In the Journals

Artificial intelligence may predict in-hospital mortality after TAVR

Researchers successfully used machine-learning methods to generate robust models for the prediction of in-hospital mortality in patients who underwent transcatheter aortic valve replacement, according to a study published in JACC: Cardiovascular Interventions.

Dagmar F. Hernandez-Suarez, MD, MSc, cardiology fellow at the University of Puerto Rico School of Medicine in San Juan, and colleagues analyzed data from 10,883 patients (mean age, 81 years; 48% women) from the National Inpatient Sample database who underwent TAVR from January 2012 to September 2015. Patients were randomly divided into a development cohort (n = 7,615) and a validation cohort (n = 3,268).

Four machine-learning algorithms were utilized in the study: artificial neural networks, logistic regression, random forest and naive Bayes. The main endpoint was all-cause in-hospital mortality.

Of the patients in the study, 3.6% died in the hospital. The main predictor of in-hospital mortality with the highest mean importance was acute kidney injury in all models.

The machine-learning algorithm with the best discrimination was logistic regression (area under the curve = 0.92; 95% CI, 0.89-0.95). The other algorithms also had good AUC for the prediction of in-hospital mortality, as shown with naive Bayes (AUC = 0.9; 95% CI, 0.88-0.92), artificial neural networks (AUC = 0.85; 95% CI, 0.82-0.88) and random forest (AUC = 0.9; 95% CI, 0.87-0.93).

After introducing 10 variables, most obtained models plateaued.

The National Inpatient Sample TAVR score had the best discrimination compared with other risk prediction models with an AUC of 0.92.

“The good discrimination of this model reveals the potential of AI in the patient risk stratification process not just for TAVR but for any novel structural intervention,” Hernandez-Suarez and colleagues wrote. “Further validation and application of [machine learning] into the day-to-day clinical practice is still warranted to better understand its true value in patients with severe aortic stenosis.”

Researchers successfully used machine-learning methods to generate robust models for the prediction of in-hospital mortality in patients who underwent transcatheter aortic valve replacement, according to a study published in JACC: Cardiovascular Interventions.
Source: Adobe Stock

In a related editorial, Thomas Modine, PhD, cardiac surgeon at the Heart Valve Center at Institut Coeur Poumon CHU de Lille in France, and Pavel Overtchouk, MD, of the department of cardiology at University Hospital of Bern in Switzerland, wrote: “Readers should remember that critical appreciation of research papers is vital, especially for topics subject to hype. Statistical significance does not guarantee clinical pertinence. Excellent results reported in ‘innovative’ studies must be verified in independent studies with different cohorts and be subjected to the test of time. Given the performance of machine learning and [deep-learning] approaches, their foreseen implementation in health care is likely. And as the hype passes, real applications should start to appear.” – by Darlene Dobkowski

Disclosures: Hernandez-Suarez and Overtchouk report no relevant financial disclosures. Modine reports he consults for Abbott, Boston Scientific, Cephea, Edwards Lifesciences, GE, Medtronic and Microport and received a research support grant from Edwards Lifesciences. Please see the study for all other authors’ relevant financial disclosures.

Researchers successfully used machine-learning methods to generate robust models for the prediction of in-hospital mortality in patients who underwent transcatheter aortic valve replacement, according to a study published in JACC: Cardiovascular Interventions.

Dagmar F. Hernandez-Suarez, MD, MSc, cardiology fellow at the University of Puerto Rico School of Medicine in San Juan, and colleagues analyzed data from 10,883 patients (mean age, 81 years; 48% women) from the National Inpatient Sample database who underwent TAVR from January 2012 to September 2015. Patients were randomly divided into a development cohort (n = 7,615) and a validation cohort (n = 3,268).

Four machine-learning algorithms were utilized in the study: artificial neural networks, logistic regression, random forest and naive Bayes. The main endpoint was all-cause in-hospital mortality.

Of the patients in the study, 3.6% died in the hospital. The main predictor of in-hospital mortality with the highest mean importance was acute kidney injury in all models.

The machine-learning algorithm with the best discrimination was logistic regression (area under the curve = 0.92; 95% CI, 0.89-0.95). The other algorithms also had good AUC for the prediction of in-hospital mortality, as shown with naive Bayes (AUC = 0.9; 95% CI, 0.88-0.92), artificial neural networks (AUC = 0.85; 95% CI, 0.82-0.88) and random forest (AUC = 0.9; 95% CI, 0.87-0.93).

After introducing 10 variables, most obtained models plateaued.

The National Inpatient Sample TAVR score had the best discrimination compared with other risk prediction models with an AUC of 0.92.

“The good discrimination of this model reveals the potential of AI in the patient risk stratification process not just for TAVR but for any novel structural intervention,” Hernandez-Suarez and colleagues wrote. “Further validation and application of [machine learning] into the day-to-day clinical practice is still warranted to better understand its true value in patients with severe aortic stenosis.”

Researchers successfully used machine-learning methods to generate robust models for the prediction of in-hospital mortality in patients who underwent transcatheter aortic valve replacement, according to a study published in JACC: Cardiovascular Interventions.
Source: Adobe Stock

In a related editorial, Thomas Modine, PhD, cardiac surgeon at the Heart Valve Center at Institut Coeur Poumon CHU de Lille in France, and Pavel Overtchouk, MD, of the department of cardiology at University Hospital of Bern in Switzerland, wrote: “Readers should remember that critical appreciation of research papers is vital, especially for topics subject to hype. Statistical significance does not guarantee clinical pertinence. Excellent results reported in ‘innovative’ studies must be verified in independent studies with different cohorts and be subjected to the test of time. Given the performance of machine learning and [deep-learning] approaches, their foreseen implementation in health care is likely. And as the hype passes, real applications should start to appear.” – by Darlene Dobkowski

Disclosures: Hernandez-Suarez and Overtchouk report no relevant financial disclosures. Modine reports he consults for Abbott, Boston Scientific, Cephea, Edwards Lifesciences, GE, Medtronic and Microport and received a research support grant from Edwards Lifesciences. Please see the study for all other authors’ relevant financial disclosures.