JMIR Publications

ABSTRACT

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 two machine learning models to predict risk and prognosis of AKD in the elderly, and to deploy the models as online applications.

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 two models for predicting the risk and in-hospital mortality of AKD, respectively. Data collected from Xiangya Hospital Central South University were for external validation. We extracted the indicators within 24 hours of the first diagnosis of AKI and the fluctuation range of some indicators, namely Delta (the third day after AKI minus the first day) as features. Six machine-learning algorithms were used for modeling; the areas under the receiver operating characteristic curve (AUC), decision curve analysis, and calibration curve for evaluating; SHapley Additive exPlanations (SHAP) for visually interpreting; and Heroku platform for deploying the best-performing models as web-based tools.

Results:

For the model of predicting the risk of AKD in elderly with AKI during hospitalization, the Light Gradient Boosting Machine (LightGBM) showed the best overall performance in the training (AUC: 0.844, 95% CI: 0.831–0.857) and external cohort (AUC: 0.755, 0.699–0.811). As well, LightGBM performed well for the AKD prognostic prediction in the training (AUC: 0.861, 0.843–0.878) and validation cohort (AUC: 0.746, 0.673–0.820). We deployed the above models as online tools. Based on the SHAP value, the importance ranking and correlation visualization of the top 10 influencing factors of the models were conducted, and partial dependence plots revealed the optimal cutoff of some interventionable indicators.

Conclusions:

We developed and externally validated two online tools for predicting the risk of AKD and its prognosis in elderly patients, respectively. There were identified and explained visually the top 10 factors that influenced the risk and outcome of AKD during hospitalization, which might provide useful tools for intelligent management and suggestions for future prospective research.