The ISACA AAIA™ manual warns against " Aggregate Accuracy " in high-risk applications like healthcare.
A model might be 95% accurate overall but have a 40% error rate for a specific minority demographic (e.g., based on age or ethnicity). This creates " Hidden Fairness and Safety Risks " where patients from certain groups are consistently misdiagnosed. Auditors must recommend " Stratified Evaluation, " where performance metrics (Precision, Recall) are calculated separately for each subgroup to ensure equitable outcomes and patient safety.