Population stability index
A number that measures how much the population a model sees has shifted away from the population it was built on. A higher value means more drift. It is one of the concrete triggers proposed for when a deployed AI tool needs re-checking: a population stability index above 0.2 is a common threshold for "investigate this". A practical early-warning signal that the data environment around a model has changed.
Three months after a readmission model goes live, the monitoring report shows the population stability index for patient age has climbed to 0.27, around the time the hospital began admitting directly from several local nursing homes. Outcomes take 30 days to arrive, so nobody yet knows whether accuracy has dropped, only that the model is now scoring an older group than the one it was built on. Treat the number as the cue to pull a recent sample, check calibration and error rates in the over-80s, and record what you found either way.
Terms like this come up in real clinical scenarios across the HelloAI courses: bite-sized modules with verifiable certificates. An account takes one minute, no password needed.
Sign in →