Use of AI in Cardiometabolic Risk Prediction in Asian Indians

Overview

The Investigators are recruiting T2DM patients (n, 500) from Fortis-CDOC Hospital. Patients' weight, BMI, lipid profile, liver and kidney function tests, EGG, glycemic parameters, blood pressure, etc. will be entered in MS Excel sheets and appropriate data coding will be performed. Additional information on sleep hygiene, self-perceived stress, environmental pollution, and socio-economic status (education, occupation, and family annual income) will be collected by phone interviews. The entered data will be filtered for outliers and missing data will be excluded from the final data sheet. Johns Hopkins Team will perform the following: 1. Mediation and moderation analysis, 2. Machine Learning methods 3. Deep Learning and Neural Networks to devise prediction models for different metrics, including diabetes, blood pressure, and lipid control. 4. Traditional statistics like Propensity Score Matching and Multivariate Linear Regression Data pre-processing The data pre-processing will be performed to standardize the variables and minimize the impact of non-normality. During this step, the raw data would be converted into appropriate transformations. Python and R programming will be used for AI and machine learning methods. Data analysis Our research collaborators are well versed in techniques like multi-fold cross-validation, Synthetic Minority Oversampling Technique for Nominal and Continuous (SMOTE-NC), a widely used technique for balancing the observations only in the training dataset and not in the testing dataset, and hyper tuning of parameters. For our research, we would require a graphic processing unit (GPU) to perform high-quality and fast computing (especially important when analyzing large data sets through neural networks and machine learning). We have an understanding with ORACLE (a large software giant), for providing GPUs at no cost on a lease basis on the submission of a feasible proposal. Key Milestones Expected * During the initial three months of the study, the plan is to obtain all requisite permissions for data gathering from the Institutional Ethics Review Committees of the respective institutions. The research assistant would be recruited from FORTIS-CDOC Hospital. * Over the next 12 months, there will be data tabulation and gathering * The last 3-4 months will be allocated to data analysis, application of AI algorithms (using training and testing datasets), and reporting of the data (meetings and manuscripts)

AI is a term that encapsulates machine learning (ML), deep learning, supervised learning, unsupervised learning, artificial neural networks (ANNs), and convolutional neural networks (CNNs). An essential component of AI, ML, is being increasingly used in health care applications, since it is an interdisciplinary field which uses techniques to enable computers to examine data sets without explicit programming. ML involves development of a model or algorithm by careful extraction of essential features from the training data which is used for testing purpose. Test data is utilized for making predictions about research problems. ML can be effectively summarized as: feature extraction, selection of technique for data analysis, training of the new model and evaluation of its efficacy, and making predictions using the trained model. Supervised and unsupervised learning are forms of machine learning. In supervised learning, known data sets are used to understand the patterns in data sets, whereas in unsupervised learning, unlabeled data sets are used to understand the data. Deep learning structures use numerous layers of computation. DL is mainly used for processing large and raw data sets. In ANN, multiple machine learning algorithms work together and process data inputs, but in CNN, hidden nodes exist in layers for information, complex data, and image processing. With rapid advancement in AI, ML, and DL, computer programs can effectively simulate neural activity of the brain's neocortex where reasoning, thinking, and cognitive functions take place at a fast pace. In the domain of cardiovascular medicine, AI has extensive applications in drug therapy, pharmacogenomics, heart failure, imaging studies, and diagnostics. Importantly, AI has the ability to provide mechanisms to apply precision medicine and big data in medicine while enhancing the effectiveness of treatment regimen advised by the cardiologist. Further, AI/ML algorithms can analyze data without assumptions for prediction and classification purposes. Thus, cardiovascular medicine can benefit remarkably from the incorporation and judicious utilization of AI. Cardiometabolic disorders (CMDs) including myocardial infarction, stroke, and type 2 diabetes mellitus (T2DM), are linked with an increased morbidity and mortality, and it is known that an individual's lifestyle and level of exercise are significant risk factors for CVD. This is especially true for patients who spend considerable periods of time doing sedentary work, have limited mobility, or find it challenging to sustain acceptable amounts of exercise, let alone reach the moderate-to-vigorous levels advised for cardiometabolic health. Asian Indians are one of the largest growing ethnic groups in the world and have one of the highest rates of cardiovascular disease (CVD), including T2DM coronary artery disease (including major adverse cardiovascular events or MACE). Several causes of CVD have been identified, such as adoption of western dietary patterns and habits, genetic predisposition, imbalance in lipoproteins, and abdominal obesity. Certain perinatal influences are also regarded as possible contributors. High consumption of omega-6 PUFAs and saturated fat are significantly correlated with fasting hyperinsulinemia and sub-clinical inflammation. Such imbalanced diets contribute to high prevalence of insulin resistance, the metabolic syndrome and T2DM in Asian Indians. Further, genetic, cultural, and socio-economic factors have also resulted in this healthcare crisis. Studies have consistently shown that Asian Indians have a high prevalence of insulin resistance which increases their tendency to develop T2DM and CVD at a younger age as compared to individuals of other ethnicities. Important reasons could be their excess body fat and adverse body fat patterning, including abdominal adiposity, even when the body mass index is within the currently defined normal limits. Some of these features have been reported at birth and childhood. The American Diabetes Association (ADA) has lowered the BMI cut-off (compared to Caucasians) for T2DM screening in South Asians. Some of the problems encountered are: variability in clinical risk stratification; lack of established guidelines for atherosclerosis imaging, and scarcity of data for formulation of meaningful clinical guidelines in persons of Asian Indian ethnicity. The population of Asian Indian is increasing in the United States at a rapid pace. A recent research study by the American Heart Association (AHA) underscores the increasing prevalence and mortality associated with CVD in this ethnicity, contributed mainly by insulin resistance and increased incidence of T2DM. Indians are representative of a highly successful group of immigrants with higher incomes and better social and economic indicators. The preliminary data obtained from this AI based study will enable our team in obtaining the necessary funds for long-term prospective research in this field.