RESEARCHERS have developed a machine learning-based algorithm to predict short-term survival outcomes in critically ill patients undergoing continuous renal replacement therapy (CRRT), a specialized form of dialysis used for patients too sick to tolerate regular haemodialysis.
CRRT, a gentler and continuous treatment, is often the last resort for patients with severe health conditions. However, nearly half of the adults placed on CRRT do not survive, raising concerns about the treatment’s efficacy and the emotional toll on patients’ families.
The new model aimed to address these concerns by providing clinicians with a data-driven tool to predict a patient’s likelihood of survival during or after CRRT. The model was trained using electronic health records from multiple institutions, analysing thousands of patients who were placed on CRRT. It achieved a strong predictive performance, with an area under the receiver operating curve (AUC) of 0.848, indicating high accuracy in distinguishing between patients who are likely to survive and those who are not.
The authors of the study emphasised the importance of this advancement, stating, “CRRT is often used as a last resort, but many patients do not survive it, leading to wasted resources and false hope for families. By making it possible to predict which patients will benefit, the model aims to improve patient outcomes and resource use.”
The model not only helped in decision-making but also shed light on key features influencing survival outcomes through feature importance, error, and subgroup analyses. These insights could guide future improvements in both the model and clinical practices.
While promising, the team noted that the model needs real-world testing in clinical trials to validate its effectiveness outside of the data it was trained on. The research has highlighted the potential of integrating machine learning into healthcare to enhance treatment outcomes and resource management, marking a significant step forward in personalised patient care.
Victoria Antoniou, EMJ
Reference
Zamanzadeh D et al. Data-driven prediction of continuous renal replacement therapy survival. Nat Commun. 2024;15(1):5440.
