Machine Learning Consensus Clustering Approach for Hospitalized Patients with Phosphate Derangements

Area of Science:

• Nephrology
• Internal Medicine
• Data Science

Background:

• Abnormal serum phosphate levels are common in hospitalized patients.
• Understanding distinct patient phenotypes is crucial for risk stratification.

Purpose of the Study:

• To categorize hospitalized patients with hypophosphatemia and hyperphosphatemia into distinct clusters using unsupervised machine learning.
• To assess the mortality risk associated with these identified clusters.

Main Methods:

• Consensus clustering applied to demographic, comorbidity, diagnosis, and laboratory data.
• Analysis of hypophosphatemia (serum phosphate ≤ 2.4 mg/dL) and hyperphosphatemia (serum phosphate ≥ 4.6 mg/dL) cohorts.
• Standardized mean difference used to identify key cluster features and assess mortality associations.

Main Results:

• Two distinct clusters were identified in both hypophosphatemia and hyperphosphatemia cohorts.
• Hypophosphatemia Cluster 2 (older, higher comorbidity, lower eGFR, more AKI) showed higher one- and five-year mortality.
• Hyperphosphatemia Cluster 2 (older, kidney disease focus, hypertension, end-stage kidney disease, AKI, higher potassium/magnesium/phosphate) exhibited higher hospital, one-, and five-year mortality.

Conclusions:

• Machine learning successfully classified distinct clinical phenotypes in patients with serum phosphate derangements.
• Age, comorbidities, and kidney function were key differentiating features of these phenotypes.
• Phenotype identification is linked to differential mortality risks, informing clinical management.