Figure 8
From: Modeling electronic health record data using an end-to-end knowledge-graph-informed topic model
Clustering of EHR codes based on their learned embedding \(\varvec{\uprho }\) by our GAT-ETM. t-SNE was applied to the embedding to reduce their dimensions from L to 2 to allow visualization of the code clustering. As shown in the legend, shape \(+\) and \(\times\) indicate ICD and ATC code, respectively; colors indicate different high-level categories. Aligned ICD and ATC categories are assigned identical or similar colors. Within ICD/ATC vocabularies, nodes of the same category are grouped together. Each group was circled and labeled with abbreviations. ICD and ATC group names are shown in regular and italic fonts, respectively.
