Traditional Risk Models Research Articles

Abstract P2-11-21: Integration of an ancestrally unbiased polygenic risk score with the Tyrer-Cuzick breast cancer risk model

Abstract Background: Polygenic risk scores (PRS) have been shown to provide genomically informed breast cancer risk assessment in both carriers and non-carriers of predisposing genetic mutations. Risk stratification is further improved by combining a PRS with risk models incorporating traditional risk factors to generate a Combined Risk Score (CRS). We recently developed and validated a breast cancer PRS for women of diverse ancestries using ancestry-informative genetic markers. Here, we combine the diverse ancestry PRS with a clinical and family history-based model to develop an integrated genomically-informed and ancestrally unbiased risk assessment tool. Methods: The study sample included women in the U.S. without a personal history of breast cancer, referred for clinical genetic testing between June 2020 and March 2021, and who tested negative for pathogenic or likely pathogenic variants in breast cancer susceptibility genes. A CRS, incorporating a validated PRS and the TC model, was generated using a previously described Fixed-Stratified method that accounts for association between PRS and clinical risk factors. Association between the PRS and each clinical risk factor included in the Tyrer-Cuzick (TC) breast cancer risk model (version 7) was tested using linear regression with PRS as the dependent variable and the TC factor as an independent variable with adjustment for age and ancestry. We examined the rate of reclassification resulting from incorporation of PRS into the CRS by classifying women as having high (>20%) remaining lifetime risk (RLR) versus low (≤20%) RLR according to both TC and CRS. Results: Among 68,803 women, 21,500 (31.2%) had one or more first degree relatives (FDR) with breast cancer. Approximately 10% of women reported only African ancestry and a similar percentage reported only Hispanic/Latina ethnicity (Table 1). Family history was significantly associated with PRS (p=1.0x10-76). After adjusting for multiple testing, no other factors showed significant association with PRS. Improved risk stratification of CRS over TC follows from two results: (1) We previously showed that PRS improved risk stratification above and beyond family history; (2) In the present study, PRS was not associated with any TC factor other than family history. Adding the PRS to the TC model significantly altered breast cancer risk estimates for women of all ancestries, with 17.3% of patients stratified differently by CRS versus TC alone. Differences in risk stratification (using the 20% threshold) for each self-reported ancestry are presented in Table 1. The CRS classified fewer patients (32.0%) as high RLR than the TC model alone (35.4%), with similar results for 5-year risk estimates. Conclusions: This is the first genomically-informed, integrated polygenic and traditional breast cancer risk model for US women referred for contemporary clinical genetic testing. This model advances the PRS component of a previously validated combined model. It effectively estimates 5-year and lifetime risk for breast cancer using a PRS with an objectively genetically determined ancestral composition, calibrated and validated for risk stratification in all ancestries. The model may reliably and responsibly inform risk reduction strategies such as enhanced surveillance and use of preventive medications. Table 1.Self-Reported Ancestry/EthnicityNumber (%) of patientsHigh TC High CRS High TC and Low CRSLow TC and High CRSAll68,803 (100%)24,332 (35.4%)22,041 (32.0%)7,080 . (10.3%)4,789 . (7.0%)Asian1,450. (2.1%)487 (33.6%)475 (32.8%)94 . (6.5%)82 . (5.7%)African7,909 (11.5%)2,540 (32.1%)2,473 (31.3%)435 . (5.5%)368 . (4.7%)Hispanic6,481 . (9.4%)1,614 (24.9%)1,345 (20.9%)606 . (9.4%)346 . (5.3%)Non-European*19,225 (27.9%)5,701 (29.7%)5,297 (27.6%)1,429 . (7.4%)1,025 . (5.3%)European**46,640 (67.8%)17,507 (37.5%)15,733 (33.7%)5,328 . (11.4%)3,554 . (7.6%)*Includes any combination of Black/African, Middle Eastern, Pacific Islander, Asian, Hispanic/Latino and/or Native American ancestry. **Includes White/Non-Hispanic and/or Ashkenazi Jewish. Citation Format: Elisha Hughes, Ryan Bernhisel, Holly Pederson, Braden Probst, Timothy Simmons, Susanne Wagner, Thaddeus Judkins, Eric Rosenthal, Benjamin Roa, Susan M. Domchek, Charis Eng, Judy Garber, Monique Gary, Ora K. Gordon, Jennifer Klemp, Semanti Mukherjee, Kenneth Offit, Funmi Olopade, Joseph Vijai, Jeffrey N. Weitzel, Pat Whitworth, Lamis Yehia, Allison Kurian, Mark Robson, Thomas P. Slavin, Alexander Gutin, Jerry S. Lanchbury. Integration of an ancestrally unbiased polygenic risk score with the Tyrer-Cuzick breast cancer risk model [abstract]. In: Proceedings of the 2021 San Antonio Breast Cancer Symposium; 2021 Dec 7-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2022;82(4 Suppl):Abstract nr P2-11-21.

Prognostic implication of serum glycated albumin for patients with non-ST-segment elevation acute coronary syndrome undergoing percutaneous coronary intervention

BackgroundIt has been demonstrated that glycated albumin (GA) is significantly associated with diabetes complications and mortality. However, among patients diagnosed with non-ST-elevation acute coronary syndrome (NSTE-ACS) administered percutaneous coronary intervention (PCI), the predictive value of GA for poor prognosis is unclear.MethodsThis study eventually included 2247 NSTE-ACS patients in Beijing Anzhen Hospital, Capital Medical University in January-December 2015 who received PCI. All patients were followed up until death or for 48 months post-discharge. The primary endpoint was major adverse cardio-cerebral events (MACCEs), including all-cause death, non-fatal myocardial infarction, ischemia-induced revascularization and non-fatal ischemic stroke.ResultsIn total, 547 (24.3%) MACCEs were recorded during the follow-up period. Upon adjusting for potential confounders, GA remained an important risk predictor of MACCEs (As nominal variate: hazard ratio [HR] 1.527, 95% confidence interval [CI] 1.236–1.886, P < 0.001; As continuous variate: HR 1.053, 95% CI 1.027–1.079, P < 0.001). GA addition significantly enhanced the predictive ability of the traditional risk model (Harrell’s C-index, GA vs. Baseline model, 0.694 vs. 0.684, comparison P = 0.002; continuous net reclassification improvement (continuous-NRI) 0.085, P = 0.053; integrated discrimination improvement (IDI) 0.007, P = 0.020).ConclusionGA is highly correlated with poor prognosis in NSTE-ACS patients undergoing PCI, suggesting that it may be a major predictive factor of adverse events among these individuals.

Machine learning prediction model of acute kidney injury after percutaneous coronary intervention

Acute kidney injury (AKI) after percutaneous coronary intervention (PCI) is associated with a significant risk of morbidity and mortality. The traditional risk model provided by the National Cardiovascular Data Registry (NCDR) is useful for predicting the preprocedural risk of AKI, although the scoring system requires a number of clinical contents. We sought to examine whether machine learning (ML) techniques could predict AKI with fewer NCDR-AKI risk model variables within a comparable PCI database in Japan. We evaluated 19,222 consecutive patients undergoing PCI between 2008 and 2019 in a Japanese multicenter registry. AKI was defined as an absolute or a relative increase in serum creatinine of 0.3 mg/dL or 50%. The data were split into training (N = 16,644; 2008–2017) and testing datasets (N = 2578; 2017–2019). The area under the curve (AUC) was calculated using the light gradient boosting model (GBM) with selected variables by Lasso and SHapley Additive exPlanations (SHAP) methods among 12 traditional variables, excluding the use of an intra-aortic balloon pump, since its use was considered operator-dependent. The incidence of AKI was 9.4% in the cohort. Lasso and SHAP methods demonstrated that seven variables (age, eGFR, preprocedural hemoglobin, ST-elevation myocardial infarction, non-ST-elevation myocardial infarction/unstable angina, heart failure symptoms, and cardiogenic shock) were pertinent. AUC calculated by the light GBM with seven variables had a performance similar to that of the conventional logistic regression prediction model that included 12 variables (light GBM, AUC [training/testing datasets]: 0.779/0.772; logistic regression, AUC [training/testing datasets]: 0.797/0.755). The AKI risk model after PCI using ML enabled adequate risk quantification with fewer variables. ML techniques may aid in enhancing the international use of validated risk models.

Application of multi-stage competing risk model to survival data

The traditional proportional hazard model is commonly used to investigate the association between main outcome and predictor variables. However, the endpoints in medical studies are often not unique. The analyses of labeling other competing outcomes other than the main outcome as censored data will theoretically lead to a biased estimate of the risk of main outcome. Although the traditional competitive risk model can adjust the influence of other outcomes on the risk of the main outcome, it can not directly compare the differences on the risks of different outcomes. The multi-state competing risk model provides a relatively suitable solution for this problem. In this study, based on a previously published follow-up data set for prostate cancer patients, we developed traditional proportional hazard model, traditional competitive risk model, and multi-state competing risk model, respectively. By comparing the advantages and disadvantages of the three models with the same survival data, we clarified the clinical application value of the multi-state competitive risk model in survival data with multiple outcomes.

Linking food production and environmental outcomes: An application of a modified relative risk model to prioritize land-management practices

CONTEXTSustainable land-use systems require clear decision-making protocols across institutions and departments. However, food production and ecological conservation are often regarded as separate or even incompatible policy goals, resulting in sustainability challenges and unclear land-management practices. More recently, with the introduction of Sustainable Development Goals (SDGs), there has been growing interests in minimizing the possible trade-offs between the two policy goals through appropriate governance. OBJECTIVEThe objective of this study was to develop a systematic framework for assessing current land-management practices and specifying future priorities. METHODSThe systematic framework included three modules and considered agroecosystems covering each administrative unit as the research objects. In the risk assessment module, we modified traditional relative risk model (RRM) to assess the risk that current management practices may affect food production and non-provisioning ecosystem services in existing agroecosystems, respectively. The framework also included a module to identify potentially available cropland (PAC) based on a combination of agricultural suitability and environmental costs for future reclamation. Finally, priorities for future land-management practices were specified based on the trade-offs between cropland expansion, ecological restoration and productive restoration. Jiangsu Province in eastern China was taken as a case study to verify the applicability of our framework. RESULTS AND CONCLUSIONSUsing Jiangsu Province as a case study, we have verified the applicability of our framework in prioritizing future land-management practices at a county level. The framework can offer a prioritization mechanism for formulating targeted strategies under alternative scenarios to correspond with varying management goals. The RRM can separate adverse effects from multiple stressors on food production and environmental outcomes rather than confounding the two interactive functions, while the risk scores and ranks generated for each sub-region, stressor and endpoint can pave the way for detailed and targeted governances. In addition, the identification of PAC can highlight areas for future cropland expansion. SIGNIFICANCEThe proposed framework can provide important insights and tools for avoiding institutional conflicts regarding land-management governance. Due to the simplicity and comparability, it can not only be applicable to our particular site, but also be used for different locations or downscaled to a landscape level.

Abstract 12965: Race-Specific Models to Predict In-Hospital Mortality in Patients With Heart Failure Using Machine Learning: The American Heart Association Get With the Guidelines Registry

Background: Prior prognostic models for acute decompensated HF (ADHF) have incorporated race as a covariate may not have completely account for the social and biological factors that underlie the racial disparities in outcomes. We developed race-specific models to improve risk prediction and better identify race-specific predictors of mortality in patients with ADHF using machine learning (ML) methods. Methods: Race-specific ML-models for in-hospital mortality were developed and validated in ADHF patients enrolled in the GWTG-HF registry between 2007 to 2020 (14,586 Black and 50,351 Non-Black adults) using over 40 candidate variables. External validation was performed among participants from the ARIC study with ADHF (n=1,115 Black and 2,028 Non-Black adults). The discrimination and calibration was compared with a previously validated risk score and cohort-specific logistic regression (LR) models. The changes in model reclassification performance with addition of zip-code level data of socioeconomic (SES) parameters was also assessed. Results: In the GWTG-HF cohort, the ML models had superior performance in both Black and Non-Black participants compared with the traditional GWTG-HF risk score ( Table ). The superior performance of the ML-model was also observed in the external validation cohort as compared with the other LR models ( Table ). Addition of measures of SES did not meaningfully change the performance but improved the reclassification metrics of the ML-models with net up-classification of risk in Black [NRI = 0.21 (0.07, 0.35)] and down-classification in Non-Black patients [NRI = -0.08 (-0.12, -0.04)]. SES was responsible for 21.7% of the total attributable mortality risk in Black adults compared to 0.4% in Non-Black adults. Conclusion: Race-specific, ML-based mortality models demonstrated superior performance when compared to traditional HF mortality risk models. Socioeconomic distress is an important contributor to risk in Black adults.

Predicting risk of progression in relapsed multiple myeloma using traditional risk models, focal lesion assessment with PET-CT and minimal residual disease status.

Predicting risk of progression in relapsed multiple myeloma using traditional risk models, focal lesion assessment with PET-CT and minimal residual disease status.

A metabolomic index based on lipoprotein subfractions and branched chain amino acids is associated with incident hypertension

ObjectiveThe present study aims to evaluate the performance of the Diabetes Risk Index (DRI), a metabolomic index based on lipoprotein particles and branched chain amino acids, on the incidence of newly developed hypertension in a large community dwelling cohort. MethodsThe DRI was calculated by combining 6 lipoprotein parameters [sizes of very-low-density lipoprotein (VLDL), low-density lipoprotein (LDL) and high-density lipoprotein (HDL), concentrations of large VLDL, small LDL, and large HDL particles], and the concentrations of valine and leucine. DRI scores were estimated in 4169 participants from the PREVEND prospective cohort. Cox proportional hazards regression was used to evaluate the association of DRI scores with incident hypertension. ResultsDuring a median follow-up of 8.6 years, 924 new hypertension cases were ascertained. In analyses adjusted for age and sex, there was a significant association between DRI and incident hypertension with a hazard ratio (HR) per 1 SD increase of 1.45 (95% CI 1.36,1.54; p < 0.001). After additional adjustment for traditional risk factors, the HR remained significant (HRadj 1.21, 95% CI 1.10, 1.33, p <0.001). Likewise, subjects in the top quartile of DRI presented with a higher risk of hypertension (HRadj 1.64, 95% CI 1.28, 2.10, p <0.001). Furthermore, the net reclassification improvement assessment improved after the addition of DRI to a traditional risk model (p <0.001), allowing proper reclassification of 34% of the participants. ConclusionHigher DRI scores were associated with an increased risk of incident hypertension. Such association was independent of traditional clinical risk factors for hypertension.

Validation of Dynamic Biomarker-Based Risk Progression Model for Smoldering Multiple Myeloma

Validation of Dynamic Biomarker-Based Risk Progression Model for Smoldering Multiple Myeloma

Research on early warning of agricultural credit and guarantee risk based on deep learning

Under the impact of agricultural industry differentiation, traditional financial risk model cannot forewarn the guarantee risk of agricultural credit with effectively. This paper proposes an early warning algorithm of agricultural credit and guarantee risk that can effectively overcome the interference of external factors. Using deep learning network, the risk algorithm of agricultural credit and guarantee was built and it could change the deep belief network into supervised learning. To train for an optimal model, two new hidden layers are added to extract image feature vectors, as well as a Softmax classifier. The model is trained and evaluated by the usage of the risk data set of L province from 2017 to 2019, reinforcing the pre-training network and data to deal with the issue of overfitting in training. The results show that the accuracy of the model reaches 92.56%, when the training sample proportion is 90%, with all the 13 factors in the test set taken as input. It shows that the training of the model worked well and that it can effectively predict the risk of agricultural credit and guarantee.

The Use of Subclinical Atherosclerosis Imaging to Guide Preventive Cardiology Management.

Clinical atherosclerotic cardiovascular disease (ASCVD) requires years to manifest, providing a window of opportunity for preventive cardiovascular management. Subclinical atherosclerosis imaging leverages this long latency period to estimate and improve future ASCVD risk. Coronary artery calcium (CAC) scoring has the most robust data in the detection of subclinical atherosclerosis. CAC scan significantly enhances cardiovascular risk stratification in addition to traditional risk models. Coronary computed tomography angiography data show similar strengths in subclinical atherosclerosis detection in addition to plaque morphology characterization with inherent limitations. Carotid intima-media thickness and ankle-brachial index are other modalities whose predictive value becomes incremental when added to the aforementioned modalities. When added to traditional risk models, subclinical atherosclerosis imaging modalities personalize future ASCVD risk stratification and assist in the initiation and rate of intensification of preventive therapies. Emerging imaging techniques exist but further research is required for primetime clinical use.

Abstract IA-21: AI in an imaging center: Challenges and opportunities

Abstract There is significant and well-founded enthusiasm for the promise of AI to improve healthcare outcomes across diverse clinical applications in oncology. However, careful analyses of the impact of AI tools in actual “real world” clinical settings are sparse in this relatively early phase of scientific discovery. The rigor applied to deep learning model development (training, testing, external validation) must be matched with equal rigor in assessing patient outcomes after implementation of deep learning tools in actual clinical settings. We can learn from our past endeavors in computer-assisted image interpretation. Tools for computer aided detection were approved by the FDA in 1998 to help radiologists find breast cancers on mammograms. However, after widespread clinical implementation (based on promising results in reader studies and reimbursement approvals), clinical trials to assess the impact of CAD on diagnostic accuracy of mammography revealed the tools did not improve human performance in findings cancers, and in many cases increased false positive readings by humans using CAD assistance. Due to severely limited healthcare resources during pandemics, and to protect patients and healthcare workers, state governments and the COVID-19 Pandemic Breast Cancer Consortium urged providers to focus cancer screening efforts on those patients at higher risk. These mandates were and continue to be necessary responses to support fair allocation of scarce resources to maximize benefits for all patients across the full spectrum of healthcare needs and acknowledge the shift in risk-benefit ratios for screening in average risk women. After screening mammography programs were closed due to the COVID-19 pandemic, we implemented an AI image-only risk model strategy at MGH to invite women at higher risk to return to screening for breast cancer. The AI model identified the top 60% of patients at risk, who were diagnosed with 90% of cancers, while the top 60% of at-risk patients based on traditional models harbored less than 50% of cancers. Our AI image-only risk model performed best in identifying women with current breast cancer, followed by strategies which used five-year traditional risk models and included all women with a personal history of breast cancer in invitations to screen. Lifetime risk models performed extremely poorly. Operating at the traditional “high risk” threshold of &amp;gt;20% lifetime risk, these models identified fewer than 10% of patients with cancer in the total population screened. AI-based risk models may support accurate and equitable risk-based screening. It may be possible with early intervention to reduce the widening disparities and later stage breast cancer diagnoses we are experiencing with COVID and mitigate the risk of delayed diagnoses and increased mortality from breast cancer during and after pandemics. Strategies for careful and rigorous assessment of clinical implementation are essential for AI to have the impact in health desired by all. Citation Format: Constance Lehman. AI in an imaging center: Challenges and opportunities [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr IA-21.

Validation of the Surgical Outcome Risk Tool (SORT) for Predicting Postoperative Mortality in Colorectal Cancer Patients Undergoing Surgery and Subgroup Analysis.

The accurate evaluation of perioperative risk is crucial to facilitate the shared decision-making process. Surgical outcome risk tool (SORT) has been developed to provide enhanced and more feasible identification of high-risk surgical patients. Nonetheless, SORT has not been validated for patients with colorectal cancer undergoing surgery. Our aim was to determine whether SORT can accurately predict mortality after surgery for colorectal cancer and to compare it with traditional risk models. 526 patients undergoing surgery performed by a colorectal surgical team in a single Greek tertiary hospital (2011-2019) were included. Five risk models were evaluated: (1) SORT, (2) Physiology and Operative Severity Score for the enumeration of Mortality and Morbidity (POSSUM), (3) Portsmouth POSSUM (P-POSSUM), (4) Colorectal POSSUM (CR-POSSUM), and (5) the Association of Great Britain and Ireland (ACPGBI) score. Model accuracy was assessed by observed to expected (O:E) ratios, and area under Receiver Operating Characteristic curve (AUC). Ten patients (1.9%) died within 30days of surgery. SORT was associated with an excellent level of discrimination [AUC:0.81 (95% CI:0.68-0.94); p = 0.001] and provided the best performing calibration of all models in the entire dataset analysis (H-L:2.82; p = 0.83). Nonetheless, SORT underestimated mortality. SORT model demonstrated excellent discrimination and calibration predicting perioperative mortality in patients undergoing (1) open surgery, (2) emergency/acute surgery, and (3) in cases with colon-located cancer. SORT is an easily adopted risk-assessment tool, associated with enhanced accuracy, that could be implemented in the perioperative pathway of patients undergoing surgery for colorectal cancer.

Association of the erythropoiesis-stimulating agent resistance index and the geriatric nutritional risk index with cardiovascular mortality in maintenance hemodialysis patients.

Hyporesponsiveness to erythropoiesis-stimulating agent (ESA) may be associated with protein-energy wasting. We investigated the relationship of the ESA resistance index (ERI) and the geriatric nutritional risk index (GNRI) for cardiovascular mortality in hemodialysis (HD) patients. A total of 180 maintenance HD patients were enrolled. The patients were stratified by the GNRI of 91.2, a previously reported cut-off value, and the ERI of 13.7 (IU/week/kg/g/dL), a cut-off value for predicting cardiovascular-specific mortality, and they were classified into four groups (group 1[G1]: higher GNRI and lower ERI, G2: higher GNRI and higher ERI, G3: lower GNRI and lower ERI, G4: lower GNRI and higher ERI). The ERI was independently associated with the GNRI (β = -0.271, p = 0.0005). During a median follow-up of 4.6 years, higher ERI and lower GNRI were independently associated with cardiovascular mortality, respectively (adjusted hazard ratio [aHR], 3.10; 95% confidence interval [CI], 1.31-7.34, and aHR, 6.64; 95%CI, 2.60-16.93, respectively). The 7-year survival rates were 96.1%, 70.3%, 77.3%, and 50.1% in G1, G2, G3, and G4, respectively. The aHR values for G4 versus G1 were 12.63 (95%CI, 3.58-44.59). With regards to model discrimination, adding the GNRI alone, the ERI alone, and both to the traditional risk model significantly improved the net reclassification improvement by 0.421, 0.662, and 0.671, respectively. Similar results were obtained for all-cause mortality. The ERI was independently associated with the GNRI, and could predict cardiovascular mortality in HD patients. Moreover, the combination of GNRI and ERI could improve the predictability for cardiovascular mortality.

Radiological society of North America: 106th annual meeting, November 29 to December 5, 2020

Radiological society of North America: 106th annual meeting, November 29 to December 5, 2020

Correlation Breakdowns, Spread Positions, and CCP Margin Models

The default of a participant at Nasdaq Clearing in 2018 and the recent COVID-19 events brought to the attention of risk managers at CCPs the importance of appropriately measuring correlation breakdowns. The sizable price dislocations registered on these occasions suggested that traditional risk models may not be fully equipped to capture the breakdowns. Because correlations are directly impacted by the statistical properties of each variable, any model that lacks the capacity to deal with non-stationarity may inappropriately represent correlation or its alterations. Using a GARCH-DCC approach to accommodate such properties, the objective of the paper is to study the correlation behaviour during adverse market conditions, and the potential subsequent impact to CCP margins. A study case for energy commodities is proposed, with the specific focus on spread positions for the electricity market. The analysis suggests that the correlation breakdowns are more frequent than traditionally expected. When different types of shocks are considered (i.e. September 2018 and March-May 2020), it becomes evident that while the magnitude of the breakdown may differ, its cycle presents a number of similarities. While elevated margin due to correlation breakdown may reduce breaching amount and improve margin coverage rate, this paper also recognizes the potentially increased margin procyclicality, and highlights the challenge of balancing margin responsiveness and stability during correlation breakdown for spread positions, and calls for further study in this area.

One-hour post-load glucose improves the prediction of cardiovascular events in the OPERA study

Background To estimate the ability of fasting, 1-h, and 2-h post-load glucose to predict cardiovascular outcomes. Methods We examined a population-based study consisting of 977 middle-aged subjects who underwent an oral glucose tolerance test with glucose values measured at 0, 60, and 120 min. Participants were followed up to 24 years, and cardiovascular outcomes were collected from national registers. Predictive abilities of fasting, 1-h, and 2-h glucose were evaluated alone and in the prediction models with traditional cardiovascular risk factors using Cox proportional hazard models, the likelihood-ratio test, Harrell's concordance index and integrated discrimination improvement. Results Cardiovascular endpoint occurred in 222 (22.7%) participants during a median follow-up of 19.8 years. In the prognostic models, 1-h glucose (HR 1.67, 95%CI 1.10–2.53), but not fasting or 2-h glucose, predicted cardiovascular events statistically significantly. In addition, when adding glucose parameters into the model including traditional cardiovascular risk factors, only 1-h glucose improved the predictive ability (LR-test p=.046). Finally, 1-h glucose found slightly over 50% more cardiovascular endpoints that were not recognized by fasting or 2-h glucose levels. Conclusions Our findings support the earlier ones suggesting that 1-h glucose would be a better long-term predictor of cardiovascular morbidity and mortality than fasting or 2-h glucose. KEY MESSAGES In addition to conventional CV risk factors,1-h but not fasting or 2-h post-load glucoses seems to be an independent predictor of cardiovascular events and seems to improve the predictive ability of the traditional cardiovascular risk model. Elevated 1-hpost-load glucose finds a large number (slightly over 50%)of cardiovascular endpoints that were not recognized by fasting or 2-h post-load glucose levels. One-hour glucose seems to be a better long-term predictor of cardiovascular morbidity and mortality than fasting or 2-h post-load glucose.

Central banks, financial stability and policy coordination in the age of climate uncertainty: a three-layered analytical and operational framework

ABSTRACT This paper explores how climate change and the transition to a low-carbon economy pose new systemic financial risks (so-called Green Swans), and the role of central banks in addressing them within their financial stability mandate. It does so by developing a three-layered analytical framework that central banks could use to shape their climate-related monetary and financial policies. First, central banks have already started to revisit their backward-looking risk models for the purpose of integrating forward-looking climate-related risks. Second, given the limitations of existing climate-economy models in a context of radical uncertainty, we argue that future climate scenario analysis should rely on systems-based approaches such as non-equilibrium models and more qualitative tools such as those provided by a socio-technical perspective. However, even these new approaches will not suffice from the perspective of financial stability: climate-related risks will remain largely unhedgeable as long as a system-wide and structural transformation is not undertaken, including an unprecedented level of cooperation between central banks, fiscal authorities, the private and public sectors at large, civil society, and the international community. Third, and as a result of the first two points, embracing climate-related uncertainty means that central banks play a new role: helping to coordinate the policies needed to fight climate change, so as to fulfil their own mandate of financial stability. To this end, we make a few specific policy proposals. Key policy insights Central banks’ traditional risk models do not enable them to identify climate-related systemic risks (Green Swans). Adopting new forward-looking and non-equilibrium modelling approaches is necessary to better appreciate the nature of climate-related risks, but it is not sufficient. To continue fulfilling their mandate of financial stability over longer time horizons than those traditionally considered, central banks must also get involved in policy coordination to mitigate climate change. This includes exploring which policy mixes (fiscal-monetary-prudential) can better address the climate imperatives ahead; considering climate stability as a global public good to be supported through reforms of the international monetary and financial system; and systematizing the integration of long-term sustainability criteria in both private and public sectors.

Comparative Analysis of Risk Stratification Scores in Atrial Fibrillation.

Atrial Fibrillation (AF) has become a major global health concern and is associated with an increased risk of poor outcomes. Identifying risk factors in patients with AF can be challenging, given the high burden of comorbidities in these patients. Risk stratification schemes appear to facilitate accurate prediction of outcomes and assist therapeutic management decisions. To summarize current evidence on risk stratification scores for patients with AF. Traditional risk models rely heavily on demographics and comorbidities, while newer tools have been gradually focusing on novel biomarkers and diagnostic imaging to facilitate more personalized risk assessment. Several studies have been conducted to compare existing risk schemes and identify specific patient populations in which the prognostic ability of each scheme excels. However, current guidelines do not appear to encourage the implementation of risk models in clinical practice, as they have not incorporated new ones in their recommendations for the management of patients with AF for almost a decade. Further work is warranted to analyze new reliable risk stratification schemes and optimally implement them into routine clinical life.

Abstract 312: A Multi-view Echocardiography Video Deep Learning Model Outperforms the Seattle Heart Failure Model in Predicting Mortality

Introduction: Managing heart failure patients relies on mortality prediction models such as the Seattle Heart Failure (SHF) model. We hypothesized that a deep learning model that leverages echocardiography video data could improve prediction performance. Methods: We trained deep neural networks (DNN) to predict 1-year all-cause mortality using echocardiography videos and 158 additional clinical variables (labs, vitals, diagnoses, etc.) acquired from 34,362 patients (812,278 videos). We tested both the DNN and the SHF model on a separate cohort of 2,404 patients with heart failure who underwent 3,384 echocardiograms. We computed the area under the receiver operating characteristic curve (AUC) and its bootstrapped 95% confidence interval (CI) in the test set. Results: The DNN model (AUC of 0.76, 95% CI [0.74, 0.77]) outperformed the SHF model (AUC of 0.70, 95% CI [0.68, 0.71]). This superior performance was observed for patients with both reduced (n=2,026, AUC of 0.76 [0.74, 0.78] vs 0.70 [0.67, 0.72]) and preserved left ventricular ejection fraction (n=1,356, AUC of 0.75 [0.72, 0.78] vs 0.69 [0.66, 0.72]). A Cox Proportional Hazard survival analysis showed that, despite the prediction being for 1-year all-cause mortality, the result held long-term predictive power over the next 9 years with a superior hazard ratio of 2.9 [2.6, 3.2] for the DNN model compared to 2.2 [2, 2.4] for the SHF model. At mid-range operating points, the DNN model also maintained a higher negative predictive value, predicting survival, compared to the SHF model (89% vs 83%, respectively), while maintaining the same positive predictive value of 40%. Conclusion: A deep neural network that automatically analyzes echocardiography videos can outperform traditional risk modeling approaches such as the Seattle Heart Failure Model. This provides additional evidence that deep learning can help improve our ability to make clinically relevant predictions by leveraging complex datasets.