Kristin M. Corey
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Kristin M. Corey has a total of 74 co-authors. Here are the top co-authors by number of shared publications:
- Jawan W. Abdulrahim
Division of Cardiology, Department of Medicine, Duke University School of Medicine
2 shared publications - Katherine Heller
Google Research, San Francisco, USA
2 shared publications - Lydia Coulter Kwee
Duke Molecular Physiology Institute, Duke University School of Medicine
2 shared publications - Megan Elizabeth Ramaker
Center for Precision Health, Duke Clinical and Translational Science Institute, Duke University, Durham, NC
2 shared publications - Suresh Balu
Duke Institute for Health Innovation, Durham, NC, USA
2 shared publications - Svati H. Shah
Duke University School of Medicine
2 shared publications
Latest Publications
- Open Access
- 25 October 2025
- Published by Elsevier in The Lancet Digital Health
- Vol. 7 (9)
Leveraging deep learning applied to chest radiograph images to identify individuals at high risk of chronic obstructive pulmonary disease: a retrospective model validation study
- 25 October 2025
- Published by Elsevier in The Lancet Digital Health
- Vol. 7 (9)
Open AccessLeveraging deep learning applied to chest radiograph images to identify individuals at high risk of chronic obstructive pulmonary disease: a retrospective model validation study
- Open Access
Comprehensive Comparison of a Genotype-Forward vs. Phenotype-Forward Approach for Monogenic Cardiovascular Diseases in the UK Biobank Cohort
Background Monogenic cardiovascular diseases (MCVD) remain substantially under-diagnosed, and thus interest is shifting from a phenotype-forward approach towards a genotype-first strategy that applies to population-wide genomic screening. We evaluated the prevalence of pathogenic and predicted pathogenic MCVD variants and assessed disease expression in the population-based UK Biobank (UKB) cohort to determine whether yield is greater in phenotype-versus genotype-forward approach. Methods Pathogenic and likely pathogenic (P/LP) variants, and variants of uncertain significance (VUS) predicted to be pathogenic (pp-VUS) in 47 MCVD genes were assessed. In the genotype-forward approach, UKB participants with P/LPs or pp-VUS were assessed for symptoms of disease using electronic health record (EHR) interrogation, emulating a clinical approach to population-based screening. In the phenotype-forward approach, participants with stricter EHR-based evidence of disease expression were identified (emulating current clinical care) and the presence of P/LPs and pp-VUS was determined. Results Following QC, 467,850 participants were included. Overall, 1 in 125 participants in UKB carry an MCVD P/LP. In the genotype-forward approach, 3,709 (0.79%) participants carried an MCVD P/LP and 42.1% of these had symptoms of the associated MCVD; another 29,269 (6.3%) carried a pp-VUS with 21.5% of these having symptoms of disease. In the phenotype-forward approach, 62,488 (13.4%) expressed an MCVD using strict evidence of disease expression and of these individuals 1% carried an associated P/LP variant and 2.4% carried an associated pp-VUS. Conclusion We show here that a genotype-forward approach leads to a 2.5 times higher diagnostic yield (3.6 times higher when considering pp-VUS) compared with the phenotype-forward approach, which missed 936 P/LP carriers with disease symptoms. While cost, access and ethics need to be considered, these results support expanded population-based genetic screening to improve diagnosis of potentially treatable MCVD.
Open AccessComprehensive Comparison of a Genotype-Forward vs. Phenotype-Forward Approach for Monogenic Cardiovascular Diseases in the UK Biobank Cohort
Background Monogenic cardiovascular diseases (MCVD) remain substantially under-diagnosed, and thus interest is shifting from a phenotype-forward approach towards a genotype-first strategy that applies to population-wide genomic screening. We evaluated the prevalence of pathogenic and predicted pathogenic MCVD variants and assessed disease expression in the population-based UK Biobank (UKB) cohort to determine whether yield is greater in phenotype-versus genotype-forward approach. Methods Pathogenic and likely pathogenic (P/LP) variants, and variants of uncertain significance (VUS) predicted to be pathogenic (pp-VUS) in 47 MCVD genes were assessed. In the genotype-forward approach, UKB participants with P/LPs or pp-VUS were assessed for symptoms of disease using electronic health record (EHR) interrogation, emulating a clinical approach to population-based screening. In the phenotype-forward approach, participants with stricter EHR-based evidence of disease expression were identified (emulating current clinical care) and the presence of P/LPs and pp-VUS was determined. Results Following QC, 467,850 participants were included. Overall, 1 in 125 participants in UKB carry an MCVD P/LP. In the genotype-forward approach, 3,709 (0.79%) participants carried an MCVD P/LP and 42.1% of these had symptoms of the associated MCVD; another 29,269 (6.3%) carried a pp-VUS with 21.5% of these having symptoms of disease. In the phenotype-forward approach, 62,488 (13.4%) expressed an MCVD using strict evidence of disease expression and of these individuals 1% carried an associated P/LP variant and 2.4% carried an associated pp-VUS. Conclusion We show here that a genotype-forward approach leads to a 2.5 times higher diagnostic yield (3.6 times higher when considering pp-VUS) compared with the phenotype-forward approach, which missed 936 P/LP carriers with disease symptoms. While cost, access and ethics need to be considered, these results support expanded population-based genetic screening to improve diagnosis of potentially treatable MCVD.
- 1 December 2024
- Published by Wolters Kluwer Health in Circulation: Genomic and Precision Medicine
- Vol. 17 (6) , e004464
Cardiovascular Disease Pathogenicity Predictor (CVD-PP): A Tissue-Specific In Silico Tool for Discriminating Pathogenicity of Variants of Unknown Significance in Cardiovascular Disease Genes
BACKGROUND: Interpretation of variants of uncertain significance (VUSs) remains a challenge in the care of patients with inherited cardiovascular diseases (CVDs); 56% of variants within CVD risk genes are VUS, and machine learning algorithms trained upon large data resources can stratify VUS into higher versus lower probability of contributing to a CVD phenotype. METHODS: We used ClinVar pathogenic/likely pathogenic and benign/likely benign variants from 47 CVD genes to build a predictive model of variant pathogenicity utilizing measures of evolutionary constraint, deleteriousness, splicogenicity, local pathogenicity, cardiac-specific expression, and population allele frequency. Performance was validated using variants for which the ClinVar pathogenicity assignment changed. Functional validation was assessed using prior studies in >900 identified VUS. The model utility was demonstrated using the Catheterization Genetics (CATHGEN) cohort. RESULTS: We identified a top-ranked model that accurately prioritized variants for which ClinVar clinical significance had changed (n=663; precision-recall area under the curve, 0.97) and performed well compared with conventional in silico methods. This model (CVD pathogenicity predictor [CVD-PP]) also had high accuracy in prioritizing VUS with functional effects in vivo (precision-recall area under the curve, 0.58). In CATHGEN, there was a greater burden of higher CVD-PP scored VUS in individuals with dilated cardiomyopathy compared with controls ( P =8.2×10 − 15 ). Of individuals in CATHGEN who harbored highly ranked CVD pathogenicity predictor VUS meeting clinical pathogenicity criteria, 27.6% had clinical evidence of disease. Variant prioritization using this model increased genetic diagnosis in CATHGEN participants with a known clinical diagnosis of hypertrophic cardiomyopathy (7.8%–27.2%). CONCLUSIONS: We present a cardiac-specific model for prioritizing variants underlying CVD syndromes with high performance in discriminating the pathogenicity of VUS in CVD genes. Variant review and phenotyping of individuals carrying VUS of pathogenic interest support the clinical utility of this model. This model could also have utility in filtering variants as part of large-scale genomic sequencing studies.
- 1 December 2024
- Published by Wolters Kluwer Health in Circulation: Genomic and Precision Medicine
- Vol. 17 (6) , e004464
Cardiovascular Disease Pathogenicity Predictor (CVD-PP): A Tissue-Specific In Silico Tool for Discriminating Pathogenicity of Variants of Unknown Significance in Cardiovascular Disease Genes
BACKGROUND: Interpretation of variants of uncertain significance (VUSs) remains a challenge in the care of patients with inherited cardiovascular diseases (CVDs); 56% of variants within CVD risk genes are VUS, and machine learning algorithms trained upon large data resources can stratify VUS into higher versus lower probability of contributing to a CVD phenotype. METHODS: We used ClinVar pathogenic/likely pathogenic and benign/likely benign variants from 47 CVD genes to build a predictive model of variant pathogenicity utilizing measures of evolutionary constraint, deleteriousness, splicogenicity, local pathogenicity, cardiac-specific expression, and population allele frequency. Performance was validated using variants for which the ClinVar pathogenicity assignment changed. Functional validation was assessed using prior studies in >900 identified VUS. The model utility was demonstrated using the Catheterization Genetics (CATHGEN) cohort. RESULTS: We identified a top-ranked model that accurately prioritized variants for which ClinVar clinical significance had changed (n=663; precision-recall area under the curve, 0.97) and performed well compared with conventional in silico methods. This model (CVD pathogenicity predictor [CVD-PP]) also had high accuracy in prioritizing VUS with functional effects in vivo (precision-recall area under the curve, 0.58). In CATHGEN, there was a greater burden of higher CVD-PP scored VUS in individuals with dilated cardiomyopathy compared with controls ( P =8.2×10 − 15 ). Of individuals in CATHGEN who harbored highly ranked CVD pathogenicity predictor VUS meeting clinical pathogenicity criteria, 27.6% had clinical evidence of disease. Variant prioritization using this model increased genetic diagnosis in CATHGEN participants with a known clinical diagnosis of hypertrophic cardiomyopathy (7.8%–27.2%). CONCLUSIONS: We present a cardiac-specific model for prioritizing variants underlying CVD syndromes with high performance in discriminating the pathogenicity of VUS in CVD genes. Variant review and phenotyping of individuals carrying VUS of pathogenic interest support the clinical utility of this model. This model could also have utility in filtering variants as part of large-scale genomic sequencing studies.
- Top Cited PapersOpen Access
- 15 July 2020
- Published by JMIR Publications Inc. in JMIR Medical Informatics
- Vol. 8 (7) , e15182
Real-World Integration of a Sepsis Deep Learning Technology Into Routine Clinical Care: Implementation Study
Background: Successful integrations of machine learning into routine clinical care are exceedingly rare, and barriers to its adoption are poorly characterized in the literature. Objective: This study aims to report a quality improvement effort to integrate a deep learning sepsis detection and management platform, Sepsis Watch, into routine clinical care. Methods: In 2016, a multidisciplinary team consisting of statisticians, data scientists, data engineers, and clinicians was assembled by the leadership of an academic health system to radically improve the detection and treatment of sepsis. This report of the quality improvement effort follows the learning health system framework to describe the problem assessment, design, development, implementation, and evaluation plan of Sepsis Watch. Results: Sepsis Watch was successfully integrated into routine clinical care and reshaped how local machine learning projects are executed. Frontline clinical staff were highly engaged in the design and development of the workflow, machine learning model, and application. Novel machine learning methods were developed to detect sepsis early, and implementation of the model required robust infrastructure. Significant investment was required to align stakeholders, develop trusting relationships, define roles and responsibilities, and to train frontline staff, leading to the establishment of 3 partnerships with internal and external research groups to evaluate Sepsis Watch. Conclusions: Machine learning models are commonly developed to enhance clinical decision making, but successful integrations of machine learning into routine clinical care are rare. Although there is no playbook for integrating deep learning into clinical care, learnings from the Sepsis Watch integration can inform efforts to develop machine learning technologies at other health care delivery systems.
- 15 July 2020
- Published by JMIR Publications Inc. in JMIR Medical Informatics
- Vol. 8 (7) , e15182
Top Cited PapersOpen AccessReal-World Integration of a Sepsis Deep Learning Technology Into Routine Clinical Care: Implementation Study
Background: Successful integrations of machine learning into routine clinical care are exceedingly rare, and barriers to its adoption are poorly characterized in the literature. Objective: This study aims to report a quality improvement effort to integrate a deep learning sepsis detection and management platform, Sepsis Watch, into routine clinical care. Methods: In 2016, a multidisciplinary team consisting of statisticians, data scientists, data engineers, and clinicians was assembled by the leadership of an academic health system to radically improve the detection and treatment of sepsis. This report of the quality improvement effort follows the learning health system framework to describe the problem assessment, design, development, implementation, and evaluation plan of Sepsis Watch. Results: Sepsis Watch was successfully integrated into routine clinical care and reshaped how local machine learning projects are executed. Frontline clinical staff were highly engaged in the design and development of the workflow, machine learning model, and application. Novel machine learning methods were developed to detect sepsis early, and implementation of the model required robust infrastructure. Significant investment was required to align stakeholders, develop trusting relationships, define roles and responsibilities, and to train frontline staff, leading to the establishment of 3 partnerships with internal and external research groups to evaluate Sepsis Watch. Conclusions: Machine learning models are commonly developed to enhance clinical decision making, but successful integrations of machine learning into routine clinical care are rare. Although there is no playbook for integrating deep learning into clinical care, learnings from the Sepsis Watch integration can inform efforts to develop machine learning technologies at other health care delivery systems.
- 25 February 2020
- Published by SAGE Publications in World Journal for Pediatric and Congenital Heart Surgery
- Vol. 11 (2) , 192-197
Pulmonary Valve Endocarditis: The Potential Utility of Multimodal Imaging Prior to Surgery
Background: The presence of echocardiographic (echo) evidence is a major criterion for the diagnosis of infective endocarditis (IE) by modified Duke criteria. Pulmonary valve (PV) IE, however, can be challenging to identify by echo. We sought to evaluate the added utility of multimodal imaging in PV IE. Methods: This is a single-center case series. We retrospectively analyzed demographic, laboratory, imaging, clinical, and surgical data from patients diagnosed with PV IE from 2008 to 2018. Results: A total of 23 patients were identified with definite PV IE by Duke criteria (83% male and ages 2 months to 70 years). Twenty-two patients had congenital heart disease, with 21 involving the right ventricular outflow tract (including three with transcatheter PV implant). Overall, 20 (87%) of 23 had positive blood cultures. A total of 17 (74%) of 23 patients demonstrated echo evidence of PV IE. In three cases, echo was negative (did not show vegetations) but showed new PV obstruction. In four cases with negative transthoracic echocardiogram and transesophageal echocardiogram, evidence of PV IE was subsequently seen by positron emission tomography/computed tomography (n = 2) or cardiac magnetic resonance imaging (n = 2). Pulmonary valve IE was confirmed at surgery by evaluation of pathologic samples in 20 cases. Conclusions: Multimodal imaging improves the ability to preoperatively identify endocardial involvement in PV IE in cases where echo is negative. Consideration should be given to revise Duke criteria to include new obstruction and endocardial involvement by multimodal imaging for PV IE.
- 25 February 2020
- Published by SAGE Publications in World Journal for Pediatric and Congenital Heart Surgery
- Vol. 11 (2) , 192-197
Pulmonary Valve Endocarditis: The Potential Utility of Multimodal Imaging Prior to Surgery
Background: The presence of echocardiographic (echo) evidence is a major criterion for the diagnosis of infective endocarditis (IE) by modified Duke criteria. Pulmonary valve (PV) IE, however, can be challenging to identify by echo. We sought to evaluate the added utility of multimodal imaging in PV IE. Methods: This is a single-center case series. We retrospectively analyzed demographic, laboratory, imaging, clinical, and surgical data from patients diagnosed with PV IE from 2008 to 2018. Results: A total of 23 patients were identified with definite PV IE by Duke criteria (83% male and ages 2 months to 70 years). Twenty-two patients had congenital heart disease, with 21 involving the right ventricular outflow tract (including three with transcatheter PV implant). Overall, 20 (87%) of 23 had positive blood cultures. A total of 17 (74%) of 23 patients demonstrated echo evidence of PV IE. In three cases, echo was negative (did not show vegetations) but showed new PV obstruction. In four cases with negative transthoracic echocardiogram and transesophageal echocardiogram, evidence of PV IE was subsequently seen by positron emission tomography/computed tomography (n = 2) or cardiac magnetic resonance imaging (n = 2). Pulmonary valve IE was confirmed at surgery by evaluation of pathologic samples in 20 cases. Conclusions: Multimodal imaging improves the ability to preoperatively identify endocardial involvement in PV IE in cases where echo is negative. Consideration should be given to revise Duke criteria to include new obstruction and endocardial involvement by multimodal imaging for PV IE.
- Top Cited PapersOpen Access
- 27 November 2018
- Published by Public Library of Science (PLoS) in PLOS Medicine
- Vol. 15 (11) , e1002701
Development and validation of machine learning models to identify high-risk surgical patients using automatically curated electronic health record data (Pythia): A retrospective, single-site study
Pythia is an automated, clinically curated surgical data pipeline and repository housing all surgical patient electronic health record (EHR) data from a large, quaternary, multisite health institute for data science initiatives. In an effort to better identify high-risk surgical patients from complex data, a machine learning project trained on Pythia was built to predict postoperative complication risk. A curated data repository of surgical outcomes was created using automated SQL and R code that extracted and processed patient clinical and surgical data across 37 million clinical encounters from the EHRs. A total of 194 clinical features including patient demographics (e.g., age, sex, race), smoking status, medications, comorbidities, procedure information, and proxies for surgical complexity were constructed and aggregated. A cohort of 66,370 patients that had undergone 99,755 invasive procedural encounters between January 1, 2014, and January 31, 2017, was studied further for the purpose of predicting postoperative complications. The average complication and 30-day postoperative mortality rates of this cohort were 16.0% and 0.51%, respectively. Least absolute shrinkage and selection operator (lasso) penalized logistic regression, random forest models, and extreme gradient boosted decision trees were trained on this surgical cohort with cross-validation on 14 specific postoperative outcome groupings. Resulting models had area under the receiver operator characteristic curve (AUC) values ranging between 0.747 and 0.924, calculated on an out-of-sample test set from the last 5 months of data. Lasso penalized regression was identified as a high-performing model, providing clinically interpretable actionable insights. Highest and lowest performing lasso models predicted postoperative shock and genitourinary outcomes with AUCs of 0.924 (95% CI: 0.901, 0.946) and 0.780 (95% CI: 0.752, 0.810), respectively. A calculator requiring input of 9 data fields was created to produce a risk assessment for the 14 groupings of postoperative outcomes. A high-risk threshold (15% risk of any complication) was determined to identify high-risk surgical patients. The model sensitivity was 76%, with a specificity of 76%. Compared to heuristics that identify high-risk patients developed by clinical experts and the ACS NSQIP calculator, this tool performed superiorly, providing an improved approach for clinicians to estimate postoperative risk for patients. Limitations of this study include the missingness of data that were removed for analysis. Extracting and curating a large, local institution’s EHR data for machine learning purposes resulted in models with strong predictive performance. These models can be used in clinical settings as decision support tools for identification of high-risk patients as well as patient evaluation and care management. Further work is necessary to evaluate the impact of the Pythia risk calculator within the clinical workflow on postoperative outcomes and to optimize this data flow for future machine learning efforts.
- 27 November 2018
- Published by Public Library of Science (PLoS) in PLOS Medicine
- Vol. 15 (11) , e1002701
Top Cited PapersOpen AccessDevelopment and validation of machine learning models to identify high-risk surgical patients using automatically curated electronic health record data (Pythia): A retrospective, single-site study
Pythia is an automated, clinically curated surgical data pipeline and repository housing all surgical patient electronic health record (EHR) data from a large, quaternary, multisite health institute for data science initiatives. In an effort to better identify high-risk surgical patients from complex data, a machine learning project trained on Pythia was built to predict postoperative complication risk. A curated data repository of surgical outcomes was created using automated SQL and R code that extracted and processed patient clinical and surgical data across 37 million clinical encounters from the EHRs. A total of 194 clinical features including patient demographics (e.g., age, sex, race), smoking status, medications, comorbidities, procedure information, and proxies for surgical complexity were constructed and aggregated. A cohort of 66,370 patients that had undergone 99,755 invasive procedural encounters between January 1, 2014, and January 31, 2017, was studied further for the purpose of predicting postoperative complications. The average complication and 30-day postoperative mortality rates of this cohort were 16.0% and 0.51%, respectively. Least absolute shrinkage and selection operator (lasso) penalized logistic regression, random forest models, and extreme gradient boosted decision trees were trained on this surgical cohort with cross-validation on 14 specific postoperative outcome groupings. Resulting models had area under the receiver operator characteristic curve (AUC) values ranging between 0.747 and 0.924, calculated on an out-of-sample test set from the last 5 months of data. Lasso penalized regression was identified as a high-performing model, providing clinically interpretable actionable insights. Highest and lowest performing lasso models predicted postoperative shock and genitourinary outcomes with AUCs of 0.924 (95% CI: 0.901, 0.946) and 0.780 (95% CI: 0.752, 0.810), respectively. A calculator requiring input of 9 data fields was created to produce a risk assessment for the 14 groupings of postoperative outcomes. A high-risk threshold (15% risk of any complication) was determined to identify high-risk surgical patients. The model sensitivity was 76%, with a specificity of 76%. Compared to heuristics that identify high-risk patients developed by clinical experts and the ACS NSQIP calculator, this tool performed superiorly, providing an improved approach for clinicians to estimate postoperative risk for patients. Limitations of this study include the missingness of data that were removed for analysis. Extracting and curating a large, local institution’s EHR data for machine learning purposes resulted in models with strong predictive performance. These models can be used in clinical settings as decision support tools for identification of high-risk patients as well as patient evaluation and care management. Further work is necessary to evaluate the impact of the Pythia risk calculator within the clinical workflow on postoperative outcomes and to optimize this data flow for future machine learning efforts.