In this paper, Alabed and colleagues present a systematic review and meta-analysis evaluating how well large language models (LLMs) can simplify radiology reports to improve patient understanding, addressing a longstanding communication gap between technical imaging interpretations and patient comprehension. The authors searched major medical databases and identified 38 eligible studies published between 2022 and 2025, in which AI systems were used to rewrite radiology reports into more accessible language. Across these studies, nearly 13,000 simplified reports were evaluated by over 500 assessors — a mix of lay people and clinical professionals. Their central finding is that LLM-rewritten reports were perceived as dramatically more understandable by patients (an 87% improvement compared to original radiologist reports), while clinicians rated them highly for accuracy and completeness. Readability also improved substantially across imaging modalities, with report complexity shifting from university level down to roughly an 11–13 year old reading level, suggesting that AI can effectively translate complex medical terminology into language accessible to a broader population.
From a patient-centred care perspective, the implications are significant. Poor understanding of radiology reports can lead to anxiety, unnecessary follow-up visits, and inefficient use of clinical time, particularly as patients increasingly access their results directly through digital portals. AI-driven simplification tools could help address these challenges by supporting shared decision-making and improving health literacy across diverse populations. However, the authors note that around 1 in 100 reports contained clinically significant errors, and that concerns around clinician oversight, trust, safety, and generalisability remain. They conclude that while LLM-based simplification shows real promise for making radiology more patient-centred and advancing more equitable healthcare delivery, wider adoption will require standardised evaluation methods and careful co-design with patients before it can be safely embedded into clinical workflows.
