AHIC Decision Support 3 — Questions and Answers
Question 1: In predictive analytics for CDS, what is a 'feature' in the context of a machine learning model?
- A user interface element
- An input variable used to train and run the model (Correct answer)
- A clinical outcome measure
- A workflow alert type
Correct answer: An input variable used to train and run the model
In machine learning, a feature is an individual measurable input variable (e.g., age, lab value, vital sign) used by the model to make predictions.
Question 2: Which evaluation measure is most appropriate when a CDS predictive model's cost of missing a true positive (e.g., sepsis) is very high?
- Specificity
- Positive predictive value
- Sensitivity (recall) (Correct answer)
- F1 score only
Correct answer: Sensitivity (recall)
High sensitivity ensures that most true positive cases (e.g., sepsis) are identified, minimizing missed diagnoses even at the cost of more false positives.
Question 3: What does 'alert fatigue' primarily lead to in clinical settings?
- Slower EHR login times
- Clinicians ignoring or overriding alerts indiscriminately, including important ones (Correct answer)
- Increased documentation burden
- Reduced system interoperability
Correct answer: Clinicians ignoring or overriding alerts indiscriminately, including important ones
Alert fatigue causes clinicians to bypass warnings automatically, increasing the risk that a clinically significant alert will be dismissed without review.
Question 4: A CDS system surfaces recommendations only after a care team member requests them rather than proactively interrupting workflow. This best describes which CDS mode?
- Active or interruptive CDS
- Passive or on-demand CDS (Correct answer)
- Hard-stop CDS
- Prescriptive CDS
Correct answer: Passive or on-demand CDS
Passive or on-demand CDS provides guidance only when the user actively requests it, reducing workflow interruptions.
Question 5: Which clinical concept is central to the ARDEN Syntax used in some CDS rule engines?
- Medical Logic Modules (MLMs) (Correct answer)
- FHIR ValueSets
- SNOMED concept hierarchies
- ICD-10 groupers
Correct answer: Medical Logic Modules (MLMs)
Arden Syntax organizes CDS logic into Medical Logic Modules (MLMs), self-contained units that define a trigger, logic, and action.
Question 6: Which approach best addresses the risk of racial or socioeconomic bias in a CDS predictive model used for care management prioritization?
- Use only age and diagnosis as input features
- Audit model performance across demographic subgroups and retrain with representative data (Correct answer)
- Deploy the model only in academic medical centers
- Limit the model to inpatient settings
Correct answer: Audit model performance across demographic subgroups and retrain with representative data
Regular subgroup performance audits and retraining with representative data help detect and mitigate bias in CDS models.
Question 7: In evidence-based CDS, what is the role of a clinical knowledge management system?
- To store patient imaging files
- To maintain, version, and govern the rules and evidence that drive CDS recommendations (Correct answer)
- To authenticate users into the EHR
- To generate patient billing codes
Correct answer: To maintain, version, and govern the rules and evidence that drive CDS recommendations
A clinical knowledge management system centralizes governance of CDS content, tracking versions, evidence sources, and approval workflows.
AHIC Decision Support 3
In predictive analytics for CDS, what is a 'feature' in the context of a machine learning model?