Evaluating a Chatbot as a Companion for Patients With Breast Cancer: Collaborative Pilot Study
Bibliographic record
Abstract
Background: Patients with breast cancer frequently experience significant uncertainty, prompting them to seek detailed, personalized, and reliable medical information to enhance adherence to prescribed treatments, medications, and recommended lifestyle adjustments. Although high-quality information exists within oncology guidelines and patient-oriented resources, the provision of tailored responses to individual patient queries remains challenging, especially for non-English-speaking populations. Objective: This study aims to evaluate the potential of an artificial intelligence-driven chatbot, specifically leveraging ChatGPT (GPT-4; OpenAI) combined with retrieval-augmented generation, to deliver personalized answers to complex breast cancer-related patient questions in German. Methods: We collaborated with one of Germany's largest breast cancer Patient Representation Groups to collect authentic patient inquiries, receiving a total of 118 questions. After initial screening, we selected 104 medical questions, organized into 7 distinct categories: aftercare, bone health, ductal carcinoma in situ, diagnostics, nutrition and supplements, complementary medicine, and therapy. A customized version of GPT-4 was configured with specific system prompts emphasizing empathetic, evidence-based responses and integrated with a comprehensive database comprising guidelines, recommendations, and patient information materials published by recognized German medical societies. To assess chatbot responses, we used 4 evaluation criteria: comprehensibility (clarity from a patient perspective), correctness (accuracy per current medical guidelines), completeness (inclusion of all relevant aspects), and potential harm (risk of undue patient harm or misinformation). Ratings were conducted using a 5-point Likert scale by a breast cancer expert (correctness, completeness, and potential harm) and patient representatives (comprehensibility). Results: The chatbot provided high-quality responses across multiple dimensions. Of the 499 responses evaluated for comprehensibility, 427 (85.6%) were rated as comprehensible. Among the 104 responses assessed for the remaining dimensions, 91 (87.5%) were rated as correct, 72 (69.2%) as complete, and 93 (89.4%) as nonharmful. Reasons for incomplete answers included omission of reimbursement details, updates from recent therapeutic guidelines, or nuanced recommendations regarding endocrine therapy and aftercare schedules. In addition, 6 (5.8%) of the answers were rated as potentially harmful due to outdated or contextually inappropriate recommendations. The chatbot also performed well in the nutrition and bone health categories despite occasionally incomplete document retrieval. Conclusions: Our findings demonstrate that an artificial intelligence-powered chatbot with GPT-4 and retrieval augmentation can effectively provide personalized, linguistically accessible, and largely accurate information to German-speaking patients with breast cancer. This approach holds considerable promise for improving patient-centered communication, empowering patients to make informed decisions. Nonetheless, observed limitations regarding response completeness and potential harm underscore the critical need for ongoing human oversight. Future research and development should prioritize regularly updated databases, advanced retrieval methods to handle complex document structures, multimodal capabilities, and clearly articulated disclaimers emphasizing the necessity of professional medical consultation. Our evaluation, along with the provided set of realistic patient questions, establishes a benchmark for future development and validation of German-language oncology chatbots.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".