JMIR Publications

ABSTRACT

Background:

Research assessing the risk of acute kidney disease (AKD) following nephrectomy (NX) is limited, and the causal relationship between it and chronic kidney disease (CKD) is unclear. It was always difficult to explain how clinical features affect the prognosis of patients with NX, while interpretable machine learning algorithms give a way that may assist in transparently identifying the risk of outcomes.

Objective:

Our objective was to deploy interpretable machine learning models and subsequently elucidate their decision-making process in identifying patients at risk of AKD and CKD following NX.

Methods:

We conducted a retrospective cohort study involving patients who underwent NX between July 2012 and June 2019. Patient data were randomly split into training, validation, and test sets, maintaining a ratio of 76.5: 8.5: 15. Eight machine learning algorithms were employed to construct predictive models for postoperative AKD and CKD. The performance of the best-performing models was assessed using various metrics. To improve model interpretation, SHAP values were used to generate summary plots, force plots, interaction plots, and decision plots. Additionally, we developed an online prediction tool utilizing the top ten most important features for both AKD and CKD prediction.

Results:

The study cohort comprised 1559 patients. Incidence rates for AKI, AKD, and CKD were 21.7%, 15.3%, and 10.6%, respectively. Among the evaluated machine learning models, the LightGBM model demonstrated superior performance, with an average AUROC of 0.955 for AKD prediction and 0.926 for CKD prediction. Performance metrics and plots highlighted the model's competence in discrimination, calibration, and clinical applicability. Operative duration, hemoglobin levels, blood loss, urine protein, and AKI grade were identified as the top five features associated with AKD. Baseline eGFR, age, pathology, renal function trajectories, and total bilirubin were the top five features associated with CKD. Additionally, we developed a web application utilizing the LightGBM model to estimate AKD and CKD risks.

Conclusions:

An interpretable machine learning model effectively elucidated its decision-making process in identifying patients at risk of AKD and CKD following NX by enumerating critical features. The web-based calculator, founded on the LightGBM model, can assist in formulating more personalized and evidence-based clinical strategies.