Davood Khalili

Purpose Cardiovascular diseases (CVD) are a leading cause of mortality in Iran and globally. This study aimed to provide more accurate estimates of associations between risk factors and CVD mortality by applying competing risk models within the Tehran Lipid and Glucose Study cohort. Methods In this prospective analysis, 7,529 individuals aged ≥ 30 years without prevalent CVD were followed for a median of 19.87 years. The primary outcome was CVD mortality (n = 311), with non-CVD death as the competing event (n = 592). Analyses were stratified by sex and age (< 65 vs. ≥65 years). Cause-specific and Fine-Gray models estimated hazard ratios for diabetes, hypertension, hypercholesterolemia, smoking, and body mass index. Results Diabetes and hypertension were the strongest predictors of CVD mortality across most subgroups. In the Fine-Gray model, diabetes showed the greatest impact in women < 65 years (HR: 4.83, p < 0.001), while hypertension showed the strongest association in women ≥ 65 years (HR: 3.32, p < 0.001). Hypercholesterolemia was associated with increased CVD mortality exclusively in women < 65 years (HR: 1.79, p = 0.02). Body mass index showed no significant association. Conclusion Diabetes and hypertension are the predominant risk factors for CVD mortality in the presence of competing risks, with effect magnitudes varying by sex and age. Applying competing risk models is essential for accurate risk estimation and targeted prevention in populations with high competing mortality burden.

Cardiovascular-kidney-metabolic (CKM) syndrome integrates metabolic, renal, and cardiovascular dysregulation into a unified construct. Current CKM staging systems rely on categorical thresholds that can obscure within-stage heterogeneity, leading to misclassification due to population-specific cut-offs. This study aimed to develop and validate a continuous CKM severity score (cCKMS-S) to mitigate these limitations of categorical staging as a continuous adjunct severity index. Using data from 7,343 adults aged 20-80 years in phase III (2005-2008) of Tehran Lipid and Glucose Study (TLGS), confirmatory factor analysis (CFA) was applied to derive age- and sex-specific latent CKM constructs. Temporal validation was performed in TLGS VI, and independent external validation in National Health and Nutrition Examination Survey (NHANES) 1999-2018. Within-stage heterogeneity was assessed by stratifying Stage 2 into tertiles. Discrimination was evaluated using receiver operating characteristic (ROC) curves and mortality associations using Cox models. In temporal external validation, area under the curves (AUCs) for stage classification were ≥ 0.83 for identifying CKM stage ≥ 2 and ≥ 0.90 for stage ≥ 3, across population subgroups. Similar discriminative performance was observed in NHANES (AUC 0.90 for stage ≥ 2 and 0.84 for stage ≥ 3). Each 1-SD increase in cCKMS-S was associated with a higher risk of all-cause (HR = 3.2;95%CI:2.7-3.9) and cardiovascular (CVD) mortality (HR = 4.7;95%CI:3.6-6.2). In NHANES, cCKMS-S also demonstrated consistent associations with mortality outcomes, with each 1-SD increase associated with a higher risk of all-cause (HR = 1.2;95%CI:1.2-1.3) and CVD mortality (HR = 1.4;95%CI:1.3-1.5), with adjusted C-indices for all-cause and CVD mortality (0.83 and 0.86, respectively) comparable to those observed in the TLGS cohort. Within Stage 2, cCKMS-S identified a marked gradient in all-cause mortality risk across tertiles (HR = 2.16, 3.33, and 9.43). Time-dependent ROC analysis for all-cause mortality in TLGS showed AUCs of 0.71 and 0.77 for cCKMS-S versus 0.66 and 0.69 for CKM staging at 10 and 18 years, improving to 0.85 and 0.84 after adjustment. cCKMS-S introduces a continuous severity score that complements CKM staging by capturing within-stage heterogeneity and demonstrating generalizability across populations. It provides a practical tool for longitudinal monitoring and assessment of intervention efficacy, and preventive risk stratification in clinical practice. The developed online calculator of cCKMS-S is available at safdar-masoumi.github.io/ckm-calc.

Polygenic risk scores (PRS) utilize genetic variants to evaluate susceptibility to diseases. While the performance of PRS has been extensively studied in E

Background Cardiovascular diseases (CVD) are a leading cause of death and disability worldwide. Self-risk assessment of these diseases can be a cornerstone of primary prevention programs. This study was designed to investigate the validity of CVD risk assessment based on self-measured data. Methods The present study is a population-based descriptive-analytic study in people 30 years and over. A sample of 475 individuals was randomly selected from a general population on the coverage of a comprehensive health center. Data were collected first through an application (self-assessment) and then through standardized measurements by trained personnel. Data analysis was based on paired t-test, ICC, and sensitivity and specificity indices. Result The results of the study showed an overestimation of height and systolic blood pressure and an underestimation of weight, BMI, waist circumference and diastolic blood pressure. However, the findings showed that Self-measurement anthropometric measures had excellent reliability (ICC > 0.90). Self-assessment tools demonstrated good agreement for CVD risk (ICC = 0.79), with high sensitivity (78–89%) and increasing specificity (76–96%) at higher risk thresholds. Conclusion Self-assessment evaluations of 10-year cardiovascular disease risk demonstrate strong validity and reliability. As a result, utilizing Self-assessment data can serve as a practical and suitable method for the early detection of at-risk individuals.

Accurate prediction of cardiovascular disease risk is crucial for prevention, but current models may not be generalizable to diverse populations. This study externally validated the predicting risk of cardiovascular disease EVENTs (PREVENT) model, which includes clinical and metabolic predictors, such as estimated glomerular filtration rate, to estimate 10-year ASCVD risk. Using data from the Tehran Lipid and Glucose Study, we assessed its performance in a Middle Eastern population of 5799 adults (ages 30–79 years) over a median of 13 years. We evaluated discrimination (AUC), calibration (pre- and post-recalibration), and decision-analytic metrics like net benefit (NB) and net reclassification improvement (NRI). The ASCVD incidence rate was 4.7 per 1000 person-years. PREVENT showed excellent discrimination in women (AUC: 0.84) and acceptable performance in men (AUC: 0.76). The model initially underestimated risk in men, which was corrected by recalibration, improving predicted risk from 4.2 to 7.6% and sensitivity from 62% to 80%. Recalibration also improved NRI (men: +61%, women: +79%) and slightly increased NB. Decision curve analysis showed clinical advantages for risk thresholds of 10–20% in women and 13–20% in men. PREVENT provides a promising tool for ASCVD risk prediction in Middle Eastern populations with local adaptation.

Purpose SCORE2 is a 10-year CVD risk prediction model developed for use in European populations. This study aimed to externally validate the SCORE2 model and assess its clinical usefulness in an Iranian adult population using data from the TLGS cohort. Methods A total of 4,942 individuals aged 40–69 years participated in the TLGS cohort, which was randomly sampled from the population. A Fine and Gray competing risk model was fitted to estimate sex-specific regression coefficients in the TLGS cohort for comparison with the original SCORE2 model. The outcome was first fatal or non-fatal CVD event (myocardial infarction, stroke, or CVD death), considering non-CVD mortality as a competing event. The predictors included age, sex, smoking status, history of diabetes, SBP, total cholesterol, and HDL cholesterol. Model performance was evaluated as statistical performance (discrimination using Harrell’s C-index , calibration via plots and O/E ratio) and clinical performance (sensitivity, specificity, decision curve analysis, and net benefit fraction) across different SCORE2 risk thresholds. Results A total of 4,579 adults without a history of CVD were included in this study. During the follow-up period, CVD events occurred in 211 (11.05%) men and 151 (5.70%) women. The 10-year cumulative incidence of CVD was 8.00% (95% CI: 6.90–9.40) in men and 3.70% (3.00–4.50) in women. The validated SCORE2 model showed higher discrimination in women (C-index: 0.80; 0.76–0.83) than in men (0.70; 0.67–0.74), but it underestimated the CVD risk in both sexes (men: O/E: 2.05; 2.02–2.08, women: O/E: 1.50; 1.40–1.50). At the risk threshold of ≤ 2.5% for individuals under 50 years and ≤ 5% for those aged 50 years and older, sensitivity in men was 0.65 (0.57–0.73) and specificity was 0.63 (0.60–0.65). In women, sensitivity was 0.54 (0.44–0.64) and specificity was 0.88 (0.87–0.89). Decision curve analysis supported its clinical usefulness, particularly at intermediate risk levels (7.5–15%) in both sexes. The net benefit fraction ranged from 0.13 to 0.49 in men and 0.07–0.42 in women. Conclusion The validated SCORE2 model demonstrated good discriminatory performance, particularly in women, but calibration tended to underestimate risk. Although net benefit indicated that the model provides meaningful clinical usefulness in identifying individuals who may benefit from preventive interventions, recalibration and updating the model according to national data is recommended.

This research uses machine learning (ML) models to determine significant predictive factors for CVD events in Iran, a country with high mortality rates. The study evaluates the effectiveness of deep learning and mixed-effects logistic models in predicting 10-year CVD incidence using longitudinal TLGS data. A total of 4,872 adults aged ≥ 30 years without a history of CVD at baseline (2006–2008) were followed until 2020. Following exclusions due to prevalent CVD or insufficient follow-up data, the final analytic sample comprised 1,942 men and 2,930 women. Baseline demographic, behavioral, and biochemical characteristics were utilized as input features. A clinical history examination was employed to identify and confirm incident CVD events, such as stroke and coronary heart disease (CHD). Using longitudinal data collected during a 10-year study period, deep learning models based on Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures were utilized to detect variations in risk factor dynamics. The area under the receiver operating characteristic curve (AUC) was employed to assess the model’s performance. During the follow-up period, 545 participants (11.2%) experienced CVD events. In both sexes, the GRU model outperformed the LSTM model. In women, the GRU achieved an AUC of 0.739, whereas in men it achieved 0.738, in both cases outperforming LSTM. Compared with the traditional mixed-effects model, discrimination was comparable in women (GRU 0.739 vs. 0.74) and higher in men (0.738 vs. 0.70). Even though the models were restricted to 21 commonly employed clinical variables, their performance was similar to that of larger-scale studies that included hundreds of variables. Deep learning models, such as GRU, can effectively leverage longitudinal medical data to predict future cardiovascular disease events. The models achieved performance comparable to studies that used far more variables, despite relying on a limited feature set.

Prediabetes is a major risk factor for the development of diabetes, defined by blood glucose levels that are elevated but not yet high enough to meet the diagnostic criteria for Diabetes Mellitus. This condition is often clinically “silent” yet it can already lead to negative effects on various organ systems and frequently indicates the impending onset of type 2 diabetes mellitus. This study aimed to compare a traditional statistical model, the Generalized Linear Mixed Model (GLMM), with two tree-based machine learning models, Random Forest (RF) and Generalized Mixed-Effects Random Forest (GMERF), for predicting prediabetes and identifying key risk indicators in longitudinal data. The study sample included 5361 individuals aged over 20 years, focusing on 32 different variables. The target variable was the presence of prediabetes in a longitudinal setting. We applied three models: RF, which is tree-based but does not account for repeated measurements; GLMM, which handles random effects but assumes linear relationships; and GMERF, a hybrid model that incorporates both random effects and the nonlinearity of decision trees. Model performance was evaluated using standard predictive metrics. Among the three models, GMERF achieved the highest predictive performance. The area under the ROC curve was 0.63 for RF, 0.70 for GLMM, and 0.74 for GMERF. In the GMERF model, the top five predictive variables were Waist-to-Hip Ratio (WHR), age, waist circumference, triglyceride level, and Waist-to-Height Ratio (WHtR). WHR was ranked as the most important feature in both the GMERF and RF models. All of these variables, except WHtR, were also found to be significant in the GLMM model. In longitudinal data, there is an inherent dependence between observations collected over time. By incorporating these considerations, models that account for this data structure are better equipped to handle the complexities of longitudinal data, leading to more reliable and accurate predictions.

Non-communicable diseases (NCDs), particularly cardiovascular diseases (CVDs), have become the leading cause of mortality worldwide, with Iran exhibiting higher-than-average incidence and mortality rates. Early detection of high-risk individuals is critical, as CVD often progresses silently. Electrocardiogram (ECG) signals may enhance risk prediction beyond Framingham risk score (FRS). This study aimed to evaluate the predictive performance of ECG signal features for incident CVD using signal processing in a large population-based cohort from the Tehran Lipid and Glucose Study (TLGS). A total of 4,637 adults aged 40 years devoid of past CVD at baseline (2006–2008) were followed up until 2018. Baseline characteristics, laboratory measurements, and ECG signal features were collected. CVD events were defined as coronary heart disease (CHD) or stroke. A recalibrated FRS (baseline) model assessed the association between ECG features and incident CVD, with model performance evaluated using Harrell’s C-index, Net Reclassification Index (NRI), and Integrated Discrimination Improvement (IDI). Over a 10-year follow-up, 483 participants (10.4%) developed CVD. The introduction of ECG signal features improved risk prediction in women, increasing the Harrell’s C-index from 0.84 to 0.85 and demonstrating significant reclassification improvement (NRI: 55.7%, IDI: 2.8%). However, no meaningful improvement was observed in men. ECG-based modeling outperformed FRS, particularly for intermediate-risk categories among women. Incorporating ECG signal features into risk models significantly enhanced CVD prediction performance in women, suggesting potential utility for improving individualized preventive strategies. Further research is warranted to refine ECG-based risk stratification tools for broader clinical application.