To enhance phenotype definition, we are developing a tool to recommend new codes to be added to the concept set. Based on existing phenotypes and the specific database to be used, the tool suggests additional codes that might be relevant but were not initially considered.
Currently, the tool uses two methods to look for recommended codes and allows you to visualize the utility of the suggested codes using Euler diagrams and UpSet plots. The tool is being tested internally at J&J (together with @Julia_Rodrigues & @Marcela ), and we plan to share it soon with the OHDSI community. If the tool proves useful, we aim to integrate it into the CohortDiagnostics package.
The new functionalities will include:
Pearson’s Correlation Calculation: This provides the Pearson’s correlation between each code from the concept set and the 30 most frequent codes within the cohort, considering the persons’ entire available history. The results will be displayed in a heat map similar to the example shown, where a new code for Chronic Congestive Heart Failure was detected to be potentially useful for the heart failure phenotype in Optum EHR.
Code Recommendation Based on Proportion Changes: For each code, it computes the proportion of patients selected for the phenotype with at least one time the code recorded in all their available history relative to the proportion with at least one time the code recorded in the total population. This indicates how much more frequently the code appears in the phenotype compared to the total population.
Visualization of the Recommended Codes: Recommended codes, based on the previous two methods and the orphan concepts from CohortDiagnostics, will be visualized. The codes will appear in tables, and users can select them to create Euler and UpSet plots that show patient overlap between the cohort and the selected recommended codes.
In the presented example for a dementia phenotype tested over Optum EHR, two of the detected codes using the proportion changes are observed to overlap with the cohort.
Eager to get feedback from the OHDSI community as soon as we can share the Code Recommender tool!