Statistical methods for constructing disease comorbidity networks from longitudinal inpatient data.

Area of Science:

• Network science
• Medical informatics
• Computational epidemiology

Background:

• Understanding disease relationships is crucial for predicting health risks and disease progression.
• Comorbidity patterns, derived from large-scale hospital data, offer insights into these inter-disease linkages.
• Network construction is the foundational step for analyzing comorbidity networks.

Purpose of the Study:

• To provide an overview of statistical approaches for constructing weighted directed networks representing disease comorbidities.
• To compare different network construction methods based on their assumed null models and extracted features.
• To apply and evaluate these methods using a large-scale longitudinal hospital discharge dataset.

Main Methods:

• Overview of statistical network science approaches for weighted directed networks.
• Analysis of null models inherent in different network construction techniques.
• Application of selected methods to a large inpatient dataset (approx. 1 million individuals, 17 years) from Montreal.

Main Results:

• Demonstration of how different network construction methods yield distinct network structures.
• Identification of varied comorbidity relationship features extracted by each method.
• Comparison of the strengths and weaknesses of each approach in representing disease linkages.

Conclusions:

• The choice of network construction method significantly impacts the resulting disease network and extracted insights.
• Understanding these methodological differences is essential for accurate comorbidity analysis and risk prediction.
• The presented methods offer valuable tools for exploring complex disease relationships in large health datasets.