Artificial intelligence-powered chatbots for oral anticancer drug patient information: an assessment of quality

The increasing use of oral anticancer drugs (OADs) in cancer therapy shifts greater responsibility towards patients, thereby also placing a higher informational burden on them. While intensified pharmacological/pharmaceutical care programs have proven beneficial for patients undergoing OAD treatment, their universal availability is currently limited. Given that patients frequently seek health information online, AI-powered chatbots may present a promising resource to address these increasing, yet often unmet information needs. This study aims to evaluate the readability, completeness of relevant information, and accuracy provided by AI-powered chatbots in response to patient questions about OAD treatment. Microsoft Bing's Copilot and Google's Gemini were queried in June 2024 on four patient questions regarding ten commonly prescribed and ten recently approved OADs in triplicate. Readability of chatbot answers was assessed using the Flesch reading-ease score (scale 0-100). Completeness of relevant information and accuracy were evaluated based on corresponding standardized written patient information materials. Both chatbots' answers demonstrated low readability according to the overall mean Flesch reading-ease scores of 38.8 (Copilot) and 50.9 (Gemini). Overall median completeness of relevant information of Copilot's and Gemini's answers was 61.1% (IQR, 35.3-78.7%) and 73.8% (IQR, 50.0-100.0%), respectively. Conversely, accuracy of chatbot answers was consistently high, with an overall median accuracy of 100.0% (IQR, 83.3-100.0%) for Copilot and 100.0% (IQR, 98.5-100.0%) for Gemini. AI-powered chatbots provide overall accurate information on OADs. However, their moderate completeness of relevant information and low readability may limit their current practical utility in meeting cancer patients' information need.