Prediction of Acute Kidney Injury in Hospitalized COVID-19 Patients Using Machine Learning

Carbonello, David J.

Full Metadata

Overview

Year:: 2021
Contributor:: Carbonello, David J (creator); Schmid, Christopher (Advisor); Sarin, Pankaj (Reader); Brown University. Department of Biostatistics (sponsor)
Genre:: theses
Subject:: Machine Learning; COVID-19
Extent:: 6 (vi), 38 p.

Files

Description

Abstract:: Prediction of Acute Kidney Injury in Hospitalized COVID-19 Patients Using Machine Learning, by David Carbonello, ScM., Brown University, May 2021 Many studies provide evidence that acute kidney injury (AKI) is associated with increased disease severity and mortality in hospitalized COVID-19 patients. The aim of this study is to identify hospitalized COVID-19 patients that are at a higher risk of AKI to aid in the appropriate triage of individuals and improve patient outcomes moving forward. This study reports the implementation of logistic regression, random forest, and neural network models to predict AKI occurrence (increase in serum creatinine of .3mg/dl or more within 48 hours or 1.5 times baseline or more within last 7 days) based on information available at or soon after admission. Information on vitals, demographics, and lab values available from the first 48 hours of admission were investigated retrospectively from electronic medical records of 5,033 hospitalized COVID-19 patients aged ≥18 years at Brigham & Women’s and Massachusetts General Hospitals in Boston, MA between January 1 and December 31, 2020. AKI occurred in 1,243 (24.7%) patients and in 362 (59.64%) patients who were admitted to the ICU upon hospital admission. Predictors found to be highly associated with AKI occurrence from the logistic regression model were admission to ICU (OR 3.83; 95% CI 3.0-4.88), absolute neutrophil count (OR 2.94; 95% CI: 1.7-5.1), and white blood cell count (OR 0.43; 95% CI 0.26-0.72). The random forest yielded the highest area under the ROC curve of the three prediction models (AUC=0.796) when evaluated on the holdout test set. Overall, all three algorithms yielded strong results when predicting AKI occurrence using information provided upon admission and could help hospitals better allocate resources to improve patient outcomes.
Notes:: Thesis (Sc. M.)--Brown University, 2021

Content

Citation

Carbonello, David J., "Prediction of Acute Kidney Injury in Hospitalized COVID-19 Patients Using Machine Learning" (2021). Biostatistics Theses and Dissertations. Brown Digital Repository. Brown University Library. https://repository.library.brown.edu/studio/item/bdr:ptez9c4h/

Relations

Collection:

Biostatistics Theses and Dissertations

Theses and Dissertations for the Biostatistics department.
...