Abstract: FR-PO026
Interpretable Machine Learning-Based Individual Analysis of AKI in Immune Checkpoint Inhibitor Therapy
Session Information
- AI, Digital Health, Data Science - II
November 03, 2023 | Location: Exhibit Hall, Pennsylvania Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Augmented Intelligence, Digital Health, and Data Science
- 300 Augmented Intelligence, Digital Health, and Data Science
Authors
- Sakuragi, Minoru, Department of Biomedical Data Intelligence, Graduate School of Medicine, Kyoto University, Kyoto, Japan
- Uchino, Eiichiro, Department of Biomedical Data Intelligence, Graduate School of Medicine, Kyoto University, Kyoto, Japan
- Sato, Noriaki, Department of Biomedical Data Intelligence, Graduate School of Medicine, Kyoto University, Kyoto, Japan
- Matsubara, Takeshi, Department of Nephrology, Graduate School of Medicine, Kyoto University, Kyoto, Japan
- Kojima, Ryosuke, Department of Biomedical Data Intelligence, Graduate School of Medicine, Kyoto University, Kyoto, Japan
- Yanagita, Motoko, Department of Nephrology, Graduate School of Medicine, Kyoto University, Kyoto, Japan
- Okuno, Yasushi, Department of Biomedical Data Intelligence, Graduate School of Medicine, Kyoto University, Kyoto, Japan
Background
Acute kidney injury (AKI) is a critical complication in immune checkpoint inhibitor (ICI) therapy. Since the etiologies of AKI in cancer therapy vary among patients, clarifying AKI causes in individuals is critical for optimal cancer treatment. Although it is essential to individually analyze ICI-treated patients for underlying pathologies existing behind each AKI onset occurring at different times, these analyses have not been realized with conventional clinical research methods.
Methods
We created a dataset from the electronic medical records (EMR) of 616 patients who received ICI therapy at the Kyoto University Hospital from July 2014 to September 2019. AKI was defined by serum creatinine changes based on the KDIGO guideline. We developed a gradient-boosting decision tree-based machine learning (ML) model continuously predicting AKI within 7 days from each time point, using 287 clinical variables obtained from EMR as input features. We noted that the temporal changes in individual predictive reasoning in AKI prediction models represented the key features contributing to each AKI prediction, and clustered AKI patients based on the pattern of features with high predictive contribution quantified in time-series by SHapley Additive exPlanations (SHAP), a model interpretation framework. We searched for common clinical backgrounds of AKI patients in each cluster, compared with annotation by three nephrologists.
Results
One hundred and twelve patients (18.2%) had at least one AKI episode. They were clustered per key features and their SHAP value patterns, and the nephrologists assessed the clusters’ clinical relevance. Receiver operating characteristic analysis revealed that the area under the curve was 0.880. Patients with AKI were categorized into four clusters with significant prognostic differences (p=0.010). The leading causes of AKI for each cluster, such as hypovolemia, drug-related, and cancer cachexia, were all clinically interpretable, which conventional approaches cannot obtain.
Conclusion
Our results enabled us to clarify the background of AKI development in ICI-treated patients with complicated AKI risks and suggested the potential for applying ML prediction models as interpretable artificial intelligence to medical care, which had been a challenge to explainability.