Constructing Disease Comorbidity Networks Using Statistical Methods from Electronic Health Records

Disease comorbidity is a prevalent and significant concern in modern medical practice. Patients in such circumstances will need specialized medical interventions. This study proposes a network-based computation model to extract the disease comorbidity patterns by utilizing the electronic health records of a “Diabetes 130-US hospitals for years 1999-2008” dataset. The estimate of disease-disease relationships using statistical methods is a crucial step in this paradigm. A Disease Comorbidity Network (DCN) was constructed based on these relationships. This model can then be used to extract the corresponding illness comorbidity patterns for various races and genders, using the endocrine disease (diabetes) as an example to determine the associations between various diseases and their comorbidities. The outcomes of this research demonstrated that the strength of some interesting and or clinically relevant disease associations (Example: musculoskeletal disorder and endocrine disorder, nervous disorder and endocrine disorder) is stronger in male patients than in female patients. Also, observed that the endocrine-specific comorbidity patterns will vary based on the race and gender of a patient. Finally, extracted comorbidities patterns are validated with medical literature as well as with the state of art method. Notably, this study exploits the network-level relationships between diseases, which will improve our perception about disease associations at the level of the patient population.