https://orcid.org/0000-0001-6764-3404
Figures
Abstract
In recent years, the prevalence of T2DM has been increasing annually, in particular, the personal and socioeconomic burden caused by multiple complications has become increasingly serious. This study aimed to screen out the high-risk complication combination of T2DM through various data mining methods, establish and evaluate a risk prediction model of the complication combination in patients with T2DM. Questionnaire surveys, physical examinations, and biochemical tests were conducted on 4,937 patients with T2DM, and 810 cases of sample data with complications were retained. The high-risk complication combination was screened by association rules based on the Apriori algorithm. Risk factors were screened using the LASSO regression model, random forest model, and support vector machine. A risk prediction model was established using logistic regression analysis, and a dynamic nomogram was constructed. Receiver operating characteristic (ROC) curves, harrell’s concordance index (C-Index), calibration curves, decision curve analysis (DCA), and internal validation were used to evaluate the differentiation, calibration, and clinical applicability of the models. This study found that patients with T2DM had a high-risk combination of lower extremity vasculopathy, diabetic foot, and diabetic retinopathy. Based on this, body mass index, diastolic blood pressure, total cholesterol, triglyceride, 2-hour postprandial blood glucose and blood urea nitrogen levels were screened and used for the modeling analysis. The area under the ROC curves of the internal and external validations were 0.768 (95% CI, 0.744−0.792) and 0.745 (95% CI, 0.669−0.820), respectively, and the C-index and AUC value were consistent. The calibration plots showed good calibration, and the risk threshold for DCA was 30–54%. In this study, we developed and evaluated a predictive model for the development of a high-risk complication combination while uncovering the pattern of complications in patients with T2DM. This model has a practical guiding effect on the health management of patients with T2DM in community settings.
Citation: Luo X, Sun J, Pan H, Zhou D, Huang P, Tang J, et al. (2023) Establishment and health management application of a prediction model for high-risk complication combination of type 2 diabetes mellitus based on data mining. PLoS ONE 18(8): e0289749. https://doi.org/10.1371/journal.pone.0289749
Editor: Yee Gary Ang, National Healthcare Group, SINGAPORE
Received: December 26, 2022; Accepted: July 26, 2023; Published: August 8, 2023
Copyright: © 2023 Luo et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are within the manuscript and its Supporting Information files.
Funding: This study was supported by a grant from Shanghai Municipal Health Commission. (The grant number is 15GWZK1002). There was no additional external funding received for this study. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Diabetes mellitus (DM), a common chronic noncommunicable disease, is a metabolic disorder caused by dysfunction in the secretion or utilization of insulin in the human body. With the improvement in the living standard of the population, its prevalence is increasing, and DM is a major global health problem [1]. According to a report by the International Diabetes Federation, 537 million adults (aged 20–79 years) worldwide will have DM in 2021, with one in 10 people living with DM, and this number is expected to increase to 783 million by 2045. Global health spending on DM amounts to $966 billion and is expected to reach $1054 billion by 2045. Currently, DM is the ninth leading cause of death in humans, and complications from persistent hyperglycemia are an important cause of death and disability in patients, with more than four million adults dying from DM complications [2, 3]. The World Health Organization reports more than 100 complications of DM, including eye, nerve, kidney, and cardiovascular diseases and pregnancy and oral complications. More than 90% of all patients with DM have type 2 DM (T2DM) [4], and Asia is the main region where the rapid prevalence of T2DM occurs, with China, India, and Pakistan topping the list of T2DM prevalence. A Lancet study predicts global age-standardized prevalence of T2DM to increase by more than 60% by 2050 [5], and the direct or indirect annual per capita cost of treating T2DM ranges from $29.91 to $237.38 [6].
Currently, there is no complete treatment for T2DM, relying mainly on existing medications and a healthy lifestyle to control the development of the disease [7]. However, more than 50% of patients with T2DM are unaware of their disease [8], leading to the development or exacerbation of complications in various parts of the body and reducing the quality of life. Therefore, based on the screening of patients with T2DM, there is an urgent need for better medical tools and interventions to prevent and stop the development of complications.
Clinical prediction models are mathematical models that use parameters (nonparametric) to predict the probability and likelihood of an outcome event occurring in the present or future of a study population. As a quantitative tool for assessing risks and benefits, it provides healthcare professionals with visual and accurate data information, and its use is becoming increasingly common [9, 10]. Through a literature review, we found that most of the studies that established prediction models for T2DM in the Chinese region were single predictions of T2DM or some type of complications [11–15], and few studies have been reported on the pattern and prediction of T2DM complications. Machine learning(ML), a branch of artificial intelligence, plays a crucial role in various applications by employing diverse algorithms and statistical models to enable computer systems to learn from data and extract meaningful patterns. This capability empowers ML to make predictions and informed decisions across a wide range of fields, including computer vision, image processing technology and healthcare, etc. [16–18]. Particularly in the realm of clinical predictive models, ML provides powerful tools and methods for their development and application.
In China, with the aggravation of the social disease burden caused by chronic diseases and the implementation of hierarchical medical systems, the management focus of chronic diseases, such as diabetes, has gradually shifted to primary medical institutions [19]. Community health workers play an important role in T2DM management. Strategies on how to further improve and strengthen the management of chronic diseases, such as T2DM, in the community, remain a current study hotspot [20–23].
In this study, based on the collected T2DM community dataset, association rule analysis based on the Apriori algorithm was used to identify the high-risk complication combination of T2DM, develop a risk prediction model, and explore the application of a prediction model in community chronic disease management.
Materials and methods
Sample collection and data processing
This was a retrospective study. The sample data were obtained from the Zhangjiang area in the Pudong New Area of Shanghai, led by the School of Public Health of Shanghai University of Traditional Chinese Medicine. Moreover, with the assistance of Huamu, Jinyang, Yinxing, Siping, Sanlin, and Daqiao community health service centers in Pudong New Area, Shanghai, the project of “The fourth round of three-year action plan for public health construction of key disciplines of health education and promotion” was completed. Through questionnaire surveys and physical, biochemical, fundus, and peripheral nerve examinations, the project collected relevant data from community residents who participated in the community T2DM cohort project in six community health service centers from December 2015 to April 2016. The study has been ethically approved by the medical ethics committee of Longhua Hospital, Shanghai University of Traditional Chinese Medicine and in accordance with the Declaration of Helsinki ethical principles and guidelines. Written informed consent of all participants were obtained after they were briefed about the study’s objectives. According to the principle of matching age and sex, randomization, and no blood relationship among all study subjects, the research group obtained data from 4937 patients with T2DM and 86 indicators in the baseline information collection stage.
According to the purpose of the study, the study firstly excluded 56 irrelevant indicators, such as patients’ dietary management, exercise management and community management, and retained the occurrence of complications, basic demographic indicators and physical examination indicators in T2DM patients, a total of 29 indicators, which involved a total of 10 diabetic complications and 19 basic variables, including diabetic ketoacidosis, diabetic nephropathy, diabetic cardio-cerebrovascular complications, hyperglycemic hyperosmolar state, diabetic retinopathy, lower extremity vascular disease, lactic acidosis, diabetic neuropathy, diabetic foot, hypoglycemia, family history of DM, gender, waist-to-hip ratio (WHR), age, body mass index (BMI), course of DM, systolic blood pressure (SBP), diastolic blood pressure (DBP), glucose (GLU), 2-hour postprandial blood glucose (GLU2H), glycosylated hemoglobin-Type A1C (HbA1c), total cholesterol (TC), triglyceride (TG), high-density lipoprotein (HDL), blood urea nitrogen (BUN), serum creatinine (SCR), uric acid (UA), urinary creatinine (UCR), and urinary microalbumin (UMA). Correlation analysis of 19 basic variables was performed using the corrplot package (version 0.8.24).
Secondly, the study only retained data on 810 cases suffering from related complications, of which a total of 138 data were missing (17%). Considering the impact of missing data on the results, the study processed the data by multiple interpolation through IBM SPSS statistics (version 25.0). The study used automatic interpolation method by default, i.e., the interpolation method was selected automatically based on the data scanning results. 5 interpolations were performed to obtain 5 datasets without missing values, and the 5 datasets were compared with the original data by reliability analysis, and the dataset with the largest Cronbach’s α value (higher value indicates higher consistency within the data) was selected as the base data for the subsequent study. The data of this study are shown in S1 File.
Finally, data analysis of the final sample of 810 cases included in this study was performed using IBM SPSS software, including basic statistical descriptions under the grouping for whether the complication combination occurred, and the chi-squared and one-sample nonparametric tests were used to analyze whether the differences between groups were statistically significant. All statistical tests used were two-sided, with a test level of 0.05.
Statistical and modeling methods
Association rule analysis primarily refers to the process of discovering potential relationships and rules between study subjects in large-scale datasets. It is widely used in medicine and economics, and the most classic algorithm is the Apriori algorithm, which can determine the frequent item set and importance in the dataset according to some indicators, such as support, confidence, and lift [24, 25]. Support indicates the proportion of rules that occur, confidence indicates the reliability of the rules, and lift indicates whether there is significance between the rules, with a value equal to 1 being no association and greater than 1 being a positive correlation. Therefore, this study used arules (version 1.7–3), arulesViz (version 1.5–1), tidyverse (version 1.3.1), kableExtra (version 1.3.4), and other packages of R software(version 4.1.0, https://www.R-project.org/) to mine association rules for the included T2DM complication dataset and to determine the high-risk complication combination. The risk prediction model was established by using the “whether patients with T2DM have concurrent complication combination” as the outcome variable.
The least absolute shrinkage and selection operator (LASSO) regression analysis mode based on a 10-fold cross-validation, random forest (RF) model, and support vector machine (SVM) was used to screen the risk factors of this high-risk complication combination in patients with T2DM. The common variables selected by the three methods were incorporated into the prediction model. The LASSO regression model is a variable selection method of the linear regression model, which is suitable for large datasets. We used the glmnet package (version 4.1–2) in R software to centralize and normalize the variables included in the analysis so that the coefficients of the variables shrank to zero and screened out the prediction variables with non-zero coefficients under the lambda value of the best penalty parameter under the condition of minimizing the prediction error [15, 26, 27]. RF is a machine learning method that can be used for classification and prediction and is applicable to all types of data. To avoid the overfitting problem, the prediction variables are measured and identified in the form of importance ranking [28, 29], which is completed by the randomForest (version 4.6–14) and varSelRF (version 0.7–8) packages in R software. SVM is a powerful class of generalized linear classifiers that classifies data in supervised learning and compared with other machine learning methods, it has more advantages in identifying subtle patterns in large-scale datasets [30]. A study review also proposed the validity of the SVM in T2DM studies [31]. Therefore, this study used an SVM based on a linear kernel function to screen predictive variables, which was completed by the e1071 (version 1.7–9), kernlab (version 0.9–29), and caret (version 6.0–88) packages in R software.
The outcome variable of this study was the dichotomous variable “whether patients with T2DM have concurrent complication combination.” Therefore, logistic regression analysis was used to construct a prediction model containing the final variables. When the screened common variables were included in the multifactor logistic regression analysis, the predictor variables with P > 0.05 were excluded so that all predictor variables included in the model were statistically significant. The rms (version 6.2–0), DynNom (version 5.0.1), and rsconnect (version 0.8.24) packages in R software were used to draw the nomogram and the online web application of the prediction model. The length of the line corresponding to each variable in the column line graph was proportional to the degree of influence of that variable on the predicted outcome [32, 33]. A nomogram can predict the possibility of disease occurrence simply and effectively and can serve as a clinical decision support tool and reduce healthcare costs [34].
Finally, multiple tests were used in the training and validation sets to assess the discrimination, calibration, and clinical application values of the prediction model. The receiver operating characteristic (ROC) curve and the area under the ROC curve (AUC) were used to assess whether the differentiation ability of the model met the requirements [35], and the closer the AUC value is to 1, the better the differentiation ability of the model [27]. A calibration curve was used to demonstrate the calibration capability of the model and assess the degree of agreement between the actual situation and the predicted results of the model. While AUC and calibration curves provide measures of model accuracy and predictive performance, they do not consider patient risk thresholds and decision consequences. In contrast, Decision Curve Analysis (DCA) can incorporate these factors to comprehensively assess and compare the utility and clinical value of predictive models, so the net benefit of the clinical decision scenarios based on this predictive model was supplemented using decision DCA [36, 37].
The flowchart of the study design is shown in Fig 1.
Results
Association rule analysis
After preprocessing, the data from the 810 sample cases were analyzed in two steps. The data containing complication conditions were first analyzed by association rules based on the Apriori algorithm, and the combination of whether or not a complication occurred (yes = 1, no = 0) was reincorporated into the sample data for modeling analysis based on identifying high-risk complication combinations.
The R software was used to mine association rules based on the Apriori algorithm for sample data containing complications. Setting the minimum confidence level, minimum support level, and minimum lift to 0.07, 0.05, and 1, respectively, 15 frequent itemset were found (Fig 2). The results were sorted according to the magnitude of the confidence level and are shown in Table 1. The highest-confidence complication combinations (lower extremity vascular disease, diabetic foot, and diabetic retinopathy) were selected, and the complication combination was used as an outcome variable for modeling analysis.
(A) Grouping matrix graph of 15 association rules mined by the algorithm. (B) Network relationship diagram of 15 association rules.
Basic statistical description
The correlation analysis of 19 variables is shown in Fig 3 and the basic statistical description under the grouping for whether the complication combination occurred is shown in Table 2 and Fig 4.
Blue indicates positive correlation, red indicates negative correlation.
It shows the differences between groups and comparative analysis for the condition " whether patients with T2DM have concurrent the complication combination ".
[n(%)/median(IQR)].
Screening results of characteristic variables
In the sample data of 810 patients in this study, 41 (5.06%) experienced a combination of complications, with relatively unbalanced positive and negative sample data. The study dataset was balanced by the Synthetic Minority Oversampling Technique in SPSSPRO software (version 1.1.1), and the balanced dataset consisted of 1538 cases, including 769 cases with the complication combination (50.0%). To ensure the accuracy and validity of the model-building results, this study used a balanced dataset (n = 1538) as the training set and the original dataset before balancing (n = 810) as the validation set.
Variables were screened by LASSO regression analysis, RF model, and SVM in the training set. Among the 19 underlying variables, 13 characteristic variables were screened based on a 10-fold cross-validated LASSO regression analysis with the best lambda of 0.017159 (Fig 5). The 19 variables included in the analysis were ranked in the order of importance using the RF model (Number of trees = 134, OOB estimate of error rate = 2.47%), and a 10-fold cross-validation was applied to determine the appropriate number of variables. The results showed that the model had the lowest error rate when the number of variables was 12; therefore, the top 12 variables in terms of importance were selected as the characteristic variables (Fig 6). The results of the SVM variable screening based on linear kernel functions showed that the highest model accuracy was achieved when the number of variables was 16, with accuracy value of 0.738 (Fig 7). The variables filtered by the three methods were intersected to obtain the following common variables: BMI, DBP, GLU, GLU2H, TC, TG and BUN (Table 3).
(A) Determination of the best penalty coefficient lambda value. (B) Coefficient profiles according to the best parameter (lambda) in Fig 5(A).
(A) Importance ranking of the included variables. (B) Ten-fold cross-validation of the random forest model, which showed the model error rate is the lowest when the number of variables is 12 among 19 variables.
Model construction and prediction results
Using the eight indicators selected above as independent variables and “whether patients with T2DM have concurrent complication combination” as the dependent variable, a model was constructed using logistic regression. Excluding GLU (P > 0.05), a risk prediction model for the complication combination in patients with T2DM was obtained (Table 4).
The nomogram drawn according to this model could simply and intuitively predict the risk of complications in patients with T2DM (Fig 8). For example, suppose a patient with T2DM has the following characteristics: BMI, 26.37 kg/m2; DBP, 79 mmHg; GLU2H, 12.10 mmol/L; TC, 4.18 mmol/L; TG, 1.94 mmol/L; BUN, 3.90 mmol/L; with a score of 258, corresponding to a probability of 0.279, indicating that the risk of concurrent complications in this patient with T2DM is 27.9%. Based on this prediction model, this study made an online application to predict the probability of combined complications in patients with T2DM, which is available at https://studentluo.shinyapps.io/DynNomapp_T2DM/, which can help medical workers better prevent and manage complications in patients with T2DM.
(A) Nomogram incorporating 6 variables of BMI, DBP, GLU2H, TC, TG and BUN for analysis. (B) One patient with T2DM was randomly selected from the sample to predict the risk of developing the complication combination based on the 6 characteristic variables of the Nomogram.
The ROC curve is shown in Fig 9, where the AUCs of the training and validation sets were 0.768 (95% confidence interval [CI], 0.744−0.792) and 0.745 (95% CI, 0.669−0.820), respectively, and the C-index and AUC value were consistent. The model had good differentiation ability. The calibration of the model was evaluated using a calibration curve (Fig 10A and 10B), and the results suggested that the model calibration capability (fitting effect) was good. The DCA is shown in (Fig 10C and 10D), and the net benefits of the model in the validation sets was good, suggesting that the model can be applied to clinical decision-making.
(A) Performance of the model in the training set, which showed an AUC value of 0.768, an optimal cut-off value of 0.541, a specificity of 0.746, and a sensitivity of 0.680. (B) Performance of the model in the validation set, which showed an AUC value of 0.745, an optimal cut-off value of 0.062, a specificity of 0.769, and a sensitivity of 0.659.
(A) Calibration curve plot in the training set, and the solid lines indicate the performance of the model in this study, and diagonal lines indicate perfect predictions of the ideal model. (B) Calibration curve plot in the validation set. (C) Decision curve plot in the training set, when the risk threshold is 30% - 54%, the net benefits from clinical decisions based on this study model. (D) Decision curve plot in the validation set, the risk threshold is 30% - 54%.
Discussion
High-risk complication combination of T2DM
Among the common complications of DM, this study found that patients with T2DM had a higher risk of three complication combinations: lower extremity vascular disease, diabetic foot, and diabetic retinopathy. The algorithm results suggest that patients with T2DM have a 97.6% risk of diabetic retinopathy in the presence of lower extremity vascular disease and diabetic foot. Some of the findings suggest a correlation between the three T2DM complications of lower extremity vascular disease, diabetic foot, and diabetic retinopathy, with the three diseases affecting each other [38]. The prognosis and risk factors for the development of diabetic foot in patients with lower extremity vascular disease and diabetic retinopathy as one of the risk factors for lower extremity vascular disease [39] suggest that the results are consistent with the pathological basis for the development of DM complications. In patients with T2DM, lower extremity vascular disease is difficult to detect at an early stage, and as the disease further deteriorates, it will lead to adverse consequences, such as amputation of the patient’s limbs [40]. As a common complication of DM, diabetic foot is a major cause of disability, death, and increased medical burden for patients [41]. Diabetic retinopathy is a common microvascular complication of DM, with 34.6% of patients experiencing both retinopathy and blindness [42], but studies on the diagnosis and treatment of diabetic retinopathy remain inadequate [43]. In summary, the occurrence of the combination of the three T2DM complications identified in this study will result in reduced quality of life, increased medical burden, and increased risk of premature death for patients. Regarding lower extremity vascular disease, diabetic foot, and diabetic retinopathy, currently, the main treatment modality is dependent on early screening and preventive management [44]. Thus, it is necessary to predict the risk of developing the complication combination in patients with T2DM at an early stage and establish targeted intervention measures.
Risk factor analysis of the complication combinations
This study screened for six risk factors for high-risk joint complications in patients with T2DM: BMI, DBP, TC, TG, GLU2H and BUN. First, the study population is patients with T2DM, and GLU2H is one of the important indicators for discovering and diagnosing diabetes. The development of cardiovascular diseases, such as hypertension evaluated by indicators of SBP and DBP, will increase the risk of lower extremity vascular disease, diabetic foot, and diabetic retinopathy [39, 45, 46], whereas obesity, an independent risk factor for cardiovascular disease [47], and abnormal BMI will also lead to an increased risk of developing all three complications. Elevated lipid levels, including TC and TG, are risk factors for the development of lower-limb peripheral artery disease (PAD) [48, 49] and lower extremity vascular disease, including PAD. In addition, some metabolic factors, such as TC, TG, and DBP, have been associated with the development and progression of diabetic foot and diabetic retinopathy [46, 50, 51]. Matsushita et al. showed that some kidney diseases are risk factors for PAD, diabetic foot, and diabetic retinopathy [52–54], whereas BUN is commonly used indicators to evaluate kidney function. Some studies have also suggested that elevated obesity, abnormal blood pressure, GLU2H, TC, TG and BUN levels are among the risk factors for T2DM [55–58], further validating the six risk factors as predictors of high-risk combined complications screened in this study.
Application of a prediction model in community chronic disease management
As the frontier and important occasion for the prevention and control of DM and other chronic diseases, the community plays the role of primary (etiological prevention) and secondary (early detection, diagnosis, and treatment) in the three-level prevention strategy of chronic diseases. The high-risk complications and risk factors of T2DM screened in this study can, to a certain extent, compensate for the study gap in community T2DM management(Complication management). Simultaneously, the risk prediction model can provide medical workers with the risk of future illness based on the current health status of residents, which provides a simple and intuitive scientific tool for community T2DM complications management. First, for T2DM patients with complications, regular monitoring and health management of lower extremity vascular disease, diabetic foot, and diabetic retinopathy should be strengthened. Second, considering the cost-benefit principle, the health management of patients with T2DM is mainly focused on prevention. Intervention measures include the establishment of personal dynamic digital files of community residents (including T2DM patients with complications and high-risk groups screened by the prediction model), focusing on the risk indicators of T2DM complications and the changes in personal disease risk, formulating and adjusting personalized management plans, and strengthening health education for patients with T2DM (Especially for patients with complications), high-risk groups, and their families. Finally, this study model can assess the current community health needs by predicting the disease risk of community residents, help community managers reasonably allocate medical resources, and provide a reference for the formulation and implementation of relevant management measures.
This study has some limitations. Some data information with high missing values and unrelated indicators were selectively excluded during the data collation phase. The results may be subject to selection bias, resulting in some potential risk factors not being included in the analysis. In addition, both the training and validation sets used for modeling in this study were derived from the same data, lacking further validation in other datasets, which may affect the extrapolation ability of the model and the application effect of the model. Therefore, further prospective studies in larger populations or regions are needed to analyze and explore the validity and utility of this study’s model and related findings.
Conclusion
In this study, a high-risk complication combination in patients with T2DM was identified among 10 common diabetic complications by association rule analysis based on the Apriori algorithm (lower extremity vascular disease, diabetic foot, and diabetic retinopathy). Divide the study population based on whether the complication combination occurred, and six risk factors (BMI, DBP, GLU2H, TC, TG and BUN) for the complication combination were screened using three methods, LASSO regression analysis, RF and SVM and established predictive models. The model performance were evaluated by using ROC curves, calibration curves and decision curves in the training and validation sets, respectively, and good performance evaluation results in all aspects.
The evaluation results showed that the model performed well in this study with high predictive power. Therefore, this model could be a useful tool for community health workers to predict the risk of high-risk complications in T2DM patients with complication and help improve the complication management of T2DM in the community. Including a focus on the development of high-risk complications such as lower extremity vascular disease, diabetic foot, and diabetic retinopathy. Through early identification of high-risk patients and timely intervention, the incidence of complications can be reduced, the quality of life of patients can be improved, and the burden on healthcare resources can be reduced. This study provides important guidance and basis for further exploration and improvement of T2DM management strategies.
Supporting information
S1 File. Original data and final analysis data used in this study.
https://doi.org/10.1371/journal.pone.0289749.s001
(XLSX)
References
- 1. Sohrabi M, Binaeizadeh MR, Iraji A, et al. A review on α-glucosidase inhibitory activity of first row transition metal complexes: a futuristic strategy for treatment of type 2 diabetes. RSC Adv 2022; 12: 12011–12052.
- 2. Aikaeli F, Njim T, Gissing S, Moyo F, Alam U, Mfinanga SG, et al. Prevalence of microvascular and macrovascular complications of diabetes in newly diagnosed type 2 diabetes in low-and-middle-income countries: A systematic review and meta-analysis. PLOS Glob Public Health. 2022 Jun 15;2(6):e0000599. pmid:36962416
- 3. Seid MA, Akalu Y, Gela YY, et al. Microvascular complications and its predictors among type 2 diabetes mellitus patients at Dessie town hospitals, Ethiopia. Diabetol Metab Syndr 2021; 13: 1–8.
- 4. Chatterjee S, Khunti K, Davies MJ. Type 2 diabetes. Lancet 2017; 389: 2239–2251. pmid:28190580
- 5. GBD 2021 Diabetes Collaborators. 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. Lancet 2023; S0140-6736(23)01301–6. pmid:37356446
- 6. Oyewole OO, Ale AO, Ogunlana MO, Gurayah T. Burden of disability in type 2 diabetes mellitus and the moderating effects of physical activity. World J Clin Cases 2023; 11(14):3128–3139. pmid:37274052
- 7. Ismail L, Materwala H, Al Kaabi J. Association of risk factors with type 2 diabetes: A systematic review. Computational and Structural Biotechnology Journal 2021; 19: 1759–1785. pmid:33897980
- 8. Saeedi P, Petersohn I, Salpea P, et al. Global and regional diabetes prevalence estimates for 2019 and projections for 2030 and 2045: Results from the International Diabetes Federation Diabetes Atlas, 9th edition. Diabetes Res Clin Pract 2019; 157: 107843. pmid:31518657
- 9. Ranstam J, Cook JA, Collins GS. Clinical prediction models. British Journal of Surgery 2016; 103: 1886. pmid:27901285
- 10. Zhou Z-R, Wang W-W, Li Y, et al. In-depth mining of clinical data: the construction of clinical prediction model with R. Ann Transl Med 2019; 7: 796–796. pmid:32042812
- 11. Shi R, Zhang T, Sun H, et al. Establishment of Clinical Prediction Model Based on the Study of Risk Factors of Stroke in Patients With Type 2 Diabetes Mellitus. Front Endocrinol (Lausanne) 2020; 11: 1–14.
- 12. Hu Y, Shi R, Mo R, et al. Nomogram for the prediction of diabetic nephropathy risk among patients with type 2 diabetes mellitus based on a questionnaire and biochemical indicators: A retrospective study. Aging (Albany NY) 2020; 12: 10317–10336. pmid:32484786
- 13. Wu B, Niu Z, Hu F. Study on risk factors of peripheral neuropathy in type 2 diabetes mellitus and establishment of prediction model. Diabetes Metab J 2021; 45: 526–538. pmid:34352988
- 14. Derevitskii I V., Kovalchuk S V. Machine Learning-Based Predictive Modeling of Complications of Chronic Diabetes. Procedia Comput Sci 2020; 178: 274–283.
- 15. Zhang Y, Shi R, Yu L, et al. Establishment of a Risk Prediction Model for Non-alcoholic Fatty Liver Disease in Type 2 Diabetes. Diabetes Ther 2020; 11: 2057–2073. pmid:32725485
- 16. Hu R., Mo Q., Xie Y., et al. AVMSN: An Audio-Visual Two Stream Crowd Counting Framework Under Low-Quality Conditions. IEEE Access 2021; 9: 80500–80510.
- 17. Hu R., Tang ZR., Wu E.Q. et al. RDC-SAL: Refine distance compensating with quantum scale-aware learning for crowd counting and localization. Appl Intell 2022; 52: 14336–14348.
- 18. Xu M, Cao L, Lu D, Hu Z, Yue Y. Application of Swarm Intelligence Optimization Algorithms in Image Processing: A Comprehensive Review of Analysis, Synthesis, and Optimization. Biomimetics (Basel) 2023; 8(2):235. pmid:37366829
- 19. Liang G, Jiang H, Huang C, et al. Diabetes health management strategy based on internet plus graded diagnosis and treatment strategy. Ann Palliat Med 2020; 9: 3915–3922. pmid:33302654
- 20. Tsolekile LP, Puoane T, Schneider H, et al. The roles of community health workers in management of non-communicable diseases in an urban township. African J Prim Heal Care Fam Med 2014; 6: 1–8.
- 21. Liu X, Gong X, Gao X, et al. Impact of Contextual Factors on the Attendance and Role in the Evidence-Based Chronic Disease Prevention Programs Among Primary Care Practitioners in Shanghai, China. Front Public Heal 2022; 9: 1–9. pmid:35186856
- 22. Firima E, Gonzalez L, Huber J, et al. Community-based models of care for management of type 2 diabetes mellitus among non-pregnant adults in sub-Saharan Africa: A scoping review protocol. F1000Research 2022; 10: 1–19.
- 23. Egbujie BA, Delobelle PA, Levitt N, et al. Role of community health workers in type 2 diabetes mellitus self-management: A scoping review. PLoS One 2018; 13: 1–18. pmid:29856846
- 24. Lu PH, Keng JL, Tsai FM, et al. An apriori algorithm-based association rule analysis to identify acupoint combinations for treating diabetic gastroparesis. Evidence-based Complement Altern Med; 2021. Epub ahead of print 2021. pmid:33833818
- 25. Lu PH, Keng JL, Kuo KL, et al. An apriori algorithm-based association rule analysis to identify herb combinations for treating uremic pruritus using Chinese herbal bath therapy. Evidence-based Complement Altern Med; 2020. Epub ahead of print 2020. pmid:33299462
- 26. Wang H, Zhang L, Liu Z, et al. Predicting medication nonadherence risk in a Chinese inflammatory rheumatic disease population: development and assessment of a new predictive nomogram. Patient Prefer Adherence 2018; 12: 1757–1765. pmid:30237698
- 27. Xue M, Liu L, Wang S, et al. A simple nomogram score for screening patients with type 2 diabetes to detect those with hypertension: A cross-sectional study based on a large community survey in China. PLoS One 2020; 15: 1–15. pmid:32764769
- 28. Alderden J, Pepper GA, Wilson A, et al. Predicting pressure injury in critical care patients: A machinelearning model. Am J Crit Care 2018; 27: 461–468. pmid:30385537
- 29. Kapwata T, Gebreslasie MT. Random forest variable selection in spatial malaria transmission modelling in Mpumalanga Province,South Africa. Geospat Health 2016; 11: 251–262. pmid:27903050
- 30. Aruna S. A Novel SVM based CSSFFS Feature Selection Algorithm for Detecting Breast Cancer. Int J Comput Appl 2011; 31: 975–8887.
- 31. Abhari S, Kalhori SRN, Ebrahimi M, et al. Artificial intelligence applications in type 2 diabetes mellitus care: Focus on machine learning methods. Healthc Inform Res 2019; 25: 248–261. pmid:31777668
- 32. Xu R, Zhou B, Hu P, et al. Development and validation of prognostic nomograms for patients with colon neuroendocrine neoplasms. World J Surg Oncol 2021; 19: 477–492. pmid:34364382
- 33. Zhang A, Luo X, Pan H, et al. Establishment and evaluation of a risk-prediction model for hypertension in elderly patients with NAFLD from a health management perspective. Sci Rep 2022; 12: 1–13.
- 34. Zhou Q, Ke X, Xue C, et al. A Nomogram for Predicting Early Recurrence in Patients with High-Grade Gliomas. World Neurosurg. Epub ahead of print 2022. pmid:35589036
- 35. Liu Q, Li W, Xie M, et al. Development and validation of prognostic nomogram for lung cancer patients below the age of 45 years. Bosn J Basic Med Sci 2021; 21: 620–631.
- 36. Steyerberg EW, Vickers AJ, Cook NR, et al. Assessing the performance of prediction models: A framework for traditional and novel measures. Epidemiology 2010; 21: 128–138. pmid:20010215
- 37. Vickers AJ, Elkin EB. Decision curve analysis: a novel method for evaluating prediction models. Med Decis Making. 2006 Nov-Dec;26(6):565–74. pmid:17099194
- 38. Nativel M, Potier L, Alexandre L, et al. Lower extremity arterial disease in patients with diabetes: A contemporary narrative review 11 Medical and Health Sciences 1103 Clinical Sciences 11 Medical and Health Sciences 1102 Cardiorespiratory Medicine and Haematology. Cardiovasc Diabetol 2018; 17: 1–14.
- 39. Buso G, Aboyans V, Mazzolai L. Lower extremity artery disease in patients with type 2 diabetes. Eur J Prev Cardiol 2019; 26: 114–124. pmid:31766923
- 40. Yang Z, Han B, Zhang H, et al. Association of Lower Extremity Vascular Disease, Coronary Artery, and Carotid Artery Atherosclerosis in Patients with Type 2 Diabetes Mellitus. Comput Math Methods Med; 2021. Epub ahead of print 2021. pmid:34697555
- 41. Wang A, Lv G, Cheng X, et al. Guidelines on multidisciplinary approaches for the prevention and management of diabetic foot disease (2020 edition). Burn Trauma; 8. Epub ahead of print 2020. pmid:32685563
- 42. Rübsam A, Parikh S, Fort PE. Role of inflammation in diabetic retinopathy. Int J Mol Sci 2018; 19: 1–31.
- 43. Sinclair SH, Schwartz SS. Diabetic Retinopathy–An Underdiagnosed and Undertreated Inflammatory, Neuro-Vascular Complication of Diabetes. Front Endocrinol (Lausanne) 2019; 10: 1–14.
- 44. Lim JZM, Ng NSL, Thomas C. Prevention and treatment of diabetic foot ulcers. J R Soc Med 2017; 110: 104–109. pmid:28116957
- 45. Al-Rubeaan K, Al Derwish M, Ouizi S, et al. Diabetic foot complications and their risk factors from a large retrospective cohort study. PLoS One 2015; 10: 1–17. pmid:25946144
- 46. Jenkins AJ, Joglekar M V., Hardikar AA, et al. Biomarkers in diabetic retinopathy. Rev Diabet Stud 2015; 12: 159–195. pmid:26676667
- 47. Bae JC, Cho NH, Kim JH, et al. Association of body mass index with the risk of incident type 2 diabetes, cardiovascular disease, and all-cause mortality: A community-based prospective study. Endocrinol Metab 2020; 35: 416–424. pmid:32615726
- 48. Bertrand C, Saulnier PJ, Potier L, et al. Plasma concentrations of lipoproteins and risk of lower-limb peripheral artery disease in people with type 2 diabetes: the SURDIAGENE study. Diabetologia 2021; 64: 668–680. pmid:33409569
- 49. Aung PP, Maxwell HG, Jepson RG, et al. Lipid-lowering for peripheral arterial disease of the lower limb. Cochrane Database Syst Rev. Epub ahead of print 2007. pmid:17943736
- 50. Yin S, Zhao P, Ai Z, et al. Sex-specific differences in blood lipids and lipid ratios in type 2 diabetic foot patients. J Diabetes Investig 2021; 12: 2203–2211.
- 51. Bryl A, Mrugacz M, Falkowski M, et al. The Effect of Diet and Lifestyle on the Course of Diabetic Retinopathy—A Review of the Literature. Nutrients 2022; 14: 1–19. pmid:35334909
- 52. Matsushita K, Ballew SH, Coresh J, et al. Measures of chronic kidney disease and risk of incident peripheral artery disease: a collaborative meta-analysis of individual participant data. Lancet Diabetes Endocrinol 2017; 5: 718–728. pmid:28716631
- 53. Al-Rubeaan K, Almashouq MK, Youssef AM, et al. All-cause mortality among diabetic foot patients and related risk factors in Saudi Arabia. PLoS One 2017; 12: 1–15. pmid:29176889
- 54. Park YH, Shin JA, Han JH, et al. The association between chronic kidney disease and diabetic retinopathy: The Korea National Health and Nutrition Examination Survey. PLoS One 2015; 10: 1–10.
- 55. Strain WD, Paldánius PM. Diabetes, cardiovascular disease and the microcirculation. Cardiovasc Diabetol 2018; 17: 1–10.
- 56. Laakso M. Biomarkers for type 2 diabetes. Mol Metab 2019; 27: S139–S146.
- 57. Li SN, Cui YF, Luo ZY, et al. Association between blood urea nitrogen and incidence of type 2 diabetes mellitus in a chinese population: A cohort study. Endocr J 2021; 68: 1057–1065. pmid:33907068
- 58. Cai X, Zhu Q, Wu T, et al. Development and Validation of a Novel Model for Predicting the 5-Year Risk of Type 2 Diabetes in Patients with Hypertension: A Retrospective Cohort Study. Biomed Res Int; 2020. Epub ahead of print 2020. pmid:33029529