Machine Learning for Predicting Risk and Prognosis of Acute Kidney Disease in Critically Ill Elderly Patients During Hospitalization: Internet-Based and Interpretable Model Study.

BACKGROUND: Acute kidney disease (AKD) affects more than half of critically ill elderly patients with acute kidney injury (AKI), which leads to worse short-term outcomes. OBJECTIVE: We aimed to establish 2 machine learning models to predict the risk and prognosis of AKD in the elderly and to deploy the models as online apps. METHODS: Data on elderly patients with AKI (n=3542) and AKD (n=2661) from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database were used to develop 2 models for predicting the AKD risk and in-hospital mortality, respectively. Data collected from Xiangya Hospital of Central South University were for external validation. A bootstrap method was used for internal validation to obtain relatively stable results. We extracted the indicators within 24 hours of the first diagnosis of AKI and the fluctuation range of some indicators, namely delta (day 3 after AKI minus day 1), as features. Six machine learning algorithms were used for modeling; the area under the receiver operating characteristic curve (AUROC), decision curve analysis, and calibration curve for evaluating; Shapley additive explanation (SHAP) analysis for visually interpreting; and the Heroku platform for deploying the best-performing models as web-based apps. RESULTS: For the model of predicting the risk of AKD in elderly patients with AKI during hospitalization, the Light Gradient Boosting Machine (LightGBM) showed the best overall performance in the training (AUROC=0.844, 95% CI 0.831-0.857), internal validation (AUROC=0.853, 95% CI 0.841-0.865), and external (AUROC=0.755, 95% CI 0.699-0.811) cohorts. In addition, LightGBM performed well for the AKD prognostic prediction in the training (AUROC=0.861, 95% CI 0.843-0.878), internal validation (AUROC=0.868, 95% CI 0.851-0.885), and external (AUROC=0.746, 95% CI 0.673-0.820) cohorts. The models deployed as online prediction apps allowed users to predict and provide feedback to submit new data for model iteration. In the importance ranking and correlation visualization of the model's top 10 influencing factors conducted based on the SHAP value, partial dependence plots revealed the optimal cutoff of some interventionable indicators. The top 5 factors predicting the risk of AKD were creatinine on day 3, sepsis, delta blood urea nitrogen (BUN), diastolic blood pressure (DBP), and heart rate, while the top 5 factors determining in-hospital mortality were age, BUN on day 1, vasopressor use, BUN on day 3, and partial pressure of carbon dioxide (PaCO2). CONCLUSIONS: We developed and validated 2 online apps for predicting the risk of AKD and its prognostic mortality in elderly patients, respectively. The top 10 factors that influenced the AKD risk and mortality during hospitalization were identified and explained visually, which might provide useful applications for intelligent management and suggestions for future prospective research.

Development and deployment of interpretable machine-learning model for predicting in-hospital mortality in elderly patients with acute kidney disease.

BACKGROUND: Acute kidney injury (AKI) is more likely to develop in the elderly admitted to the intensive care unit (ICU). Acute kidney disease (AKD) affects ∼45% of patients with AKI and increases short-term mortality. However, there are no studies on the prognosis of AKD in the elderly. METHODS: Data from 2666 elderly patients with AKD in the Medical Information Mart for Intensive Care IV were used for model development and 535 in the eICU Collaborative Research Database for external validation. Based on 5 machine learning algorithms, 33 noninvasive parameters were extracted as features for modeling. RESULTS: In-hospital mortality of AKD in the elderly was 29.6% and 31.8% in development and validation cohorts, respectively. The comprehensive best-performing algorithm was the support vector machine (SVM), and a simplified online application included only 10 features employing SVM (AUC: 0.810 and 0.776 in the training and external validation cohorts, respectively) was deployed. Model interpretation by SHapley Additive exPlanation (SHAP) values revealed that the difference (AKD day - ICU day) in sequential organ failure assessment (delta SOFA), Glasgow coma scale (GCS), delta GCS, delta peripheral oxygen saturation (SpO2), and SOFA were the top five features associated with prognosis. The optimal target was determined by SHAP values from partial dependence plots. CONCLUSIONS: A web-based tool was externally validated and deployed to predict the early prognosis of AKD in the elderly based on readily available noninvasive parameters, assisting clinicians in intervening with precision and purpose to save lives to the greatest extent.

Establishment and external validation of an online dynamic nomogram for predicting in-hospital death risk in sepsis-associated acute kidney disease.

OBJECTIVES: Approximately one-third of patients with sepsis-associated acute kidney injury (AKI) progress to acute kidney disease (AKD) with higher short-term mortality. We aimed to identify the clinical characteristics that influence in-hospital death in sepsis-associated AKD and develop a nomogram to facilitate early warning. METHODS: Logical regression was applied to screen variables based on clinical data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. A nomogram was established to predict in-hospital death risk in patients with sepsis-associated AKD. The eICU Collaborative Research Database (eICU-CRD) was used for external validation. The receiver operating characteristic and calibration curves were used to determine the model's performance. RESULTS: A total of 1,779 patients with sepsis-associated AKD were included from the MIMIC-IV and 344 from the eICU-CRD. Age, Glasgow coma scale score, systolic blood pressure, peripheral oxygen saturation, platelet count, white blood cell count, and bicarbonate levels were significantly correlated with death. The nomogram demonstrated high discrimination in the training (C-index, 0.829; 95% confidence interval [CI] [0.807-0.852]) and testing sets (C-index: 0.760; 95% CI [0.706-0.814]). At the optimal cut-off value of 0.270, the model's sensitivity in the training and validation datasets was 72.8% (95% CI [68.3-76.9%]) and 64.5% (95% CI [54.9-73.4%]), while the specificity was 79.2% (95% CI [76.9-81.4%]) and 74.8% (95% CI [68.7-80.2%]), respectively. CONCLUSION: We identified seven predictors of in-hospital death in patients with sepsis-associated AKD. In addition, we developed an online dynamic nomogram to accurately and conveniently predict short-term outcomes, which performed well in the external dataset.