Work

Predictive model for acute kidney injury in critical patients with pulmonary embolism based on machine learning: A multicenter clinical study

Year: 2026
Type: article
Study design: Observational Study
Abstract: Pulmonary embolism (PE) is recognized as an acute, critical cardiovascular condition, and acute kidney injury (AKI) is frequently observed as a complication in this clinical context. Currently, no predictive tools are available to accurately assess AKI risk in critical patients with PE. A predictive model was developed in this study to estimate AKI risk in critical PE. Data obtained from Medical Information Mart for Intensive Care IV were utilized for the development of training and internal validation sets, while data derived from Emergency Intensive Care Unit Collaborative Research Database were employed for external validation. Eight machine learning models were developed through least absolute shrinkage and selection operator regression and multivariate logistic analysis, and a web-based calculator was subsequently deployed. Model performance was evaluated by receiver operating characteristic curve analysis, calibration curve assessment, and decision curve analysis to determine discrimination, calibration accuracy, and clinical utility. Feature importance within the optimal performing models was quantified through Shapley Additive eXplanation analysis. Incremental predictive value relative to the pulmonary embolism severity index score was evaluated via net reclassification improvement and integrated discrimination improvement. A total of 2187 and 2183 patients diagnosed with critical PE were enrolled by 2 centers. In the training set, 1725 (78.88%) patients had AKI, whereas 1185 (54.28%) patients had AKI in the validation set. Hypertension, heart failure, pneumonia, sepsis, ventilation, respiratory rate, low-density lipoprotein cholesterol, partial pressure of oxygen, lactate, and Oxford Acute Severity of Illness Score were ultimately identified as risk factors for AKI in critical PE. The extreme gradient boosting model demonstrated the most stable discrimination, calibration, and clinical benefit across training (area under curve [AUC] = 0.825), internal validation (AUC = 0.771), and external validation (AUC = 0.757) sets. Shapley Additive eXplanation analysis of the optimal model revealed that Oxford Acute Severity of Illness Score, low-density lipoprotein cholesterol, and partial pressure of oxygen contributed substantially to AKI risk. Relative to the pulmonary embolism severity index model, the proposed model exhibited superior performance in predicting AKI. A machine learning prediction model that integrates clinical indicators with disease activity scores was constructed and externally validated to estimate the risk of AKI in critical PE. A web-based calculator was subsequently developed to enable clinicians to identify high-risk cohorts and furnish evidence-based guidance for early intervention and individualized management.
Source: Medicine
Language: English
Subfield: Nephrology
Field: Medicine
Open Access status: gold
Estimated APC paid: $1,950
OpenAlex ID: W7225994520
FWCI:
Cites: 38
Cited by:
Related to: 10