Dataset Information

Machine Learning Prediction Models for In-Hospital Mortality After Transcatheter Aortic Valve Replacement.

ABSTRACT: OBJECTIVES:This study sought to develop and compare an array of machine learning methods to predict in-hospital mortality after transcatheter aortic valve replacement (TAVR) in the United States. BACKGROUND:Existing risk prediction tools for in-hospital complications in patients undergoing TAVR have been designed using statistical modeling approaches and have certain limitations. METHODS:Patient data were obtained from the National Inpatient Sample database from 2012 to 2015. The data were randomly divided into a development cohort (n = 7,615) and a validation cohort (n = 3,268). Logistic regression, artificial neural network, naive Bayes, and random forest machine learning algorithms were applied to obtain in-hospital mortality prediction models. RESULTS:A total of 10,883 TAVRs were analyzed in our study. The overall in-hospital mortality was 3.6%. Overall, prediction models' performance measured by area under the curve were good (>0.80). The best model was obtained by logistic regression (area under the curve: 0.92; 95% confidence interval: 0.89 to 0.95). Most obtained models plateaued after introducing 10 variables. Acute kidney injury was the main predictor of in-hospital mortality ranked with the highest mean importance in all the models. The National Inpatient Sample TAVR score showed the best discrimination among available TAVR prediction scores. CONCLUSIONS:Machine learning methods can generate robust models to predict in-hospital mortality for TAVR. The National Inpatient Sample TAVR score should be considered for prognosis and shared decision making in TAVR patients.

SUBMITTER: Hernandez-Suarez DF

PROVIDER: S-EPMC6650265 | biostudies-literature | 2019 Jul

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Machine Learning Prediction Models for In-Hospital Mortality After Transcatheter Aortic Valve Replacement.

Hernandez-Suarez Dagmar F DF Kim Yeunjung Y Villablanca Pedro P Gupta Tanush T Wiley Jose J Nieves-Rodriguez Brenda G BG Rodriguez-Maldonado Jovaniel J Feliu Maldonado Roberto R da Luz Sant'Ana Istoni I Sanina Cristina C Cox-Alomar Pedro P Ramakrishna Harish H Lopez-Candales Angel A O'Neill William W WW Pinto Duane S DS Latib Azeem A Roche-Lima Abiel A

JACC. Cardiovascular interventions 20190701 14

<h4>Objectives</h4>This study sought to develop and compare an array of machine learning methods to predict in-hospital mortality after transcatheter aortic valve replacement (TAVR) in the United States.<h4>Background</h4>Existing risk prediction tools for in-hospital complications in patients undergoing TAVR have been designed using statistical modeling approaches and have certain limitations.<h4>Methods</h4>Patient data were obtained from the National Inpatient Sample database from 2012 to 201 ...[more]

PMID: 31320027

Similar Datasets

Project description:ImportanceThe Centers for Medicare & Medicaid Services national coverage determination for transcatheter aortic valve replacement (TAVR) includes volume requirements for surgical aortic valve replacement (SAVR) for hospitals seeking to initiate or continue TAVR programs. Evidence regarding the association between SAVR volume and TAVR outcomes is limited.ObjectiveTo examine the association of hospital SAVR and combined SAVR and TAVR volumes with patient outcomes of TAVR procedures performed within 1 year, 2 years, and for the entire period after initiation of TAVR programs.Design, setting, and participantsThis observational cohort study included 60 538 TAVR procedures performed in 438 hospitals between October 1, 2011, and December 31, 2015, among Medicare beneficiaries.Main outcomes and measuresThe associations between SAVR volume, SAVR and TAVR volumes, and risks of death, death or stroke, and readmissions within 30 days were determined using a hierarchical logistic regression model adjusting for patient and hospital characteristics. The association between SAVR and SAVR and TAVR volumes and 1-year and 2-year mortality after TAVR procedures was determined using a multivariable proportional hazard model with a robust variance estimator. The associations for procedures performed within 1 year, 2 years, and for the entire period after initiation of TAVR programs were examined.ResultsAmong the 60 538 patients, 29 173 were women and 31 365 were men, with a mean (SD) age of 82.3 (8.0) years. Hospitals with high SAVR volume (mean annual volume, ≥97 per year) were more likely to adopt TAVR early and had a higher growth in TAVR volumes over time (median TAVR volume by hospitals with high SAVR volume and low SAVR volume: year 1, 32 vs 19; year 2, 48 vs 28; year 3, 82 vs 38; year 4, 118 vs 54; P < .001). In adjusted analysis, high hospital SAVR volume alone was not associated with better patient outcomes after TAVR. When hospital TAVR and SAVR volumes were jointly analyzed, patients treated in hospitals with high TAVR volume had lower 30-day mortality after TAVR (high TAVR and low SAVR vs low TAVR and low SAVR: odds ratio, 0.85; 95% CI, 0.72-0.99; high TAVR and high SAVR vs low TAVR and high SAVR: odds ratio, 0.81; 95% CI, 0.69-0.95), the effect of which was more pronounced when hospitals also had high SAVR volume. Patients treated in hospitals with high SAVR volume and high TAVR volume had the lowest 30-day mortality (vs hospitals with low SAVR volume and TAVR volume: odds ratio, 0.77; 95% CI, 0.66-0.89).Conclusions and relevanceHospitals with high SAVR volume are most likely to be fast adopters of TAVR. Hospital SAVR volume alone is not associated with better TAVR outcomes. Accumulating high volumes of TAVR is associated with lower mortality after TAVR, particularly when hospitals have high SAVR volumes. Hospitals with high caseloads of both SAVR and TAVR are likely to achieve the best outcomes.

Project description:ImportanceHospital outcomes for transcatheter aortic valve replacement (TAVR) may be dependent on the quality of evaluation, personnel, and procedural and postprocedural care common to patients undergoing cardiac surgery.ObjectivesWe sought to assess whether those hospitals with better patient outcomes for surgical aortic valve replacement (SAVR) subsequently achieved better TAVR outcomes after launching TAVR programs.Design, setting, and participantsThis national cohort included US patients 65 years and older. The analysis used the Centers for Medicare and Medicaid Services' Medicare Provider and Review data collected between January 1, 2010, and September 29, 2015. Only hospitals performing at least 1 SAVR prior to September 1, 2011, and performing at least 1 TAVR after this date were included in the analysis. Data analysis was completed from June 2018 to August 2018.InterventionsIsolated aortic valve replacements.Main outcomes and measuresHospital risk-adjusted 30-day mortality for SAVR in the pre-TAVR period was used as a surrogate for SAVR quality. Thirty-day and 1-year TAVR mortality rates were examined after stratification by quartile of baseline hospital risk-adjusted SAVR mortality.ResultsA total of 51 924 TAVR procedures were performed in 519 hospitals, of which 19 798 were performed at hospitals in quartile 1 (the lowest risk-adjusted SAVR mortality rate), 7663 were performed in quartile 2, 10 180 were performed in quartile 3, and 14 283 were performed in quartile 4 (the highest risk-adjusted SAVR mortality rate). Observed mortality rates at 30 days consistently increased with increasing baseline hospital SAVR risk-adjusted mortality (quartile 1, 917 patients [4.6%]; quartile 2, 381 [5.0%]; quartile 3, 521 [5.1%]; quartile 4, 800 [5.6%]; P < .001). The same pattern was observed in 1-year mortality (quartile 1, 3359 [17.0%]; quartile 2, 1337 [17.5%]; quartile 3, 1852 [18.2%]; quartile 4, 2652 [18.6%]; P < .001). After multivariable analysis, compared with the lowest quartile of SAVR mortality, undergoing TAVR at a hospital with higher baseline SAVR mortality continued to be associated with higher 30-day mortality (odds ratios: quartile 2, 1.02 [95% CI, 0.87-1.21]; quartile 3, 1.13 [95% CI, 1.02-1.26]; quartile 4, 1.23 [95% CI, 1.07-1.40]; P = .02) and 1-year mortality (hazard ratios: quartile 2, 1.04 [95% CI, 0.92-1.17]; quartile 3, 1.14 [95% CI, 1.02-1.28]; quartile 4, 1.16 [95% CI, 1.05-1.28]; P = .02).Conclusions and relevanceHospitals with higher SAVR mortality rates also had higher short-term and long-term TAVR mortality after initiating TAVR programs. Quality of cardiac surgical care may be associated with a hospital's performance with new structural heart disease programs.

Project description:BACKGROUNDA series of models have been developed to identify patients at high risk for poor outcomes after transcatheter aortic valve replacement (TAVR) to help guide treatment choices, offer patients realistic expectations of long-term outcomes, and support decision making.OBJECTIVESThis study examined the performance of the previously developed TAVR Poor Outcome risk models in an external dataset and explored the incremental contribution of geriatric domains to model performance.METHODSPoor outcome after TAVR was defined as death, poor quality of life (QOL), or decline in QOL, as assessed using the Kansas City Cardiomyopathy Questionnaire. We tested 4 TAVR Poor Outcome risk models: 6-month and 1-year full and clinical (reduced) models. We examined each model's discrimination and calibration in the CoreValve trial dataset, and then tested the incremental contribution of frailty and disability markers to the model's discrimination using the incremental discrimination index.RESULTSAmong 2,830 patients who underwent TAVR in the CoreValve US Pivotal Extreme and High Risk trials and associated continued access registries, 31.2% experienced a poor outcome at 6 months following TAVR (death, 17.6%; very poor QOL, 11.6%; QOL decline, 2.0%) and 50.8% experienced a poor outcome at 1 year (death, 30.2%; poor QOL, 19.6%; QOL, decline 1.0%). The models demonstrated similar discrimination as in the Placement of Aortic Transcatheter Valves Trial cohorts (c-indexes, 0.637 to 0.665) and excellent calibration. Adding frailty as a syndrome increased the c-indexes by 0.000 to 0.004 (incremental discrimination index, p < 0.01 for all except the 1-year clinical model), with the most important individual components being disability and unintentional weight loss.CONCLUSIONSAlthough discrimination of the TAVR Poor Outcome risk models was generally moderate, calibration was excellent among patients with different risk profiles and treated with a different TAVR device. These findings demonstrated the value of these models for individualizing outcome predictions in high-risk patients undergoing TAVR.

Dataset Information

Machine Learning Prediction Models for In-Hospital Mortality After Transcatheter Aortic Valve Replacement.

Publications

Machine Learning Prediction Models for In-Hospital Mortality After Transcatheter Aortic Valve Replacement.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure