Abstract

Objectives: Focal segmental glomerulosclerosis (FSGS) exhibits substantial heterogeneity in clinical and histologic presentation. This study aimed to explore latent patient subgroups through unsupervised machine learning and assess whether these clusters offer meaningful stratification beyond conventional histopathological classifications. Methods & Results: A retrospective cohort of 57 histologically confirmed FSGS patients was studied. Of these, 30 patients with complete biochemical data were included in the clustering analysis. Immune complex deposition was analyzed in the full 57-patient cohort. Thirty (30) FSGS patients was analyzed using four quantitative clinical variables: age, proteinuria, serum albumin, and serum creatinine. Principal component analysis (PCA) followed by k-means clustering (k=3) identified three distinct patient clusters. Cluster 0 comprised older patients with moderate proteinuria and hypoalbuminemia but the highest serum creatinine, suggesting chronic renal dysfunction. Cluster 1 featured the youngest patients with mild proteinuria and preserved renal function, consistent with early or less aggressive disease. Cluster 2 exhibited the most severe nephrotic phenotype with high proteinuria and hypoalbuminemia but moderate renal impairment. Histopathological parameters, including glomerular sclerosis and tubulointerstitial injury, showed only weak associations with biochemical severity. Additionally, patients with C3 deposition demonstrated significantly higher serum creatinine compared to those without (p = 0.031), implicating complement activation in renal injury. Conclusion: Unsupervised clustering uncovered three clinically meaningful FSGS phenotypes that may reflect disease stage and severity more accurately than histologic scoring alone. These findings support the utility of data-driven patient stratification in augmenting risk assessment and guiding therapeutic strategies.