Development and Validation of a Machine Learning-Based Risk Prediction Model for Ischemic Stroke-Diabetes Comorbidity

Development and Validation of a Machine Learning-Based Risk Prediction Model for Ischemic Stroke-Diabetes Comorbidity | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Development and Validation of a Machine Learning-Based Risk Prediction Model for Ischemic Stroke-Diabetes Comorbidity Litian Hu, Hongyu Sun This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8265041/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Aims: This study aimed to develop and validate machine learning-based risk prediction models for ischemic stroke-diabetes mellitus (IS-DM) comorbidity using routinely available clinical data, and to compare the performance of traditional logistic regression with backpropagation neural networks (BPNN). ​ Methods Health records of 16,406 community-dwelling adults from Beijing, China, we analyzed. From 41 initial candidate predictors across five categories, seven optimal predictors were selected through ​univariate analysis​ followed by ​multivariate analysis. The dataset was randomly split into training (70%) and validation (30%) sets. We developed prediction models using both logistic regression and BPNN approaches, with model performance evaluated through confusion matrix, AUC, and 10-fold cross-validation. ​ Results The single-hidden-layer BPNN model with three hidden nodes demonstrated superior predictive performance, achieving an AUC of 0.921 (95% CI: 0.92-0.93), outperforming logistic regression. Key predictors included age, marital status, fasting glucose, HbA1c, systolic blood pressure, serum creatinine, and serum sodium. However, the BPNN required significantly more computational resources. ​ Conclusion Machine learning approaches, particularly BPNN, can effectively predict IS-DM comorbidity risk using routine clinical parameters. These models could enhance early comorbidity detection in community settings and inform targeted prevention strategies. Despite it predictive efficacy, the computational demands of BPNN should be considered for clinical implementation. ischaemic stroke diabetes co-morbidities machine learning risk prediction models Figures Figure 1 Figure 2 Figure 3 1. Introduction Ischemic stroke (IS) and diabetes mellitus (DM) are two prevalent chronic diseases. IS is the most predominant type of stroke, accounting for approximately 71% of strokes worldwide 1 . Globally, there are over 500 million people with DM 2 , and this number is projected to rise to 1.3 billion by 2050 3 . IS and DM impose a substantial burden on society and healthcare systems. As, two prevalent chronic conditions demanding urgent attention, their prevention and treatment become an urgent public health challenge worldwide, particularly in developing countries, ​​Cardiometabolic multimorbidity (CMM) refers to the concurrent presence of cardiovascular diseases, cerebrovascular diseases, and metabolic disorders, representing a prevalent chronic disease multimorbidity pattern 4 . Extensive research 5 – 7 has demonstrated that DM and cardiovascular-cerebrovascular diseases (CCVD) constitute a prevalent multimorbidity cluster. Compared to single chronic conditions, CMM demonstrates synergistic disease interactions that increase systemic fragility and disrupt homeostatic control. These mechanisms promote multisystem dysregulation, accelerating the progression of CCVD and metabolic diseases 8 . The IS-DM comorbidity dyad is a common comorbidity pattern of CMM. DM is associated with extensive metabolic alterations that may contribute to stroke predisposition 9 , with a study 10 identifying DM as an independent risk factor for IS.​ A meta-analysis 11 revealed that nearly one-third of stroke patients have DM. Furthermore, among individuals with DM, the prevalence of IS is significantly higher compared to hemorrhagic stroke. Additionally, DM is associated with worse stroke outcomes.​ Evidence suggests that individuals with DM face a twofold increased risk of stroke compared to non-diabetic populations 12 . The IS-DM comorbidity dyad contributes significantly to the escalating global burden and mortality of cardiovascular diseases 13 . ​In this comorbidity state, patients face significantly higher risks of disability and mortality compared to those with either condition alone, leading to increased healthcare resource utilization and socioeconomic burden. This poses major challenges to health systems, underscoring the critical need to better understand and manage the CMM of DM and IS.​ Singer et al. 14 demonstrated that the development of comorbidity involves not only mutual influences between chronic diseases but also multifactorial interactions across biological, psychological, and social domains. These complex and diverse factors may exhibit nonlinear and synergistic effects, resulting in highly heterogeneous clinical manifestations of comorbidity.​ Risk prediction model is a mathematical framework designed to estimate the probability of specific populations developing particular diseases or experiencing defined clinical outcomes, which can fully leverage healthcare big data to enable early forecasting of disease onset and progression. Machine learning (ML) is a discipline focused on enabling computers to learn from data and extract meaningful patterns. 15 It excels at uncovering latent value within large-scale datasets and is particularly adept at analyzing complex, high-dimensional healthcare data. These capabilities give ML broad applications in exploring disease etiology, predicting chronic conditions, and advancing preventive strategies 16 .​ ML-based risk prediction model can assist clinicians in identifying high-risk patients, thereby reducing the societal burden of excessive screening while mitigating potential risks associated with underscreening 17 .​ In recent years, ML has been increasingly applied to construct predictive models for comorbidity.​ ​Studies by Md Ekramul Hossain et al. 18 and Ahmad Shaker Abdalrada et al. 19 employed ML algorithms to develop prediction models for DM- cardiovascular disease comorbidity. Vasilis Nikolaou et al. 20 utilized decision tree, random forest, and logistic regression to construct predictive models for chronic obstructive pulmonary disease- cardiovascular disease comorbidity. Additional research has established ML-based prediction models for cancer-depression comorbidity 21 .​ Although numerous studies have developed predictive models for various comorbidities, most rely on data from European and North American populations 22 .​ Current research predominantly focuses on single-disease prediction, with limited studies addressing comorbidity risk assessment. Moreover, most existing models rely on traditional linear approaches, which fail to capture the complex nonlinear interactions inherent in comorbid conditions.​ Current research on ML-based risk prediction models for CMM patterns - particularly the IS and DM comorbidity cluster - remains limited. The Backpropagation Neural Network (BPNN) is currently the most widely used multilayer feedforward neural network. It derives its name from the error backpropagation algorithm employed for adjusting network weights, and represents the most maturely developed, extensively applied, and stably performing neural network architecture to date. Thus, this study will utilize clinical databases from Chinese community hospitals to construct and evaluate CMM risk prediction models for IS-DM comorbidity using both traditional logistic regression and BPNN. Through comparative analysis, we aim to identify the optimal predictive model. The findings will: a) provide theoretical guidance for healthcare professionals to identify potential comorbid patients; b) enrich research on IS-DM CMM risk prediction; c) serve as a reference for future related studies; and d) offer valuable insights for developing prediction models for similar disease clusters.​ 2. Materials and methods 2.1 Data source and study cohort This study used data from residents' health records and service records of Ganjiakou Community Health Service Centre, Haidian District, Beijing, China. The center is located in the southeastern part of Haidian District, Beijing, serving a population of 123,000 residents. Among them, 28,500 are elderly aged 60 and above, including over 8,000 individuals aged 80+. With an aging population ratio approaching 20% in its service area, the region is transitioning into a moderate aging phase. This demographic profile provides a robust dataset of chronic disease patients for our study.​ Each resident contains a unique patient ID, age, sex, examination dates and examination data. The dataset includes documented diagnoses of chronic diseases, including DM, IS, hypertension, and related conditions.​ The study utilised documented diagnoses of chronic diseases to identify data for patients with DM and IS as our goal is to predict the risk of comorbidity. Several filtering criteria were applied to the initial dataset to collect the research dataset. The criteria for filtering included: a) selecting health examination data collected between January 1, 2023 and August 19, 2024, b) selecting participants aged > 18 years at the time of health examination, c) removing duplicate data, d)removing participants with malignancies, psychiatric disorders, disabilities, mortality, or outmigration, e) removing deceased or relocated participants, f) removing participants with missing data rates exceeding 48% (Given that this study adopts a retrospective cohort design, the dataset contains varying degrees of missing data across both study subjects and variables. To ensure adequate sample size for analysis, we have pragmatically relaxed the criteria for acceptable missingness rates in the dataset). 2.2 Data processing We utilized R version 4.4.1 for all data processing and analyses.​ This study employed a retrospective cohort design. Missing data were observed across both study participants and variables to varying degrees. To address this, we implemented the following protocol based on the actual data characteristics: participants with missing data rates exceeding 48% were excluded, while those with missing rates below 48% underwent imputation. Specifically, continuous variables were imputed using mean substitution, whereas categorical variables were handled via multiple imputation. The multiple imputation was performed using the "mice" R package with the Multivariate Imputation by Chained Equations method 23 , 24 . The parameter configurations for the multiple imputation in this study are detailed in Table 1 .​ Table 1 Parameter Specifications for Multiple Imputation Parameter Setting Imputation method Continuous variables: "pmm" Categorical variables: "polyreg" Number of imputed datasets (m) 5 (default) Iterations of chained equations(maxit) 10 Random seed 1234 Imbalanced data refers to datasets where sample sizes across different classes exhibit significant disparities. This imbalance in feature classification can bias models toward predicting majority-class samples to achieve higher accuracy, thereby neglecting minority-class samples of greater research interest and ultimately compromising predictive performance. In this study, which focuses on predicting the comorbidity of DM and IS, the dataset is inherently imbalanced.To address this, we employed the ​Synthetic Minority Oversampling Technique (SMOTE)​​ 25 via the ​​"SmoteClassif​​" function from the ​​"UBL" R package. This approach synthetically augments minority-class (comorbidity) samples to achieve data balance, after careful consideration of dataset size, memory usage, and computational efficiency. 2.3 Construction of risk prediction models This study developed risk prediction models for cardio-cerebrovascular metabolic comorbidity patterns between DM and IS using traditional logistic regression and BPNN.We randomly divided the final included data into training and validation sets at a 7:3 ratio after balancing treatment. The training set was used for model construction, while the validation set served for model verification. The study focuses on identifying common risk factors for the comorbidity of DM and IS. Accordingly, these common risk factors were incorporated as predictors in the predictive model, while the presence/absence of DM and IS served as outcome measures. This study employed the ​​"glm​​" function from the ​​"caret" R package​ to construct the logistic regression-based risk prediction model. A typical BPNN consists of three layers: an input layer, a hidden layer (intermediate layer), and an output layer. Generally, each layer contains multiple neurons (nodes), which are fully interconnected between adjacent layers (meaning each neuron in the preceding layer connects to all neurons in the subsequent layer). Neurons from one layer exclusively influence those in the next layer, while no interactions exist among neurons within the same layer (i.e., intra-layer neurons operate independently). The magnitude of connection weights reflects the degree of influence exerted by upper-layer neurons on lower-layer neurons. The most common BPNN architecture employs a three-layer structure with a single hidden layer. The machine learning process of a BP neural network can be summarized as follows: First, signals propagate forward from the input layer through the hidden layer to the output layer, where the error signal is obtained by calculating the difference between the expected and actual output values. Subsequently, the error is backpropagated from the output layer through the hidden layer to the input layer, adjusting the connection weights (ω) layer by layer. This iterative process alternates between forward propagation and backward weight adjustment, continuously optimizing the connection weights (ω) and thresholds (b). Finally, the network undergoes a "learning convergence" phase, where the global error approaches a minimum, indicating that the network is converging toward an optimal staet 26 . 2.4 ​Identification of predictors Based on literature review and data availability, we initially selected 41 candidate predictive features across five categories for model construction: a) sociodemographic factors, b) family history, c) physiological and biochemical parameters, d) lifestyle behaviors, and e) mental health factors. Supplementary Table 1 details the sources and coding schemes of these risk factors. In the original dataset, dietary habits were categorized into six types: balanced meat-vegetable diet, meat-dominated diet, vegetarian-dominated diet, salt-preference, oil-preference, and sugar-preference. For this study, we reclassified these into two groups: unhealthy dietary preferences and healthy eating habits. Participants were classified as having unhealthy dietary preferences if they exhibited any of the following characteristics: salt-preference, oil-preference, or sugar-preference. Conversely, those maintaining a balanced meat-vegetable diet without exhibiting any of these three preferences were categorized as having healthy eating habits. This study employed ​univariate analysis​ followed by ​multivariate analysis​ to screen for predictive factors. First, the relationship between each independent variable and the dependent variable (outcome) was examined using ​univariate analysis. Variables showing ​statistically significant differences ( P < 0.05)​​ were preliminarily selected as candidate predictors. ​Spearman correlation analysis​ was then performed to assess multicollinearity among the selected predictors. Variables with high correlation coefficients (indicating collinearity) were excluded to avoid redundancy in the model.​ The remaining candidate predictors were entered into a ​multivariate logistic regression model. Only variables that retained ​statistical significance ( P < 0.05)​​ in the multivariate analysis were retained as ​final predictive variables. 2.5 Model Evaluation and Comparison The confusion matrix is widely employed to assess the classification accuracy of machine learning models, providing an intuitive representation of the discrepancy between predicted and actual outcomes. In this study, we initially evaluated model performance by calculating key metrics derived from the confusion matrix, including Accuracy, Precision, Sensitivity (Recall), and F1-score. Furthermore, we utilized the "pROC" and "plot" packages to generate Receiver Operating Characteristic (ROC) curves and compute the Area Under the Curve (AUC) values, enabling comparative analysis of the models' discriminative capabilities. Cross-validation serves as a robust method for evaluating model performance, particularly for assessing predictive accuracy and generalization ability. Among various cross-validation techniques, k-fold cross-validation has emerged as one of the most prevalent approaches 27 . Following established conventions 28 , we set k = 10, partitioning the dataset into 10 distinct subsets or "folds". This procedure involved conducting 10 independent training and validation iterations, with the model's predictive and generalization performance ultimately determined by averaging the results across all 10 validation cycles. For implementation, we employed the "caret" package to perform 10-fold cross-validation in this study. 3. Results 3.1 Baseline Characteristics of Study Participants After data cleaning and screening, a total of 16,406 participants were finally included in this study, with the participant selection flowchart shown in Fig. 1 . Among them, there were 157 cumulative cases of ischemic stroke, with an incidence rate of 96 per 10,000 population; 5,305 cumulative cases of diabetes, with an incidence rate of 3,234 per 10,000 population; and 72 cumulative cases of comorbid ischemic stroke and diabetes, with an incidence rate of 44 per 10,000 population. The features of the included study participants and the results of univariate analysis are presented in Supplementary Table 2 . 3.2 Predictor Screening Results Fourteen predictors showing statistically significant differences ( P < 0.05) in univariate analysis were initially selected, including age, marital status, waist circumference, systolic blood pressure, fasting glucose, glycated hemoglobin, serum creatinine, white blood cell count, serum sodium, alanine aminotransferase (HbA1c), total cholesterol, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, and exercise frequency. Collinearity testing among these 14 predictors revealed that two predictor pairs (total cholesterol and low-density lipoprotein cholesterol) exhibited correlation coefficients exceeding the predefined threshold of 0.8, while all other predictor pairs showed correlation coefficients below 0.43. After excluding these two collinear predictors, the remaining 13 predictors were subjected to multivariate logistic regression analysis, ultimately yielding seven predictors that maintained statistical significance ( P < 0.05), as presented in Table 2 . 3.3 Model Construction Results The seven selected predictors were used as predictive variables for the comorbidity of IS and DM. A traditional logistic regression model was constructed, with the risk prediction equation as follows: ​Logit(P) = -1.13 + 0.24 × Age + 0.14 × Marital status + 0.21 × Systolic blood pressure − 0.60 × Fasting glucose + 1.47 × Glycated hemoglobin (HbA1c) + 0.44 × Serum creatinine − 0.79 × Blood sodium.​​ Details are presented in ​ Table 3 . In this study, the input layer of the BPNN was configured with 7 nodes based on the included predictors. Since the outcome was binary (presence or absence of comorbidity), the output layer consisted of 1 neuron. A single hidden layer was adopted, with the number of hidden nodes ranging from 1 to 21, determined by empirical node calculation methods. The learning rate was set within the range of 0.08–0.1. The final results indicated that the model with 3 hidden nodes demonstrated the best performance, whereas models with other node numbers either failed to converge during training or exhibited inferior predictive capability. A schematic diagram of the optimal BPNN-based risk prediction model is illustrated in ​Fig. 2 . Table 2 Final Inclusion of Predictors (7) Predictor t Value P Value Age 2.55 0.0107 Systolic Blood Pressure(mmHg) 2.17 0.0303 Fasting Glucose(mmol/L) -5.99 < 0.001 HbA1c(%) 11.66 < 0.001 Serum Creatinine(µmol/L) 4.33 < 0.001 Serum Sodium(mmol/L) -6.52 < 0.001 Marital Status 3.28 0.00105 Table 3 Results of Risk Prediction Models Constructed by Traditional Logistic Regression Variable β SE Waldχ 2 P OR 95%CI Constant Term -1.13 0.022 2672.63 <0.001 0.32 0.31 ~ 0.34 Age 0.24 0.022 112.15 <0.001 1.27 1.21 ~ 1.32 Systolic Blood Pressure (mmHg) 0.21 0.021 96.05 <0.001 1.23 1.18 ~ 1.28 Fasting Glucose(mmol/L) -0.60 0.024 642.55 <0.001 0.55 0.52 ~ 0.58 HbA1c(%) 1.47 0.025 3461.02 <0.001 4.33 4.13 ~ 4.55 Serum Creatinine(µmol/L) 0.44 0.019 512.81 <0.001 1.55 1.50 ~ 1.61 Serum Sodium(mmol/L) -0.79 0.019 1745.77 <0.001 .46 0.44 ~ 0.47 Marital Status 0.14 0.017 70.93 <0.001 1.16 1.12 ~ 1.19 3.4 Model Comparison Results The accuracy, precision, recall, F1-score, specificity, AUC, and their 95% confidence intervals (95% CI) of the two prediction models on the validation set are presented in ​ Table 4 .The accuracy of the logistic regression and BP neural network models was 0.822 and 0.862, respectively, indicating that the BP neural network achieved better classification performance. The precision was 0.816 for logistic regression and 0.938 for the BP neural network, suggesting that the BP neural network exhibited higher accuracy in predicting positive cases. The sensitivity (recall) was 0.833 for logistic regression and 0.776 for the BP neural network, demonstrating that logistic regression had a better ability to identify true positive cases. The F1-scores were 0.824 and 0.849, respectively, indicating that the BP neural network model was more robust.In terms of specificity, logistic regression and the BP neural network achieved 0.811 and 0.949, respectively, suggesting that the BP neural network performed better in correctly identifying negative-class samples (i.e., true negatives). Finally, the AUC values were 0.876 for logistic regression and 0.921 for the BP neural network, confirming that the BP neural network exhibited superior overall classification performance. The ROC curves of the two prediction models are presented in ​Fig. 3 , where the curve of the BPNN-based risk prediction model completely envelops that of the traditional logistic regression-based model. This demonstrates that the classification performance of the BPNN-based model is significantly superior to that of the logistic regression-based model. The performance of both models after k-fold cross-validation (k = 10) is presented in ​Table 5 . The cross-validation results show that the BPNN-based model achieved slightly higher mean accuracy and mean specificity compared to the logistic regression-based model, while the logistic regression model demonstrated marginally better mean sensitivity and mean AUC than the neural network model. Both models performed well in cross-validation; however, the BPNN-based risk prediction model did not exhibit a clear advantage over the logistic regression-based model in the cross-validation results. Table 4 Performance Metrics of Risk Prediction Models on the Validation Set Model Accuracy Precision Recall F1-Score Specificity AUC 95%CI LR 0.822 0.816 0.833 0.824 0.811 0.876 0.88 ~ 0.89 BPNN 0.862 0.938 0.776 0.849 0.949 0.921 0.92 ~ 0.93 Table 5 Cross-Validation Results of the Models Model Mean Accuracy Mean Recall Mean Specificity Mean AUC LR 0.817 0.830 0.803 0.881 BPNN 0.821 0.818 0.823 0.880 The computational efficiency and resource consumption of both models are presented in ​Table 6 , where training time refers to the duration required for model training (measured in seconds) and computational resource consumption indicates the memory usage during model construction (measured in megabytes, MB). The results show that the BPNN model required nearly ten times the training time of the logistic regression model. Additionally, the memory consumption of the BPNN was eighteen times greater than that of logistic regression. Table 6 Computational Efficiency and Resource Consumption of Model Training Model Training Time(s) Memory Usage(MB) LR 3.04 492.8 BNN 30.12 8892.8 Overall, the BPNN-based risk prediction model demonstrates superior predictive performance compared to the logistic regression-based approach. However, this enhanced capability comes at a substantial computational cost, as the neural network requires significantly greater training time and memory resources. 4. Discussion 4.1 Predictors of IS and DM Co-Morbidity Age and marital status in sociodemographic factors have long been important influences on chronic diseases. Studies have shown that population aging is one of the risk factors for stroke in China 29 , with a particularly noticeable increase in stroke cases among those over 50 years old 30 . A meta-analysis evaluating the relationship between marital status and cardiovascular diseases indicates that, compared to married individuals, being unmarried, divorced, or widowed increases the likelihood of stroke 28 . Possible reasons include delayed medical care-seeking among unmarried patients 31 , which to some extent contributes to disease progression. Other studies suggest that divorce or widowhood may activate chronic inflammatory responses, thereby increasing the risk of cardiovascular and metabolic diseases 32 . Stress-related theories propose that losing a partner or poor marital quality may negatively impact an individual's economic, behavioral, and emotional health, reducing their ability to prevent, detect, and treat diseases 33 . The increased stress from divorce or widowhood may further amplify the effects of inflammation on the body 34 , ultimately influencing the progression of stroke and diabetes 35 . Additionally, married individuals may have greater financial resources, better access to healthcare 36 , and are less likely to experience delays or unmet medical and mental health needs due to costs 37 , thereby reducing the risk of chronic diseases to some extent. Systolic blood pressure is a significant influencing factor for chronic diseases, particularly stroke. Research indicates that hypertension is the leading risk factor for stroke 38 , and it remains a major risk factor in cases treated at tertiary and secondary public hospitals or private hospitals 39 . Furthermore, among diabetic patients, systolic blood pressure ≥ 130 mmHg is associated with a higher risk of cardiovascular mortality 40 . A study have demonstrated that for hypertensive patients at high cardiovascular risk, intensive blood pressure-lowering strategies targeting lower systolic thresholds can significantly prevent major vascular events with acceptable additional risks, regardless of diabetes comorbidity or prior stroke history 41 . Therefore, controlling systolic blood pressure is crucial for managing the comorbidity of ischemic stroke and diabetes. Fasting glucose and glycated hemoglobin (HbA1c) reflect blood sugar levels, with fasting glucose being a key indicator for DM diagnosis and daily monitoring, while HbA1c reflects long-term trends in glycemic control. These two indicators are not only directly related to DM but also closely linked to IS. For instance, acute hyperglycemia and DM are both associated with poor outcomes after IS 11 , and elevated HbA1c levels are linked to an increased risk of first-time stroke in both diabetic and non-diabetic populations 42 . Earlier studies have conclusively demonstrated that chronic hyperglycemia and advanced glycation end-products (AGE) collectively elevate the risk of cardio-cerebrovascular events by inducing endothelial dysfunction and cellular damage 13 . Thus, blood sugar control is equally critical for managing the comorbidity of IS and DM. Serum creatinine and sodium levels are typically used as biochemical indicators of renal function, yet numerous studies have identified their association with IS and DM. A meta-analysis found a significant correlation between the urine albumin-to-creatinine ratio (ACR) and stroke incidence 43 , with hypertensive patients having an ACR ≥ 10 mg/g showing a significantly higher risk of first-time IS 44 . Other studies indicate that ACR exhibits high individual variability in type II diabetes patients 45 . Some research suggests that low serum creatinine is a risk factor for type II diabetes 46 , and serum creatinine levels can serve as a muscle mass indicator to predict diabetes progression 47 . Sodium levels are also associated with increased chronic disease risk, with studies showing that serum sodium > 142 mmol/L or < 138 mmol/L is linked to a higher risk of chronic diseases, including DM 48 . Therefore, serum creatinine and sodium levels can, to some extent, serve as predictive factors for the comorbidity of IS and DM. 4.2 Risk Prediction Models for Comorbidity Screening Insulin resistance or deficient insulin secretion in diabetic patients not only impairs glycemic control but also disrupts adipocyte function, leading to elevated circulating free fatty acids. This metabolic derangement precipitates a cascade of pathological processes including chronic inflammation, endothelial dysfunction, and atherosclerosis - all established risk factors for cardiovascular disease 49 . Notably, DM is frequently diagnosed only after the manifestation of complications such as IS 50 , indicating that many patients may have already developed substantial vascular damage or other serious health impairments at the time of diagnosis. China's current preventive framework primarily employs community-based strategies targeting high-risk populations for stroke prevention 51 . However, epidemiological evidence demonstrates that only 11% of stroke events are prevented through such high-risk approaches 52 , with the majority of cerebrovascular and cardiovascular incidents occurring in individuals classified as low-risk 53 . Of particular clinical concern is the bidirectional pathological relationship between IS and DM: IS represents the second leading cause of mortality following cancer in diabetic populations 54 , while acute hyperglycemia during IS exacerbates poor outcomes through potentiation of inflammatory responses and oxidative stress 55 . These findings underscore the critical importance of developing predictive models for early identification of stroke-diabetes comorbidity. Our risk prediction model incorporates routinely available clinical parameters that enable accurate identification of at-risk individuals, thereby facilitating several key clinical applications: a)​ Early risk stratification: Enables timely implementation of preventive measures by both patients and healthcare providers, potentially reducing disease incidence and improving quality of life 18 , b)​ Clinical utility: Provides community and clinical nurses with a tool for initiating preemptive interventions based on individualized risk assessment 56 , c) ​Healthcare optimization: Supports medical institutions in resource allocation and service planning 57 , d)​ Public health policymaking: Informs the development of targeted chronic disease management programs for at-risk populations, potentially alleviating overall healthcare burdens 58 . The implementation of this predictive model may substantially enhance current preventive strategies by shifting the paradigm from reactive to proactive care, ultimately mitigating the significant clinical and socioeconomic impacts of this high-morbidity comorbidity. 4.3 Limitations and Potential Future Works This study has several limitations that should be acknowledged. First, the sample was limited to the Ganjiaokou Community in Haidian District, Beijing, which may affect the generalizability of our findings. Second, while we implemented strict quality control measures, the variables exhibited varying degrees of missing data. To maintain an adequate sample size for model development, we extended the acceptable missing data threshold from the conventional < 30% to 48%. This adjustment may have influenced the quality of data imputation 59 and consequently impacted model performance, as potentially reflected in the relatively low sensitivity of our BPNN model. Future studies could compare different imputation methods and conduct sensitivity analyses to minimize the effects of high missingness rates on model performance. Although BPNN can effectively capture complex nonlinear relationships, their inherent complexity may lead to overfitting during training, potentially compromising generalizability to test datasets 60 . Subsequent research could incorporate techniques such as regularization and dropout to mitigate overfitting risks. Additionally, our prediction models require external validation in independent cohorts to fully assess their generalizability, transportability, and clinical applicability. Another consideration is the substantial computational resources required for BPNN training compared to logistic regression, suggesting the need for careful evaluation of the cost-performance trade-off in practical applications. Most existing studies develop risk prediction models using retrospective cohort data 61 . While such data are more readily available, prospective cohort studies that measure candidate predictors before outcome occurrence may yield more reliable results 62 . Future research could therefore benefit from using prospective cohort data. Furthermore, due to data limitations, several established risk factors for IS or DM - including the acute-to-chronic glycemic ratio 63 , urinary albumin-to-creatinine ratio (UACR) 45 , and residential location (urban/rural) 38 - were not included as candidate predictors in our analysis. Subsequent studies could incorporate these and other potential predictors, along with multimodal data such as imaging, genomics, and dynamic monitoring parameters, to further enhance model performance. Emerging methodologies such as Federated Learning 64 , a distributed machine learning approach that enables model training across client devices while preserving data privacy, could facilitate multi-center data sharing in future risk prediction model development. The rapid development of "Internet + Nursing" in China, including evidence-based studies on its application for type 2 diabetes management 65 , presents opportunities for integrating risk prediction models into digital healthcare platforms. Future research could also explore the development of risk prediction models for other cardiometabolic multimorbidity patterns, providing theoretical foundations for chronic disease prevention and control strategies to alleviate the burden of chronic disease management. 5. Conclusions This study focused on the comorbidity of IS and DM. Through comprehensive literature review and assessment of data availability, we systematically evaluated 41 candidate predictors spanning five categories: sociodemographic factors, family history factors, physiological and biochemical parameters, lifestyle behaviors, and mental health indicators. Following rigorous screening, seven optimal predictors were selected for model construction, including age, marital status, fasting glucose, HbA1c, systolic blood pressure, serum creatinine, and serum sodium. We developed and compared risk prediction models using both traditional logistic regression and BPNN approaches. Comprehensive model evaluation demonstrated that a single-hidden-layer BPNN architecture with three hidden nodes yielded optimal predictive performance. The final model exhibited excellent discrimination ability, achieving an accuracy of 0.938 and an area under the receiver operating characteristic curve (AUC) of 0.921, indicating robust predictive capability for IS-DM comorbidity risk assessment. Declarations Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Funding This study was funded by the Natural Science Foundation of Beijing Municipality (L222103). Ethics approval and consent to participate This study was performed in accordance with the Helsinki standard and the study’s protocol was approved by Peking University Institutional Review Board (PU IRB) (ref No: IRB00001052-23097). Participation in the study was voluntary and their informed consent was obtained. Author details 1 School of Nursing, Peking University, Beijing, China References Campbell B, Khatri P. Stroke. Lancet . 2020;396(10244):129-142. doi:10.1016/S0140-6736(20)31179-X Sun H, Saeedi P, Karuranga S, et al. IDF Diabetes Atlas: Global, regional and country-level diabetes prevalence estimates for 2021 and projections for 2045. Diabetes Research and Clinical Practice . 2022;183:109119. doi:10.1016/j.diabres.2021.109119 Ong KL, Stafford LK, McLaughlin SA, et al. Global, regional, and national burden of diabetes from 1990 to 2021, with projections of prevalence to 2050: a systematic analysis for the Global Burden of Disease Study 2021. The Lancet . 2023;402(10397):203-234. doi:10.1016/S0140-6736(23)01301-6 Gupta P, Patel SA, Sharma H, et al. Burden, patterns, and impact of multimorbidity in North India: findings from a rural population-based study. BMC Public Health . 2022;22(1):1101. doi:10.1186/s12889-022-13495-0 刘茜, 张静怡, 李玉静. 唐山市老年住院患者共病模式系统聚类分析. 中国预防医学杂志. 2023;24(12):1333-1338. doi:10.16506/j.1009-6639.2023.12.011 张海波, 温雯婷, 谢佳东, 姚玲, 马靓, 赵俊. 三级公立医院老年糖尿病共病患者疾病特征与住院费用分析. 中国慢性病预防与控制. 2024;32(07):534-537. doi:10.16386/j.cjpccd.issn.1004-6194.2024.07.012 纪倩云, 李曼, 崔雯霞, et al. 老年2型糖尿病住院病人共病情况分析. 实用老年医学. 2021;35(11):1170-1173+1177. Wu Z, Luo S, Cai D, et al. The causal relationship between metabolic syndrome and its components and cardiovascular disease: A mendelian randomization study. Diabetes Research and Clinical Practice . 2024;211:111679. doi:10.1016/j.diabres.2024.111679 Babel RA, Dandekar MP. A Review on Cellular and Molecular Mechanisms Linked to the Development of Diabetes Complications. CDR . 2021;17(4):457-473. doi:10.2174/1573399816666201103143818 Misirli H, Somay G, Ozbal N, Yasar EN. Relation of lipid and lipoprotein(a) to ischaemic stroke. J Clin Neurosci . 2002;9(2):127-132. doi:10.1054/jocn.2001.1030 Lau L, Lew J, Borschmann K, Thijs V, Ekinci EI. Prevalence of diabetes and its effects on stroke outcomes: A meta‐analysis and literature review. J of Diabetes Invest . 2019;10(3):780-792. doi:10.1111/jdi.12932 Davis TME. Risk Factors for Stroke in Type 2 Diabetes Mellitus: United Kingdom Prospective Diabetes Study (UKPDS) 29. Arch Intern Med . 1999;159(10):1097. doi:10.1001/archinte.159.10.1097 Maida CD, Daidone M, Pacinella G, Norrito RL, Pinto A, Tuttolomondo A. Diabetes and Ischemic Stroke: An Old and New Relationship an Overview of the Close Interaction between These Diseases. IJMS . 2022;23(4):2397. doi:10.3390/ijms23042397 Singer M, Bulled N, Ostrach B, Mendenhall E. Syndemics and the biosocial conception of health. Lancet . 2017;389(10072):941-950. doi:10.1016/S0140-6736(17)30003-X Deo RC. Machine Learning in Medicine: Will This Time Be Different? Circulation . 2020;142(16):1521-1523. doi:10.1161/CIRCULATIONAHA.120.050583 孙涛, 徐秀林. 基于机器学习的医疗大数据分析与临床应用. 软件导刊. 2019;18(11):10-14. doi:10.11907/rjdk.191047 Van Der Heijden AA, Nijpels G, Badloe F, et al. Prediction models for development of retinopathy in people with type 2 diabetes: systematic review and external validation in a Dutch primary care setting. Diabetologia . 2020;63(6):1110-1119. doi:10.1007/s00125-020-05134-3 Hossain ME, Uddin S, Khan A. Network analytics and machine learning for predictive risk modelling of cardiovascular disease in patients with type 2 diabetes. Expert Systems with Applications . 2021;164:113918. doi:10.1016/j.eswa.2020.113918 Abdalrada AS, Abawajy J, Al-Quraishi T, Islam SMS. Machine learning models for prediction of co-occurrence of diabetes and cardiovascular diseases: a retrospective cohort study. J Diabetes Metab Disord . 2022;21(1):251-261. doi:10.1007/s40200-021-00968-z Nikolaou V, Massaro S, Garn W, Fakhimi M, Stergioulas L, Price D. The cardiovascular phenotype of Chronic Obstructive Pulmonary Disease (COPD): Applying machine learning to the prediction of cardiovascular comorbidities. Respiratory Medicine . 2021;186:106528. doi:10.1016/j.rmed.2021.106528 Wang X, Eichhorn J, Haq I, Baghal A. Resting-state brain metabolic fingerprinting clusters (biomarkers) and predictive models for major depression in multiple myeloma patients. Bauckneht M, ed. PLoS ONE . 2021;16(5):e0251026. doi:10.1371/journal.pone.0251026 Alsaleh MM, Allery F, Choi JW, et al. Prediction of disease comorbidity using explainable artificial intelligence and machine learning techniques: A systematic review. International Journal of Medical Informatics . 2023;175:105088. doi:10.1016/j.ijmedinf.2023.105088 Chawla NV, Bowyer KW, Hall LO, Kegelmeyer WP. SMOTE: Synthetic Minority Over-sampling Technique. jair . 2002;16:321-357. doi:10.1613/jair.953 Haibo He, Garcia EA. Learning from Imbalanced Data. IEEE Trans Knowl Data Eng . 2009;21(9):1263-1284. doi:10.1109/TKDE.2008.239 Guo J, Wu H, Chen X, Lin W. Adaptive SV-Borderline SMOTE-SVM algorithm for imbalanced data classification. Applied Soft Computing . 2024;150:110986. doi:10.1016/j.asoc.2023.110986 马锐. 人工神经网络原理. 机械工业出版社; 2010. Kuhn M, Johnson K. Applied Predictive Modeling . Springer New York; 2013. doi:10.1007/978-1-4614-6849-3 Wong CW, Kwok CS, Narain A, et al. Marital status and risk of cardiovascular diseases: a systematic review and meta-analysis. Heart . 2018;104(23):1937-1948. doi:10.1136/heartjnl-2018-313005 Wang L, Zhou B, Zhao Z, et al. Body-mass index and obesity in urban and rural China: findings from consecutive nationally representative surveys during 2004–18. The Lancet . 2021;398(10294):53-63. doi:10.1016/S0140-6736(21)00798-4 Wang W, Jiang B, Sun H, et al. Prevalence, Incidence, and Mortality of Stroke in China: Results from a Nationwide Population-Based Survey of 480 687 Adults. Circulation . 2017;135(8):759-771. doi:10.1161/CIRCULATIONAHA.116.025250 Austin D, Yan AT, Spratt JC, et al. Patient characteristics associated with self-presentation, treatment delay and survival following primary percutaneous coronary intervention. European Heart Journal: Acute Cardiovascular Care . 2014;3(3):214-222. doi:10.1177/2048872614527011 Yang YC, Boen C, Gerken K, Li T, Schorpp K, Harris KM. Social relationships and physiological determinants of longevity across the human life span. Proc Natl Acad Sci USA . 2016;113(3):578-583. doi:10.1073/pnas.1511085112 Quinones PA, Kirchberger I, Heier M, et al. Marital status shows a strong protective effect on long-term mortality among first acute myocardial infarction-survivors with diagnosed hyperlipidemia – findings from the MONICA/KORA myocardial infarction registry. BMC Public Health . 2014;14(1):98. doi:10.1186/1471-2458-14-98 Brown RL, LeRoy AS, Chen MA, et al. Grief Symptoms Promote Inflammation During Acute Stress Among Bereaved Spouses. Psychol Sci . 2022;33(6):859-873. doi:10.1177/09567976211059502 Vujcic I, Vlajinac H, Dubljanin E, et al. Long-term prognostic significance of living alone and other risk factors in patients with acute myocardial infarction. Ir J Med Sci . 2015;184(1):153-158. doi:10.1007/s11845-014-1079-2 Dupre ME, Nelson A. Marital history and survival after a heart attack. Social Science & Medicine . 2016;170:114-123. doi:10.1016/j.socscimed.2016.10.013 Elton E, Gonzales G. Health Insurance Coverage and Access to Care by Sexual Orientation and Marital/Cohabitation Status: New Evidence from the 2015–2018 National Health Interview Survey. Popul Res Policy Rev . 2022;41(2):479-493. doi:10.1007/s11113-021-09670-7 Tu WJ, Zhao Z, Yin P, et al. Estimated Burden of Stroke in China in 2020. JAMA Netw Open . 2023;6(3):e231455. doi:10.1001/jamanetworkopen.2023.1455 Wang YJ, Li ZX, Gu HQ, et al. China Stroke Statistics: an update on the 2019 report from the National Center for Healthcare Quality Management in Neurological Diseases, China National Clinical Research Center for Neurological Diseases, the Chinese Stroke Association, National Center for Chronic and Non-communicable Disease Control and Prevention, Chinese Center for Disease Control and Prevention and Institute for Global Neuroscience and Stroke Collaborations. Stroke Vasc Neurol . 2022;7(5):415-450. doi:10.1136/svn-2021-001374 Seng LL, Hai Kiat TP, Bee YM, Jafar TH. Real‐World Systolic and Diastolic Blood Pressure Levels and Cardiovascular Mortality in Patients With Type 2 Diabetes—Results From a Large Registry Cohort in Asia. JAHA . 2023;12(23):e030772. doi:10.1161/JAHA.123.030772 Liu J, Li Y, Ge J, et al. Lowering systolic blood pressure to less than 120 mm Hg versus less than 140 mm Hg in patients with high cardiovascular risk with and without diabetes or previous stroke: an open-label, blinded-outcome, randomised trial. The Lancet . 2024;404(10449):245-255. doi:10.1016/S0140-6736(24)01028-6 Mitsios JP, Ekinci EI, Mitsios GP, Churilov L, Thijs V. Relationship Between Glycated Hemoglobin and Stroke Risk: A Systematic Review and Meta‐Analysis. JAHA . 2018;7(11):e007858. doi:10.1161/JAHA.117.007858 Huang R, Chen X. Increased Spot Urine Albumin-to-Creatinine Ratio and Stroke Incidence: A Systematic Review and Meta-Analysis. Journal of Stroke and Cerebrovascular Diseases . 2019;28(10):104260. doi:10.1016/j.jstrokecerebrovasdis.2019.06.018 He P, Yang Y, Tian J, et al. Urinary albumin-to-creatinine ratio and the risk of first stroke in Chinese hypertensive patients treated with angiotensin-converting enzyme inhibitors. Hypertens Res . 2022;45(1):116-124. doi:10.1038/s41440-021-00780-5 Rasaratnam N, Salim A, Blackberry I, et al. Urine Albumin-Creatinine Ratio Variability in People With Type 2 Diabetes: Clinical and Research Implications. American Journal of Kidney Diseases . 2024;84(1):8-17.e1. doi:10.1053/j.ajkd.2023.12.018 Harita N, Hayashi T, Sato KK, et al. Lower Serum Creatinine Is a New Risk Factor of Type 2 Diabetes. Diabetes Care . 2009;32(3):424-426. doi:10.2337/dc08-1265 Kashima S, Inoue K, Matsumoto M, Akimoto K. Low serum creatinine is a type 2 diabetes risk factor in men and women: The Yuport Health Checkup Center cohort study. Diabetes & Metabolism . 2017;43(5):460-464. doi:10.1016/j.diabet.2017.04.005 Stookey JD, Kavouras SA, Suh H, Lang F. Underhydration Is Associated with Obesity, Chronic Diseases, and Death Within 3 to 6 Years in the U.S. Population Aged 51–70 Years. Nutrients . 2020;12(4):905. doi:10.3390/nu12040905 Schmidt AM. Highlighting Diabetes Mellitus: The Epidemic Continues. ATVB . 2018;38(1). doi:10.1161/ATVBAHA.117.310221 Magliano DJ, Boyko EJ, IDF Diabetes Atlas 10th edition scientific committee. IDF DIABETES ATLAS . 10th ed. International Diabetes Federation; 2021. Accessed March 14, 2025. http://www.ncbi.nlm.nih.gov/books/NBK581934/ Chao BH, Tu WJ, Wang LD. Initial establishment of a stroke management model in China: 10 years (2011–2020) of Stroke Prevention Project Committee, National Health Commission. Chinese Medical Journal . 2021;134(20):2418-2420. doi:10.1097/CM9.0000000000001856 Ma Q, Li R, Wang L, et al. Temporal trend and attributable risk factors of stroke burden in China, 1990–2019: an analysis for the Global Burden of Disease Study 2019. The Lancet Public health . 2021;6(12):e897-e906. doi:10.1016/S2468-2667(21)00228-0 Dalton AR, Soljak M, Samarasundera E, Millett C, Majeed A. Prevalence of cardiovascular disease risk amongst the population eligible for the NHS Health Check Programme. Eur J Prev Cardiolog . 2013;20(1):142-150. doi:10.1177/1741826711428797 Katakura M, Naka M, Kondo T, et al. Prospective analysis of mortality, morbidity, and risk factors in elderly diabetic subjects: Nagano study. Diabetes Care . 2003;26(3):638-644. doi:10.2337/diacare.26.3.638 Dhindsa S, Tripathy D, Mohanty P, et al. Differential effects of glucose and alcohol on reactive oxygen species generation and intranuclear nuclear factor-κB in mononuclear cells. Metabolism . 2004;53(3):330-334. doi:10.1016/j.metabol.2003.10.013 Khan A, Uddin S, Srinivasan U. Chronic disease prediction using administrative data and graph theory: The case of type 2 diabetes. Expert Systems with Applications . 2019;136:230-241. doi:10.1016/j.eswa.2019.05.048 Paparello J, Kshirsagar A, Batlle D. Comorbidity and Cardiovascular Risk Factors in Patients With Chronic Kidney Disease. Seminars in Nephrology . 2002;22(6):494-506. doi:10.1053/snep.2002.35969 Yach D, Hawkes C, Gould CL, Hofman KJ. The Global Burden of Chronic Diseases: Overcoming Impediments to Prevention and Control. JAMA . 2004;291(21):2616. doi:10.1001/jama.291.21.2616 Demirtas H. Flexible Imputation of Missing Data. J Stat Soft . 2018;85(Book Review 4). doi:10.18637/jss.v085.b04 Hastie T, Tibshirani R, Friedman J. The Elements of Statistical Learning: Data Mining, Inference, and Prediction . Springer; 2013. 伍丽华, 吴心雨, 赖湘瑜, et al. 人工关节置换术后患者深静脉血栓风险预测模型的系统评价. 护理学报. 2025;32(03):12-16. doi:10.16460/j.issn1008-9969.2025.03.012 莫航沣, 陈亚萍, 韩慧, et al. 临床预测模型研究方法与步骤. 中国循证医学杂志. 2024;24(02):228-236. Climent E, Rodriguez-Campello A, Jiménez-Balado J, et al. Acute-to-chronic glycemic ratio as an outcome predictor in ischemic stroke in patients with and without diabetes mellitus. Cardiovasc Diabetol . 2024;23(1):206. doi:10.1186/s12933-024-02260-9 Qi P, Chiaro D, Guzzo A, Ianni M, Fortino G, Piccialli F. Model aggregation techniques in federated learning: A comprehensive survey. Future Generation Computer Systems . 2024;150:272-293. doi:10.1016/j.future.2023.09.008 Hu L, Chen H, Mo W, Han Y, Sun H. Best Evidence Summary for Management of Older People With Type 2 Diabetes Mellitus Using “Internet Plus Nursing Services.” Int J Older People Nurs . 2024;19(6):e12657. doi:10.1111/opn.12657 Additional Declarations No competing interests reported. Supplementary Files SupplementaryTable1CandidatePredictiveFeaturesandTheirCodingSchemes.docx SupplementaryTable2TheFeaturesoftheIncludedStudyParticipantsandtheResultsofUnivariateAnalysis.docx SupplementaryTable3originaldataafterfiltering.xlsx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8265041","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":563466206,"identity":"8a188962-33bd-4a16-8073-6231f1b3e67f","order_by":0,"name":"Litian Hu","email":"","orcid":"","institution":"Peking University","correspondingAuthor":false,"prefix":"","firstName":"Litian","middleName":"","lastName":"Hu","suffix":""},{"id":563466207,"identity":"337a7da8-4dcb-458d-92e2-c9132961b873","order_by":1,"name":"Hongyu Sun","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4UlEQVRIiWNgGAWjYHACNgjF3sNgAGYcIFoLzxmStUjkQPmEtMjPyD324OOO2sQNN98eKLrZxiDHdyOB8XMBHi0GN/LSDWeeOZ644XZegnFuG4Ox5I0EZukZ+LRI5JhJ87YdA2rJMQBpSdxwI4GNmQevw2Babp4Ba6knqIXhBlhLDdBwHrCWBANCWgzOvDGTnNl2wHjmGaBfcs5JAD32sFkar8Pac8wkPrbVyfYdP3vMOKfMRp7vePLBz3gdBgGHHRuAEQSMSgkgh7GBsAYGhjp7IMH8gBilo2AUjIJRMPIAADoTT3j76XnCAAAAAElFTkSuQmCC","orcid":"","institution":"Peking University","correspondingAuthor":true,"prefix":"","firstName":"Hongyu","middleName":"","lastName":"Sun","suffix":""}],"badges":[],"createdAt":"2025-12-03 02:23:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8265041/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8265041/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":99308570,"identity":"f20a25d0-4767-4f21-b6fd-46e4734e4336","added_by":"auto","created_at":"2025-12-31 16:08:47","extension":"png","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":32221,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.1.ParticipantSelectionFlowchart.png","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/fbb7b5eb819f52e1c339e948.png"},{"id":99308749,"identity":"b3c9d14a-eea3-4115-8b2f-ca045fc2bc9b","added_by":"auto","created_at":"2025-12-31 16:09:06","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":249141,"visible":true,"origin":"","legend":"","description":"","filename":"Manuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/fa5897d5a2b45188b369b1db.docx"},{"id":98865527,"identity":"ea680a21-3944-4e8f-9fbb-54fcfc2a6a42","added_by":"auto","created_at":"2025-12-23 10:50:31","extension":"png","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":69939,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.2.SchematicDiagramoftheRiskPredictionModelConstructedBasedonBPNN.png","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/4c11575d8bcb9bc40ffd9464.png"},{"id":98865530,"identity":"6aead083-732c-4aee-9e97-6c99b4877c1a","added_by":"auto","created_at":"2025-12-23 10:50:31","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16412,"visible":true,"origin":"","legend":"","description":"","filename":"Table1.ParameterSpecificationsforMultipleImputation.docx","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/3de6fd334ece38f9d36d0d1a.docx"},{"id":99308779,"identity":"a71a0fbb-aa34-4020-aa83-31491dd5e2f6","added_by":"auto","created_at":"2025-12-31 16:09:08","extension":"png","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":86079,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.3.TheROCCurvesoftheTwoPredictionModels.png","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/96c9fefa06b9be07841ea15c.png"},{"id":98865533,"identity":"8278e6d9-c847-43bb-b895-d482dff3246f","added_by":"auto","created_at":"2025-12-23 10:50:31","extension":"docx","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":17166,"visible":true,"origin":"","legend":"","description":"","filename":"Table2.FinalInclusionofPredictors7.docx","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/c5a9f49975c7dd03ea8ce812.docx"},{"id":99309019,"identity":"4e2c93d2-da35-491e-b0b6-3ab6117f7c46","added_by":"auto","created_at":"2025-12-31 16:09:40","extension":"docx","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":18166,"visible":true,"origin":"","legend":"","description":"","filename":"Table3.ResultsofRiskPredictionModelsConstructedbyTraditionalLogisticRegression.docx","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/e0d71fb2bca0183d48a1e0d6.docx"},{"id":98865532,"identity":"036bc751-8cb0-4547-aa98-3884e7b6f34c","added_by":"auto","created_at":"2025-12-23 10:50:31","extension":"docx","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16972,"visible":true,"origin":"","legend":"","description":"","filename":"Table4.PerformanceMetricsofRiskPredictionModelsontheValidationSet.docx","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/6197cc0daa4c59b823c87b69.docx"},{"id":99308871,"identity":"13a27f9a-2d80-4fc5-85b1-27114d17cbc8","added_by":"auto","created_at":"2025-12-31 16:09:23","extension":"docx","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16628,"visible":true,"origin":"","legend":"","description":"","filename":"Table5.CrossValidationResultsoftheModels.docx","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/7c395fc2aef1fb4dabeea20c.docx"},{"id":99309215,"identity":"cdb37b74-b5d0-4151-9e95-6cac38675bbb","added_by":"auto","created_at":"2025-12-31 16:09:54","extension":"docx","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16607,"visible":true,"origin":"","legend":"","description":"","filename":"Table6.ComputationalEfficiencyandResourceConsumptionofModelTraining.docx","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/dd3e1da4ea2601038f2a409e.docx"},{"id":98865539,"identity":"ffe732f0-5f77-47ac-a30f-d1b8b8f064d5","added_by":"auto","created_at":"2025-12-23 10:50:31","extension":"json","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4480,"visible":true,"origin":"","legend":"","description":"","filename":"ecaa756ce4374b3781a76b649ffa40e4.json","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/46fe151d719f5c2de09067c1.json"},{"id":99308780,"identity":"e796cdec-d2a1-4894-ad61-e8aa7e089245","added_by":"auto","created_at":"2025-12-31 16:09:08","extension":"docx","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":20024,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTable1CandidatePredictiveFeaturesandTheirCodingSchemes.docx","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/428d5af3875b8557d91e04e2.docx"},{"id":99309147,"identity":"c29770f6-860a-4582-8bed-1ab9c94f1544","added_by":"auto","created_at":"2025-12-31 16:09:48","extension":"docx","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":25818,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTable2TheFeaturesoftheIncludedStudyParticipantsandtheResultsofUnivariateAnalysis.docx","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/802f63d5d69c08bd56526477.docx"},{"id":99308960,"identity":"a89cf626-765e-4d1a-8b5f-b6da93ca920b","added_by":"auto","created_at":"2025-12-31 16:09:35","extension":"xlsx","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3673279,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTable3originaldataafterfiltering.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/3a11336f86b1059bf1041956.xlsx"},{"id":98865546,"identity":"895406a6-438b-4999-b140-cdc368c468c1","added_by":"auto","created_at":"2025-12-23 10:50:31","extension":"xml","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":177081,"visible":true,"origin":"","legend":"","description":"","filename":"ecaa756ce4374b3781a76b649ffa40e41enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/8795b6a2dd0bf33f5633be52.xml"},{"id":98865545,"identity":"5dc25462-f839-40da-a768-50a4a6ebe085","added_by":"auto","created_at":"2025-12-23 10:50:31","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":32221,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.1.ParticipantSelectionFlowchart.png","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/076d0c7cabcd47d05a15b257.png"},{"id":98865544,"identity":"a4ee6f3f-2155-4e4c-b8f0-7dc254441526","added_by":"auto","created_at":"2025-12-23 10:50:31","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":69939,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.2.SchematicDiagramoftheRiskPredictionModelConstructedBasedonBPNN.png","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/9ce9f0cbaf1c3e72ea54b898.png"},{"id":99308741,"identity":"9227e8e8-0dd5-480b-a085-880cc0faf4ab","added_by":"auto","created_at":"2025-12-31 16:09:03","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":86079,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.3.TheROCCurvesoftheTwoPredictionModels.png","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/2f71ac5e8c0e9e08b1e22231.png"},{"id":98865541,"identity":"f60325d3-0506-452b-bd71-2215cd01dbd0","added_by":"auto","created_at":"2025-12-23 10:50:31","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":32221,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.1.ParticipantSelectionFlowchart.png","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/23eae7b8bba56def1ff79591.png"},{"id":99308975,"identity":"56b3ea5e-554c-4654-bb9c-3946704055d1","added_by":"auto","created_at":"2025-12-31 16:09:36","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":70056,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/6006a318f72b1a6f1d7abd92.png"},{"id":98865554,"identity":"444d4f22-055e-4623-8f50-7ca0279112e6","added_by":"auto","created_at":"2025-12-23 10:50:32","extension":"jpeg","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":45711,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/b4e4725291d6d84989664365.jpeg"},{"id":98865548,"identity":"ec7392ed-aea2-4145-814b-c939c867f7f9","added_by":"auto","created_at":"2025-12-23 10:50:31","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":17109,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig.1.ParticipantSelectionFlowchart.png","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/aa05dbd74bff6973683ea3f9.png"},{"id":99308568,"identity":"238a8c9e-e2a8-4f4f-8a02-8907e8a0d0b6","added_by":"auto","created_at":"2025-12-31 16:08:47","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":18607,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig.2.SchematicDiagramoftheRiskPredictionModelConstructedBasedonBPNN.png","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/501643c791960683eee6f8e2.png"},{"id":98865549,"identity":"0d16c86a-4ab6-487e-b022-cfb35ec36d61","added_by":"auto","created_at":"2025-12-23 10:50:32","extension":"png","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":13750,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig.3.TheROCCurvesoftheTwoPredictionModels.png","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/937dee71c8daf8ca3d6ca6fe.png"},{"id":98865558,"identity":"5dc6708d-046a-4299-b852-309652cff739","added_by":"auto","created_at":"2025-12-23 10:50:32","extension":"png","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":17109,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/6ad42626791b4fa3d3f8d965.png"},{"id":98865551,"identity":"7ca6981a-f7b5-4c9e-8da9-68488b5ab632","added_by":"auto","created_at":"2025-12-23 10:50:32","extension":"png","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":18607,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/34f3900a37b26e871426d28f.png"},{"id":99309039,"identity":"562eee94-1354-4a57-9b54-957648cffb81","added_by":"auto","created_at":"2025-12-31 16:09:41","extension":"png","order_by":26,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":13750,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/f27d9d9f654b538df0f8fdfc.png"},{"id":98865555,"identity":"0fd689c8-55cc-4b86-b79f-eb182a7e1e8e","added_by":"auto","created_at":"2025-12-23 10:50:32","extension":"xml","order_by":27,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":174713,"visible":true,"origin":"","legend":"","description":"","filename":"ecaa756ce4374b3781a76b649ffa40e41structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/5364e86d0e7e969c08c05908.xml"},{"id":98865552,"identity":"e4138b98-2fe2-4cd3-9281-93ddc4d0a44e","added_by":"auto","created_at":"2025-12-23 10:50:32","extension":"html","order_by":28,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":187212,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/43de0be9e71e598b0fa97867.html"},{"id":98865523,"identity":"98151764-43d9-4622-8de6-f025138331db","added_by":"auto","created_at":"2025-12-23 10:50:31","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":120879,"visible":true,"origin":"","legend":"\u003cp\u003eParticipant Selection Flowchart\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/93f17dad3ecf7eee785bd8bd.png"},{"id":98865524,"identity":"721a979a-14e4-4028-a964-162a274cb643","added_by":"auto","created_at":"2025-12-23 10:50:31","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":100577,"visible":true,"origin":"","legend":"\u003cp\u003eSchematic Diagram of the Risk Prediction Model Constructed Based on BPNN\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/9d1a05c8f6073a635bddabfa.png"},{"id":98865526,"identity":"7b205f8b-9450-4d07-aeec-a336e7cd02b1","added_by":"auto","created_at":"2025-12-23 10:50:31","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":74997,"visible":true,"origin":"","legend":"\u003cp\u003eThe ROC Curves of the Two Prediction Models\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/e8a32437c466577da35289db.png"},{"id":99322538,"identity":"c271f3fe-d54e-4c24-b573-f9a562f52fef","added_by":"auto","created_at":"2025-12-31 16:43:44","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1154788,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/9e8bc4d5-97e2-466e-bd83-8ae9ea25a333.pdf"},{"id":98865536,"identity":"6a798782-ccd7-4dae-a742-2c58afa00943","added_by":"auto","created_at":"2025-12-23 10:50:31","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":20024,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTable1CandidatePredictiveFeaturesandTheirCodingSchemes.docx","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/ee59eb32a00dd3c77d785377.docx"},{"id":98865529,"identity":"35e2732a-49f7-4467-a70b-32580aef0c6b","added_by":"auto","created_at":"2025-12-23 10:50:31","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":25818,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTable2TheFeaturesoftheIncludedStudyParticipantsandtheResultsofUnivariateAnalysis.docx","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/466a7f230a44809ee88daffe.docx"},{"id":98865538,"identity":"47c30335-46d7-4552-ab73-0517c3e2ecbd","added_by":"auto","created_at":"2025-12-23 10:50:31","extension":"xlsx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":3673279,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTable3originaldataafterfiltering.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8265041/v1/89bfbc7227408c11fe015c03.xlsx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Development and Validation of a Machine Learning-Based Risk Prediction Model for Ischemic Stroke-Diabetes Comorbidity","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eIschemic stroke (IS) and diabetes mellitus (DM) are two prevalent chronic diseases. IS is the most predominant type of stroke, accounting for approximately 71% of strokes worldwide\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. Globally, there are over 500\u0026nbsp;million people with DM\u003csup\u003e2\u003c/sup\u003e, and this number is projected to rise to 1.3\u0026nbsp;billion by 2050\u003csup\u003e3\u003c/sup\u003e. IS and DM impose a substantial burden on society and healthcare systems. As, two prevalent chronic conditions demanding urgent attention, their prevention and treatment become an urgent public health challenge worldwide, particularly in developing countries,\u003c/p\u003e \u003cp\u003e​​Cardiometabolic multimorbidity (CMM) refers to the concurrent presence of cardiovascular diseases, cerebrovascular diseases, and metabolic disorders, representing a prevalent chronic disease multimorbidity pattern\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. Extensive research\u003csup\u003e\u003cspan additionalcitationids=\"CR6\" citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e has demonstrated that DM and cardiovascular-cerebrovascular diseases (CCVD) constitute a prevalent multimorbidity cluster. Compared to single chronic conditions, CMM demonstrates synergistic disease interactions that increase systemic fragility and disrupt homeostatic control. These mechanisms promote multisystem dysregulation, accelerating the progression of CCVD and metabolic diseases\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe IS-DM comorbidity dyad is a common comorbidity pattern of CMM. DM is associated with extensive metabolic alterations that may contribute to stroke predisposition\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e, with a study\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e identifying DM as an independent risk factor for IS.​ A meta-analysis\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e revealed that nearly one-third of stroke patients have DM. Furthermore, among individuals with DM, the prevalence of IS is significantly higher compared to hemorrhagic stroke. Additionally, DM is associated with worse stroke outcomes.​ Evidence suggests that individuals with DM face a twofold increased risk of stroke compared to non-diabetic populations\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. The IS-DM comorbidity dyad contributes significantly to the escalating global burden and mortality of cardiovascular diseases\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. ​In this comorbidity state, patients face significantly higher risks of disability and mortality compared to those with either condition alone, leading to increased healthcare resource utilization and socioeconomic burden. This poses major challenges to health systems, underscoring the critical need to better understand and manage the CMM of DM and IS.​\u003c/p\u003e \u003cp\u003eSinger et al.\u003csup\u003e14\u003c/sup\u003e demonstrated that the development of comorbidity involves not only mutual influences between chronic diseases but also multifactorial interactions across biological, psychological, and social domains. These complex and diverse factors may exhibit nonlinear and synergistic effects, resulting in highly heterogeneous clinical manifestations of comorbidity.​ Risk prediction model is a mathematical framework designed to estimate the probability of specific populations developing particular diseases or experiencing defined clinical outcomes, which can fully leverage healthcare big data to enable early forecasting of disease onset and progression. Machine learning (ML) is a discipline focused on enabling computers to learn from data and extract meaningful patterns.\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e It excels at uncovering latent value within large-scale datasets and is particularly adept at analyzing complex, high-dimensional healthcare data. These capabilities give ML broad applications in exploring disease etiology, predicting chronic conditions, and advancing preventive strategies\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e.​ ML-based risk prediction model can assist clinicians in identifying high-risk patients, thereby reducing the societal burden of excessive screening while mitigating potential risks associated with underscreening\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e.​\u003c/p\u003e \u003cp\u003eIn recent years, ML has been increasingly applied to construct predictive models for comorbidity.​ ​Studies by Md Ekramul Hossain et al.\u003csup\u003e18\u003c/sup\u003e and Ahmad Shaker Abdalrada et al.\u003csup\u003e19\u003c/sup\u003e employed ML algorithms to develop prediction models for DM- cardiovascular disease comorbidity. Vasilis Nikolaou et al.\u003csup\u003e20\u003c/sup\u003e utilized decision tree, random forest, and logistic regression to construct predictive models for chronic obstructive pulmonary disease- cardiovascular disease comorbidity. Additional research has established ML-based prediction models for cancer-depression comorbidity\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e.​ Although numerous studies have developed predictive models for various comorbidities, most rely on data from European and North American populations\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e.​ Current research predominantly focuses on single-disease prediction, with limited studies addressing comorbidity risk assessment. Moreover, most existing models rely on traditional linear approaches, which fail to capture the complex nonlinear interactions inherent in comorbid conditions.​ Current research on ML-based risk prediction models for CMM patterns - particularly the IS and DM comorbidity cluster - remains limited.\u003c/p\u003e \u003cp\u003eThe Backpropagation Neural Network (BPNN) is currently the most widely used multilayer feedforward neural network. It derives its name from the error backpropagation algorithm employed for adjusting network weights, and represents the most maturely developed, extensively applied, and stably performing neural network architecture to date. Thus, this study will utilize clinical databases from Chinese community hospitals to construct and evaluate CMM risk prediction models for IS-DM comorbidity using both traditional logistic regression and BPNN. Through comparative analysis, we aim to identify the optimal predictive model. The findings will: a) provide theoretical guidance for healthcare professionals to identify potential comorbid patients; b) enrich research on IS-DM CMM risk prediction; c) serve as a reference for future related studies; and d) offer valuable insights for developing prediction models for similar disease clusters.​\u003c/p\u003e"},{"header":"2. Materials and methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Data source and study cohort\u003c/h2\u003e \u003cp\u003eThis study used data from residents' health records and service records of Ganjiakou Community Health Service Centre, Haidian District, Beijing, China. The center is located in the southeastern part of Haidian District, Beijing, serving a population of 123,000 residents. Among them, 28,500 are elderly aged 60 and above, including over 8,000 individuals aged 80+. With an aging population ratio approaching 20% in its service area, the region is transitioning into a moderate aging phase. This demographic profile provides a robust dataset of chronic disease patients for our study.​ Each resident contains a unique patient ID, age, sex, examination dates and examination data. The dataset includes documented diagnoses of chronic diseases, including DM, IS, hypertension, and related conditions.​ The study utilised documented diagnoses of chronic diseases to identify data for patients with DM and IS as our goal is to predict the risk of comorbidity. Several filtering criteria were applied to the initial dataset to collect the research dataset. The criteria for filtering included: a) selecting health examination data collected between January 1, 2023 and August 19, 2024, b) selecting participants aged\u0026thinsp;\u0026gt;\u0026thinsp;18 years at the time of health examination, c) removing duplicate data, d)removing participants with malignancies, psychiatric disorders, disabilities, mortality, or outmigration, e) removing deceased or relocated participants, f) removing participants with missing data rates exceeding 48% (Given that this study adopts a retrospective cohort design, the dataset contains varying degrees of missing data across both study subjects and variables. To ensure adequate sample size for analysis, we have pragmatically relaxed the criteria for acceptable missingness rates in the dataset).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Data processing\u003c/h2\u003e \u003cp\u003eWe utilized R version 4.4.1 for all data processing and analyses.​ This study employed a retrospective cohort design. Missing data were observed across both study participants and variables to varying degrees. To address this, we implemented the following protocol based on the actual data characteristics: participants with missing data rates exceeding 48% were excluded, while those with missing rates below 48% underwent imputation. Specifically, continuous variables were imputed using mean substitution, whereas categorical variables were handled via multiple imputation. The multiple imputation was performed using the \"mice\" R package with the Multivariate Imputation by Chained Equations method\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e,\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e. The parameter configurations for the multiple imputation in this study are detailed in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.​\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eParameter Specifications for Multiple Imputation\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParameter\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSetting\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eImputation method\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eContinuous variables: \"pmm\"\u003c/p\u003e \u003cp\u003eCategorical variables: \"polyreg\"\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNumber of imputed datasets (m)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5 (default)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIterations of chained equations(maxit)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom seed\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1234\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eImbalanced data refers to datasets where sample sizes across different classes exhibit significant disparities. This imbalance in feature classification can bias models toward predicting majority-class samples to achieve higher accuracy, thereby neglecting minority-class samples of greater research interest and ultimately compromising predictive performance. In this study, which focuses on predicting the comorbidity of DM and IS, the dataset is inherently imbalanced.To address this, we employed the ​Synthetic Minority Oversampling Technique (SMOTE)​​\u003csup\u003e25\u003c/sup\u003e via the ​​\"SmoteClassif​​\" function from the ​​\"UBL\" R package. This approach synthetically augments minority-class (comorbidity) samples to achieve data balance, after careful consideration of dataset size, memory usage, and computational efficiency.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Construction of risk prediction models\u003c/h2\u003e \u003cp\u003eThis study developed risk prediction models for cardio-cerebrovascular metabolic comorbidity patterns between DM and IS using traditional logistic regression and BPNN.We randomly divided the final included data into training and validation sets at a 7:3 ratio after balancing treatment. The training set was used for model construction, while the validation set served for model verification. The study focuses on identifying common risk factors for the comorbidity of DM and IS. Accordingly, these common risk factors were incorporated as predictors in the predictive model, while the presence/absence of DM and IS served as outcome measures.\u003c/p\u003e \u003cp\u003eThis study employed the ​​\"glm​​\" function from the ​​\"caret\" R package​ to construct the logistic regression-based risk prediction model.\u003c/p\u003e \u003cp\u003eA typical BPNN consists of three layers: an input layer, a hidden layer (intermediate layer), and an output layer. Generally, each layer contains multiple neurons (nodes), which are fully interconnected between adjacent layers (meaning each neuron in the preceding layer connects to all neurons in the subsequent layer). Neurons from one layer exclusively influence those in the next layer, while no interactions exist among neurons within the same layer (i.e., intra-layer neurons operate independently). The magnitude of connection weights reflects the degree of influence exerted by upper-layer neurons on lower-layer neurons. The most common BPNN architecture employs a three-layer structure with a single hidden layer.\u003c/p\u003e \u003cp\u003eThe machine learning process of a BP neural network can be summarized as follows: First, signals propagate forward from the input layer through the hidden layer to the output layer, where the error signal is obtained by calculating the difference between the expected and actual output values. Subsequently, the error is backpropagated from the output layer through the hidden layer to the input layer, adjusting the connection weights (ω) layer by layer. This iterative process alternates between forward propagation and backward weight adjustment, continuously optimizing the connection weights (ω) and thresholds (b). Finally, the network undergoes a \"learning convergence\" phase, where the global error approaches a minimum, indicating that the network is converging toward an optimal staet\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 ​Identification of predictors\u003c/h2\u003e \u003cp\u003eBased on literature review and data availability, we initially selected 41 candidate predictive features across five categories for model construction: a) sociodemographic factors, b) family history, c) physiological and biochemical parameters, d) lifestyle behaviors, and e) mental health factors. \u003cb\u003eSupplementary Table\u0026nbsp;1\u003c/b\u003e details the sources and coding schemes of these risk factors.\u003c/p\u003e \u003cp\u003eIn the original dataset, dietary habits were categorized into six types: balanced meat-vegetable diet, meat-dominated diet, vegetarian-dominated diet, salt-preference, oil-preference, and sugar-preference. For this study, we reclassified these into two groups: unhealthy dietary preferences and healthy eating habits. Participants were classified as having unhealthy dietary preferences if they exhibited any of the following characteristics: salt-preference, oil-preference, or sugar-preference. Conversely, those maintaining a balanced meat-vegetable diet without exhibiting any of these three preferences were categorized as having healthy eating habits.\u003c/p\u003e \u003cp\u003eThis study employed ​univariate analysis​ followed by ​multivariate analysis​ to screen for predictive factors. First, the relationship between each independent variable and the dependent variable (outcome) was examined using ​univariate analysis. Variables showing ​statistically significant differences (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05)​​ were preliminarily selected as candidate predictors. ​Spearman correlation analysis​ was then performed to assess multicollinearity among the selected predictors. Variables with high correlation coefficients (indicating collinearity) were excluded to avoid redundancy in the model.​ The remaining candidate predictors were entered into a ​multivariate logistic regression model. Only variables that retained ​statistical significance (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05)​​ in the multivariate analysis were retained as ​final predictive variables.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5 Model Evaluation and Comparison\u003c/h2\u003e \u003cp\u003eThe confusion matrix is widely employed to assess the classification accuracy of machine learning models, providing an intuitive representation of the discrepancy between predicted and actual outcomes. In this study, we initially evaluated model performance by calculating key metrics derived from the confusion matrix, including Accuracy, Precision, Sensitivity (Recall), and F1-score.\u003c/p\u003e \u003cp\u003eFurthermore, we utilized the \"pROC\" and \"plot\" packages to generate Receiver Operating Characteristic (ROC) curves and compute the Area Under the Curve (AUC) values, enabling comparative analysis of the models' discriminative capabilities.\u003c/p\u003e \u003cp\u003eCross-validation serves as a robust method for evaluating model performance, particularly for assessing predictive accuracy and generalization ability. Among various cross-validation techniques, k-fold cross-validation has emerged as one of the most prevalent approaches\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e. Following established conventions\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e, we set k\u0026thinsp;=\u0026thinsp;10, partitioning the dataset into 10 distinct subsets or \"folds\". This procedure involved conducting 10 independent training and validation iterations, with the model's predictive and generalization performance ultimately determined by averaging the results across all 10 validation cycles. For implementation, we employed the \"caret\" package to perform 10-fold cross-validation in this study.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Baseline Characteristics of Study Participants\u003c/h2\u003e \u003cp\u003eAfter data cleaning and screening, a total of 16,406 participants were finally included in this study, with the participant selection flowchart shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Among them, there were 157 cumulative cases of ischemic stroke, with an incidence rate of 96 per 10,000 population; 5,305 cumulative cases of diabetes, with an incidence rate of 3,234 per 10,000 population; and 72 cumulative cases of comorbid ischemic stroke and diabetes, with an incidence rate of 44 per 10,000 population. The features of the included study participants and the results of univariate analysis are presented in \u003cb\u003eSupplementary Table\u0026nbsp;2\u003c/b\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Predictor Screening Results\u003c/h2\u003e \u003cp\u003eFourteen predictors showing statistically significant differences (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05) in univariate analysis were initially selected, including age, marital status, waist circumference, systolic blood pressure, fasting glucose, glycated hemoglobin, serum creatinine, white blood cell count, serum sodium, alanine aminotransferase (HbA1c), total cholesterol, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, and exercise frequency. Collinearity testing among these 14 predictors revealed that two predictor pairs (total cholesterol and low-density lipoprotein cholesterol) exhibited correlation coefficients exceeding the predefined threshold of 0.8, while all other predictor pairs showed correlation coefficients below 0.43. After excluding these two collinear predictors, the remaining 13 predictors were subjected to multivariate logistic regression analysis, ultimately yielding seven predictors that maintained statistical significance (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05), as presented in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e\u003cb\u003e3.3 Model Construction Results\u003c/b\u003e\u003c/h2\u003e \u003cp\u003eThe seven selected predictors were used as predictive variables for the comorbidity of IS and DM. A traditional logistic regression model was constructed, with the risk prediction equation as follows: ​Logit(P) = -1.13\u0026thinsp;+\u0026thinsp;0.24 \u0026times; Age\u0026thinsp;+\u0026thinsp;0.14 \u0026times; Marital status\u0026thinsp;+\u0026thinsp;0.21 \u0026times; Systolic blood pressure \u0026minus;\u0026thinsp;0.60 \u0026times; Fasting glucose\u0026thinsp;+\u0026thinsp;1.47 \u0026times; Glycated hemoglobin (HbA1c)\u0026thinsp;+\u0026thinsp;0.44 \u0026times; Serum creatinine \u0026minus;\u0026thinsp;0.79 \u0026times; Blood sodium.​​ Details are presented in \u003cb\u003e​\u003c/b\u003eTable\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eIn this study, the input layer of the BPNN was configured with 7 nodes based on the included predictors. Since the outcome was binary (presence or absence of comorbidity), the output layer consisted of 1 neuron. A single hidden layer was adopted, with the number of hidden nodes ranging from 1 to 21, determined by empirical node calculation methods. The learning rate was set within the range of 0.08\u0026ndash;0.1.\u003c/p\u003e \u003cp\u003eThe final results indicated that the model with 3 hidden nodes demonstrated the best performance, whereas models with other node numbers either failed to converge during training or exhibited inferior predictive capability. A schematic diagram of the optimal BPNN-based risk prediction model is illustrated in ​Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eFinal Inclusion of Predictors (7)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePredictor\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cem\u003et\u003c/em\u003e Value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cem\u003eP\u003c/em\u003e Value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0107\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSystolic Blood Pressure(mmHg)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.0303\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFasting Glucose(mmol/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-5.99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHbA1c(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e11.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSerum Creatinine(\u0026micro;mol/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4.33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSerum Sodium(mmol/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-6.52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMarital Status\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.00105\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eResults of Risk Prediction Models Constructed by Traditional Logistic Regression\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eβ\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eWaldχ\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eOR\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003e95%CI\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eConstant Term\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-1.13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.022\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2672.63\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.31\u0026thinsp;~\u0026thinsp;0.34\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.24\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.022\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e112.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.21\u0026thinsp;~\u0026thinsp;1.32\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSystolic Blood Pressure\u003c/p\u003e \u003cp\u003e(mmHg)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.021\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e96.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.18\u0026thinsp;~\u0026thinsp;1.28\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFasting Glucose(mmol/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.024\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e642.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.52\u0026thinsp;~\u0026thinsp;0.58\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHbA1c(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.025\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3461.02\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e4.33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e4.13\u0026thinsp;~\u0026thinsp;4.55\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSerum Creatinine(\u0026micro;mol/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.44\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.019\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e512.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.50\u0026thinsp;~\u0026thinsp;1.61\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSerum Sodium(mmol/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.019\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1745.77\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e.46\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.44\u0026thinsp;~\u0026thinsp;0.47\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMarital Status\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.017\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e70.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.12\u0026thinsp;~\u0026thinsp;1.19\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e\u003cb\u003e3.4 Model Comparison Results\u003c/b\u003e\u003c/h2\u003e \u003cp\u003eThe accuracy, precision, recall, F1-score, specificity, AUC, and their 95% confidence intervals (95% CI) of the two prediction models on the validation set are presented in \u003cb\u003e​\u003c/b\u003eTable\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.The accuracy of the logistic regression and BP neural network models was 0.822 and 0.862, respectively, indicating that the BP neural network achieved better classification performance. The precision was 0.816 for logistic regression and 0.938 for the BP neural network, suggesting that the BP neural network exhibited higher accuracy in predicting positive cases. The sensitivity (recall) was 0.833 for logistic regression and 0.776 for the BP neural network, demonstrating that logistic regression had a better ability to identify true positive cases. The F1-scores were 0.824 and 0.849, respectively, indicating that the BP neural network model was more robust.In terms of specificity, logistic regression and the BP neural network achieved 0.811 and 0.949, respectively, suggesting that the BP neural network performed better in correctly identifying negative-class samples (i.e., true negatives). Finally, the AUC values were 0.876 for logistic regression and 0.921 for the BP neural network, confirming that the BP neural network exhibited superior overall classification performance.\u003c/p\u003e \u003cp\u003eThe ROC curves of the two prediction models are presented in ​Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, where the curve of the BPNN-based risk prediction model completely envelops that of the traditional logistic regression-based model. This demonstrates that the classification performance of the BPNN-based model is significantly superior to that of the logistic regression-based model.\u003c/p\u003e \u003cp\u003eThe performance of both models after k-fold cross-validation (k\u0026thinsp;=\u0026thinsp;10) is presented in ​Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e. The cross-validation results show that the BPNN-based model achieved slightly higher mean accuracy and mean specificity compared to the logistic regression-based model, while the logistic regression model demonstrated marginally better mean sensitivity and mean AUC than the neural network model. Both models performed well in cross-validation; however, the BPNN-based risk prediction model did not exhibit a clear advantage over the logistic regression-based model in the cross-validation results.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance Metrics of Risk Prediction Models on the Validation Set\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eF1-Score\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSpecificity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eAUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003e95%CI\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.822\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.816\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.833\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.824\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.811\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.876\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.88\u0026thinsp;~\u0026thinsp;0.89\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBPNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.862\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.938\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.776\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.849\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.949\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.921\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.92\u0026thinsp;~\u0026thinsp;0.93\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCross-Validation Results of the Models\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean Accuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMean Recall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMean Specificity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMean AUC\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.817\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.830\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.803\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.881\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBPNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.821\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.818\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.823\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.880\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe computational efficiency and resource consumption of both models are presented in ​Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, where training time refers to the duration required for model training (measured in seconds) and computational resource consumption indicates the memory usage during model construction (measured in megabytes, MB). The results show that the BPNN model required nearly ten times the training time of the logistic regression model. Additionally, the memory consumption of the BPNN was eighteen times greater than that of logistic regression.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComputational Efficiency and Resource Consumption of Model Training\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTraining Time(s)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMemory Usage(MB)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.04\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e492.8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e30.12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e8892.8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eOverall, the BPNN-based risk prediction model demonstrates superior predictive performance compared to the logistic regression-based approach. However, this enhanced capability comes at a substantial computational cost, as the neural network requires significantly greater training time and memory resources.\u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Predictors of IS and DM Co-Morbidity\u003c/h2\u003e \u003cp\u003eAge and marital status in sociodemographic factors have long been important influences on chronic diseases. Studies have shown that population aging is one of the risk factors for stroke in China\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e, with a particularly noticeable increase in stroke cases among those over 50 years old\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e. A meta-analysis evaluating the relationship between marital status and cardiovascular diseases indicates that, compared to married individuals, being unmarried, divorced, or widowed increases the likelihood of stroke\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e. Possible reasons include delayed medical care-seeking among unmarried patients\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e, which to some extent contributes to disease progression. Other studies suggest that divorce or widowhood may activate chronic inflammatory responses, thereby increasing the risk of cardiovascular and metabolic diseases\u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e. Stress-related theories propose that losing a partner or poor marital quality may negatively impact an individual's economic, behavioral, and emotional health, reducing their ability to prevent, detect, and treat diseases\u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e. The increased stress from divorce or widowhood may further amplify the effects of inflammation on the body\u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e, ultimately influencing the progression of stroke and diabetes\u003csup\u003e\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e. Additionally, married individuals may have greater financial resources, better access to healthcare\u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e, and are less likely to experience delays or unmet medical and mental health needs due to costs\u003csup\u003e\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e, thereby reducing the risk of chronic diseases to some extent.\u003c/p\u003e \u003cp\u003eSystolic blood pressure is a significant influencing factor for chronic diseases, particularly stroke. Research indicates that hypertension is the leading risk factor for stroke\u003csup\u003e\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u003c/sup\u003e, and it remains a major risk factor in cases treated at tertiary and secondary public hospitals or private hospitals\u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e. Furthermore, among diabetic patients, systolic blood pressure\u0026thinsp;\u0026ge;\u0026thinsp;130 mmHg is associated with a higher risk of cardiovascular mortality\u003csup\u003e\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e. A study have demonstrated that for hypertensive patients at high cardiovascular risk, intensive blood pressure-lowering strategies targeting lower systolic thresholds can significantly prevent major vascular events with acceptable additional risks, regardless of diabetes comorbidity or prior stroke history\u003csup\u003e\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e. Therefore, controlling systolic blood pressure is crucial for managing the comorbidity of ischemic stroke and diabetes.\u003c/p\u003e \u003cp\u003eFasting glucose and glycated hemoglobin (HbA1c) reflect blood sugar levels, with fasting glucose being a key indicator for DM diagnosis and daily monitoring, while HbA1c reflects long-term trends in glycemic control. These two indicators are not only directly related to DM but also closely linked to IS. For instance, acute hyperglycemia and DM are both associated with poor outcomes after IS\u003csup\u003e11\u003c/sup\u003e, and elevated HbA1c levels are linked to an increased risk of first-time stroke in both diabetic and non-diabetic populations\u003csup\u003e\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u003c/sup\u003e. Earlier studies have conclusively demonstrated that chronic hyperglycemia and advanced glycation end-products (AGE) collectively elevate the risk of cardio-cerebrovascular events by inducing endothelial dysfunction and cellular damage\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. Thus, blood sugar control is equally critical for managing the comorbidity of IS and DM.\u003c/p\u003e \u003cp\u003eSerum creatinine and sodium levels are typically used as biochemical indicators of renal function, yet numerous studies have identified their association with IS and DM. A meta-analysis found a significant correlation between the urine albumin-to-creatinine ratio (ACR) and stroke incidence\u003csup\u003e\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e, with hypertensive patients having an ACR\u0026thinsp;\u0026ge;\u0026thinsp;10 mg/g showing a significantly higher risk of first-time IS\u003csup\u003e44\u003c/sup\u003e. Other studies indicate that ACR exhibits high individual variability in type II diabetes patients\u003csup\u003e\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e\u003c/sup\u003e. Some research suggests that low serum creatinine is a risk factor for type II diabetes\u003csup\u003e\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e\u003c/sup\u003e, and serum creatinine levels can serve as a muscle mass indicator to predict diabetes progression\u003csup\u003e\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e\u003c/sup\u003e. Sodium levels are also associated with increased chronic disease risk, with studies showing that serum sodium\u0026thinsp;\u0026gt;\u0026thinsp;142 mmol/L or \u0026lt;\u0026thinsp;138 mmol/L is linked to a higher risk of chronic diseases, including DM\u003csup\u003e48\u003c/sup\u003e. Therefore, serum creatinine and sodium levels can, to some extent, serve as predictive factors for the comorbidity of IS and DM.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Risk Prediction Models for Comorbidity Screening\u003c/h2\u003e \u003cp\u003eInsulin resistance or deficient insulin secretion in diabetic patients not only impairs glycemic control but also disrupts adipocyte function, leading to elevated circulating free fatty acids. This metabolic derangement precipitates a cascade of pathological processes including chronic inflammation, endothelial dysfunction, and atherosclerosis - all established risk factors for cardiovascular disease\u003csup\u003e\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e\u003c/sup\u003e. Notably, DM is frequently diagnosed only after the manifestation of complications such as IS\u003csup\u003e50\u003c/sup\u003e, indicating that many patients may have already developed substantial vascular damage or other serious health impairments at the time of diagnosis.\u003c/p\u003e \u003cp\u003eChina's current preventive framework primarily employs community-based strategies targeting high-risk populations for stroke prevention\u003csup\u003e\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e\u003c/sup\u003e. However, epidemiological evidence demonstrates that only 11% of stroke events are prevented through such high-risk approaches\u003csup\u003e\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e\u003c/sup\u003e, with the majority of cerebrovascular and cardiovascular incidents occurring in individuals classified as low-risk\u003csup\u003e\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e\u003c/sup\u003e. Of particular clinical concern is the bidirectional pathological relationship between IS and DM: IS represents the second leading cause of mortality following cancer in diabetic populations\u003csup\u003e\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e\u003c/sup\u003e, while acute hyperglycemia during IS exacerbates poor outcomes through potentiation of inflammatory responses and oxidative stress\u003csup\u003e\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThese findings underscore the critical importance of developing predictive models for early identification of stroke-diabetes comorbidity. Our risk prediction model incorporates routinely available clinical parameters that enable accurate identification of at-risk individuals, thereby facilitating several key clinical applications: a)​ Early risk stratification: Enables timely implementation of preventive measures by both patients and healthcare providers, potentially reducing disease incidence and improving quality of life\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e, b)​ Clinical utility: Provides community and clinical nurses with a tool for initiating preemptive interventions based on individualized risk assessment\u003csup\u003e\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e\u003c/sup\u003e, c) ​Healthcare optimization: Supports medical institutions in resource allocation and service planning\u003csup\u003e\u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e\u003c/sup\u003e, d)​ Public health policymaking: Informs the development of targeted chronic disease management programs for at-risk populations, potentially alleviating overall healthcare burdens\u003csup\u003e\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe implementation of this predictive model may substantially enhance current preventive strategies by shifting the paradigm from reactive to proactive care, ultimately mitigating the significant clinical and socioeconomic impacts of this high-morbidity comorbidity.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Limitations and Potential Future Works\u003c/h2\u003e \u003cp\u003eThis study has several limitations that should be acknowledged. First, the sample was limited to the Ganjiaokou Community in Haidian District, Beijing, which may affect the generalizability of our findings. Second, while we implemented strict quality control measures, the variables exhibited varying degrees of missing data. To maintain an adequate sample size for model development, we extended the acceptable missing data threshold from the conventional\u0026thinsp;\u0026lt;\u0026thinsp;30% to 48%. This adjustment may have influenced the quality of data imputation\u003csup\u003e\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e\u003c/sup\u003e and consequently impacted model performance, as potentially reflected in the relatively low sensitivity of our BPNN model. Future studies could compare different imputation methods and conduct sensitivity analyses to minimize the effects of high missingness rates on model performance.\u003c/p\u003e \u003cp\u003eAlthough BPNN can effectively capture complex nonlinear relationships, their inherent complexity may lead to overfitting during training, potentially compromising generalizability to test datasets\u003csup\u003e\u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e\u003c/sup\u003e. Subsequent research could incorporate techniques such as regularization and dropout to mitigate overfitting risks. Additionally, our prediction models require external validation in independent cohorts to fully assess their generalizability, transportability, and clinical applicability. Another consideration is the substantial computational resources required for BPNN training compared to logistic regression, suggesting the need for careful evaluation of the cost-performance trade-off in practical applications.\u003c/p\u003e \u003cp\u003eMost existing studies develop risk prediction models using retrospective cohort data\u003csup\u003e\u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e61\u003c/span\u003e\u003c/sup\u003e. While such data are more readily available, prospective cohort studies that measure candidate predictors before outcome occurrence may yield more reliable results\u003csup\u003e\u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e62\u003c/span\u003e\u003c/sup\u003e. Future research could therefore benefit from using prospective cohort data. Furthermore, due to data limitations, several established risk factors for IS or DM - including the acute-to-chronic glycemic ratio\u003csup\u003e\u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e63\u003c/span\u003e\u003c/sup\u003e, urinary albumin-to-creatinine ratio (UACR)\u003csup\u003e\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e\u003c/sup\u003e, and residential location (urban/rural)\u003csup\u003e38\u003c/sup\u003e - were not included as candidate predictors in our analysis. Subsequent studies could incorporate these and other potential predictors, along with multimodal data such as imaging, genomics, and dynamic monitoring parameters, to further enhance model performance.\u003c/p\u003e \u003cp\u003eEmerging methodologies such as Federated Learning\u003csup\u003e\u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e64\u003c/span\u003e\u003c/sup\u003e, a distributed machine learning approach that enables model training across client devices while preserving data privacy, could facilitate multi-center data sharing in future risk prediction model development. The rapid development of \"Internet\u0026thinsp;+\u0026thinsp;Nursing\" in China, including evidence-based studies on its application for type 2 diabetes management\u003csup\u003e\u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e65\u003c/span\u003e\u003c/sup\u003e, presents opportunities for integrating risk prediction models into digital healthcare platforms. Future research could also explore the development of risk prediction models for other cardiometabolic multimorbidity patterns, providing theoretical foundations for chronic disease prevention and control strategies to alleviate the burden of chronic disease management.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Conclusions","content":"\u003cp\u003eThis study focused on the comorbidity of IS and DM. Through comprehensive literature review and assessment of data availability, we systematically evaluated 41 candidate predictors spanning five categories: sociodemographic factors, family history factors, physiological and biochemical parameters, lifestyle behaviors, and mental health indicators. Following rigorous screening, seven optimal predictors were selected for model construction, including age, marital status, fasting glucose, HbA1c, systolic blood pressure, serum creatinine, and serum sodium.\u003c/p\u003e \u003cp\u003eWe developed and compared risk prediction models using both traditional logistic regression and BPNN approaches. Comprehensive model evaluation demonstrated that a single-hidden-layer BPNN architecture with three hidden nodes yielded optimal predictive performance. The final model exhibited excellent discrimination ability, achieving an accuracy of 0.938 and an area under the receiver operating characteristic curve (AUC) of 0.921, indicating robust predictive capability for IS-DM comorbidity risk assessment.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eDeclaration of Competing Interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was funded by the Natural Science Foundation of Beijing Municipality (L222103).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was performed in accordance with the Helsinki standard and the study’s protocol was approved by Peking University Institutional Review Board (PU IRB) (ref No: IRB00001052-23097). Participation in the study was voluntary\u0026nbsp;and their informed consent was obtained.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor details\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003csup\u003e1\u003c/sup\u003eSchool of Nursing, Peking University, Beijing, China\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eCampbell B, Khatri P. Stroke. \u003cem\u003eLancet\u003c/em\u003e. 2020;396(10244):129-142. doi:10.1016/S0140-6736(20)31179-X\u003c/li\u003e\n\u003cli\u003eSun H, Saeedi P, Karuranga S, et al. IDF Diabetes Atlas: Global, regional and country-level diabetes prevalence estimates for 2021 and projections for 2045. \u003cem\u003eDiabetes Research and Clinical Practice\u003c/em\u003e. 2022;183:109119. doi:10.1016/j.diabres.2021.109119\u003c/li\u003e\n\u003cli\u003eOng KL, Stafford LK, McLaughlin SA, et al. Global, regional, and national burden of diabetes from 1990 to 2021, with projections of prevalence to 2050: a systematic analysis for the Global Burden of Disease Study 2021. \u003cem\u003eThe Lancet\u003c/em\u003e. 2023;402(10397):203-234. doi:10.1016/S0140-6736(23)01301-6\u003c/li\u003e\n\u003cli\u003eGupta P, Patel SA, Sharma H, et al. Burden, patterns, and impact of multimorbidity in North India: findings from a rural population-based study. \u003cem\u003eBMC Public Health\u003c/em\u003e. 2022;22(1):1101. doi:10.1186/s12889-022-13495-0\u003c/li\u003e\n\u003cli\u003e刘茜, 张静怡, 李玉静. 唐山市老年住院患者共病模式系统聚类分析. 中国预防医学杂志. 2023;24(12):1333-1338. doi:10.16506/j.1009-6639.2023.12.011\u003c/li\u003e\n\u003cli\u003e张海波, 温雯婷, 谢佳东, 姚玲, 马靓, 赵俊. 三级公立医院老年糖尿病共病患者疾病特征与住院费用分析. 中国慢性病预防与控制. 2024;32(07):534-537. doi:10.16386/j.cjpccd.issn.1004-6194.2024.07.012\u003c/li\u003e\n\u003cli\u003e纪倩云, 李曼, 崔雯霞, et al. 老年2型糖尿病住院病人共病情况分析. 实用老年医学. 2021;35(11):1170-1173+1177.\u003c/li\u003e\n\u003cli\u003eWu Z, Luo S, Cai D, et al. The causal relationship between metabolic syndrome and its components and cardiovascular disease: A mendelian randomization study. \u003cem\u003eDiabetes Research and Clinical Practice\u003c/em\u003e. 2024;211:111679. doi:10.1016/j.diabres.2024.111679\u003c/li\u003e\n\u003cli\u003eBabel RA, Dandekar MP. A Review on Cellular and Molecular Mechanisms Linked to the Development of Diabetes Complications. \u003cem\u003eCDR\u003c/em\u003e. 2021;17(4):457-473. doi:10.2174/1573399816666201103143818\u003c/li\u003e\n\u003cli\u003eMisirli H, Somay G, Ozbal N, Yasar EN. Relation of lipid and lipoprotein(a) to ischaemic stroke. \u003cem\u003eJ Clin Neurosci\u003c/em\u003e. 2002;9(2):127-132. doi:10.1054/jocn.2001.1030\u003c/li\u003e\n\u003cli\u003eLau L, Lew J, Borschmann K, Thijs V, Ekinci EI. Prevalence of diabetes and its effects on stroke outcomes: A meta‐analysis and literature review. \u003cem\u003eJ of Diabetes Invest\u003c/em\u003e. 2019;10(3):780-792. doi:10.1111/jdi.12932\u003c/li\u003e\n\u003cli\u003eDavis TME. Risk Factors for Stroke in Type 2 Diabetes Mellitus: United Kingdom Prospective Diabetes Study (UKPDS) 29. \u003cem\u003eArch Intern Med\u003c/em\u003e. 1999;159(10):1097. doi:10.1001/archinte.159.10.1097\u003c/li\u003e\n\u003cli\u003eMaida CD, Daidone M, Pacinella G, Norrito RL, Pinto A, Tuttolomondo A. Diabetes and Ischemic Stroke: An Old and New Relationship an Overview of the Close Interaction between These Diseases. \u003cem\u003eIJMS\u003c/em\u003e. 2022;23(4):2397. doi:10.3390/ijms23042397\u003c/li\u003e\n\u003cli\u003eSinger M, Bulled N, Ostrach B, Mendenhall E. Syndemics and the biosocial conception of health. \u003cem\u003eLancet\u003c/em\u003e. 2017;389(10072):941-950. doi:10.1016/S0140-6736(17)30003-X\u003c/li\u003e\n\u003cli\u003eDeo RC. Machine Learning in Medicine: Will This Time Be Different? \u003cem\u003eCirculation\u003c/em\u003e. 2020;142(16):1521-1523. doi:10.1161/CIRCULATIONAHA.120.050583\u003c/li\u003e\n\u003cli\u003e孙涛, 徐秀林. 基于机器学习的医疗大数据分析与临床应用. 软件导刊. 2019;18(11):10-14. doi:10.11907/rjdk.191047\u003c/li\u003e\n\u003cli\u003eVan Der Heijden AA, Nijpels G, Badloe F, et al. Prediction models for development of retinopathy in people with type 2 diabetes: systematic review and external validation in a Dutch primary care setting. \u003cem\u003eDiabetologia\u003c/em\u003e. 2020;63(6):1110-1119. doi:10.1007/s00125-020-05134-3\u003c/li\u003e\n\u003cli\u003eHossain ME, Uddin S, Khan A. Network analytics and machine learning for predictive risk modelling of cardiovascular disease in patients with type 2 diabetes. \u003cem\u003eExpert Systems with Applications\u003c/em\u003e. 2021;164:113918. doi:10.1016/j.eswa.2020.113918\u003c/li\u003e\n\u003cli\u003eAbdalrada AS, Abawajy J, Al-Quraishi T, Islam SMS. Machine learning models for prediction of co-occurrence of diabetes and cardiovascular diseases: a retrospective cohort study. \u003cem\u003eJ Diabetes Metab Disord\u003c/em\u003e. 2022;21(1):251-261. doi:10.1007/s40200-021-00968-z\u003c/li\u003e\n\u003cli\u003eNikolaou V, Massaro S, Garn W, Fakhimi M, Stergioulas L, Price D. The cardiovascular phenotype of Chronic Obstructive Pulmonary Disease (COPD): Applying machine learning to the prediction of cardiovascular comorbidities. \u003cem\u003eRespiratory Medicine\u003c/em\u003e. 2021;186:106528. doi:10.1016/j.rmed.2021.106528\u003c/li\u003e\n\u003cli\u003eWang X, Eichhorn J, Haq I, Baghal A. Resting-state brain metabolic fingerprinting clusters (biomarkers) and predictive models for major depression in multiple myeloma patients. Bauckneht M, ed. \u003cem\u003ePLoS ONE\u003c/em\u003e. 2021;16(5):e0251026. doi:10.1371/journal.pone.0251026\u003c/li\u003e\n\u003cli\u003eAlsaleh MM, Allery F, Choi JW, et al. Prediction of disease comorbidity using explainable artificial intelligence and machine learning techniques: A systematic review. \u003cem\u003eInternational Journal of Medical Informatics\u003c/em\u003e. 2023;175:105088. doi:10.1016/j.ijmedinf.2023.105088\u003c/li\u003e\n\u003cli\u003eChawla NV, Bowyer KW, Hall LO, Kegelmeyer WP. SMOTE: Synthetic Minority Over-sampling Technique. \u003cem\u003ejair\u003c/em\u003e. 2002;16:321-357. doi:10.1613/jair.953\u003c/li\u003e\n\u003cli\u003eHaibo He, Garcia EA. Learning from Imbalanced Data. \u003cem\u003eIEEE Trans Knowl Data Eng\u003c/em\u003e. 2009;21(9):1263-1284. doi:10.1109/TKDE.2008.239\u003c/li\u003e\n\u003cli\u003eGuo J, Wu H, Chen X, Lin W. Adaptive SV-Borderline SMOTE-SVM algorithm for imbalanced data classification. \u003cem\u003eApplied Soft Computing\u003c/em\u003e. 2024;150:110986. doi:10.1016/j.asoc.2023.110986\u003c/li\u003e\n\u003cli\u003e马锐. 人工神经网络原理. 机械工业出版社; 2010.\u003c/li\u003e\n\u003cli\u003eKuhn M, Johnson K. \u003cem\u003eApplied Predictive Modeling\u003c/em\u003e. Springer New York; 2013. doi:10.1007/978-1-4614-6849-3\u003c/li\u003e\n\u003cli\u003eWong CW, Kwok CS, Narain A, et al. Marital status and risk of cardiovascular diseases: a systematic review and meta-analysis. \u003cem\u003eHeart\u003c/em\u003e. 2018;104(23):1937-1948. doi:10.1136/heartjnl-2018-313005\u003c/li\u003e\n\u003cli\u003eWang L, Zhou B, Zhao Z, et al. Body-mass index and obesity in urban and rural China: findings from consecutive nationally representative surveys during 2004\u0026ndash;18. \u003cem\u003eThe Lancet\u003c/em\u003e. 2021;398(10294):53-63. doi:10.1016/S0140-6736(21)00798-4\u003c/li\u003e\n\u003cli\u003eWang W, Jiang B, Sun H, et al. Prevalence, Incidence, and Mortality of Stroke in China: Results from a Nationwide Population-Based Survey of 480 687 Adults. \u003cem\u003eCirculation\u003c/em\u003e. 2017;135(8):759-771. doi:10.1161/CIRCULATIONAHA.116.025250\u003c/li\u003e\n\u003cli\u003eAustin D, Yan AT, Spratt JC, et al. Patient characteristics associated with self-presentation, treatment delay and survival following primary percutaneous coronary intervention. \u003cem\u003eEuropean Heart Journal: Acute Cardiovascular Care\u003c/em\u003e. 2014;3(3):214-222. doi:10.1177/2048872614527011\u003c/li\u003e\n\u003cli\u003eYang YC, Boen C, Gerken K, Li T, Schorpp K, Harris KM. Social relationships and physiological determinants of longevity across the human life span. \u003cem\u003eProc Natl Acad Sci USA\u003c/em\u003e. 2016;113(3):578-583. doi:10.1073/pnas.1511085112\u003c/li\u003e\n\u003cli\u003eQuinones PA, Kirchberger I, Heier M, et al. Marital status shows a strong protective effect on long-term mortality among first acute myocardial infarction-survivors with diagnosed hyperlipidemia \u0026ndash; findings from the MONICA/KORA myocardial infarction registry. \u003cem\u003eBMC Public Health\u003c/em\u003e. 2014;14(1):98. doi:10.1186/1471-2458-14-98\u003c/li\u003e\n\u003cli\u003eBrown RL, LeRoy AS, Chen MA, et al. Grief Symptoms Promote Inflammation During Acute Stress Among Bereaved Spouses. \u003cem\u003ePsychol Sci\u003c/em\u003e. 2022;33(6):859-873. doi:10.1177/09567976211059502\u003c/li\u003e\n\u003cli\u003eVujcic I, Vlajinac H, Dubljanin E, et al. Long-term prognostic significance of living alone and other risk factors in patients with acute myocardial infarction. \u003cem\u003eIr J Med Sci\u003c/em\u003e. 2015;184(1):153-158. doi:10.1007/s11845-014-1079-2\u003c/li\u003e\n\u003cli\u003eDupre ME, Nelson A. Marital history and survival after a heart attack. \u003cem\u003eSocial Science \u0026amp; Medicine\u003c/em\u003e. 2016;170:114-123. doi:10.1016/j.socscimed.2016.10.013\u003c/li\u003e\n\u003cli\u003eElton E, Gonzales G. Health Insurance Coverage and Access to Care by Sexual Orientation and Marital/Cohabitation Status: New Evidence from the 2015\u0026ndash;2018 National Health Interview Survey. \u003cem\u003ePopul Res Policy Rev\u003c/em\u003e. 2022;41(2):479-493. doi:10.1007/s11113-021-09670-7\u003c/li\u003e\n\u003cli\u003eTu WJ, Zhao Z, Yin P, et al. Estimated Burden of Stroke in China in 2020. \u003cem\u003eJAMA Netw Open\u003c/em\u003e. 2023;6(3):e231455. doi:10.1001/jamanetworkopen.2023.1455\u003c/li\u003e\n\u003cli\u003eWang YJ, Li ZX, Gu HQ, et al. China Stroke Statistics: an update on the 2019 report from the National Center for Healthcare Quality Management in Neurological Diseases, China National Clinical Research Center for Neurological Diseases, the Chinese Stroke Association, National Center for Chronic and Non-communicable Disease Control and Prevention, Chinese Center for Disease Control and Prevention and Institute for Global Neuroscience and Stroke Collaborations. \u003cem\u003eStroke Vasc Neurol\u003c/em\u003e. 2022;7(5):415-450. doi:10.1136/svn-2021-001374\u003c/li\u003e\n\u003cli\u003eSeng LL, Hai Kiat TP, Bee YM, Jafar TH. Real‐World Systolic and Diastolic Blood Pressure Levels and Cardiovascular Mortality in Patients With Type 2 Diabetes\u0026mdash;Results From a Large Registry Cohort in Asia. \u003cem\u003eJAHA\u003c/em\u003e. 2023;12(23):e030772. doi:10.1161/JAHA.123.030772\u003c/li\u003e\n\u003cli\u003eLiu J, Li Y, Ge J, et al. Lowering systolic blood pressure to less than 120 mm Hg versus less than 140 mm Hg in patients with high cardiovascular risk with and without diabetes or previous stroke: an open-label, blinded-outcome, randomised trial. \u003cem\u003eThe Lancet\u003c/em\u003e. 2024;404(10449):245-255. doi:10.1016/S0140-6736(24)01028-6\u003c/li\u003e\n\u003cli\u003eMitsios JP, Ekinci EI, Mitsios GP, Churilov L, Thijs V. Relationship Between Glycated Hemoglobin and Stroke Risk: A Systematic Review and Meta‐Analysis. \u003cem\u003eJAHA\u003c/em\u003e. 2018;7(11):e007858. doi:10.1161/JAHA.117.007858\u003c/li\u003e\n\u003cli\u003eHuang R, Chen X. Increased Spot Urine Albumin-to-Creatinine Ratio and Stroke Incidence: A Systematic Review and Meta-Analysis. \u003cem\u003eJournal of Stroke and Cerebrovascular Diseases\u003c/em\u003e. 2019;28(10):104260. doi:10.1016/j.jstrokecerebrovasdis.2019.06.018\u003c/li\u003e\n\u003cli\u003eHe P, Yang Y, Tian J, et al. Urinary albumin-to-creatinine ratio and the risk of first stroke in Chinese hypertensive patients treated with angiotensin-converting enzyme inhibitors. \u003cem\u003eHypertens Res\u003c/em\u003e. 2022;45(1):116-124. doi:10.1038/s41440-021-00780-5\u003c/li\u003e\n\u003cli\u003eRasaratnam N, Salim A, Blackberry I, et al. Urine Albumin-Creatinine Ratio Variability in People With Type 2 Diabetes: Clinical and Research Implications. \u003cem\u003eAmerican Journal of Kidney Diseases\u003c/em\u003e. 2024;84(1):8-17.e1. doi:10.1053/j.ajkd.2023.12.018\u003c/li\u003e\n\u003cli\u003eHarita N, Hayashi T, Sato KK, et al. Lower Serum Creatinine Is a New Risk Factor of Type 2 Diabetes. \u003cem\u003eDiabetes Care\u003c/em\u003e. 2009;32(3):424-426. doi:10.2337/dc08-1265\u003c/li\u003e\n\u003cli\u003eKashima S, Inoue K, Matsumoto M, Akimoto K. Low serum creatinine is a type 2 diabetes risk factor in men and women: The Yuport Health Checkup Center cohort study. \u003cem\u003eDiabetes \u0026amp; Metabolism\u003c/em\u003e. 2017;43(5):460-464. doi:10.1016/j.diabet.2017.04.005\u003c/li\u003e\n\u003cli\u003eStookey JD, Kavouras SA, Suh H, Lang F. Underhydration Is Associated with Obesity, Chronic Diseases, and Death Within 3 to 6 Years in the U.S. Population Aged 51\u0026ndash;70 Years. \u003cem\u003eNutrients\u003c/em\u003e. 2020;12(4):905. doi:10.3390/nu12040905\u003c/li\u003e\n\u003cli\u003eSchmidt AM. Highlighting Diabetes Mellitus: The Epidemic Continues. \u003cem\u003eATVB\u003c/em\u003e. 2018;38(1). doi:10.1161/ATVBAHA.117.310221\u003c/li\u003e\n\u003cli\u003eMagliano DJ, Boyko EJ, IDF Diabetes Atlas 10th edition scientific committee. \u003cem\u003eIDF DIABETES ATLAS\u003c/em\u003e. 10th ed. International Diabetes Federation; 2021. Accessed March 14, 2025. http://www.ncbi.nlm.nih.gov/books/NBK581934/\u003c/li\u003e\n\u003cli\u003eChao BH, Tu WJ, Wang LD. Initial establishment of a stroke management model in China: 10 years (2011\u0026ndash;2020) of Stroke Prevention Project Committee, National Health Commission. \u003cem\u003eChinese Medical Journal\u003c/em\u003e. 2021;134(20):2418-2420. doi:10.1097/CM9.0000000000001856\u003c/li\u003e\n\u003cli\u003eMa Q, Li R, Wang L, et al. Temporal trend and attributable risk factors of stroke burden in China, 1990\u0026ndash;2019: an analysis for the Global Burden of Disease Study 2019. \u003cem\u003eThe Lancet Public health\u003c/em\u003e. 2021;6(12):e897-e906. doi:10.1016/S2468-2667(21)00228-0\u003c/li\u003e\n\u003cli\u003eDalton AR, Soljak M, Samarasundera E, Millett C, Majeed A. Prevalence of cardiovascular disease risk amongst the population eligible for the NHS Health Check Programme. \u003cem\u003eEur J Prev Cardiolog\u003c/em\u003e. 2013;20(1):142-150. doi:10.1177/1741826711428797\u003c/li\u003e\n\u003cli\u003eKatakura M, Naka M, Kondo T, et al. Prospective analysis of mortality, morbidity, and risk factors in elderly diabetic subjects: Nagano study. \u003cem\u003eDiabetes Care\u003c/em\u003e. 2003;26(3):638-644. doi:10.2337/diacare.26.3.638\u003c/li\u003e\n\u003cli\u003eDhindsa S, Tripathy D, Mohanty P, et al. Differential effects of glucose and alcohol on reactive oxygen species generation and intranuclear nuclear factor-\u0026kappa;B in mononuclear cells. \u003cem\u003eMetabolism\u003c/em\u003e. 2004;53(3):330-334. doi:10.1016/j.metabol.2003.10.013\u003c/li\u003e\n\u003cli\u003eKhan A, Uddin S, Srinivasan U. Chronic disease prediction using administrative data and graph theory: The case of type 2 diabetes. \u003cem\u003eExpert Systems with Applications\u003c/em\u003e. 2019;136:230-241. doi:10.1016/j.eswa.2019.05.048\u003c/li\u003e\n\u003cli\u003ePaparello J, Kshirsagar A, Batlle D. Comorbidity and Cardiovascular Risk Factors in Patients With Chronic Kidney Disease. \u003cem\u003eSeminars in Nephrology\u003c/em\u003e. 2002;22(6):494-506. doi:10.1053/snep.2002.35969\u003c/li\u003e\n\u003cli\u003eYach D, Hawkes C, Gould CL, Hofman KJ. The Global Burden of Chronic Diseases: Overcoming Impediments to Prevention and Control. \u003cem\u003eJAMA\u003c/em\u003e. 2004;291(21):2616. doi:10.1001/jama.291.21.2616\u003c/li\u003e\n\u003cli\u003eDemirtas H. Flexible Imputation of Missing Data. \u003cem\u003eJ Stat Soft\u003c/em\u003e. 2018;85(Book Review 4). doi:10.18637/jss.v085.b04\u003c/li\u003e\n\u003cli\u003eHastie T, Tibshirani R, Friedman J. \u003cem\u003eThe Elements of Statistical Learning: Data Mining, Inference, and Prediction\u003c/em\u003e. Springer; 2013.\u003c/li\u003e\n\u003cli\u003e伍丽华, 吴心雨, 赖湘瑜, et al. 人工关节置换术后患者深静脉血栓风险预测模型的系统评价. 护理学报. 2025;32(03):12-16. doi:10.16460/j.issn1008-9969.2025.03.012\u003c/li\u003e\n\u003cli\u003e莫航沣, 陈亚萍, 韩慧, et al. 临床预测模型研究方法与步骤. 中国循证医学杂志. 2024;24(02):228-236.\u003c/li\u003e\n\u003cli\u003eCliment E, Rodriguez-Campello A, Jim\u0026eacute;nez-Balado J, et al. Acute-to-chronic glycemic ratio as an outcome predictor in ischemic stroke in patients with and without diabetes mellitus. \u003cem\u003eCardiovasc Diabetol\u003c/em\u003e. 2024;23(1):206. doi:10.1186/s12933-024-02260-9\u003c/li\u003e\n\u003cli\u003eQi P, Chiaro D, Guzzo A, Ianni M, Fortino G, Piccialli F. Model aggregation techniques in federated learning: A comprehensive survey. \u003cem\u003eFuture Generation Computer Systems\u003c/em\u003e. 2024;150:272-293. doi:10.1016/j.future.2023.09.008\u003c/li\u003e\n\u003cli\u003eHu L, Chen H, Mo W, Han Y, Sun H. Best Evidence Summary for Management of Older People With Type 2 Diabetes Mellitus Using \u0026ldquo;Internet Plus Nursing Services.\u0026rdquo; \u003cem\u003eInt J Older People Nurs\u003c/em\u003e. 2024;19(6):e12657. doi:10.1111/opn.12657\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"ischaemic stroke, diabetes, co-morbidities, machine learning, risk prediction models","lastPublishedDoi":"10.21203/rs.3.rs-8265041/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8265041/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eAims: \u003c/strong\u003eThis study aimed to develop and validate machine learning-based risk prediction models for ischemic stroke-diabetes mellitus (IS-DM) comorbidity using routinely available clinical data, and to compare the performance of traditional logistic regression with backpropagation neural networks (BPNN). \u003cstrong\u003e​\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods \u003c/strong\u003eHealth records of 16,406 community-dwelling adults from Beijing, China, we analyzed. From 41 initial candidate predictors across five categories, seven optimal predictors were selected through ​univariate analysis​ followed by ​multivariate analysis. The dataset was randomly split into training (70%) and validation (30%) sets. We developed prediction models using both logistic regression and BPNN approaches, with model performance evaluated through confusion matrix, AUC, and 10-fold cross-validation. \u003cstrong\u003e​\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults \u003c/strong\u003eThe single-hidden-layer BPNN model with three hidden nodes demonstrated superior predictive performance, achieving an AUC of 0.921 (95% CI: 0.92-0.93), outperforming logistic regression. Key predictors included age, marital status, fasting glucose, HbA1c, systolic blood pressure, serum creatinine, and serum sodium. However, the BPNN required significantly more computational resources. \u003cstrong\u003e​\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion \u003c/strong\u003eMachine learning approaches, particularly BPNN, can effectively predict IS-DM comorbidity risk using routine clinical parameters. These models could enhance early comorbidity detection in community settings and inform targeted prevention strategies. Despite it predictive efficacy, the computational demands of BPNN should be considered for clinical implementation.\u003c/p\u003e","manuscriptTitle":"Development and Validation of a Machine Learning-Based Risk Prediction Model for Ischemic Stroke-Diabetes Comorbidity","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-23 10:50:26","doi":"10.21203/rs.3.rs-8265041/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"1de925cd-708a-4a0a-880f-5d643584f0c0","owner":[],"postedDate":"December 23rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-12-29T06:39:05+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-23 10:50:26","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8265041","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8265041","identity":"rs-8265041","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}