Data-driven phenotypes across the full AKI severity spectrum in patients admitted to the ICU

Acute kidney injury (AKI) is a frequent and heterogeneous complication among critically ill patients in the intensive care unit (ICU), often associated with adverse outcomes. This study aimed to identify phenotypic subtypes of ICU patients with AKI and to evaluate their association with clinical outcomes.

A secondary analysis was conducted using the MIMIC-IV database, including a cohort of adults with varying stages of AKI, as well as patients without AKI. Factorial analysis of mixed data, followed by hierarchical clustering, was used to identify patient phenotypes based on a wide range of clinical, demographic, laboratory, and treatment variables. Cluster profiling was conducted using a multivariable logistic regression model.

Among 1372 patients evenly distributed across stages 0 (non-AKI) to 3 (n=343 per stage), two distinct clusters were identified. Cluster 2 (n=671) had significantly higher in-hospital mortality (54.7% vs. 21.9%, p<0.001), and a greater prevalence of higher AKI stages (p<0.001). Moreover, cluster 2 showed a significantly greater frequency of sepsis, vasopressors and diuretics administration, chronic kidney disease, heart failure, and also higher respiratory and heart rate, and phosphorus. Patients in cluster 2 were a little younger and had a lower arterial O₂ pressure and blood pH. A logistic regression profiling model achieved an accuracy (95% CI) of 91.4% (89.8%, 92.8%) in predicting cluster assignment.

There are two clinically distinct phenotypes in patients admitted to the ICU concerning AKI with strong prognostic implications. The findings highlight the potential of routine ICU data to enable phenotype-based risk stratification in AKI.