Development A Novel Classification Based on Serum Sodium Level Integrated with Comorbid Conditions (BASIC) in Hyponatremia Patients Via Data-Driven Cluster Analysis.

Journal: The Tohoku journal of experimental medicine
Published Date:

Abstract

Hyponatremia was defined as a serum sodium level below 136 mmol/L. The current classification identified serum sodium level below 120 mmol/L as high-risk patients, which ignored the impact of comorbidities on outcomes. We aim to develop a novel classification based on the integration of serum sodium level and comorbid conditions to aid precise clinical management. Data were collected from patients admitted to Peking Union Medical College Hospital. Non-supervised machine learning method TwoStep and k-means analysis were used to classify patients into homozygous clusters. Decision tree modelling derived a clinical classifier from the novel classification, which was further validated in an internal dataset. During the development of novel classification based on serum sodium level Integrated with comorbid conditions (BASIC), TwoStep and k-means analysis yielded similar results. Cluster analysis identified 5 stable and replicable clusters which had distinct clinical characters: (i) mild hyponatremia and at least one comorbidity; (ii) mild hyponatremia and high prevalence of cancer; (iii) younger patients with mild hyponatremia; (iv) mild hyponatremia, poor renal function, and high prevalence of chronic kidney disease; (v) moderate and severe hyponatremia patients with other characters similar to Cluster 1. BASIC is a machine learning originated best cluster. However, it lacked specific classification criteria. Therefore, we developed a simple and practical classifier, BASIC-C, which assigned patients to their corresponding subgroups with a combined accuracy of 93.1% (95% CI: 91.5%-94.7%) in the validation cohort. The outcomes of identified subgroups were similar to their corresponding clusters in BASIC. We proposed BASIC, a novel classification of hyponatremia via data-driven cluster analysis, and derived a practical clinical classifier, BASIC-C. Our results revealed that serum sodium of 130-135 mmol/L increased in-hospital mortality when patients were combined with cancer or chronic kidney disease, and it was necessary to provide careful management.

Authors

Keywords

No keywords available for this article.