Inspiration
Disease comorbidity prediction has acquired the consideration of numerous scientists in the previous years. Mass production of clinical information as electronic well-being records and natural information made way to investigate disease affiliations and comorbidity designs. This prompted the advancement of scientific devices to recognize disease comorbidities and examine their causal hereditary source.
What it does
We present a clever methodology considering connection prediction and contingent likelihood calculation for anticipating comorbidities patients might insight in the wake of recuperating from a specific disease utilizing clinical patient information. Results showed that the framework outflanked existing frameworks in disease comorbidity prediction.
How we built it
The dataset utilized for prediction is MIMIC-III (Medical Information Mart for Intensive Care), an enormous, single-focus data set comprising data connecting with patients owned up to basic consideration units at a huge tertiary consideration clinic. The dataset consists of 112,000 clinical reports records (normal length 709.3 tokens) and 1,159 high-level ICD-9 codes on average. Out of all these records we only use Patient Details, Patient Problems, and Problems from Doctor Notes. We create connections between the Common Unique Identifier (CUID) and the illness name.Using the conditional probability approach, we find the number of patients who had a certain disease from their CUIDs and then map it with the list of other comorbidities that patients experienced. We continue doing the same for all the patients with the same CUIDs and then map them with all the patients who experienced the same symptoms after recovery
Challenges we ran into
Getting permissions and real patient records
Accomplishments that we're proud of
We constructed a disease comorbidity prediction network derived from a large number of electronic clinical records with symptomatic codes from the MIMIC dataset and tracked down fascinating topological examples for this network. Besides, we distinguished clinically significant disease comorbidity networks considering patient similitude.
What we learned
New technologies such as Neo4j and Graph database
What's next for PostRecCom
By figuring out the illness comorbid directions into a parallel characterization issue, we can research the practicality of predicting the disease comorbidities utilizing our proposed algorithm and accomplish promising outcomes.
Log in or sign up for Devpost to join the conversation.