Document resource
Background Patient and public involvement and engagement is vital to research by identifying interests, methods and outcomes which are important and acceptable to stakeholders. Speech processing has great potential to enhance these interactions using artificial intelligence to analyse research encounters and focus proposals and amendments to pressing issues across larger groups.Methods 260 people with neurodegenerative disorders and their carers and relatives who completed a research mobile app also provided feedback and suggestions in speech recordings and the System Usability Scale (SUS) in 360 sessions. Recordings were automatically transcribed using whisper large-v3. Transcript sentiment analysis and emotion classification was conducted using pre-trained natural language processing models. Embeddings from sentiment analysis were used to predict SUS scores. A large language model (LLM) was used to summarise key responses.Results Mean SUS score was 78.2, significantly higher than published average of 68 (p<0.001). Sentiment was positive in 74% of responses. Predominant emotions expressed were joy, approval and gratitude. Sentiment embeddings from recordings predicted SUS scores (RMSE=9.6, R 2=0.53). LLM summary highlighted actionable improvements participants raised, which were discussed and implemented in future study designs.Conclusions These methods provide a valid and useful approach to participant involvement in neurological research.jtam@ed.ac.uk