Fig. 6: Harm assessment of model outputs by board-certified ophthalmologists. | npj Digital Medicine

Fig. 6: Harm assessment of model outputs by board-certified ophthalmologists.

From: Enhancing clinical documentation with voice processing and large language models: a study on the LAOS system

Fig. 6: Harm assessment of model outputs by board-certified ophthalmologists.

Distribution of harm assessment across likelihood and severity classifications based on evaluations by five board-certified ophthalmologists reviewing 100 paired comparisons using the standardized questionnaire (Tab. 9). Horizontal axes denote percentage from 0 to 100. Vertical axes present two pairs of comparisons: harm likelihood assessment (Human vs. Model) and harm degree classification (Human vs. Model). The assessment used AHRQ harm scale metrics, with each category divided into three levels: High/Severe or death (for likelihood/degree), Medium/Mild, and Low/None, represented by different shades in the bars.

Back to article page