AI in Sepsis Prediction Quiz

Answers and explanations appear as you take the quiz.

1. What does an AI sepsis early-warning model primarily produce?
   • A confirmed sepsis diagnosis that replaces clinical assessment
   • A risk score that flags patients for clinicians to assess
   • An automatic antibiotic prescription
   • A billing code for sepsis treatment
2. What did the 2021 Michigan Medicine external validation of the Epic Sepsis Model find?
   • It performed better than the developer had reported
   • It performed exactly as the developer had reported
   • It performed substantially worse, missing about two thirds of sepsis cases
   • It could not be tested because the hospital had no sepsis cases
3. What is alert fatigue?
   • Staff becoming desensitized to warnings after frequent false alarms
   • A hardware fault that stops alerts from being sent
   • Patients feeling tired after sepsis treatment
   • A model's accuracy improving as it sends more alerts
4. Why can defining sepsis onset using antibiotic or blood culture orders cause problems for a model?
   • Those orders are never recorded in the EHR
   • It makes the model too slow to run in real time
   • It prevents the model from using vital signs
   • The model can learn that clinicians already suspect infection, inflating its apparent accuracy
5. Why is the evidence from the TREWS studies still debated?
   • The system was never used on real patients
   • The results were never published
   • The studies were observational rather than randomized
   • The studies only measured cost, not outcomes
6. Why should practitioners report positive predictive value at their local prevalence?
   • Because a high AUROC can coexist with most alerts being false
   • Because AUROC cannot be calculated for sepsis
   • Because prevalence has no effect on alert accuracy
   • Because regulators ban the use of AUROC
7. Which rule-based sepsis screens did hospitals use before AI models?
   • Sepsis-3 and FHIR
   • AUROC and PPV
   • TREWS and ImmunoScore
   • SIRS criteria and qSOFA
8. What is the purpose of running a sepsis model in silent mode before go-live?
   • To send alerts only to patients instead of staff
   • To compare its would-be alerts against real cases without affecting care
   • To train clinicians to ignore alerts
   • To reduce the number of lab tests ordered