How can a machine win a test and still miss the point?
A prediction system can look impressive in one evaluation and disappoint in another. A 2021 study of a historical sepsis model at Michigan Medicine makes the questions concrete: which patients, which outcome, and how early must a warning arrive?
Questions you could explore
- What changes when a prediction is scored before illness begins rather than after it is already apparent?
Start reading
- jamanetwork.comExternal Validation of a Widely Implemented Proprietary Sepsis Prediction Model in Hospitalized Patients
This external validation uses 2018–2019 hospitalizations at one US academic center. Its results show why evaluation timing and setting matter, without establishing the performance of today’s product.
Ready to try your ideas aloud?
Your topic carries over. Record up to 2½ minutes, depending on your plan.