An AI model used for clinical trial participant selection achieves excellent overall performance but the audit team discovers it excludes patients with multiple comorbidities at a disproportionately high rate. The model was trained on historically enrolled trial participants, who typically had fewer comorbidities. What AI risk does this BEST illustrate?
Select an answer to reveal the explanation.
Short Explanation
You're looking at selection bias in the training data: the model learned from past enrollments that underrepresented comorbid patients, so it keeps excluding them even when the actual protocol would allow them.
Full Explanation
Selection bias in training data creating a self-perpetuating pattern that excludes comorbid patients from trials who might have been eligible under the actual trial protocol is correct. The model learned eligibility patterns from historical enrollments that were themselves biased: few comorbid patients were enrolled historically, possibly due to overly conservative human decisions. It perpetuates this bias by predicting comorbid patients as ineligible even when the actual protocol would allow them, creating a feedback loop. Overfitting to complex patient profiles with multiple comorbidities describes memorizing training noise, not systematic exclusion. Concept drift from changes in clinical trial eligibility criteria since the training data was collected relates to temporal changes, not historical underrepresentation. Feature engineering error where comorbidity indicators were incorrectly coded as exclusion criteria would be consistent across all patients, not specific to comorbid patients.