10-year Risk Prediction Research Articles

CKD Progression Prediction in a Diverse US Population: A Machine-Learning Model

Chronic kidney disease (CKD) is a major cause of morbidity and mortality. To date, there are no widely used machine-learning models that can predict progressive CKD across the entire disease spectrum, including the earliest stages. The objective of this study was to use readily available demographic and laboratory data from Sonic Healthcare USA laboratories to train and test the performance of machine learning-based predictive risk models for CKD progression. Retrospective observational study. The study population was composed of deidentified laboratory information services data procured from a large US outpatient laboratory network. The retrospective data set included 110,264 adult patients over a 5-year period with initial estimated glomerular filtration rate (eGFR) values between 15-89mL/min/1.73m2. Patient demographic and laboratory characteristics. Accelerated (ie,>30%) eGFR decline associated with CKD progression within 5 years. Machine-learning models were developed using random forest survival methods, with laboratory-based risk factors analyzed as potential predictors of significant eGFR decline. The 7-variable risk classifier model accurately predicted an eGFR decline of>30% within 5 years and achieved an area under the curve receiver-operator characteristic of 0.85. The most important predictor of progressive decline in kidney function was the eGFR slope. Other key contributors to the model included initial eGFR, urine albumin-creatinine ratio, serum albumin (initial and slope), age, and sex. The cohort study did not evaluate the role of clinical variables (eg, blood pressure) on the performance of the model. Our progressive CKD classifier accurately predicts significant eGFR decline in patients with early, mid, and advanced disease using readily obtainable laboratory data. Although prospective studies are warranted, our results support the clinical utility of the model to improve timely recognition and optimal management for patients at risk for CKD progression. Defined by a significant decrease in estimated glomerular filtration rate (eGFR), chronic kidney disease (CKD) progression is strongly associated with kidney failure. However, to date, there are no broadly used resources that can predict this clinically significant event. Using machine-learning techniques on a diverse US population, this cohort study aimed to address this deficiency and found that a 5-year risk prediction model for CKD progression was accurate. The most important predictor of progressive decline in kidney function was the eGFR slope, followed by the urine albumin-creatinine ratio and serum albumin slope. Although further study is warranted, the results showed that a machine-learning model using readily obtainable laboratory information accurately predicts CKD progression, which may inform clinical diagnosis and management for this at-risk population.

517-P: Machine Learning Algorithms Using Electronic Medical Record and Ultrasonography of Lower Extremities for 10-Year Risk Assessment of Diabetic Foot Ulcer in China

Objective: Accurate prediction of diabetic foot ulcer (DFU) occurrence in patients with type 2 diabetes mellitus (T2DM) is crucial for determining appropriate treatment strategy and avoiding amputation and infection. Here, we explored machine learning algorithms (ML) of electronic medical record (EMR) for predicting the incident of DFU. Methods: In this retrospective longitudinal observational cohort, 26579 cases from T2DM database of The Third Affiliated Hospital of Sun Yatsen University from Jan 2011 to Dec 2019 were screened and finally 2770 patients without DFU were enrolled. During 5.6 years of follow-up, 178 patients occurred DFU. 44 clinical parameters including EMR and imaging finding were used to develop a DFU risk prediction model using XGBoost, RF, and LR. Results: ML models were established in 178 patients with DFU (110 male, mean 61.1±12.4 years), and 2591 patients without DFU (1424 male, mean 59.0±12.1 years). XGBoost model achieved the best performance in predicting DFU occurrence with an AUC of 0.94, which is better than the performance of RF with AUC of 0.91and LR with AUC of 0.89. A DFU incident significantly increased in high-risk patients (>10%) in the next 10 years. Conclusion: Our study provided a simple and effective risk tool for the prediction of 10-year DFU risk, which can help professionals clinical screening and decision-making for T2DM patients. Disclosure L.Wu: None. Y.Liang: None. X.He: None. Y.Chen: None.

Laboratory-based versus non-laboratory-based World Health Organization risk equations for assessment of cardiovascular disease risk

BackgroundThe WHO model has laboratory-based and non-laboratory-based versions for 10-year risk prediction of cardiovascular diseases. Due to the fact that in some settings, there may not be the necessary facilities for risk assessment with a laboratory-based model, the present study aimed to determine the agreement between laboratory-based and non-laboratory-based WHO cardiovascular risk equations.MethodsIn this cross-sectional study, we used the baseline data of 6796 individuals without a history of cardiovascular disease and stroke who participated in the Fasa cohort study. The risk factors of the laboratory-based model included age, sex, systolic blood pressure (SBP), diabetes, smoking and total cholesterol, while the non-laboratory-based model included age, sex, SBP, smoking and BMI. Kappa coefficients was used to determine the agreement between the grouped risk and Bland–Altman plots were used to determine the agreement between the scores of the two models. Sensitivity and specificity of non-laboratory-based model were measured at the high-risk threshold.ResultsIn the whole population, the agreement between the grouped risk of the two models was substantial (percent agreement = 79.0%, kappa = 0.68). The agreement was better in males than in females. A substantial agreement was observed in all males (percent agreement = 79.8%, kappa = 0.70) and males < 60 years old (percent agreement = 79.9%, kappa = 0.67). The agreement in males ≥ 60 years old was moderate (percent agreement = 79.7%, kappa = 0.59). The agreement among females was also substantial (percent agreement = 78.3%, kappa = 0.66). The agreement for females < 60 years old, (percent agreement = 78.8%, kappa = 0.61) was substantial and for females ≥ 60 years old, (percent agreement = 75.8%, kappa = 0.46) was moderate. According to Bland–Altman plots, the limit of agreement was (95%CI: -4.2% to 4.3%) for males and (95%CI: -4.1% to 4.6%) for females. The range of agreement was suitable for both males < 60 years (95%CI: -3.8% to 4.0%) and females < 60 years (95%CI: -3.6% to 3.9%). However, it was not suitable for males ≥ 60 years (95% CI: -5.8% to 5.5%) and females ≥ 60 years (95%CI: -5.7% to 7.4%). At the high-risk threshold of 20% in non-laboratory and laboratory-based models, the sensitivity of the non-laboratory-based model was 25.7%, 70.7%, 35.7%, and 35.4% for males < 60 years, males ≥ 60 years, females < 60 years, and females ≥ 60 years, respectively. At the high-risk threshold of 10% in non-laboratory-based and 20% in laboratory-based models, the non-laboratory model has high sensitivity of 100% for males ≥ 60 years, females < 60 years, females ≥ 60 years, and 91.4% for males < 60 years.ConclusionA good agreement was observed between laboratory-based and non-laboratory-based versions of the WHO risk model. Also, at the risk threshold of 10% to detect high-risk individuals, the non-laboratory-based model has acceptable sensitivity for practical risk assessment and the screening programs in settings where resources are limited and people do not have access to laboratory tests.

Improving 10-year cardiovascular risk prediction in apparently healthy people: flexible addition of risk modifiers on top of SCORE2.

In clinical practice, factors associated with cardiovascular disease (CVD) like albuminuria, education level, or coronary artery calcium (CAC) are often known, but not incorporated in cardiovascular risk prediction models. The aims of the current study were to evaluate a methodology for the flexible addition of risk modifying characteristics on top of SCORE2 and to quantify the added value of several clinically relevant risk modifying characteristics. Individuals without previous CVD or DM were included from the UK Biobank; Atherosclerosis Risk in Communities (ARIC); Multi-Ethnic Study of Atherosclerosis (MESA); European Prospective Investigation into Cancer, The Netherlands (EPIC-NL); and Heinz Nixdorf Recall (HNR) studies (n = 409 757) in whom 16 166 CVD events and 19 149 non-cardiovascular deaths were observed over exactly 10.0 years of follow-up. The effect of each possible risk modifying characteristic was derived using competing risk-adjusted Fine and Gray models. The risk modifying characteristics were applied to individual predictions with a flexible method using the population prevalence and the subdistribution hazard ratio (SHR) of the relevant predictor. Risk modifying characteristics that increased discrimination most were CAC percentile with 0.0198 [95% confidence interval (CI) 0.0115; 0.0281] and hs-Troponin-T with 0.0100 (95% CI 0.0063; 0.0137). External validation was performed in the Clinical Practice Research Datalink (CPRD) cohort (UK, n = 518 015, 12 675 CVD events). Adjustment of SCORE2-predicted risks with both single and multiple risk modifiers did not negatively affect calibration and led to a modest increase in discrimination [0.740 (95% CI 0.736-0.745) vs. unimproved SCORE2 risk C-index 0.737 (95% CI 0.732-0.741)]. The current paper presents a method on how to integrate possible risk modifying characteristics that are not included in existing CVD risk models for the prediction of CVD event risk in apparently healthy people. This flexible methodology improves the accuracy of predicted risks and increases applicability of prediction models for individuals with additional risk known modifiers.

Comparison of 5- and 10-year breast cancer risk predictions.

10529 Background: Breast cancer risk assessment is valuable in personalizing risk-reduction and screening approaches for the general population. Maximizing the benefits while reducing the harms associated with current breast cancer screening and risk-reduction recommendations should increase the efficacy of these paradigms. Because healthcare providers have limited time with their patients, we developed a risk prediction model, BRISK, which uses the most important clinical risk factors for breast cancer (in an easily collected format), polygenic risk and mammographic density. We previously validated BRISK 5-year risk predictions and found that it performed better than IBISv7 and BCRAT. Methods: The dataset comprised 241,633 women (mean age at baseline assessment = 56.6 years). We included all races/ethnicities available in the UK Biobank. There were 3726 women affected during the first five years of follow-up and 7482 women affected during the full 10 years of follow-up. We compared the performance of the 5-year and 10-year risk predictions by assessing the association with breast cancer using Cox regression to estimate the hazard ratio (HR) per SD of risk, discrimination using calculating Harrell’s C-index and calibration using logistic regression to estimate the slope and the intercept of the logit of the risk predictions. Nelson–Aalen estimates of the hazard function by quintiles of risk and standardized incidence ratios (SIRs) of the number of breast cancer cases expected were used to illustrate performance. Results: The predictive performance of BRISK was consistent for 5-year and 10-year risk predictions. Discrimination was similar, with a Harrell’s C-index of 0.645 (95% CI = 0.636, 0.654) for the 5-year risk and of 0.638 (95% CI = 0.632, 0.645) for the 10-year risk. We used Nelson_Aalen cumulative hazard estimates to compare breast cancer incidence by risk quintile for both 5 and 10-year risk. Both risk scores overestimated risk in the top two deciles with the 5-year expected risks of 2.8% and 5.0% corresponding to observed risks of 2.3% and 3.3%. The 10-year expected risks for the top two deciles were 5.6% and 9.8% while the corresponding observed risks were 4.6% and 6.4%. Finally, we used clinically actionable thresholds to show the standard incidence ratios (SIR) of 1.77 (95% CI = 1.69, 1.85) for women ≥3% 5-year risk and 2.05 (95% CI = 1.93, 2.18) ≥8% 10-year risk. Conclusions: The BRISK 5-year and 10-year risk prediction perform equally well. There is an overestimation of risk for the top decile in both models. This is important to consider in risk-management conversations between clinician and patient. While the observed risk of the top decile remained above the 3% 5-year risk threshold used in the United States, the top decile observed risk dropped below the 8% 10-year risk threshold used in the United Kingdom. Risk scores are not used in isolation, but it may be preferable to err on the side of overestimation rather than underestimation.

Differences in 10-year cardiovascular risk estimation using SCORE and SCORE2 risk prediction tools: a moderate risk country population analysis

Abstract Funding Acknowledgements Type of funding sources: None. Background Cardiovascular disease (CVD) risk assessment plays a central role in nowadays clinical practice given its burden in morbidity and mortality worldwide. In the 2021 European Society of Cardiology Guidelines on cardiovascular disease prevention a new CVD risk prediction algorithm was presented, the SCORE2. This tool overcomes some of old SCORE limitations by predicting 10-year fatal and non-fatal CVD risk in individuals without previous CVD or diabetes aged 40-65 years and by including contemporary data from epidemiological data of 13 European countries. Differences on 10-year cardiovascular risk category prediction between SCORE and SCORE2 is still scarce. Purpose We aimed to compare differences in 10-year cardiovascular disease risk prediction in a Portuguese population using SCORE and SCORE2. Methods We conducted a cross-sectional study in a Portuguese population sample. Individuals aged 40-65 years old without known Atherosclerotic Cardiovascular Disease, Diabetes, Chronic Kidney Disease or Familial Hypercholesterolemia were included. The 10-year CVD risk was calculated using SCORE and SCORE2. Based on CVD risk category, patients were stratified into 4 categories - low, moderate, high and very high risk – according to SCORE and in 3 categories - low to moderate, high and very high risk - according to SCORE2 model. Primary outcome was 10-year CVD risk prediction difference between above mentioned models. According to the data distribution, appropriate statistical tests were conducted. Results 117 individuals were included in the study cohort, 79 (67.5%) of which were women. In our study, the 10-year risk prediction of fatal and non-fatal cardiovascular events was significantly different between SCORE and SCORE2. When assessing 10-year risk of CVD through SCORE model, 97.1% (n= 100) of the individuals were classified into low and moderate risk categories. On the other hand, when evaluating CVD risk with SCORE2 only 66% (n=62) of the participants were classified into those risk categories, meaning that 30.9% (n=29) of patients were in high and very high-risk categories with SCORE2 (vs. 2.9% with SCORE). Correlation between SCORE and SCORE2 was verified (R=0.572; p&amp;lt; 0.0001). Regarding cardiovascular risk factors, active smoking was the only independent predictor for 10-year CVD risk in SCORE2 (RR: 3.28, 95% CI: 1.93-4.66, p=0.001). There were no independent predictors for CVD risk in SCORE. Conclusions Ten-year cardiovascular risk assessment may be underestimated by SCORE model. In the present study, through SCORE CVD risk classification most of patients were classified into lower risk categories than those obtained through updated SCORE2. SCORE2 is a more up-to-date and more calibrated CVD risk assessment tool than old SCORE, so SCORE2 may contribute to a better reclassification of patients and hereafter allow intensification of CVD protection measures.

Is there a role for conicity index in 10-year risk cardiovascular disease prediction?

Abstract Funding Acknowledgements Type of funding sources: None. Background Cardiovascular disease (CVD) is the leading cause of death globally and one of the most frequent causes of disability. Therefore, the complete and correct evaluation of cardiovascular risk factors is essential to timely prevent CVD. Once central obesity is related with the amount of visceral adipose tissue - an independent predictor of CVD risk - its assessment may play an important role in 10-year CVD prediction. The conicity index (C-Index) assesses central adiposity but there is limited data regarding its role in 10-year CVD risk prediction. Purpose To evaluate the role of C-Index in predicting 10-year risk of fatal and non-fatal CVD according to SCORE2 and SCORE2-OP models in a moderate-risk country population. Methods We conducted a cross-sectional study in a Portuguese population sample. Individuals aged 40-90 years without known established Atherosclerotic Cardiovascular Disease, Diabetes mellitus, Chronic Kidney Disease or Familial Hypercholesterolemia were included in the study through a local cardiovascular (CV) screening event in Portugal, during May 2022. The C-Index was calculated for each individual according to the Valdez’s formula. For CVD risk assessment, we used the cut-offs values proposed by Pitanga et al. 1.18 for women and 1.25 for men. The population was divided into two groups according to the C-Index (risk high versus low). The 10-year risk of fatal and non-fatal CVD was calculated using SCORE2 and SCORE2-OP models. Results A total of 431 individuals were enrolled in this study. Median age was 71 years (Q1-Q3: 65-75) and 66.8% of were women. A total of 48.1% had hypertension, 38.3% dyslipidemia, 25.5% obesity and 6.6% were active smokers. Overall, 387 (90.2%) individuals were found to have elevated C-Index (SCORE2/2-OP mean± DP: 14.99± 7.56 versus 09.81± 3.37), of which 259 were female. In the univariate analysis, there was a correlation between C-Index and estimation of the Risk Cardiovascular Disease according to SCORE2/2-OP but was not statistically significant (Exp (beta) 95 % (CI):2.52 (-2.06-7.10, p=0.28)). However, in the multivariable linear regression analyzes the variable waist circumference (Exp (beta) 95 % (CI)18.47 (4.69 – 32.24, p&amp;lt;0.001)) was independent predictive factor related to the Risk of Cardiovascular Disease according to SCORE2 and SCORE2-OP. Conclusions In the present study most of the patients were categorized into high-risk category. Conicity index was not significantly correlated with estimation of the Risk of Cardiovascular Disease according to SCORE2 and SCORE2-OP models, but the variable waist circumference (Exp (beta) 95 % (CI)18.47 (4.69 – 32.24, p&amp;lt;0.001)) was an independent predictive factor related. More studies are needed to assess the role of C-Index in CVD risk prediction.

Prediction of the 10-year risk of atherosclerotic cardiovascular disease in the Korean population.

Proper risk assessment is important for the primary prevention of atherosclerotic cardiovascular disease (ASCVD). However, no validated risk prediction tools are currently in use in Korea. This study sought to develop a 10-year risk prediction model for incident ASCVD. Using the National Sample Cohort of Korea, 325,934 subjects aged 20-80 years without previous ASCVD were enrolled. ASCVD was defined as a composite of cardiovascular death, myocardial infarction, and stroke. The Korean atherosclerotic cardiovas cular disease risk prediction (K-CVD) model was developed separately for men and women using the development dataset and validated in the validation dataset. Furthermore, the model performance was compared with the Framingham risk score (FRS) and pooled cohort equation (PCE). Over 10 years of follow-up, 4,367 ASCVD events occurred in the overall population. The predictors of ASCVD included in the model were age, smoking status, diabetes, systolic blood pressure, lipid profiles, urine protein, and lipid-lowering and blood pressure-lowering treatment. The K-CVD model had good discrimination and strong calibration in the validation dataset (time-dependent area under the curve=0.846; 95% confidence interval, 0.828 to 0.864; calibration χ2=4.73, goodness-of-fit p=0.32). Compared with our model, both FRS and PCE showed worse calibration, overestimating ASCVD risk in the Korean population. Through a nationwide cohort, we developed a model for 10-year ASCVD risk prediction in a contemporary Korean population. The K-CVD model showed excellent discrimination and calibration in Koreans. This population-based risk prediction tool would help to appropriately identify high-risk individuals and provide preventive interventions in the Korean population.

Combining serum metabolomic profiles with traditional risk factors improves 10-year cardiovascular risk prediction in people with type 2 diabetes.

To identify a group of metabolites associated with incident cardiovascular disease (CVD) in people with type 2 diabetes and assess its predictive performance over-and-above a current CVD risk score (QRISK3). A panel of 228 serum metabolites was measured at baseline in 1066 individuals with type 2 diabetes (Edinburgh Type 2 Diabetes Study) who were then followed up for CVD over the subsequent 10 years. We applied 100 repeats of Cox least absolute shrinkage and selection operator to select metabolites with frequency >90% as components for a metabolites-based risk score (MRS). The predictive performance of the MRS was assessed in relation to a reference model that was based on QRISK3 plus prevalent CVD and statin use at baseline. Of 1021 available individuals, 255 (25.0%) developed CVD (median follow-up: 10.6 years). Twelve metabolites relating to fluid balance, ketone bodies, amino acids, fatty acids, glycolysis, and lipoproteins were selected to construct the MRS that showed positive association with 10-year cardiovascular risk following adjustment for traditional risk factors [hazard ratio (HR) 2.67; 95% confidence interval (CI) 1.96, 3.64]. The c-statistic was 0.709 (95%CI 0.679, 0.739) for the reference model alone, increasing slightly to 0.728 (95%CI 0.700, 0.757) following addition of the MRS. Compared with the reference model, the net reclassification index and integrated discrimination index for the reference model plus the MRS were 0.362 (95%CI 0.179, 0.506) and 0.041 (95%CI 0.020, 0.071), respectively. Metabolomics data might improve predictive performance of current CVD risk scores based on traditional risk factors in people with type 2 diabetes. External validation is warranted to assess the generalizability of improved CVD risk prediction using the MRS.

Estimating the Cardiovascular Disease Risk Reduction of a Quality Improvement Initiative in Primary Care: Findings from EvidenceNOW.

This study estimates reductions in 10-year atherosclerotic cardiovascular disease (ASCVD) risk associated with EvidenceNOW, a multi-state initiative that sought to improve cardiovascular preventive care in the form of (A)spirin prescribing for high-risk patients, (B)lood pressure control for people with hypertension, (C)holesterol management, and (S)moking screening and cessation counseling (ABCS) among small primary care practices by providing supportive interventions such as practice facilitation. We conducted an analytic modeling study that combined (1) data from 1,278 EvidenceNOW practices collected 2015 to 2017; (2) patient-level information of individuals ages 40 to 79 years who participated in the 2015 to 2016 National Health and Nutrition Examination Survey (n = 1,295); and (3) 10-year ASCVD risk prediction equations. The primary outcome measure was 10-year ASCVD risk. EvidenceNOW practices cared for an estimated 4 million patients ages 40 to 79 who might benefit from ABCS interventions. The average 10-year ASCVD risk of these patients before intervention was 10.11%. Improvements in ABCS due to EvidenceNOW reduced their 10-year ASCVD risk to 10.03% (absolute risk reduction: -0.08, P ≤ .001). This risk reduction would prevent 3,169 ASCVD events over 10 years and avoid $150 million in 90-day direct medical costs. Small preventive care improvements and associated reductions in absolute ASCVD risk levels can lead to meaningful life-saving benefits at the population level.

Development and validation of DNA methylation scores in two European cohorts augment 10-year risk prediction of type 2 diabetes.

Type 2 diabetes mellitus (T2D) presents a major health and economic burden that could be alleviated with improved early prediction and intervention. While standard risk factors have shown good predictive performance, we show that the use of blood-based DNA methylation information leads to a significant improvement in the prediction of 10-year T2D incidence risk. Previous studies have been largely constrained by linear assumptions, the use of cytosine-guanine pairs one-at-a-time and binary outcomes. We present a flexible approach (via an R package, MethylPipeR) based on a range of linear and tree-ensemble models that incorporate time-to-event data for prediction. Using the Generation Scotland cohort (training set ncases = 374, ncontrols = 9,461; test set ncases = 252, ncontrols = 4,526) our best-performing model (area under the receiver operating characteristic curve (AUC) = 0.872, area under the precision-recall curve (PRAUC) = 0.302) showed notable improvement in 10-year onset prediction beyond standard risk factors (AUC = 0.839, precision-recall AUC = 0.227). Replication was observed in the German-based KORA study (n = 1,451, ncases = 142, P = 1.6 × 10-5).

Predicting risk on cardiovascular or cerebrovascular disease based on a physical activity cohort: Results from APAC study.

Commonly used prediction models have been primarily constructed without taking physical activity into account. Using the Kailuan physical activity cohorts from Asymptomatic Polyvascular Abnormalities in Community (APAC) study, we developed a 9-year cardiovascular or cerebrovascular disease (CVD) risk prediction equation. Participants in this study were included from APAC cohort, which included 5440 participants from the Kailuan cohort in China. Cox proportional hazard regression model was applied to construct sex-specific risk prediction equations for the physical activity cohort (PA equation). Proposed equations were compared with the 10-year risk prediction model, which is developed for atherosclerotic cardiovascular disease risk in Chinese cohorts (China-PAR equation). C statistics of PA equations were 0.755 (95% confidence interval, 0.750-0.758) for men and 0.801 (95% confidence interval, 0.790-0.813) for women. The estimated area under the receiver operating characteristic curves in the validation set shows that the PA equations perform as good as the China-PAR. From calibration among four categories of predicted risks, the predicted risk rates by PA equations were almost identical to the Kaplan-Meier observed rates. Therefore, our developed sex-specific PA equations have effective performance for predicting CVD for physically active cohorts in the physical activity cohort in Kailuan.

Prediction of Suicidal Behaviors in the Middle-aged Population: Machine Learning Analyses of UK Biobank.

Suicidal behaviors, including suicide deaths and attempts, are major public health concerns. However, previous suicide models required a huge amount of input features, resulting in limited applicability in clinical practice. We aimed to construct applicable models (ie, with limited features) for short- and long-term suicidal behavior prediction. We further validated these models among individuals with different genetic risks of suicide. Based on the prospective cohort of UK Biobank, we included 223 (0.06%) eligible cases of suicide attempts or deaths, according to hospital inpatient or death register data within 1 year from baseline and randomly selected 4460 (1.18%) controls (1:20) without such records. We similarly identified 833 (0.22%) cases of suicidal behaviors 1 to 6 years from baseline and 16,660 (4.42%) corresponding controls. Based on 143 input features, mainly including sociodemographic, environmental, and psychosocial factors; medical history; and polygenic risk scores (PRS) for suicidality, we applied a bagged balanced light gradient-boosting machine (LightGBM) with stratified 10-fold cross-validation and grid-search to construct the full prediction models for suicide attempts or deaths within 1 year or between 1 and 6 years. The Shapley Additive Explanations (SHAP) approach was used to quantify the importance of input features, and the top 20 features with the highest SHAP values were selected to train the applicable models. The external validity of the established models was assessed among 50,310 individuals who participated in UK Biobank repeated assessments both overall and by the level of PRS for suicidality. Individuals with suicidal behaviors were on average 56 years old, with equal sex distribution. The application of these full models in the external validation data set demonstrated good model performance, with the area under the receiver operating characteristic (AUROC) curves of 0.919 and 0.892 within 1 year and between 1 and 6 years, respectively. Importantly, the applicable models with the top 20 most important features showed comparable external-validated performance (AUROC curves of 0.901 and 0.885) as the full models, based on which we found that individuals in the top quintile of predicted risk accounted for 91.7% (n=11) and 80.7% (n=25) of all suicidality cases within 1 year and during 1 to 6 years, respectively. We further obtained comparable prediction accuracy when applying these models to subpopulations with different genetic susceptibilities to suicidality. For example, for the 1-year risk prediction, the AUROC curves were 0.907 and 0.885 for the high (>2nd tertile of PRS) and low (<1st) genetic susceptibilities groups, respectively. We established applicable machine learning-based models for predicting both the short- and long-term risk of suicidality with high accuracy across populations of varying genetic risk for suicide, highlighting a cost-effective method of identifying individuals with a high risk of suicidality.

Comparison of State-of-the-Art Neural Network Survival Models with the Pooled Cohort Equations for Cardiovascular Disease Risk Prediction

BackgroundThe Pooled Cohort Equations (PCEs) are race- and sex-specific Cox proportional hazards (PH)-based models used for 10-year atherosclerotic cardiovascular disease (ASCVD) risk prediction with acceptable discrimination. In recent years, neural network models have gained increasing popularity with their success in image recognition and text classification. Various survival neural network models have been proposed by combining survival analysis and neural network architecture to take advantage of the strengths from both. However, the performance of these survival neural network models compared to each other and to PCEs in ASCVD prediction is unknown.MethodsIn this study, we used 6 cohorts from the Lifetime Risk Pooling Project (with 5 cohorts as training/internal validation and one cohort as external validation) and compared the performance of the PCEs in 10-year ASCVD risk prediction with an all two-way interactions Cox PH model (Cox PH-TWI) and three state-of-the-art neural network survival models including Nnet-survival, Deepsurv, and Cox-nnet. For all the models, we used the same 7 covariates as used in the PCEs. We fitted each of the aforementioned models in white females, white males, black females, and black males, respectively. We evaluated models’ internal and external discrimination power and calibration.ResultsThe training/internal validation sample comprised 23216 individuals. The average age at baseline was 57.8 years old (SD = 9.6); 16% developed ASCVD during average follow-up of 10.50 (SD = 3.02) years. Based on 10 × 10 cross-validation, the method that had the highest C-statistics was Deepsurv (0.7371) for white males, Deepsurv and Cox PH-TWI (0.7972) for white females, PCE (0.6981) for black males, and Deepsurv (0.7886) for black females. In the external validation dataset, Deepsurv (0.7032), Cox-nnet (0.7282), PCE (0.6811), and Deepsurv (0.7316) had the highest C-statistics for white male, white female, black male, and black female population, respectively. Calibration plots showed that in 10 × 10 validation, all models had good calibration in all race and sex groups. In external validation, all models overestimated the risk for 10-year ASCVD.ConclusionsWe demonstrated the use of the state-of-the-art neural network survival models in ASCVD risk prediction. Neural network survival models had similar if not superior discrimination and calibration compared to PCEs.

POL-RISK: an algorithm for 10-year fracture risk prediction in postmenopausal women from the RAC-OST-POL study.

Fragility fractures are the major consequence of osteoporosis. Thus, fracture risk assessment is an essential part of the diagnostic process in osteoporosis. The aim of the study was to develop an algorithm for fracture risk prediction. Bone status was evaluated in a population‑basedcohort of postmenopausal women at a mean (SD) age of 66.4 (7.8) years. Subsequently, all participants were contacted by phone once a year (for 10 consecutive years) to update their history of fractures. At the end of the 10‑year follow‑up, the number of the study participants was 640, of whom 129 had a history of 190 osteoporotic fractures recorded during the study period. Statistical analysis included multistep data preprocessing, feature selection, identification of fracture risk factors, and design of the final model. Logistic regression models were fitted and used for the evaluation of variables from determined feature sets, including global fit measures, as well as individual parameters, such as the Wald statistic and P value, odds ratio, and 95% CI. The 10‑year risk for any fracture depended on the age of the patient, the number of recorded fractures after the age of 40 years, femoral neck bone mass, and the occurrence of falls during the previous year. The achieved equation was incorporated into an algorithm, available at the www.fracture‑risk.pl webpage. A fracture prediction algorithm was developed in a longitudinal study to enable calculation of the 10‑year fracture risk. Identification of patients at a high risk of fracture should be followed by implementation of appropriate treatment strategy to reduce the number of future fractures.

Angiographic Findings in Unstable Angina and Prediction of 1-Year Risk of Recurrent Acute Coronary Syndrome or Death: A Nationwide Machine Learning Study

Background: We studied whether machine learning could predict survival, risk of future acute coronary syndrome (ACS) and coronary angiographic (CAG) findings in patients with unstable angina (UA). We also studied whether high-sensitivity troponin levels within normal range can predict the presence of obstructive coronary lesions, risk of future ACS and death. Methods: We used the SWEDEHEART registry to include patients admitted to the coronary care unit due to chest pain, with normal highsensitivity cardiac troponin T or I (hs-cTnI, hs-cTnT), who underwent CAG and did not receive a final diagnosis of acute myocardial infarction (AMI). We studied CAG findings on segmental level, developed machine learning models predicting the risk of ACS or death within 1-year, and angiographic findings. The latter model predicted CAG resulting in interventions (any) or eliciting further assessments after CAG. Models for ACS and death included 130 candidate predictors and models for angiographic findings included 110 predictors. Results: We included 9’314 patients; 1-year rate of death was 0.9% (n=78) and ACS was 2.7% (n=251). A total of 5455 (61.5%) of CAG resulted in no intervention and no further assessment afterwards, with 40% without significant stenosis. There was a strong association between hscTnI (within normal range) and severity of coronary atherosclerosis; e.g 32.4% in patients with hs-cTnI 26-35 ng/L had >50% stenosis in segment 6, as compared with 12.6% in those with hs-cTnI 0-5 ng/L. All segments displayed similar associations with troponin levels. Among 17 predictors for atherosclerosis, hs-cTnI was the strongest predictor of having >50% stenosis in the left anterior descending artery (LAD). Mortality increased at hs-cTnI levels above 10 ng/L for men, but not women. Age and sex adjusted hazard ratio for hs-cTnI 25-35 vs hs-TnI 0-5 was 5.73 (2.14- 15.35) for 1-year mortality. No association was noted for hs-cTnT. The strongest predictors of 1-year mortality were C-reactive protein, body mass index, estimated glomerular filtration rate. Conclusion: Troponin levels within normal range exhibit a strong association with obstructive CAD and survival.

P333: Polygenic risk scores improve 10-year risk prediction of coronary artery disease in individuals at borderline and intermediate clinical risk*

P333: Polygenic risk scores improve 10-year risk prediction of coronary artery disease in individuals at borderline and intermediate clinical risk*

AW-AD-2: ASSOCIATION BETWEEN AMBULATORY BLOOD PRESSURE AND RISK OF HOME HYPERTENSION IN A NORMOTENSIVE POPULATION

Objective: Most of the hypertension management guidelines recommend out-of-office blood pressure (BP) measurement such as self-measured BP at home (home BP) and ambulatory BP monitoring for the accurate diagnosis and management of hypertension. Both out-of-office BP measurements are more useful than office BP as prognostic indicators of cardiovascular risk. However, no prospective study has assessed the longitudinal association between the two types of out-of-office BP measurements. Therefore, we investigated the association between ambulatory BP and the risk of home BP-based hypertension in a normotensive population and whether considering ambulatory BP improves the 10-year prediction model for home hypertension risk, which was developed in the previous study. Desigh and method: We followed up 410 participants (83.2% women; age, 53.6 years) without home and ambulatory hypertension in the general population of Ohasama, Japan. The Cox model was used to assess the hazard ratios for home hypertension (home BP ≧135/≧85 mmHg or initiation of antihypertensive treatment). In addition, the predictive ability for home hypertension incidence between the models adjusted for the covariates with 24-h, daytime, or nighttime SBP was examined using Harrell's C-statistics, continuous net reclassification improvement (NRI), and integrated discrimination improvement (IDI). The NRI considers changes in the estimated prediction probabilities that imply a change from one category to another between different models. IDI is defined as the difference between the discrimination slopes of a basic model and an extended model. Results: During a mean 14.2-year follow-up, 225 home hypertension incidences occurred. The hazard ratio (95% confidence interval) for home hypertension incidence per 1-standard deviation increase in 24-h systolic BP (SBP) was 1.59 (1.33–1.90) after adjustments for possible confounding factors, including baseline home SBP. Harrell's C-statistic was 0.72 (0.69–0.75) for the basic 10-year risk prediction model, which includes sex, age, BMI, current or ex-smoker status, office SBP, and baseline home SBP. By adding 24-h, daytime, or nighttime SBP to the basic model, continuous NRI and IDI significantly improved although no significant improvement in Harrell's C-statistics was observed. Conclusions: Independent of the baseline home SBP, 24-h SBP was associated with the risk of home hypertension in a normotensive population. The prediction model for home hypertension risk was improved by the addition of 24-h SBP to the basic model. It could be useful to measure ambulatory BP for the early detection of home hypertension.

Development and validation of a novel 10-year cardiovascular risk prediction nomogram for the United Arab Emirates national population

ObjectivesAtherosclerotic cardiovascular disease (ASCVD) is the leading cause of morbidity and mortality among United Arab Emirates (UAE) nationals. Recent studies have shown that current tools are poor in predicting the...

Development and Validation of a Lifetime Risk Model for Kidney Failure and Treatment Benefit in Type 2 Diabetes: 10-Year and Lifetime Risk Prediction Models.

Individuals with type 2 diabetes are at a higher risk of developing kidney failure. The objective of this study was to develop and validate a decision support tool for estimating 10-year and lifetime risks of kidney failure in individuals with type 2 diabetes as well as estimating individual treatment effects of preventive medication. The prediction algorithm was developed in 707,077 individuals with prevalent and incident type 2 diabetes from the Swedish National Diabetes Register for 2002-2019. Two Cox proportional regression functions for kidney failure (first occurrence of kidney transplantation, long-term dialysis, or persistent eGFR <15 ml/min per 1.73 m2) and all-cause mortality as respective end points were developed using routinely available predictors. These functions were combined into life tables to calculate the predicted survival without kidney failure while using all-cause mortality as the competing outcome. The model was externally validated in 256,265 individuals with incident type 2 diabetes from the Scottish Care Information Diabetes database between 2004 and 2019. During a median follow-up of 6.8 years (interquartile range, 3.2-10.6), 8004 (1%) individuals with type 2 diabetes in the Swedish National Diabetes Register cohort developed kidney failure, and 202,078 (29%) died. The model performed well, with c statistics for kidney failure of 0.89 (95% confidence interval, 0.88 to 0.90) for internal validation and 0.74 (95% confidence interval, 0.73 to 0.76) for external validation. Calibration plots showed good agreement in observed versus predicted 10-year risk of kidney failure for both internal and external validation. This study derived and externally validated a prediction tool for estimating 10-year and lifetime risks of kidney failure as well as life years free of kidney failure gained with preventive treatment in individuals with type 2 diabetes using easily available clinical predictors. This article contains a podcast at https://dts.podtrac.com/redirect.mp3/www.asn-online.org/media/podcast/CJASN/2022_11_04_CJN05020422.mp3.