IMPLEMENTATION OF FEEDFORWARD NEURAL NETWORK FOR CARDIOVASCULAR DISEASE PREDICTION WITH PERFORMANCE EVALUATION

Background/Introduction Accurate cardiovascular disease (CVD) risk prediction models can facilitate the identification and cost-effective treatment of high-risk individuals. Discrimination and calibration are the most commonly recommended metrics for evaluating model performance. However, both are population-level measures, and it remains unclear whether these metrics are sufficiently sensitive to assess the performance of models designed for use with individual patients. Purpose Using a real-world dataset, we aimed to compare CVD risk estimates for example individuals (individual-level measures) and discrimination and calibration metrics (population-level measures) from two CVD risk prediction models; one with "age" as the only predictor and one with multiple predictors. Methods We constructed a cohort of almost all New Zealanders without CVD or heart failure, alive and aged 30–79 years in 2014, with follow-up linkage to hospitalisations and mortality until 2018 (N = 2,098,359). We derived two sets of sex-specific Cox regression models to predict 5-year CVD risk. The age-only models included "age" as the only predictor, and the full models used nine pre-defined routinely available CVD risk predictors with interaction terms. We calculated and compared individuals’ 5-year CVD risk from the age-only and full models and assessed model performance using Harrell’s C-statistics and calibration plots. Results Risk estimates for individuals using the two models showed substantial differences. A 65-year-old woman had predicted 5-year risks of 4.9% with the age-only model but 16.1% with the full model; a 74-year-old man had predicted risks of 15.2% and 9.2%, respectively, with age-only and full models (Figure 1). However, the models had similar discrimination and calibration (Figure 2). Age-only models had good discrimination with C-statistics of 0.773 (95% CI: 0.771–0.775) in women and 0.740 (0.738–0.742) in men, and calibrated well in both sexes. The full models had slightly better discrimination (C-statistics: 0.814 [0.812–0.816] in women and 0.769 [0.767–0.771] in men), and very similar calibration to the age-only models in both sexes. Conclusions Age-only models showed good discrimination and were well calibrated, but predicted very different risks in individual case studies compared to models with multiple predictors. This study calls into question the use of discrimination and calibration metrics as the main measures for comparing the performance of different CVD risk prediction models. We recommend that examples of predicted risk for individuals and an assessment of the magnitude and prevalence of cardiovascular predictors should also be considered when comparing the performance of prognostic models.

To develop 10-year cardiovascular disease (CVD) risk prediction models in Chinese patients with type 2 diabetes mellitus (T2DM) managed in primary care using machine learning (ML) methods. In this 10-year population-based retrospective cohort study, 141 516 Chinese T2DM patients aged 18 years or above, without history of CVD or end-stage renal disease and managed in public primary care clinics in 2008, were included and followed up until December 2017. Two-thirds of the patients were randomly selected to develop sex-specific CVD risk prediction models. The remaining one-third of patients were used as the validation sample to evaluate the discrimination and calibration of the models. ML-based methods were applied to missing data imputation, predictor selection, risk prediction modelling, model interpretation, and model evaluation. Cox regression was used to develop the statistical models in parallel for comparison. During a median follow-up of 9.75 years, 32 445 patients (22.9%) developed CVD. Age, T2DM duration, urine albumin-to-creatinine ratio (ACR), estimated glomerular filtration rate (eGFR), systolic blood pressure variability and glycated haemoglobin (HbA1c) variability were the most important predictors. ML models also identified nonlinear effects of several predictors, particularly the U-shaped effects of eGFR and body mass index. The ML models showed a Harrell's C statistic of >0.80 and good calibration. The ML models performed significantly better than the Cox regression models in CVD risk prediction and achieved better risk stratification for individual patients. Using routinely available predictors and ML-based algorithms, this study established 10-year CVD risk prediction models for Chinese T2DM patients in primary care. The findings highlight the importance of renal function indicators, and variability in both blood pressure and HbA1c as CVD predictors, which deserve more clinical attention. The derived risk prediction tools have the potential to support clinical decision making and encourage patients towards self-care, subject to further research confirming the models' feasibility, acceptability and applicability at the point of care.

Cardiovascular disease (CVD) is the leading cause of death in patients with chronic kidney disease (CKD). However, traditional CVD risk prediction equations do not work well in patients with CKD, and inclusion of kidney disease metrics such as albuminuria and estimated glomerular filtration rate have a modest to no benefit in improving prediction. As CKD progresses, the strength of traditional CVD risk factors in predicting clinical outcomes weakens. A pooled cohort equation used for CVD risk prediction is a useful tool for guiding clinicians on management of patients with CVD risk, but these equations do not calibrate well in patients with CKD, although a number of studies have developed modifications of the traditional equations to improve risk prediction. The reason for the poor calibration may be related to the fact that as CKD progresses, associations of traditional risk factors such as BMI, lipids and blood pressure with CVD outcomes are attenuated or reverse, and other risk factors may become more important. Large national cohorts such as the US Veteran cohort with many patients with evolving CKD may be useful resources for the developing CVD prediction models; however, additional considerations are needed for the unique composition of patients receiving care in these healthcare systems, including those with multiple comorbidities, as well as mental health issues, homelessness, posttraumatic stress disorders, frailty, malnutrition and polypharmacy. Machine learning over conventional risk prediction models may be better suited to handle the complexity needed for these CVD prediction models.

12023 Background: Previous efforts in predicting cardiotoxicity risk among patients with breast cancer have mostly focused on preventing and monitoring of cardiovascular complications during cancer therapy. We aimed to develop prediction models for the risk of chronic cardiovascular complications in breast cancer survivors during long-term follow-up. Methods: We used the Korean National Health Insurance Service databases between 2005 and 2021, including 5,131 5-year female breast cancer survivors diagnosed in 2006. The study cohort was randomized on a 4:1 ratio into the derivation and validation cohort. The primary outcome was the occurrence of major adverse cardiovascular events (MACEs), a composite of acute myocardial infarction, congestive heart failure, and stroke, at any time before the final follow-up at 10 years. We used the Cox proportional hazards model (CoxPH) with the least absolute shrinkage and selection operator penalty to determine the order of clinical factors based on their absolute coefficient value. We developed a deep learning survival model (DeepSurv) and compared its performance with traditional models such as CoxPH and random survival forest (RSF) using the same dataset. Model performance was assessed by discrimination and calibration. Results: During the 48054.1 person-years of follow-up, 325 (6.3%) patients developed MACE. We identified 18 relevant clinical factors without zero coefficients, including age, hypertension, atrial fibrillation, stroke, diabetes mellitus, myocardial infarction, household income, congestive heart failure, hemoglobin level, systolic blood pressure, peripheral arterial occlusive disease, chronic kidney disease, aromatase inhibitor use, alcohol consumption, cigarette smoking, tamoxifen use, radiotherapy, and total cholesterol level. The CoxPH, RSF, and DeepSurv model yielded time-dependent concordance index values of 0.713, 0.721, and 0.729, respectively, in the validation cohort. all models demonstrated good integrated Brier scores of 0.031 or less. Conclusions: We developed and validated a deep learning survival model that predict MACEs in individual 5-year breast cancer survivors, incorporating both conventional and breast cancer treatment-related cardiovascular risk factors, and demonstrated good calibration and discrimination. These models can assist breast cancer survivors and clinicians in optimally selecting risk-reducing strategies based on individual MACEs risk.

Risk prediction models play an important role in the primary prevention of cardiovascular diseases (CVD) in the elderly population. There are fifteen papers about CVD risk prediction models developed for the elderly domestically and internationally, of which the definitions of disease outcome vary widely. Ten models were reported with insufficient information about study methods or results. Ten models were at high risk of bias. Thirteen models presented moderate discrimination in internal validation, and only four models have undertaken external validation. The CVD risk prediction models for the elderly differed from those for the general population in terms of model algorithm and the effect size of association between predictor and outcome, and the prediction performance of the models for the elderly attenuated. In the future, high-quality external validation researches are necessary to provide more solid evidence. Different ways, including adding new predictors, using competing risk model algorithms, machine learning methods, or joint models, and altering the prediction time horizon, should be explored to optimize the current models.

AimsA diverse set of factors influence cardiovascular diseases (CVDs), but a systematic investigation of the interplay between these determinants and the contribution of each to CVD incidence prediction is largely missing from the literature. In this study, we leverage one of the most comprehensive biobanks worldwide, the UK Biobank, to investigate the contribution of different risk factor categories to more accurate incidence predictions in the overall population, by sex, different age groups, and ethnicity.Methods and resultsThe investigated categories include the history of medical events, behavioural factors, socioeconomic factors, environmental factors, and measurements. We included data from a cohort of 405 257 participants aged 37–73 years and trained various machine learning and deep learning models on different subsets of risk factors to predict CVD incidence. Each of the models was trained on the complete set of predictors and subsets where each category was excluded. The results were benchmarked against QRISK3. The findings highlight that (i) leveraging a more comprehensive medical history substantially improves model performance. Relative to QRISK3, the best performing models improved the discrimination by 3.78% and improved precision by 1.80%. (ii) Both model- and data-centric approaches are necessary to improve predictive performance. The benefits of using a comprehensive history of diseases were far more pronounced when a neural sequence model, BEHRT, was used. This highlights the importance of the temporality of medical events that existing clinical risk models fail to capture. (iii) Besides the history of diseases, socioeconomic factors and measurements had small but significant independent contributions to the predictive performance.ConclusionThese findings emphasize the need for considering broad determinants and novel modelling approaches to enhance CVD incidence prediction.

Risk estimation is key for primary prevention of cardiovascular diseases (CVD). The European Society of Cardiology recommends SCORE2, SCORE2-OP (Older Persons), and SCORE2-Diabetes (collectively, the SCORE2 family-of-models (FOMs)) for 10-year CVD-risk estimation. As the COVID-19 pandemic was associated with higher CVD incidences, we externally validated the SCORE2-FOMs using contemporary nationwide data and assessed the need for adaptation. Using population-wide electronic health records (EHRs) for England, we defined a cohort of 26.5 million individuals aged 40 to 89 years without prior CVD on 1st January 2022 followed-up until 31st December 2023. We calculated annual incidence rates of CVD events and non-CVD mortality by sex and one-year age-groups to estimate expected 10-year, competing-risk adjusted, CVD-risk using a life-table approach. We assessed SCORE2-FOMs calibration and discrimination, overall and by subgroups. Within two years of follow-up, 239,815 incident CVD events occurred. For SCORE2, SCORE2-Diabetes, and SCORE2-OP the C-index values were 0.75 (95%CI: 0.75-0.76), 0.71 (0.71-0.71) and 0.65 (0.65-0.65), respectively. SCORE2-FOMs overpredicted CVD risks by 3% to 4% in males, and by 8% to 14% in females, with calibration variations by English region and ethnic group. Elevated CVD-risk associated with a recent COVID-19 hospitalisation led SCORE2 to underestimate risks by 2% in affected males aged 40-69 years. The SCORE2-FOMs demonstrated good calibration and discrimination, suggesting that no adaptation is needed in response to the COVID-19 pandemic. Short-term EHRs provide a valuable resource for timely contemporary risk model validation and, if necessary, for adaptation according to shifts in CVD trends.

Abstract: The increasing pressure on health resources has led to the emergence of risk assessment as an essential tool in the management of cardiovascular disease (CVD). Concern exists regarding the validity of their generalization to all populations. Existing risk scoring models do not incorporate emerging ‘novel’ risk factors. In this context, the aim of the study was to examine the relevance of British, European, and Framingham predictive CVD risk scores to the asymptomatic high risk Indian population. Blood samples drawn from the participants were analyzed for various ‘traditional’ and ‘novel’ biomarkers, and their CVD risk factor profiling was also done. The Framingham model defined only 5% of the study cohort to be at high risk, which appears to be an underestimation of CVD risk in this genetically predisposed population. These subjects at high risk had significantly elevated levels of lipid, pro-inflammatory, pro-thrombotic, and serological markers. It is more relevant to develop risk predictive scores for application to the Indian population. This study substantiates the argument that alternative approaches to risk stratification are required in order to make them more adaptable and applicable to different populations with varying risk factor and disease patterns. Keywords: atherosclerosis, risk factors, risk score, Framingham, plasma biomarkers

Statistical Approach to Neural Network Model Building for Gentamicin Peak Predictions

Introduction and objectivesThere are no cardiovascular (CV) risk prediction models for Sri Lankans. Different risk prediction models not validated for Sri Lankans are being used to predict CV risk of Sri Lankans. We validated the WHO/ISH (SEAR-B) risk prediction charts prospectively in a population-based cohort of Sri Lankans.MethodWe selected 40–64 year-old participants from the Ragama Medical Officer of Health (MOH) area in 2007 by stratified random sampling and followed them up for 10 years. Ten-year risk predictions of a fatal/non-fatal cardiovascular event (CVE) in 2007 were calculated using WHO/ISH (SEAR-B) charts with and without cholesterol. The CVEs that occurred from 2007–2017 were ascertained. Risk predictions in 2007 were validated against observed CVEs in 2017.ResultsOf 2517 participants, the mean age was 53.7 year (SD: 6.7) and 1132 (45%) were males. Using WHO/ISH chart with cholesterol, the percentages of subjects with a 10-year CV risk <10%, 10–19%, 20%-29%, 30–39%, ≥40% were 80.7%, 9.9%, 3.8%, 2.5% and 3.1%, respectively. 142 non-fatal and 73 fatal CVEs were observed during follow-up. Among the cohort, 9.4% were predicted of having a CV risk ≥20% and 8.6% CVEs were observed in the risk category. CVEs were within the predictions of WHO/ISH charts with and without cholesterol in both high (≥20%) and low(<20%) risk males, but only in low(<20%) risk females. The predictions of WHO/ISH charts, with-and without-cholesterol were in agreement in 81% of subjects (ĸ = 0.429; p<0.001).ConclusionsWHO/ISH (SEAR B) risk prediction charts with-and without-cholesterol may be used in Sri Lanka. Risk charts are more predictive in males than in females and for lower-risk categories. The predictions when stratifying into 2 categories, low risk (<20%) and high risk (≥20%), are more appropriate in clinical practice.

Introduction: Allogeneic hematopoietic cell transplantation (HCT) is a curative treatment for malignant and non-malignant hematologic diseases. However, long-term HCT survivors have a &amp;gt;4-fold risk of developing cardiovascular disease (CVD) compared to the general population, and CVD is a leading cause of excess mortality after HCT. Comorbidities that emerge after HCT such as hypertension (HTN), diabetes mellitus (DM), and dyslipidemia (DL) are important modifiers of CVD risk, but the impact of social determinants of health (SDOH) on the development of these risk factors and resultant CVD has not been well-characterized. Methods: This retrospective cohort study included 889 patients who underwent an allogenic HCT between 2013 and 2019 at City of Hope and survived ≥1y after HCT. Demographic, disease, treatment, comorbidity, and health outcomes data were abstracted from medical records. The social vulnerability index (SVI), obtained from the Center for Disease Control, represented overall social vulnerability and included 4 themes: socioeconomic status, household composition &amp; disability, minority status &amp; language, and housing &amp; transportation. SVI scores were derived using patients’ addresses at time of HCT. SVI scores ranged from 0 (least vulnerable) to 100 (most vulnerable). Fine-Gray regression analysis evaluated the association between SVI and risk of HTN, DM, and DL, adjusted for age at HCT and treating death as competing risk. We also evaluated the association between having 0, 1, ≥2 cardiovascular risk factors on incidence of clinically significant CVD (heart failure, coronary artery disease, pericarditis, stroke, intracranial hemorrhage, arrhythmia, and/or valvular disease) after HCT. Results: Mean age at HCT was 44.2y (standard deviation 19.8y); 56.7% were male, 43.8% were non-Hispanic White, 35.4% were Hispanic, and 12.9% were Asian. The most common diagnosis was acute myeloid leukemia (39.6%); 48.6% received reduced intensity conditioning. Higher overall SVI was associated with increased risk of HTN (adjusted Hazard Ratio per 10 unit increase in SVI [aHR] 1.10, 95%CI 1.02-1.19) and DM (aHR 1.13, 95%CI 1.03-1.25). There was also an increased risk for DL for the highest SVI tertile for the housing &amp; transportation theme (aHR 1.61, 95%CI 1.10-2.37). Accumulation of de novo risk factors after HCT was associated with an incremental 5y incidence of CVD (None: 6.1%, 95% CI 3.4%-9.9%; 1 risk factor: 15.7%, 95% CI 7.1%-27.3%; &amp;gt;2 risk factors: 31.6%, 95%CI 12.8%-52.4%; p&amp;lt;0.001). In the adjusted multivariable model, survivors with &amp;gt;2 risk factors had 7.0-fold (aHR 6.99, 95%CI 2.86-17.05) risk of CVD compared to those without (ref). Conclusion: Among allogeneic HCT recipients, higher SVI was associated with increased risk of HTN, DM, and DL after HCT, which in turn were associated with incrementally higher 5y incidence of CVD. These findings underscore the importance of SDOH on clinically important outcomes after allogeneic HCT, and may ultimately inform tailored approaches for risk-based screening and resource allocation in long-term survivors. Citation Format: Osariemen V Ogiamien, Sitong Chen, Lizel Atencio, Alysia Bosworth, Meagan Echevarria, Caitlyn Estrada, Mareen Kassabian, Lanie Lindenfeld, F. Lennie Wong, Saro H Armenian, Rusha Bhandari. Social vulnerability is an important predictor of cardiovascular disease risk after allogeneic hematopoietic cell transplantation [abstract]. In: Proceedings of the 17th AACR Conference on the Science of Cancer Health Disparities in Racial/Ethnic Minorities and the Medically Underserved; 2024 Sep 21-24; Los Angeles, CA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2024;33(9 Suppl):Abstract nr C044.

Applying Framingham Risk Score 2008 to Predict the 10-Year Risk of Cardiovascular Disease in a Group of Office Workers in Hanoi, Vietnam

A feed-forward multilayer perceptron neural network model with Levenberg-Marquardt learning algorithm has been developed to predict rheological properties of water-based drilling fluids. The developed model was based on a three-layer network containing input, hidden, and output layer. The sigmoid function was applied as the transfer function in the hidden layer and linear transfer function in the output layer. Two statistical parameters, mean square errors and correlation coefficient, were used as a criterion for evaluating artificial neural network modeling performance. Accordingly, a comparison between experimental values and those predicted by the artificial neural network showed a good coincidence proving its high accuracy in estimating target value.

Introduction High-density lipoprotein cholesterol (HDL-C) has traditionally been included in atherosclerotic cardiovascular disease (ASCVD) risk prediction models, such as the pooled cohort equations. However, growing evidence suggests that lipoprotein particle concentration may serve as a more reliable predictor of cardiovascular outcomes than cholesterol content alone. Studies, including the MESA trial, indicate that HDL particle concentration (HDL-P) is a stronger determinant of cardiovascular risk than HDL-C. Larger HDL particles may be dysfunctional, leading to misleading HDL-C levels that do not accurately reflect cardiovascular protection. Adjusting HDL-C for HDL-P may improve risk stratification. We hypothesize that incorporating HDL-P into risk prediction models will refine cardiovascular risk assessment. Purpose This study evaluates the limitations of HDL-C in risk prediction and assesses whether particle-adjusted values provide a more accurate measure of cardiovascular risk when integrated into standard risk calculations. Methods A total of 67 lipid profiles from Scripps Health were analyzed, including traditional lipid concentrations, HDL and LDL particle concentrations, and age at the time of lipid profile collection. Cardiovascular risk was calculated using unadjusted baseline lipid values within the PREVENT risk prediction calculator. To isolate the impact of lipid parameter adjustments on risk estimation, risk factors such as diabetes, smoking, eGFR, SBP, and BMI were held constant across all calculations. The calculations were then repeated using adjusted HDL-C and Total Cholesterol values, accounting for LDL and HDL particle concentrations. Adjusted values were derived using statistical regression models based on real-world data from the Framingham Offspring Study and pooled cohort analyses (DHS, ARIC, MESA, PREVEND). Results After adjustment, 47 patients (70.1%) demonstrated an increase in predicted cardiovascular risk, with an average absolute risk increase of 0.834% and an average relative change of 15.31%. Conversely, 20 patients (29.9%) exhibited a decrease in predicted risk, with an average absolute risk decrease of 0.2785% and an average relative change of 6.72%. The total combined relative risk change across both groups was 22.03%, suggesting that incorporating HDL-P and LDL-P meaningfully alters risk assessment. Conclusions Adjusting for particle concentration significantly influences cardiovascular risk prediction, with the majority of patients experiencing an increase in estimated risk. This supports the hypothesis that HDL-P, rather than HDL-C, provides a more precise assessment of cardiovascular risk. Current risk models relying solely on HDL-C and total cholesterol may misclassify risk, underscoring the need to incorporate particle-adjusted values into predictive algorithms for more accurate cardiovascular risk stratification.

This study investigates the effectiveness of an Artificial Neural Network (ANN) model in enhancing the classification of dry beans varieties. The dry beans dataset, comprising 13,611 samples and 16 numerical features, was preprocessed using Z-score normalization for outlier removal and Principal Component Analysis (PCA) to remove 702 outliers and reduce dimensionality to 10 components while preserving 99.99% of the original variance. A feedforward ANN architecture was designed, constructed as a feedforward neural network with three fully connected layers namely, an input layer that receives 10 features (as reduced by PCA), a first hidden layer with 16 neurons and ReLU activation, a second hidden layer with 32 neurons and ReLU activation, and an output layer with neurons equal to the number of classes, using softmax activation for multi-class probability prediction. The model was trained over 60 epochs using TensorFlow/Keras. Evaluation on a held-out test set demonstrated that the ANN model achieved a classification accuracy of 94.50%. Performance metrics, including accuracy, precision, recall, F1-score, and confusion matrix analysis, confirmed the model's robustness and generalization capability across all dry beans’ classes. The findings affirm that ANN models, when supported by appropriate preprocessing, can serve as scalable and accurate alternatives to conventional machine learning approaches in agricultural classification tasks.