Fig. 3: Prioritize patients by disease mixture from MIMIC-III dataset.
From: Inferring multimodal latent topics from electronic health records
a Word clouds of the three topics with colors indicating different data types and font size proportional to the probabilities of the EHR codes under that topic. b We selected top 50 patients under three of the learned topics for illustration. These topics are displayed as word clouds where the size of the word is proportional to their probabilities under the topic. The heatmap indicates patient mixture memberships for the topics. The color bar on the right of the heatmap indicates the actual diagnoses of the diseases related to each of the topics, namely leukemia, pulmonary embolism, and cirrhosis. c Top EHR codes of the high-risk patients. The top ten EHR codes (rows) from each topic were displayed for the top 50 patients (columns) under each topic. We highlighted some evidence (asterisks) as why some of the patients in absence of ICD-9 codes were prioritized as high risk by our method based on their relevance to the diseases of interests (i.e., leukemia, pulmonary embolism, and cirrhosis).
