Evaluation of GPT-4 for 10-year cardiovascular risk prediction: Insights from the UK Biobank and KoGES data.
Ontology highlight
ABSTRACT:
SUBMITTER: Han C
PROVIDER: S-EPMC10865411 | biostudies-literature | 2024 Feb
REPOSITORIES: biostudies-literature
Publications
Evaluation of GPT-4 for 10-year cardiovascular risk prediction: Insights from the UK Biobank and KoGES data.
Han Changho C Kim Dong Won DW Kim Songsoo S Chan You Seng S Park Jin Young JY Bae SungA S Yoon Dukyong D
iScience 20240124 2
Cardiovascular disease (CVD) remains a pressing global health concern. While traditional risk prediction methods such as the Framingham and American College of Cardiology/American Heart Association (ACC/AHA) risk scores have been widely used in the practice, artificial intelligence (AI), especially GPT-4, offers new opportunities. Utilizing large scale of multi-center data from 47,468 UK Biobank participants and 5,718 KoGES participants, this study quantitatively evaluated the predictive capabil ...[more]
Similar Datasets
Project description:AimsCardiovascular diseases (CVDs) are among the leading causes of death worldwide. Predictive scores providing personalized risk of developing CVD are increasingly used in clinical practice. Most scores, however, utilize a homogenous set of features and require the presence of a physician. The aim was to develop a new risk model (DiCAVA) using statistical and machine learning techniques that could be applied in a remote setting. A secondary goal was to identify new patient-centric variables that could be incorporated into CVD risk assessments.Methods and resultsAcross 466 052 participants, Cox proportional hazards (CPH) and DeepSurv models were trained using 608 variables derived from the UK Biobank to investigate the 10-year risk of developing a CVD. Data-driven feature selection reduced the number of features to 47, after which reduced models were trained. Both models were compared to the Framingham score. The reduced CPH model achieved a c-index of 0.7443, whereas DeepSurv achieved a c-index of 0.7446. Both CPH and DeepSurv were superior in determining the CVD risk compared to Framingham score. Minimal difference was observed when cholesterol and blood pressure were excluded from the models (CPH: 0.741, DeepSurv: 0.739). The models show very good calibration and discrimination on the test data.ConclusionWe developed a cardiovascular risk model that has very good predictive capacity and encompasses new variables. The score could be incorporated into clinical practice and utilized in a remote setting, without the need of including cholesterol. Future studies will focus on external validation across heterogeneous samples.
| S-EPMC9707906 | biostudies-literature
Project description:BackgroundCardiovascular diseases (CVDs) are a leading global health concern. Emerging evidence suggests a potential protective role of well-being in reducing CVD risk.Methods and resultsWe conducted a cohort analysis using the UK Biobank data set, encompassing 121 317 participants. We assessed the well-being of participants using a well-being index derived from baseline questionnaires. Well-being categories were derived by latent class analysis using general happiness and satisfaction with family, friendships, health, and finance situations. The relationship between well-being and 4 major CVDs was analyzed using Cox proportional hazards models and Mendelian randomization. The study also examined the impacts of well-being on lifestyle factors and inflammatory markers, and its mediating role in the well-being-CVD relationship. Higher well-being was associated with a significantly reduced risk of various CVDs. Latent class analysis identified 4 distinct well-being groups (low, variable, moderate-to-high, and high satisfaction), with higher satisfaction levels generally associated with lower risk of CVDs. Mendelian randomization suggested potential causal relationships between well-being and reduced risk of CVDs. Participants with greater well-being demonstrated healthier behaviors and lower levels of inflammatory markers. Mediation analysis indicated that lifestyle and inflammatory markers partially mediated the relationship between well-being and CVDs.ConclusionsThis study demonstrates a robust inverse association between well-being and the risks of CVDs, suggesting that enhancing well-being may be a viable strategy for CVD prevention. The role of lifestyle factors and inflammation as a mediator provides insight into possible biological pathways linking psychological states and cardiovascular health.
| S-EPMC11681495 | biostudies-literature
Project description:BackgroundTreatment guidelines for prediabetes primarily focus on glycemic control and lifestyle management. Few evidence-based cardiovascular and kidney risk-reduction strategies are available in this population.ObjectivesThis study sought to characterize cardiovascular and kidney outcomes across the glycemic spectrum.MethodsAmong participants in the UK Biobank without prevalent type 1 diabetes, cardiovascular disease, or kidney disease, Cox models tested the association of glycemic exposures (type 2 diabetes [T2D], prediabetes, normoglycemia) with outcomes (atherosclerotic cardiovascular disease [ASCVD], chronic kidney disease [CKD], and heart failure), adjusting for demographic, lifestyle, and cardiometabolic risk factors.ResultsAmong 336,709 individuals (mean age: 56.3 years, 55.4% female), 46,911 (13.9%) had prediabetes and 12,717 (3.8%) had T2D. Over median follow-up of 11.1 years, 6,476 (13.8%) individuals with prediabetes developed ≥1 incident outcome, of whom only 802 (12.4%) developed T2D prior to an incident diagnosis. Prediabetes and T2D were independently associated with ASCVD (prediabetes: adjusted HR [aHR]: 1.11; 95% CI: 1.08-1.15; P < 0.001; T2D: aHR: 1.44; 95% CI: 1.37-1.51; P < 0.001), CKD (prediabetes: aHR: 1.08; 95% CI: 1.02-1.14; P < 0.001; T2D: aHR: 1.57; 95% CI: 1.46-1.69; P < 0.001), and heart failure (prediabetes: aHR: 1.07; 95% CI: 1.01-1.14; P = 0.03; T2D: aHR: 1.25; 95% CI: 1.14-1.37; P < 0.001). Compared with hemoglobin A1c (HbA1c) <5.0%, covariate-adjusted risks increased significantly for ASCVD above HbA1c of 5.4%, CKD above HbA1c of 6.2%, and heart failure above HbA1c of 7.0%.ConclusionsPrediabetes and T2D were associated with ASCVD, CKD, and heart failure, but a substantial gradient of risk was observed across HbA1c levels below the threshold for diabetes. These findings highlight the need to design risk-reduction strategies across the glycemic spectrum.
| S-EPMC8324525 | biostudies-literature
Project description:Cardiovascular diseases (CVDs) are the primary cause of all death globally. Timely and accurate identification of people at risk of developing an atherosclerotic CVD and its sequelae is a central pillar of preventive cardiology. One widely used approach is risk prediction models; however, currently available models consider only a limited set of risk factors and outcomes, yield no actionable advice to individuals based on their holistic medical state and lifestyle, are often not interpretable, were built with small cohort sizes or are based on lifestyle data from the 1960s, e.g. the Framingham model. The risk of developing atherosclerotic CVDs is heavily lifestyle dependent, potentially making many occurrences preventable. Providing actionable and accurate risk prediction tools to the public could assist in atherosclerotic CVD prevention. Accordingly, we developed a benchmarking pipeline to find the best set of data preprocessing and algorithms to predict absolute 10-year atherosclerotic CVD risk. Based on the data of 464,547 UK Biobank participants without atherosclerotic CVD at baseline, we used a comprehensive set of 203 consolidated risk factors associated with atherosclerosis and its sequelae (e.g. heart failure). Our two best performing absolute atherosclerotic risk prediction models provided higher performance, (AUROC: 0.7573, 95% CI: 0.755-0.7595) and (AUROC: 0.7544, 95% CI: 0.7522-0.7567), than Framingham (AUROC: 0.680, 95% CI: 0.6775-0.6824) and QRisk3 (AUROC: 0.725, 95% CI: 0.7226-0.7273). Using a subset of 25 risk factors identified with feature selection, our reduced model achieves similar performance (AUROC 0.7415, 95% CI: 0.7392-0.7438) while being less complex. Further, it is interpretable, actionable and highly generalizable. The model could be incorporated into clinical practice and might allow continuous personalized predictions with automated intervention suggestions.
| S-EPMC8836294 | biostudies-literature
Project description:We assigned 329,908 UK Biobank participants into six subgroups based on a self-organizing map of 51 biochemical measures (blinded for clinical outcomes). The subgroup with the most favorable metabolic traits was chosen as the reference. Hazard ratios (HR) for incident disease were modeled by Cox regression. Enrichment ratios (ER) of incident multi-morbidity versus randomly expected co-occurrence were evaluated by permutation tests; ER is like HR but captures co-occurrence rather than event frequency. The subgroup with high urinary excretion without kidney stress (HR = 1.24) and the subgroup with the highest apolipoprotein B and blood pressure (HR = 1.52) were associated with ischemic heart disease (IHD). The subgroup with kidney stress, high adiposity and inflammation was associated with IHD (HR = 2.11), cancer (HR = 1.29), dementia (HR = 1.70) and mortality (HR = 2.12). The subgroup with high liver enzymes and triglycerides was at risk of diabetes (HR = 15.6). Multimorbidity was enriched in metabolically favorable subgroups (3.4 ≤ ER ≤ 4.0) despite lower disease burden overall; the relative risk of co-occurring disease was higher in the absence of obvious metabolic dysfunction. These results provide synergistic insight into metabolic health and its associations with cardiovascular disease in a large population sample.
| S-EPMC9124207 | biostudies-literature
Project description:Background: Polygenic risk score (PRS) has proved useful in predicting the risk of cardiovascular diseases (CVD) based on the genotypes of an individual, but most analyses have focused on disease onset in the general population. The usefulness of PRS to predict CVD risk among type 2 diabetes (T2D) patients remains unclear. Methods: We built a meta-PRSCVD upon the candidate PRSs developed from state-of-the-art PRS methods for three CVD subtypes of significant importance: coronary artery disease (CAD), ischemic stroke (IS), and heart failure (HF). To evaluate the prediction performance of the meta-PRSCVD, we restricted our analysis to 21,092 white British T2D patients in the UK Biobank, among which 4,015 had CVD events. Results: Results showed that the meta-PRSCVD was significantly associated with CVD risk with a hazard ratio per standard deviation increase of 1.28 (95% CI: 1.23-1.33). The meta-PRSCVD alone predicted the CVD incidence with an area under the receiver operating characteristic curve (AUC) of 0.57 (95% CI: 0.54-0.59). When restricted to the early-onset patients (onset age ≤ 55), the AUC was further increased to 0.61 (95% CI 0.56-0.67). Conclusion: Our results highlight the potential role of genomic screening for secondary preventions of CVD among T2D patients, especially among early-onset patients.
| S-EPMC10794561 | biostudies-literature
Project description:BackgroundWhile risk stratification for atherosclerotic cardiovascular disease (ASCVD) is essential for primary prevention, current clinical risk algorithms demonstrate variability and leave room for further improvement. The plasma proteome holds promise as a future diagnostic and prognostic tool that can accurately reflect complex human traits and disease processes. We assessed the ability of plasma proteins to predict ASCVD.MethodClinical, genetic, and high-throughput plasma proteomic data were analyzed for association with ASCVD in a cohort of 41,650 UK Biobank participants. Selected features for analysis included clinical variables such as a UK-based cardiovascular clinical risk score (QRISK3) and lipid levels, 36 polygenic risk scores (PRSs), and Olink protein expression data of 2,920 proteins. We used least absolute shrinkage and selection operator (LASSO) regression to select features and compared area under the curve (AUC) statistics between data types. Randomized LASSO regression with a stability selection algorithm identified a smaller set of more robustly associated proteins. The benefit of plasma proteins over standard clinical variables, the QRISK3 score, and PRSs was evaluated through the derivation of Δ AUC values. We also assessed the incremental gain in model performance using proteomic datasets with varying numbers of proteins. To identify potential causal proteins for ASCVD, we conducted a two-sample Mendelian randomization (MR) analysis.ResultThe mean age of our cohort was 56.0 years, 60.3% were female, and 9.8% developed incident ASCVD over a median follow-up of 6.9 years. A protein-only LASSO model selected 294 proteins and returned an AUC of 0.723 (95% CI 0.708-0.737). A clinical variable and PRS-only LASSO model selected 4 clinical variables and 20 PRSs and achieved an AUC of 0.726 (95% CI 0.712-0.741). The addition of the full proteomic dataset to clinical variables and PRSs resulted in a Δ AUC of 0.010 (95% CI 0.003-0.018). Fifteen proteins selected by a stability selection algorithm offered improvement in ASCVD prediction over the QRISK3 risk score [Δ AUC: 0.013 (95% CI 0.005-0.021)]. Filtered and clustered versions of the full proteomic dataset (consisting of 600-1,500 proteins) performed comparably to the full dataset for ASCVD prediction. Using MR, we identified 11 proteins as potentially causal for ASCVD.ConclusionA plasma proteomic signature performs well for incident ASCVD prediction but only modestly improves prediction over clinical and genetic factors. Further studies are warranted to better elucidate the clinical utility of this signature in predicting the risk of ASCVD over the standard practice of using the QRISK3 score.
| S-EPMC11419231 | biostudies-literature
Project description:BackgroundRobust and accurate prediction of cardiovascular disease (CVD) risk facilitates early intervention to benefit patients. The intricate relationship between mental health disorders and CVD is widely recognized. However, existing models often overlook psychological factors, relying on a limited set of clinical and lifestyle parameters, or being developed on restricted population subsets.ObjectivesThis study aims to assess the impact of integrating psychological data into a novel machine learning (ML) approach on enhancing CVD prediction performance.MethodsUsing a comprehensive UK Biobank data set (n = 375,145), the correlation between CVD and traditional and psychological risk factors was examined. CVD included hypertensive disease, ischemic heart disease, heart failure, and arrhythmias. An ensemble ML model containing 5 constituent algorithms (decision tree, random forest, XGBoost, support vector machine, and deep neural network) was tested for its ability to predict CVD based on 2 training data sets: using traditional CVD risk factors alone, or using a combination of traditional and psychological risk factors.ResultsA total of 375,145 subjects with normal health status and with CVD were included. The ensemble ML model could predict CVD with 71.31% accuracy using traditional CVD risk factors alone. However, by adding psychological factors to the training data, accuracy increased to 85.13%. The accuracy and robustness of the ensemble ML model outperformed all 5 constituent learning algorithms.ConclusionsIncorporating mental health assessment data within an ensemble ML model results in a significantly improved, highly accurate, CVD prediction model, outperforming traditional risk factor prediction alone.
| S-EPMC11450915 | biostudies-literature
Project description:BackgroundIdentifying people at risk of cardiovascular diseases (CVD) is a cornerstone of preventative cardiology. Risk prediction models currently recommended by clinical guidelines are typically based on a limited number of predictors with sub-optimal performance across all patient groups. Data-driven techniques based on machine learning (ML) might improve the performance of risk predictions by agnostically discovering novel risk predictors and learning the complex interactions between them. We tested (1) whether ML techniques based on a state-of-the-art automated ML framework (AutoPrognosis) could improve CVD risk prediction compared to traditional approaches, and (2) whether considering non-traditional variables could increase the accuracy of CVD risk predictions.Methods and findingsUsing data on 423,604 participants without CVD at baseline in UK Biobank, we developed a ML-based model for predicting CVD risk based on 473 available variables. Our ML-based model was derived using AutoPrognosis, an algorithmic tool that automatically selects and tunes ensembles of ML modeling pipelines (comprising data imputation, feature processing, classification and calibration algorithms). We compared our model with a well-established risk prediction algorithm based on conventional CVD risk factors (Framingham score), a Cox proportional hazards (PH) model based on familiar risk factors (i.e, age, gender, smoking status, systolic blood pressure, history of diabetes, reception of treatments for hypertension and body mass index), and a Cox PH model based on all of the 473 available variables. Predictive performances were assessed using area under the receiver operating characteristic curve (AUC-ROC). Overall, our AutoPrognosis model improved risk prediction (AUC-ROC: 0.774, 95% CI: 0.768-0.780) compared to Framingham score (AUC-ROC: 0.724, 95% CI: 0.720-0.728, p < 0.001), Cox PH model with conventional risk factors (AUC-ROC: 0.734, 95% CI: 0.729-0.739, p < 0.001), and Cox PH model with all UK Biobank variables (AUC-ROC: 0.758, 95% CI: 0.753-0.763, p < 0.001). Out of 4,801 CVD cases recorded within 5 years of baseline, AutoPrognosis was able to correctly predict 368 more cases compared to the Framingham score. Our AutoPrognosis model included predictors that are not usually considered in existing risk prediction models, such as the individuals' usual walking pace and their self-reported overall health rating. Furthermore, our model improved risk prediction in potentially relevant sub-populations, such as in individuals with history of diabetes. We also highlight the relative benefits accrued from including more information into a predictive model (information gain) as compared to the benefits of using more complex models (modeling gain).ConclusionsOur AutoPrognosis model improves the accuracy of CVD risk prediction in the UK Biobank population. This approach performs well in traditionally poorly served patient subgroups. Additionally, AutoPrognosis uncovered novel predictors for CVD disease that may now be tested in prospective studies. We found that the "information gain" achieved by considering more risk factors in the predictive model was significantly higher than the "modeling gain" achieved by adopting complex predictive models.
| S-EPMC6519796 | biostudies-literature
Project description:IntroductionRisk prediction is important for preventing and managing cardiovascular disease (CVD). CVD risk prediction tools designed for the general population may be inaccurate in people with inflammatory diseases.ObjectivesTo investigate the performance of four cardiovascular risk prediction tools (QRISK3, Framingham Risk Score, Reynolds Risk Score and SCORE) in psoriatic arthritis (PsA) and psoriasis. We also compare performance in participants with no inflammatory conditions and in people with rheumatoid arthritis (RA).MethodsThis research utilised the UK Biobank Resource. We identified participants with PsA, psoriasis and RA and calculated their cardiovascular risk using each risk tool. We assessed model calibration by comparing observed and predicted outcomes. Discrimination of 10-year risk prediction was assessed using time-dependent area under ROC curve (AUC), sensitivity, specificity, positive and negative predictive values.ResultsWe included 769 individuals with PsA, 8062 with psoriasis and 4772 with RA when assessing the QRISK3 tool. Predictions for individuals with psoriasis were roughly as accurate as those with no inflammatory conditions with time-dependent AUC of 0.74 (95%CI, 0.72, 0.76) and of 0.74 (95%CI, 0.72, 0.77) respectively. In contrast, individuals with PsA obtained the least accurate predictions with an AUC of 0.70 (95%CI, 0.64, 0.76). Individuals with RA also obtained less accurate predictions with AUC of 0.72 (0.69,0.74). For the Framingham risk score, AUCs varied between 0.61 (95%CI, 0.55, 0.68) for participants with PsA and 0.71 (95%CI, 0.68, 0.74) for individuals with no inflammatory condition.ConclusionsIn general, CVD risk prediction accuracy was similar for individuals with psoriasis or no inflammatory condition, but lower for individuals with PsA or RA.
| S-EPMC11865138 | biostudies-literature