Network Analysis of Human Disease Comorbidity Patterns Based on Large-Scale Data Mining

Disease comorbidity is an important aspect of phenotype associations and reflects overlapping pathogenesis between diseases. Existing comorbidity studies usually focused on specific diseases and patient populations. In this study, we systematically mined and analyzed disease comorbidity patterns without restricting disease types and patient populations. We presented a data mining approach and extracted comorbidity patterns from a patient-disease database in the drug adverse event reporting system. The database contains records of 3,354,043 patients. We first demonstrated that the data are not severely biased towards specific patient populations and valuable for comorbidity mining. Then we developed an automatic pipeline to process the data, and applied an association rule mining algorithm to mine comorbidity relationships among multiple diseases. Our approach extracted 8,576 comorbidity patterns for 613 diseases. We constructed a disease comorbidity network from these patterns and demonstrated that the comorbidity clusters reflect genetic associations between diseases. Different from previous studies based on relative risk, which tends to identify comorbidities for rare diseases, our approach extracted many patterns for common diseases. We applied the approach on colorectal cancer, and found interesting relationships between colorectal cancer and metabolic disorders, which may lead to promising pathogenesis discoveries.