Large language model–generated plain-language summaries were associated with higher self-reported comprehension of lung cancer screening reports compared with standard radiology reports alone, according to a vignette survey study published in JAMA Network Open.

As low-dose computed tomography screening becomes more common, patients often receive detailed radiology reports that include technical terminology and structured findings. The study examined whether adding a large language model (LLM)-generated summary could improve patient understanding of screening results.

Researchers conducted a randomized vignette survey among US adults recruited through an online platform. Participants were shown simulated lung cancer screening results representing different Lung-RADS categories. They were randomized to view either a standard radiology report or the same report accompanied by an LLM-generated plain-language summary.

The primary outcome was self-reported comprehension, measured using a Likert scale assessing how well participants felt they understood the screening results. Secondary outcomes included perceived clarity, satisfaction with the report, and anxiety related to the findings.

Participants who received reports with LLM-generated summaries reported higher levels of understanding compared with those who viewed standard reports alone. Perceived clarity and satisfaction were also higher in the summary group across multiple Lung-RADS categories.

The association between summaries and improved self-reported comprehension remained after adjustment for age, education level, and health literacy. Anxiety levels did not significantly differ between groups, including in scenarios describing findings that required short-term follow-up.

The study used hypothetical screening scenarios rather than real clinical reports, and outcomes were based on self-reported measures rather than objective testing of understanding or subsequent clinical decisions. Participants were recruited online, which may limit generalizability to populations undergoing lung cancer screening.

The authors concluded that LLM-generated plain-language summaries may support patient understanding of lung cancer screening results. They noted that further evaluation in clinical settings is needed to assess real-world performance, workflow integration, and mechanisms to ensure accuracy of AI-generated content.

Source: JAMA Network Open.