Key result
Machine learning models, particularly the Extreme Gradient Boosting Machine, outperformed traditional clinical scores in predicting intradialytic hypotension with an AUROC of 0.828.
Why the study?
Predictors of intradialytic hypotension in critically ill patients undergoing kidney replacement therapy for acute kidney injury needed to be identified.
Do predictive models and clinical indicators accurately forecast intradialytic hypotension in critically ill adults undergoing kidney replacement therapy for acute kidney injury?
Systematic Review (n=2,873)
Do predictive models and clinical indicators accurately forecast intradialytic hypotension in critically ill adults undergoing kidney replacement therapy for acute kidney injury?
Effect estimate: AUROC 0.828 (95% CI 0.796-0.861)
Machine learning models and specific clinical indices like the SOCRATE score offer superior predictive accuracy for intradialytic hypotension in critically ill AKI patients compared to traditional clinical scoring systems.
Machine learning models and SOCRATE score predict IDH in AKI-KRT; supports proactive risk stratification but leaves validation and outcome impact open.
BACKGROUND: This systematic review aims to identify predictors of intradialytic hypotension (IDH) in critically ill patients undergoing kidney replacement therapy (KRT) for acute kidney injury (AKI). METHODS: A comprehensive search of PubMed was conducted from 2002 to April 2024. Studies included critically ill adults undergoing KRT for AKI, excluding pediatric patients, non-critically ill individuals, those with chronic kidney disease, and those not undergoing KRT. The primary outcome was identifying predictive tools for hypotensive episodes during KRT sessions. RESULTS: The review analyzed data from 8 studies involving 2873 patients. Various machine learning models were assessed for their predictive accuracy. The Extreme Gradient Boosting Machine (XGB) model was the top performer with an area under the receiver operating characteristic curve (AUROC) of 0.828 (95% CI 0.796-0.861), closely followed by the deep neural network (DNN) with an AUROC of 0.822 (95% CI 0.789-0.856). All machine learning models outperformed other predictors. The SOCRATE score, which includes cardiovascular SOFA score, index capillary refill, and lactate level, had an AUROC of 0.79 (95% CI 0.69-0.89, p < 0.0001). Peripheral perfusion index (PPI) and heart rate variability (HRV) showed AUROCs of 0.721 (95% CI 0.547-0.857) and 0.761 (95% CI 0.59-0.887), respectively. Pulmonary vascular permeability index (PVPI) and mechanical ventilation also demonstrated significant diagnostic performance. A PVPI ≥ 1.6 at the onset of intermittent hemodialysis (IHD) sessions predicted IDH associated with preload dependence with a sensitivity of 91% (95% CI 59-100%) and specificity of 53% (95% CI 42-63%). CONCLUSION: This systematic review shows how combining predictive models with clinical indicators can forecast IDH in critically ill AKI patients undergoing KRT, with validation in diverse settings needed to improve accuracy and patient care strategies.
No takes yet. Share an insight, caveat, or question.
Lyrio et al. (2024) conducted a systematic review in Acute kidney injury requiring kidney replacement therapy in critically ill patients (n=2,873). Machine learning models and clinical scores (e.g., XGB, SOCRATE) vs. Traditional clinical scores (e.g., SOFA, APACHE II) was evaluated on Predictive accuracy (AUROC) for hypotensive episodes during KRT sessions (AUROC 0.828, 95% CI 0.796-0.861). Machine learning models, particularly the Extreme Gradient Boosting Machine, outperformed traditional clinical scores in predicting intradialytic hypotension with an AUROC of 0.828.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: