Acute kidney injury phenotypes: clinical reality or statistical artifact?

Acute kidney injury (AKI) is a common syndrome associated with poor outcomes; however, definitions based only on serum creatinine and urine output obscure patient-level differences relevant to prognosis and treatment. Sub-phenotyping strategies seek to resolve this heterogeneity. We evaluate recent advances across sub-phenotyping strategies, including creatinine-trajectory modeling, biomarker analysis, and EHR data-driven strategies, assessing reproducibility, biological plausibility, incremental value over AKI staging, and clinical utility.

Sub-phenotyping has advanced across all strategies. Creatinine-trajectory studies have applied more rigorous longitudinal methods and consistently link nonresolving or worsening trajectories to adverse outcomes, independent from AKI stage. Biomarker-based work has moved toward consensus recommendations for profiling, reproduced endothelial-inflammatory subtypes across independent cohorts, and has demonstrated in trials that biomarker-positive patients can be targeted with a preventive care bundle. Data-driven approaches using routine EHR data have reproducibly recovered a high-acuity sepsis-AKI phenotype and increasingly apply causal methods to estimate treatment-effect heterogeneity.

The convergence of similar sub-phenotypes across independent methods and cohorts suggests some groupings reflect real biology rather than statistical artifact. Realizing their clinical value will require linking these classifications to targeted interventions, advancing AKI sub-phenotyping from descriptive categorization toward precision nephrology.