinol. 2024;15:1371682. doi:10.3389/fendo.2024.1371682

47. Braffett BH, Gubitosi-Klug RA, Albers JW, et al. Risk factors for diabetic peripheral neuropathy and cardiovascular autonomic neuropathy in the diabetes control and complications trial/epidemiology of diabetes interventions and complications (DCCT/EDIC) study. Diabetes. 2020;69(5):1000–1010. doi:10.2337/db19-1046

48. Clair C, Cohen MJ, Eichler F, Selby KJ, Rigotti NA. The effect of cigarette smoking on diabetic peripheral neuropathy: a systematic review and meta-analysis. J Gen Intern Med. 2015;30(8):1193–1203. doi:10.1007/s11606-015-3354-y

49. Chastin S, Gardiner PA, Harvey JA, et al. Interventions for reducing sedentary behaviour in community-dwelling older adults. Cochrane Database Syst Rev. 2021;6(6):CD012784. doi:10.1002/14651858.CD012784.pub2