Data-driven identification of complex disease phenotypes.

Area of Science:

• Network science
• Machine learning
• Computational phenotyping
• Medical informatics

Background:

• Disease interactions in multimorbid patients are poorly understood, impacting treatment and prognosis.
• Current methods often fail to capture complex, synergistic disease relationships.

Purpose of the Study:

• To develop a transparent method for assessing interactions between multiple diseases.
• To characterize patient health states using higher-order disease interaction features.
• To identify distinct disease phenotypes and their underlying etiologies.

Main Methods:

• Applied network science, machine learning, and computational phenotyping.
• Analyzed a population-wide medical claims dataset (N=9 million).
• Constructed a generalized disease network linking higher-order diagnosis features.

Main Results:

• Health states are better characterized by including interactions among more than two diseases.
• Identified synergistic disease interactions using higher-order diagnosis features.
• Data-driven detection of two distinct obesity phenotypes: metabolically healthy and unhealthy.
• Demonstrated progression from metabolically healthy to unhealthy obesity over time.

Conclusions:

• Higher-order disease interactions provide a more comprehensive understanding of patient health.
• The generalized disease network facilitates the identification of distinct disease phenotypes.
• The findings offer insights into the dynamic nature of obesity phenotypes and their clinical implications.