Table 2.

Applying 4 dimensions of amplification to the clinical data pipeline for empiric antibiotic prediction.

Domain expert task				Amplification
Data curation
	Identify variables of interest, validate patients included in the cohort, and make domain-specific exclusionary rules			Summarization: present distribution of variables of interest Guidance: suggesting additional variables based on the selected ones Interactions: allow expert to select and remove data points Acceleration: suggest criteria based on the domain expert’s inclusion and exclusion
Data cleaning
		Augmentation
			Fill in unreported microbiology susceptibilities with rules		Summarization: preview a rule by showing distribution of the cells that will be impacted Guidance: show high-impact data subsets for edits Interactions: direct edits on interface and indirect edits via rules Acceleration: suggest general rules based on the domain expert’s single edit
		Validation
			Validate data augmentation by examining rule set and consolidating them to remove conflicts		Summarization: visual summary of rules and their relations Guidance: node size guides user to high-conflict areas Interactions: edit rule set by accepting and rejecting rules Acceleration: automatically remove redundant rules
Data analysis
	Understand the model and its predictions for individuals and different patient subpopulations			Summarization: show probability of coverage with confidence interval Guidance: highlight covariates of concern Interactions: allow domain expert to select covariates to include Acceleration: show similar patients for who the model should be updated