AI in Suicide Risk Prediction Quiz

Answers and explanations appear as you take the quiz.

1. A model with 90 percent sensitivity and specificity is applied where 100 of 100,000 patients attempt suicide. Roughly what share of flagged patients will actually attempt?
   • About 90 percent
   • About 50 percent
   • Under 1 percent
   • About 10 percent
2. According to the guide, what are suicide risk models best understood as?
   • Tools that identify who will die by suicide
   • Tools for prioritizing limited clinical attention
   • Replacements for clinical assessment
   • Legal evidence for involuntary commitment
3. Roughly what share of risk scores does the VA's REACH VET program flag for review?
   • The top 50 percent
   • The top 10 percent
   • The top 0.1 percent
   • Every veteran with a mental health diagnosis
4. Why do health systems often set flagging thresholds by capacity?
   • So they flag only as many patients as staff can meaningfully contact
   • Because regulators require exactly 100 flags per month
   • To maximize the number of false positives
   • Because the models cannot produce scores below a set value
5. Which response to a risk flag fits the ethical practice described in the guide?
   • Sharing the flag with the patient's employer
   • Voluntary outreach, safety planning and discussing safe storage of lethal means
   • Automatically canceling the patient's firearm license
   • Not telling the patient or clinician anything
6. What does calibration mean for a suicide risk model?
   • How fast the model runs
   • Whether predicted risk levels match observed rates, such as a 2 percent prediction corresponding to about 2 percent observed
   • Whether the model ranks patients in order
   • How many variables the model uses
7. Why do these models need periodic revalidation?
   • Changes in coding practice, screening workflows or care patterns can quietly degrade performance
   • Software licenses expire yearly
   • Patients' names change
   • Revalidation increases the base rate of suicide
8. Which kind of evidence does the guide describe as strongest for these programs?
   • High retrospective accuracy on historical records
   • Vendor marketing materials
   • Prospective evaluation showing that flag-triggered outreach reduces attempts
   • A large number of input variables