Researchers Create Machine Learning Model to Predict Dialysis Treatment or Inpatient Death COVID-19

Title of the paper: Approaches to predicting the need for acute dialysis and death in COVID-19

Journal: The American Society of Nephrology Clinical Journal (published online May 24, 2021)

Authors: Girish Nadkarni, MD, Associate Professor in the Department of Medicine (Nephrology), Clinical Director of the Hasso Plattner Institute for Digital Health, and Co-chair of the Mount Sinai Clinical Intelligence Center at the Icahn School of Medicine at Mount Sinai; Lili Chan, MD, assistant professor in the Department of Medicine (Nephrology) at Icahn School of Medicine at Mount Sinai; Akhil Vaid, MD, postdoctoral researcher in the Department of Genetics and Genomic Sciences, Icahn School of Medicine, Mount Sinai, and member of the Mount Sinai Clinical Intelligence Center and the Hasso Plattner Institute for Digital Health of Mount Sinai; and other co-authors.

Bottom Line: SARS-CoV-2, the virus that causes COVID-19, has infected more than 103 million people worldwide. Acute kidney injury (AKI) treated with dialysis was a common complication in patients hospitalized with COVID-19. Acute kidney injury is associated with increased risk of morbidity and mortality. Early prediction of which patients will need dialysis or who will suffer from a serious illness resulting in death during hospital care can improve appropriate monitoring and better inform conversations with patients and their caregivers.

Results: The Mount Sinai team developed and tested five different algorithms to predict patients requiring dialysis treatment or severe illness resulting in death on Days 1, 3, 5, and 7 of the hospital stay, using the data from the first 12 hours of admission to hospital. the health system of Mount Sinai. Of the five models, the XGBoost without an imputation method outperformed all others with higher precision and recall.

Why the research is interesting: While the Mount Sinai model requires further external review, such machine learning models can potentially be deployed across healthcare systems to help determine which COVID-19 patients are most at risk. risk of adverse effects from coronavirus. Early detection of patients at risk can improve closer monitoring of patients and spark earlier discussions about goals of care.

Who: More than 6,000 adults with COVID-19 admitted to five hospitals in the Mount Sinai health system.

When: COVID-19 patients admitted from March 10 to December 26, 2020.

What: The study uses a machine learning model to determine which COVID-19 patients are most at risk for treatment requiring dialysis or severe illness resulting in death.

How: The team used data from adults hospitalized with COVID-19 across the Mount Sinai health system to develop and validate five models for predicting dialysis treatment or death at different time periods – 1, 3, 5 and 7 days – after hospital admission. Patients admitted to Manhattan’s Mount Sinai Hospital were used for internal validation, while the other four hospital sites were part of the external validation cohort. Characteristics assessed included demographics, comorbidities, laboratory results, and vital signs within 12 hours of hospital admission.

The five models created and tested were: logistic regression, LASSO, random forest, and XGBoost with and without imputation. Of the total model approaches used, XGBoost without imputation had the largest area under the receiver curve and the area under the precision recall curve during internal validation for all time points. This model also had the highest test settings on external validation over all time windows. Characteristics such as red blood cell distribution width, creatinine, and blood urea nitrogen were the main drivers of the model’s prediction.

Study conclusions: Mount Sinai researchers developed and validated a machine learning model to identify COVID-19 hospital patients at risk for acute kidney injury and death. The non-imputation XGBoost model had the best performance compared to standard models and other machine learning models. The widespread use of electronic health records makes it possible to deploy prediction models like this one.

According to Dr Girish Nadkarni of Mount Sinai, on the research: The near universal use of electronic health records has created a huge amount of data, which has allowed us to generate prediction models that can directly aid in the patient care. A version of this model is currently being deployed at Mount Sinai Hospital in patients admitted with COVID-19.

Dr Lili Chan from Mount Sinai, on the research, said: As a nephrologist, we were overwhelmed by the increase in the number of AKI patients during the initial outbreak of the COVID-19 pandemic . Prediction models like this allow us to identify, from the start of the hospital journey, people at risk of severe ARI (those requiring dialysis) and death. This information will facilitate the clinical care of patients and inform discussions with patients and their families.

Dr Akhil Vaid of Mount Sinai said of the research: Machine learning allows us to discern complex patterns in large amounts of data. For COVID-19 hospital patients, this means being able to more easily identify new patients at risk, while also identifying the underlying factors that make them better or worse. The underlying algorithm, XGBoost, excels in accuracy, speed, and other features under the hood that allow easier deployment and understanding of model predictions.


To request a full copy of the paper or to schedule an interview with the research team, please contact the Mount Sinai Press Office at [email protected] or 347-346-3390.

Warning: AAAS and EurekAlert! are not responsible for the accuracy of any press releases posted on EurekAlert! by contributing institutions or for the use of any information via the EurekAlert system.

About Hector Hedgepeth

Check Also

Study reveals how lack of technology can impact food insecurity

According to Professor Di Fang of the University of Arkansas, the psychological effects of food …

Leave a Reply

Your email address will not be published.