New AI model accurately detects early cardiovascular aging risks

In evaluating different AI techniques, the researchers trained models using logistic regression, random forest, k-NN, and XGBoost. While all models demonstrated reasonable capacity to detect cardiovascular risk, random forest achieved a classification accuracy of 81%, with logistic regression and k-NN trailing behind. However, the XGBoost model, boosted by the use of the SMOTE technique to address data imbalance, significantly outperformed the others. It correctly classified 10 out of 11 test cases, with only one false positive.

A team of researchers at Al-Farabi Kazakh National University has developed a powerful artificial intelligence (AI) model capable of accurately predicting cardiovascular aging by analyzing immunological and clinical biomarkers in individuals over 60 years of age. The model, built using the XGBoost machine learning algorithm, achieved a 91% accuracy rate and an area under the ROC curve (AUC) of 0.8333, outperforming traditional statistical models such as logistic regression and k-nearest neighbors (k-NN).

The peer-reviewed study, titled "A Predictive Model of Cardiovascular Aging by Clinical and Immunological Markers Using Machine Learning" and published in Diagnostics, involved 52 elderly participants from medical centers in Almaty, Kazakhstan, and aimed to explore the role of immune system markers in the progression of age-related cardiovascular disease (CVD). It represents one of the most comprehensive efforts to date to integrate behavioral, clinical, and immunological data into a predictive framework powered by AI.

The model's strength lies in its ability to identify correlations between immune activity and cardiovascular health status. Among the most significant findings was the positive correlation between the immune activation marker HLA-DR and body mass index (BMI), measured at r = 0.39. This suggests a link between obesity and elevated immune activity. Likewise, the anti-inflammatory cytokine IL-10 showed a negative correlation with BMI (r = -0.52), underscoring how systemic inflammation intensifies with increasing weight - a key risk factor in metabolic and cardiovascular diseases.

CD14+, a monocyte activation marker, emerged as another critical indicator. It showed a 0.37 correlation with post-infarction cardiosclerosis (PICS), a condition often associated with myocardial tissue damage and fibrotic scarring. Researchers noted that elevated CD14+ levels point to systemic inflammation that may worsen myocardial remodeling post-heart attack. Conversely, CD19+—a B-lymphocyte marker—displayed a negative correlation with acute cerebral circulatory failure (r = -0.43), suggesting that reductions in adaptive immune function may raise the risk of cerebrovascular complications.

In evaluating different AI techniques, the researchers trained models using logistic regression, random forest, k-NN, and XGBoost. While all models demonstrated reasonable capacity to detect cardiovascular risk, random forest achieved a classification accuracy of 81%, with logistic regression and k-NN trailing behind. However, the XGBoost model, boosted by the use of the SMOTE technique to address data imbalance, significantly outperformed the others. It correctly classified 10 out of 11 test cases, with only one false positive.

The analysis also included feature importance rankings within the XGBoost model. The most influential markers in predicting cardiovascular aging were CD59+ (a cell protection protein), VEGF-2 (linked to vascular growth and inflammation), and TNF-α (a pro-inflammatory cytokine). Markers like CD8+ and IL-10, while significant in correlation matrices, were found to contribute less to the predictive model’s decision-making tree.

Lead author Madina Suleimenova of the Department of Big Data and Artificial Intelligence emphasized the multidimensional nature of the study. “We demonstrated how AI can integrate biological, clinical, and lifestyle data to assess individual cardiovascular aging risk,” she said. “Our findings pave the way for precision medicine interventions focused on aging-related diseases.”

The study identified specific associations between immune profiles and behavioral variables as well. Smoking showed a strong negative correlation with CD8+ T-lymphocyte levels (r = -0.4), indicating suppression of adaptive immune function among smokers. Education level correlated positively with HLA-DR expression (r = 0.43), suggesting that socioeconomic or lifestyle factors tied to education may influence immune system regulation.

Patients were divided into two groups: those with a history of cardiovascular disease and those without. Machine learning models were trained on data labeled accordingly. Each model's performance was measured using confusion matrices and ROC curves. Random forest and k-NN performed well in predicting class 1 (patients with CVD) but struggled to distinguish class 0 (patients without CVD), likely due to dataset imbalance. SMOTE sampling was employed to mitigate this issue in the XGBoost model.

In the XGBoost evaluation, the model demonstrated high sensitivity and specificity across all classes. The final confusion matrix showed four correct classifications for non-CVD patients and six for CVD patients, with only one misclassification. These results were achieved after dividing the dataset into training (80%) and test (20%) sets with class stratification.

The broader implications of the research are profound. By identifying inflammation-related immune markers as key predictors of cardiovascular aging, the study provides empirical support for “inflammaging” as a mechanism behind age-related decline. CD95+, a receptor associated with programmed cell death (apoptosis), showed a negative correlation with chronic inflammatory diseases, reinforcing the theory that impaired cell clearance mechanisms contribute to prolonged inflammation and tissue damage in aging populations.

Beyond prediction, the study raises potential for early intervention. For example, recognizing elevated VEGF-2 or TNF-α levels could trigger preventive measures in patients at risk of arterial stiffening or plaque buildup. Similarly, IL-10 suppression in obese individuals suggests that weight management may restore anti-inflammatory capacity, potentially decelerating the aging process.

While the dataset was relatively small, the study lays a strong foundation for future research into AI-assisted diagnosis of aging-related cardiovascular risks. It calls for larger longitudinal studies and further refinement of features to enhance classification for underrepresented groups. The authors also suggest developing an integrative aging model that combines immunological, metabolic, and behavioral dimensions to guide clinical decision-making.