Journal of thoracic disease · 2026
Construction and validation of a machine learning-based prediction model for in-hospital acute kidney injury in patients with lung cancer complicated with sepsis: clinical and nursing applications.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Zhang Huiping, Zhang Jinyan
Аннотация
BACKGROUND: Acute kidney injury (AKI) represents a critical complication in patients with lung cancer (LC) complicated by sepsis, exerting a substantial adverse impact on their prognosis. However, to date, no studies have been conducted to predict the risk of AKI in this specific patient population. The primary objective of this study is to develop and rigorously validate seven machine learning (ML) algorithms tailored to predict the probability of in-hospital AKI in patients with LC complicated by sepsis, while simultaneously evaluating their utility in clinical and nursing workflows.
METHODS: We retrieved retrospective clinical records pertaining to LC patients presenting with concurrent sepsis directly from the Medical Information Mart for Intensive Care (MIMIC)-IV dataset. All the independent variables were derived from the first measurement records within 24 h after the patient's admission, the AKI diagnosis period was from 24 h after admission until discharge. Subsequent to a rigorous data preprocessing phase, we employed a randomized splitting protocol to segregate the dataset into distinct training (70%) and testing (30%) cohorts. To authenticate model generalizability, an external validation phase was executed utilizing records from the electronic Intensive Care Unit Collaborative Research Database (eICU-CRD). Feature selection was accomplished through the deployment of both univariate and multivariate logistic regression (LR) methodologies. Seven ML models were developed model performance was evaluated using a comprehensive suite of metrics, prominently including the area under the curve (AUC), sensitivity and accuracy. Additionally, the optimal model was subjected to interpretability analysis utilizing Shapley additive explanations (SHAP).
RESULTS: In the MIMIC-IV dataset, 523 cases developed AKI, accounting for 79.00%. Among them, 362 cases were male (54.68%), with an average age of (standard deviation) 68.76 (10.89) years, and body mass index (BMI) was 31.34 (141.57) kg/m2, etc. Our analysis identified a distinct profile of independent predictors for in-hospital AKI within this population, specifically: age, body weight, prothrombin time (PT), Acute Physiology Score III (APSIII) score, calcium, glucose, sodium, and potassium. Of the seven ML architectures evaluated, the neural network (NNET) yielded the highest predictive performance, securing AUC values of 0.864 (training), 0.760 (testing), and 0.785 (validation).
CONCLUSIONS: By integrating variables such as age, weight, PT, APSIII score, and key electrolytes, the NNET model functions as a potential prognostic instrument for estimating in-hospital AKI risk in patients with LC complicated by sepsis.
METHODS: We retrieved retrospective clinical records pertaining to LC patients presenting with concurrent sepsis directly from the Medical Information Mart for Intensive Care (MIMIC)-IV dataset. All the independent variables were derived from the first measurement records within 24 h after the patient's admission, the AKI diagnosis period was from 24 h after admission until discharge. Subsequent to a rigorous data preprocessing phase, we employed a randomized splitting protocol to segregate the dataset into distinct training (70%) and testing (30%) cohorts. To authenticate model generalizability, an external validation phase was executed utilizing records from the electronic Intensive Care Unit Collaborative Research Database (eICU-CRD). Feature selection was accomplished through the deployment of both univariate and multivariate logistic regression (LR) methodologies. Seven ML models were developed model performance was evaluated using a comprehensive suite of metrics, prominently including the area under the curve (AUC), sensitivity and accuracy. Additionally, the optimal model was subjected to interpretability analysis utilizing Shapley additive explanations (SHAP).
RESULTS: In the MIMIC-IV dataset, 523 cases developed AKI, accounting for 79.00%. Among them, 362 cases were male (54.68%), with an average age of (standard deviation) 68.76 (10.89) years, and body mass index (BMI) was 31.34 (141.57) kg/m2, etc. Our analysis identified a distinct profile of independent predictors for in-hospital AKI within this population, specifically: age, body weight, prothrombin time (PT), Acute Physiology Score III (APSIII) score, calcium, glucose, sodium, and potassium. Of the seven ML architectures evaluated, the neural network (NNET) yielded the highest predictive performance, securing AUC values of 0.864 (training), 0.760 (testing), and 0.785 (validation).
CONCLUSIONS: By integrating variables such as age, weight, PT, APSIII score, and key electrolytes, the NNET model functions as a potential prognostic instrument for estimating in-hospital AKI risk in patients with LC complicated by sepsis.