AI Prediction of Acute Kidney Injury Quiz

Answers and explanations appear as you take the quiz.

1. Under the KDIGO criteria described in this guide, which creatinine change within 48 hours qualifies as acute kidney injury?
   • A rise of 0.1 mg/dL
   • A rise of at least 0.3 mg/dL
   • A fall of at least 0.3 mg/dL
   • Any value above 1.0 mg/dL
2. Why do AKI models try to predict injury instead of waiting for creatinine to rise?
   • Creatinine tests are too expensive to repeat
   • Creatinine can only be measured in the ICU
   • Creatinine is a lagging marker, so the injury may already be a day or more old when it rises
   • Creatinine falls during kidney injury
3. In the 2019 DeepMind and VA study, roughly how many false alerts came with each correct AKI prediction?
   • About two
   • About twenty
   • None
   • About one false alert per hundred true ones
4. Which group did the VA-trained AKI model perform worse for, and why?
   • Older patients, because they have less lab data
   • Women, because they were a small share of the VA training population
   • ICU patients, because their creatinine is measured too often
   • Surgical patients, because anesthesia hides creatinine changes
5. What did DeepMind's Streams app at the Royal Free Hospital actually deliver to clinicians?
   • Deep learning predictions of AKI 48 hours ahead
   • Automated dialysis orders
   • Genetic risk scores for kidney disease
   • Alerts from the rule-based NHS England AKI algorithm
6. What did the Yale randomized trials of electronic AKI alerts find?
   • Alerts halved the need for dialysis
   • Alerts did not improve outcomes such as dialysis or death overall
   • Alerts only worked when sent by text message
   • Alerts made creatinine measurements more accurate
7. Why does the guide call missing lab data 'informative' in AKI models?
   • Missing values are always errors in the database
   • Labs are missing only for patients who died
   • A lab that was never ordered often means the clinician wasn't worried, so the ordering pattern itself predicts risk
   • Missing values are filled in by the laboratory later
8. Why should an AKI model's evaluation exclude patients who already meet AKI criteria when the prediction is made?
   • To avoid giving the model credit for spotting injuries that are already known
   • Because those patients have no creatinine data
   • Because the law forbids scoring them
   • Because they always go on to need dialysis