Machine Learning Consensus Clustering Approach for Hospitalized Patients with Dysmagnesemia

Area of Science:

• Internal Medicine
• Biostatistics
• Medical Informatics

Background:

• Serum magnesium derangement (dysmagnesemia) is common in hospitalized patients.
• Understanding distinct patient phenotypes and their associated mortality risks is crucial for effective clinical management.
• Unsupervised machine learning offers a novel approach to identify these phenotypes.

Purpose of the Study:

• To classify hospitalized patients with hypomagnesemia and hypermagnesemia into distinct clusters using machine learning.
• To evaluate the differential mortality risks associated with these identified clusters.

Main Methods:

• Consensus cluster analysis was applied to demographic, diagnostic, comorbidity, and laboratory data.
• Analysis was conducted separately for hypomagnesemia (serum magnesium ≤ 1.6 mg/dL) and hypermagnesemia (serum magnesium ≥ 2.4 mg/dL) cohorts.
• Associations between clusters and hospital/one-year mortality were assessed.

Main Results:

• In hypomagnesemia (n=13,320), three clusters emerged: Cluster 1 (high comorbidity, low magnesium), Cluster 2 (youngest, low comorbidity, high kidney function), and Cluster 3 (oldest, low kidney function). Clusters 1 and 3 showed higher mortality than Cluster 2.
• In hypermagnesemia (n=4671), two clusters were identified. Cluster 2 exhibited older age, higher comorbidity, more kidney disease admissions, acute kidney injury, and lower kidney function compared to Cluster 1.
• Cluster 2 in the hypermagnesemia cohort was associated with increased hospital and one-year mortality.

Conclusions:

• Machine learning-based cluster analysis successfully identified clinically distinct phenotypes of dysmagnesemia.
• These phenotypes are associated with significantly different mortality risks in hospitalized patients.
• Further research is warranted to explore the clinical utility of this machine learning approach in managing dysmagnesemia.