Machine learning model predicts need for renal therapy in critically ill COVID-19 patients 

Even five years after the pandemic began, COVID-19 remains a global health concern. With new variants emerging and long-term effects still under investigation, the disease continues to challenge healthcare systems. 

In a study published in the Journal of Critical Care, researchers from the D’Or Institute for Research and Education (IDOR) and the Federal University of Rio de Janeiro (UFRJ) applied machine learning to predict which critically ill COVID-19 patients would require renal replacement therapy (RRT) during intensive care. 

The study aimed to develop and validate a model capable of identifying, upon ICU admission, which COVID-19 patients would need RRT — a therapy often required in cases of acute kidney injury, a common complication in severe COVID-19 that demands significant planning and resources. 

Using retrospective data from over 14,000 patients admitted to 126 ICUs within the Rede D’Or hospital network between February 2020 and December 2021, the team analyzed clinical and laboratory characteristics available within the first hours of ICU admission. Eight machine learning models were trained and tested, with the most accurate model selected for validation. 

Predicting Outcomes with Data 

The analysis focused on adult patients with confirmed COVID-19 who required respiratory support upon ICU admission. Variables included age, comorbidities such as hypertension and diabetes, frailty index, and a range of lab tests. 

The dataset was split into 80% for training and 20% for testing. Because most of the data were collected during critical phases of the pandemic—before widespread vaccination—the model was later validated with data from 946 patients admitted to ICUs between January and May 2022, reflecting changes due to new variants and immunization. 

Findings and Implications 

From 2020 to 2021, 13% of the patients analyzed required renal replacement therapy. These patients were generally older, had more comorbidities, and higher clinical severity. In the 2022 validation cohort, 11% needed RRT—demonstrating consistency in the model’s predictive performance. 

The results show that machine learning is a powerful tool for anticipating the need for renal therapy in critically ill COVID-19 patients. Early identification enables better planning and clinical decision-making. 

This research analyzed one of the largest multicenter cohorts of COVID-19 ICU patients to date, highlighting the value of integrating cutting-edge technology into clinical care. The model’s accuracy and practical applicability make it a valuable resource not only for managing COVID-19 but also for improving responses to future public health crises.