Answers are grounded in this paper and cite their sources.
Background Patients with breast cancer are increasingly expected to engage in treatment and supportive care decisions, yet evidence-based medical information is often difficult to understand. Plain language summaries (PLSs) aim to improve accessibility of systematic review findings, but their usefulness for patients remains variable. Artificial intelligence (AI)-generated summaries may simplify complex health evidence while preserving key messages. Objective To compare AI-generated PLSs with original Cochrane PLSs regarding patients’ intention to apply described interventions, comprehension, perceived trustworthiness, clarity, and confidence in the decision to change behavior. Methods A randomized, double-blinded study was conducted among patients with non-metastatic breast cancer at the Department of Oncology, University Hospital of Split, Croatia from October 2025 to February 2026. Participants were randomly assigned to receive either AI-generated Croatian-language PLSs or original Croatian Cochrane PLSs on two supportive oncology topics: exercise during adjuvant therapy and psychological interventions for patients with non-metastatic breast cancer. Chat generative pre-trained transformer (ChatGPT)-4o was used to generate AI-summaries. The primary outcome was intention to apply the interventions, measured using a visual analog scale (0–100). Secondary outcomes included comprehension, trustworthiness, clarity, and confidence in the decision-making. Because each participant evaluated both topics, the primary analysis used linear mixed-effects models with participant as a random intercept and summary group, topic, and group-by-topic interaction as fixed effects. Participant-level averaged outcomes were analyzed in sensitivity analyses. Exploratory post-hoc equivalence analyses used two one-sided tests. Results The sample included 133 participants, 67 in the original-summary-group and 66 in the AI-generated-summary group. In the primary repeated-measures analysis, no statistically significant overall differences were detected between groups for primary or secondary outcomes. However, the association between summary type and outcomes differed by topic for comprehension, intention to change behavior, and confidence in decision-making. In topic-specific analyses, confidence in decision-making was significantly higher for the AI-generated psychological intervention summary than for the original. Post-hoc equivalence analyses supported equivalence for primary outcome, comprehension, trust and clarity, but not for confidence in decision-making. Conclusion AI-generated summaries may represent an assistive tool for producing patient-facing information, when combined with expert review and quality control. Further studies using longer-term behavioral outcomes are needed. Clinical trial registration https://clinicaltrials.gov/ , identifier: NCT07140107.
No takes yet. Share an insight, caveat, or question.
Pavičić et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: