Special Reports – Features
Biased AI impacts diagnostic accuracy and explanation strategies
by Michael DePeau-Wilson, Enterprise & Investigative Writer, MedPage Today
December 19, 2023
Providing clinicians with artificial intelligence (AI) predictions and model explanations can enhance diagnostic accuracy, but a biased AI model can significantly reduce that accuracy, and explanations do not fully mitigate these negative effects, according to a randomized clinical vignette survey study.
In the study, clinicians were asked to differentiate between pneumonia, heart failure, or chronic obstructive pulmonary disease (COPD), and the results showed that diagnostic accuracy increased with AI predictions, and explanations. However, when a systematically biased AI model was used, diagnostic accuracy declined and explanations were not effective in restoring it.
The research also emphasized the potential of AI to improve clinical care but called for a careful integration of AI into clinical workflows to prevent systematic errors that could harm patients.
To delve deeper into the issue, the study involved analyzing vignettes of patients hospitalized with acute respiratory failure and diagnosing their underlying conditions. Clinicians’ diagnostic accuracy was established, and they were then randomized to see different vignettes with or without AI input and explanations.
Overall, the study concluded that while AI models have the potential to enhance clinical decision-making, they must be thoughtfully integrated to avoid introducing errors or harming patients. Additionally, the study emphasized the need to establish the effectiveness of explanation strategies for AI models. For more information, you can read the full article from MedPage Today.
