Precise electronic phenotyping is critical for leveraging clinical databases. We identified a condition to extract cases of heart failure (HF) accurately using machine learning. Using a large clinical database in Japan, we retrieved 200 potential cases with HF and verified the actual cases by reviewing electronic medical records. In the initial condition, the precision rate was 0.350 and the recall rate was 0.912. In the additional condition using machine learning, the precision rate improved to 0.878.

Learning Objective: Learn challenges and possible solutions in precise electronic phenotyping from a large amount of clinical data

Authors:

Masaharu Nakayama (Presenter)
Tohoku University Hospital

Presentation Materials:

• Abstract/Manuscript - 1