Fig. 1 | Nature Communications

Fig. 1

From: Integrating biomedical research and electronic health records to create knowledge-based biologically meaningful machine-readable embeddings

Fig. 1

Embedding EHR concepts in a knowledge network. a Distribution of patient BMIs at UCSF. Four BMI cohorts were created using the natural boundaries of the BMI distribution (boxes I–IV: <18, 18–24.5, 24.6–29.5, and >29.6). Arrows at the bottom correspond to the BMIs that separate the standardize weight classes. b Step 1: find the overlapping concepts between SPOKE and the patient data (EHRs). These are called SPOKE Entry Points (SEPs). Step 2: choose any code or concept in the EHR to make cohort. Here, we have chosen patients with a high BMI (Cohort IV). Then connect each patient in the cohort to all of the SEPs in their records. Step 3: perform PageRank such that the walker restarts in the patient cohort. Iterate until desired threshold is reached. Step 4: final node ranks are then used to create the weights in the Propagated SPOKE Entry Vector (PSEV)

Back to article page