Skip to main content
Fig. 1 | Critical Care

Fig. 1

From: Pharmacophenotype identification of intensive care unit medications using unsupervised cluster analysis of the ICURx common data model

Fig. 1

Pharmacophenotype derivation workflow. a Medication administration records (including drug name, dose, formulation, route, and time) as well as other relevant patient data are recorded in the electronic health record (EHR) system. b The MAR data was processed to indicate 0 for not receiving that medication and 1 for receiving that medication for each patient. c Using restricted Boltzmann machine, six pharmacophenotypes were generated. If the medication from the visible layer was not assigned to a hidden layer, that medication was grouped in the sixth or unassigned cluster. d The pharmacophenotypes are displayed in a Venn diagram describing the degree of overlap between the clusters and how the medications are distributed among the clusters. e The frequency of every pharmacophenotype is counted and normalized by considering the total medications taken by every patient during their stay. f The resulting normalized pharmacophenotype distribution of every patient was used as a feature in the agglomerative hierarchical clustering method to develop novel pharmacophenotypes of critically ill patients. g The Uniform Manifold Approximation and Projection (UMAP) for Dimension Reduction of the five patient clusters was performed. h These novel pharmacophenotypes were associated with unique patterns of patient outcomes. MRC-ICU – medication regimen complexity in the intensive care unit

Back to article page