Clustering datasets with demographics and diagnosis codes.

Area of Science:

• Health Informatics
• Data Science
• Computational Biology

Background:

• Clustering Electronic Health Record (EHR) data is crucial for understanding patient clinical profiles and aiding downstream analysis like classification.
• The heterogeneity of EHR data presents significant challenges for traditional clustering algorithms, necessitating novel approaches.
• Existing methods struggle with the complex, multi-faceted nature of patient data, including demographic information and diagnosis codes.

Purpose of the Study:

• To propose and evaluate a novel clustering approach specifically designed for heterogeneous EHR data.
• To enable the discovery of relationships between patient demographics and diagnosis codes within large datasets.
• To provide an efficient and scalable method for patient data clustering as a preprocessing step for analysis.

Main Methods:

• Representing EHR data in a binary format, incorporating selected demographic values.
• Identifying and utilizing combinations of frequent and correlated diagnosis codes as features.
• Employing cosine similarity for measuring record similarity in the binary representation.
• Applying hierarchical clustering to identify compact and well-separated patient clusters.

Main Results:

• The proposed approach successfully constructed clusters exhibiting correlated demographics and diagnosis codes.
• Experiments on two large, publicly available EHR datasets (over 26,000 and 52,000 records) validated the method's effectiveness.
• The clustering approach demonstrated efficiency and scalability, suitable for large-scale EHR data analysis.

Conclusions:

• The novel binary representation and hierarchical clustering method effectively addresses the challenges of EHR data heterogeneity.
• This approach facilitates the discovery of meaningful patient subgroups based on integrated demographic and diagnostic information.
• The method offers a scalable and efficient solution for preprocessing EHR data, improving subsequent analytical tasks.