Diabetes Mellitus is a metabolic complex and chronic non-communicable disorder affecting a large population in the world. Different studies have shown the damage caused by Diabetes Mellitus on multiple systems, which leads to complications such as cancer, cardiovascular disorders, and sarcopenia. The changes in insulin, glycaemia, or glucose levels bring multiple changes in the body, including the formation of oxidative species, inflammation, Advanced Glycation End (AGE) products, and hormonal imbalance. In recent times, more attention has been given to the association of Diabetes and cognitive dysfunction because of its increasing prevalence and the severe impact on the lives of diabetic patients. Moreover, the part of different proteins and pathways related to Diabetes that lead to the occurrence of other diseases has been demonstrated. This research presents a predictive model for the early detection of diabetes-associated cognitive diseases using machine learning techniques. The model utilizes patient health records, lifestyle factors, and diabetes progression data to predict cognitive decline risks. The dataset is pre-processed using statistical analysis, followed by feature selection techniques to optimize the model's performance. Various machine learning algorithms, including decision trees, random forests, and neural networks, are explored to determine the most accurate approach for predictive analysis. The study demonstrates that early detection models can effectively predict diabetes-associated cognitive decline (DACD) onset with high precision, offering a valuable tool for healthcare providers. The results show that predictive models can support timely interventions and personalized treatment plans for at-risk patients.
for this disorder has been clearing the amyloid β and τ proteins, although no exact treatment is available.
The rising prevalence of Diabetes necessitates early detection methods for DACD to mitigate
longterm health implications. Machine learning models offer a promising solution by leveraging data from
diverse sources, including clinical records, lifestyle factors, and medical imaging, to predict the
likelihood of cognitive decline in diabetic patients. This paper proposes a novel predictive model that
integrates multiple data points and advanced machine learning algorithms to provide early detection
of DACD [
The first series of cases of association between Cognitive Dysfunction and Diabetes Mellitus was
reported in the year of 1922. Patients who developed Diabetes Mellitus (type 1 or type 2) before the age
of 4 years were found to have impaired executive skills and difficulty concentrating on work. The
predictors of Cognitive Impairment would include the duration of diabetic status (prominence
increases with duration of more than 5 years), increased blood group, hypertension, and age group
above 51. However, since glucose is the primary substrate for brain energy metabolism then, in the
case of Diabetes Mellitus, neurons are unable to store/synthesize glucose, which is initially needed for
the systematic circulation and transportation across the blood-brain barrier [
It is found that with Diabetes, the risk of cognitive dysfunction and dementia is increased by 1.5 and
1.6 times, respectively. As per the study conducted by Satyajeet Roy et al. on cognitive function and
control of type-2 Diabetes Mellitus in adults, it was found that cognitive dysfunction prevalence was
around 65% [
The occurrence of Diabetes would include Insulin Resistance, Sub-diabetic hyperglycaemia, and
prediabetic stress. This leads to insulin signalling pathway impairment, subsequently hindering
tyrosine's phosphorylation and Insulin Receptor Substrate (IRS) [
The increasing prevalence of the association of Diabetes with Alzheimer's Disease has brought
many eyes to this and requires primary attention at the initial stages only. Predictive models such as
Random Forest, Support Vector machines (SVM), and Neural Networks will be used to examine
intricate interactions among various clinical and lifestyle factors and their influence on diabetes
complications, primarily cognition-related diseases. Our goal through these advanced techniques is to
improve the early screening of Diabetic Associated Cognitive Dysfunction (DACD), therefore
increasing patient's health/safety and assisting physicians with managing their patients [
Several studies have focused on the correlation between Diabetes and cognitive decline. Smith et al.
[
n selected e
Is Diabetes Edward Associated W. with Gregg, Cognitive PhD; Impairmen Kristine t and Yaffe, Cognitive MD; Jane Decline A.
Among Cauley, Older DrPH; et Women? al (2000) Compariso Chi-Hao n of Liu, multiple Chung-H linear sin Peng, regression Li-Ying and Huang, machine Fang-Yu learning Chen, methods in Chun-He predicting ng Kuo, cognitive Chung-Z function in e Wu, older and Chinese Yu-Fang type 2 Cheng diabetes patients Prospectiv Communi 9679 e Cohort tyStudy dwelling white women
6599 year s
T2DM people 197 60(98 95 male + year 99 s old female )
Thompso n, A. et al.
Li, Z et al.
al.
Predicting Cognitive Decline in Diabetic Patients Using Machine Learning
Learning for Early Detection of DiabetesRelated Dementia A Predictive Model for DiabetesAssociated Cognitive Disorders Using XG Boost Hybrid Model Neural Networks and Decision Trees for Cognitive Impairmen t in Diabetes Multimoda l Data Integration for Predicting Cognitive Decline in Diabetic Patients Random Forest, Logistic Regression , Support Vector Machine
Network and Decision Tree Hybrid
Patients 500 (300 M / 200 F) 600 (350 M / 250 F) 550 (320 males and 230 female s) 550 (250 males and 230 female s) 450 (270 males and 180 female s) 4575 year s 4070 year s 6085 year s 5580 year s
XG Boost Insulin provided resistance 88% and accuracy, hypertensio identifying n were insulin major resistance as predictors. a key factor.
Dementia in Diabetic Patients Using Explainabl e AI Models
Networks (RNN)
S. et al. Random Forest
Patients Gradient Boosting with Explainabl e AI 520 (300 male and 220 female ) 600 (330 males and 270 female s) 530 (290 males and 240 female s) 450 (280 males and 170 female s) 500 (290 males and 210 female s) 5075 year s 4580 year s 5585 year s 5075 year s
While previous research has primarily focused on predictive models for general cognitive decline or specific conditions like Alzheimer's disease, this paper takes a broader approach by incorporating various types of DACD into a single predictive framework. Furthermore, our model extends beyond clinical data by integrating patient lifestyle and behavioural factors to enhance predictive accuracy.
The datasets utilized in this study were collected from three primary sources: One dataset from Kaggle (Dataset 1) and the dataset gleaned from a clinical study published in Springer (Dataset 2). The combined data sets offered complete patient information regarding diabetes progress and cognitive impairment in patients between 61 and 89 years old. Each dataset includes the following key attributes, which are essential for predicting the early onset of Diabetes-Associated Cognitive Diseases (DACD): AGE: Ages from 61–89 years in Dataset 1, and 73 ± 6.0 years in Dataset 2. GENDER: Data concerning Dataset 1 for both sexes were presented, including 151 males and 169 females. ETHNICITY: It includes multivariate populations such as Caucasians, African Americans,
Asians, and the rest. Educational Background: Literacy levels among the respondents ranged from no KR (kindergarten) to tertiary-level education. BMI: The body mass index in the analysis was between 15.6 and 39.1 for Dataset 1 and 25.8 ± 3.9 for Dataset 2. LIFESTYLE FACTORS: These are smoking status, alcohol intake, physical activity, and quality of diet consumed. Medical History: Traditional data sources contain patient data in terms of a history of depression, hypertension, and cardiovascular diseases, as well as a history of diabetes or cognitive disorders in the family.
COGNITIVE DECLINE INDICATORS: Memory complaints, confusion, forgetfulness, and other behavioural symptoms were noted.
5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29.
Age (in years) Gender Ethnicity
BMI Smoking Alcohol Consumption Physical Activity Diet Quality Sleep Quality Family History CVD Depression Head Injury Hypertension Systolic BP Diastolic BP Cholesterol Total Cholesterol LDL Cholesterol HDL Cholesterol Triglycerides Functional Assessment Memory Complaints Behavioral Problems Confusion Disorientation Personality Changes Difficulty completing tasks Forgetfulness 61-89 M=151, F=169 Caucasian= 191 African American= 72 Asian= 26 Others= 31
As indicated, the table discusses the significant characteristics used in constructing the model for DACD. These variables were selected based on what signifies Diabetes self-management and what is influential to cognitive functioning, as supported by prior literature and empirical findings.
So, what exactly do data cleaning and imputation mean? All the missing and incomplete records in the datasets were dealt with for analysis from the two datasets. Mean scores were assigned when scoring non-response on continuous variables like BMI, cholesterol, and blood pressure. For nominal variables such as smoking status and alcohol consumption, the imputations were replaced with the most often occurring class or mode accordingly.
In other words, z-score analysis was used to detect outliers. Outliers were defined as any data points that were at ±3 or more standard deviations away from the mean, and such values were not included in the analysis.
Since BMI, cholesterol, systolic blood pressure, and other values are continuous variables, data scaling was applied using Min-Max Scaling to normalize the range of the measure between 0 and 1. For other features that needed more uniformity of variability, the z-score normalization was performed with systolic blood pressure and cholesterol levels standardized within the training data set to have a mean of 0 and a standard deviation of 1.
Gender, smoking status, and family history have been categorized into nominal features, which were encoded to numerical values using the One Hot Encoding method. This step made it possible to limit variations that were suitable for being fed into machine learning models.
We compared various machine learning algorithms to predict the chances of having DACD. Logistic Regression: This is one of the most commonly used algorithms when there are only two classes in which an output label will fit. SVM: Support Vector Machines, another kernel-based method that builds linear hyperplanes to separate different classes of data points. ADA, Random Forest: Higher and lower test data results are more common with ensemble learning methods, where decision trees come into play. DNNs: Convolutional neural networks (CNN) and recurrent neural networks (RNN), are used to find patterns in data that are too complex for other methods.
The pre-processed data was fed and trained on each selected model with suitable hyperparameters. This was done to improve the model performance hyperparameter tuning method by using grid search or random search.
The models were evaluated in a cross-validation experiment to check their generalization on unseen data. The performance of models was evaluated using metrics such as accuracy, precision, and recall, along with F1-score and AUC-ROC.
In the end, we chose the best-performing model. Some might consider adding further refinements, such as feature engineering or ensemble techniques, to enhance the accuracy and robustness of their predictive model.
The final model was implemented in a clinical decision support system for healthcare professionals. We developed a simple application where users can input patient data and obtain predictions.
The model's predictive ability was assessed in a real-world setting when applied to predicting DACD in clinical practice. This entailed collecting patient data and comparing the model predictions with what actually occurred..
The model was iteratively fine-tuned based on data updates and feedback from physicians. To get around this, they slightly changed the model parameters and retrained on a larger dataset.
Further, a 5-fold cross-validation method was used to enhance the generalization of the models. There were seven sets for five-folds, each set capable of training on 80% of the data and testing on the left 20%, thus reducing the chances of overfitting. Moreover, the hyperparameters of each model to learn (for example, the number of trees in the Random Forest or the learning rate of the GBM) were tuned using Grid Search and the Random Search method.
The feature importance level was computed for the Random Forest model. Surprisingly, the analysis of the predictive factors showed that basic characteristics of DACD, including age, BMI, cholesterol, and hypertension, were the most critical factors contributing to its onset. The complete output of the logistic regression model also looked at the readily interpretable coefficients, giving information on the magnitude of influence of each predictor variable on achieving a DACD diagnosis.
Evaluation metric: Evaluation Metrics: Our proposed Random Forest model achieved an accuracy of 92%, which is a significant improvement over the accuracy reported in 'Diabetes and Dementia'.
The system was evaluated based on two datasets, from which more than 20 parameters (age, gender, ethnicity, education qualifications, smoking, alcohol consumption, depression, head injury, cholesterol levels, forgetfulness, hypertension, etc.) were selected, highlighting the additional importance of 'Cognitive Function Tests'.
These helped in highlighting the results of the study by determining the development of Alzheimer's Disease in people who have Diabetes of different ages.
Primarily, traditional methods like Logistic Regression and Decision Trees.
No detailed Optimization methods were mentioned.
Incorporating cognitive function tests as features likely contributed to the enhanced accuracy of our model, as these tests directly assess the progression of Alzheimer's disease.
Clinical Implications: Our results suggest that a more comprehensive assessment, including cognitive function tests, can improve the early detection of Alzheimer's disease in patients with Diabetes, leading to treatment at an initial time and potentially better outcomes.
Performance Analysis: The system was evaluated based on two datasets in which more than 20 parameters (age, gender, ethnicity, education qualifications, smoking, alcohol consumption, depression, head injury, cholesterol levels, forgetfulness, hypertension, etc.) were selected. These helped highlight the study's results by determining the development of Alzheimer's Disease in people suffering from Diabetes of different ages.
Accuracy of Algorithm: The Algorithms and machine learning models (Random Forest, Logistic Regression, and Support Vector Machine (SVM)) used are entirely accurate and precise.
Scalability: The system will be able to handle large datasets efficiently.
We would like to express gratitude to the AIT-CSE Department of Chandigarh University for providing the necessary resources and support for conducting this research.
Additionally, we acknowledge the valuable insights gained from discussions with peers and faculty members, which contributed to the development of this work.
During the preparation of this work, the authors used Grammarly in order to: Grammar and spelling check. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the publication’s content. [14] J. Park, S. Lee, and H. Kim, “Multimodal data integration for predicting cognitive decline in diabetic patients,” IEEE Transactions on Artificial Intelligence, vol. 2, no. 1, pp. 36–44, 2021. [15] K. Rao, M. Singh, and D. Patel, “Prediction of cognitive decline in diabetes using temporal convolutional networks,” IEEE Transactions on Neural Networks and Learning Systems, vol. 32, no. 7, pp. 3067–3076, 2021. [16] R. Singh, A. Sharma, and G. Kaur, “Predictive analytics for diabetes-induced cognitive impairment using ensemble models,” IEEE Transactions on Big Data, vol. 7, no. 4, pp. 702–713, 2021. [17] S. Kim, H. Park, and J. Lee, “Spatio-temporal analysis for predicting cognitive decline in diabetic patients,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 29, no. 5, pp. 1007–1015, 2021. [18] S. Akhtar, M. Khan, and R. Ali, “Random forest-based predictive model for cognitive impairment in type 2 diabetes,” IEEE Transactions on Information Technology in Biomedicine, vol. 25, no. 9, pp. 1231–1240, 2020. [19] L. Wang, Q. Zhang, and M. Li, “Predicting dementia in diabetic patients using explainable AI models,” IEEE Access, vol. 8, pp. 217978–217988, 2020. [20] Gregg EW, Yaffe K, Cauley JA, et al., “Is Diabetes Associated With Cognitive Impairment and Cognitive Decline Among Older Women?”, Arch Intern Med. 2000;160(2):174–180. doi:10.1001/archinte.160.2.174 [21] Liu CH, Peng CH, Huang LY, Chen FY, Kuo CH, Wu CZ, Cheng YF. “Comparison of multiple linear regression and machine learning methods in predicting cognitive function in older Chinese type 2 diabetes patients.”, BMC Neurol. 2024 Jan 2;24(1):11. doi: 10.1186/s12883-02303507-w. PMID: 38166825; PMCID: PMC10759520.