OBJECTIVES: Our primary aim was to examine non-traditional clinical phenotypes of Acute Kidney Injury (AKI) leading to Chronic Kidney Disease (CKD) using an unsupervised data mining and graphical network approach.

METHODS: A retrospective observational study of 90,602 critically ill adults from February 1, 2010, to June 30, 2022, was performed. Excluded patients were if they had renal replacement therapy prior to or had an AKI event less than 3 years prior to CKD diagnosis. AKI status was determined according to the 2012 Kidney Disease Improving Global Outcomes guidelines. Hospital-Acquired AKI (HA-AKI) and Community-Acquired AKI (CA-AKI) were defined as an AKI event within or 90-days after hospitalization prior to CKD diagnosis, respectively. A hierarchical clustering method was used to identify, and a graphical network was used to confirm phenotype clusters, respectively. A centrality metric of ‘betweenness’ was used to identify key nodes between components within the network.

RESULTS: Among 58,606 patients, 49, 229 (84.0%) were White, 30,475 female (52%), and had a mean age of 61 years. The non-AKI cohort had a higher number of comorbidities (mean 2.84 vs. 2.04; p < 0.05; 74.6% vs. 53.6%), respectively. The AKI cohort compared to the non-AKI cohort had a higher number of nodes and edges (71 vs. 59, 971 vs. 465), respectively. Common HA and CA-AKI comorbidity features were long-term opiate use, atelectasis, ischemic heart disease, lactic acidosis, and type 2 diabetes. A higher degree and betweenness for sepsis, tachycardia, dizziness, and type 2 diabetes were found in the HA-AKI cohort and hypoosmolality and hyponatremia, hypertensive heart disease with heart failure, COPD, and pleural effusion were found in the CA-AKI cohort.

CONCLUSIONS: The findings of our study suggest the incorporation of data mining methodologies may lead to the identification of otherwise unknown clinical phenotypes to reduce the transition of HA and CA-AKI to CKD.