Dispatch from the Digital Health Frontier: Shift Happens

Dataset shifts, where the data collected during the development of an algorithm changes over time, can compromise the accuracy of AI-based algorithms in healthcare. Changes in technology, population, setting, and behaviour can all lead to this issue. Ongoing dialogue between medical and IT teams, updating variable mapping, retraining algorithms, and multidisciplinary root cause analysis can all help address these issues. However, artificial intelligence can still misdirect and cannot replace clinical experience and common sense.