Abstract 1296: Evaluating the accuracy and reproducibility of ChatGPT models in answering lung cancer patient queries
Bibliographic record
Abstract
Abstract Large language models (LLMs) such as ChatGPT can imitate human conversation and produce rapid, coherent responses, which may mask their potential for inaccuracies. With patients increasingly turning to the internet for medical information, the use of LLM chatbots for cancer-related queries risks spreading misinformation. Our study assessed ChatGPT’s accuracy and reproducibility in offering valid information and treatment advice for lung cancer in line with established guidelines. In the evolving landscape of AI-driven healthcare support, the ability of language models to provide accurate and reliable information is crucial. Our study delves into the effectiveness of OpenAI's ChatGPT models (versions 3.5 and 4.0) in responding to patient inquiries about lung cancer across various domains including general information, clinical presentation, risk factors, screening, diagnosis, staging, treatment options, prognosis, post-treatment follow-up, lifestyle recommendations, and psychosocial/educational aspects. We conducted a structured assessment, posing identical sets of questions to both ChatGPT 3.5 and 4.0. A total of 47 questions were posed with each query being repeated twice per model to evaluate both the accuracy and reproducibility of the responses. The scoring system focused on the accuracy and comprehensiveness of each response. Our findings revealed a notable disparity in the performance of the two models. GPT 4.0 demonstrated higher consistency and accuracy, with 41 out of 47 (87.2%) responses deemed accurate and comprehensive, compared to 36 out of 47 (76.6%) for GPT 3.5. In terms of reproducibility, both models exhibited strong performance: 42 out of 47 (89.3%) for GPT 3.5 and 45 out of 47 for GPT 4.0 (95.7%). When comparing responses between the models, we observed good reproducibility in 38 out of 47 questions (80.8%). A key observation was that GPT 4.0 significantly outperformed its predecessor GPT 3.5 in terms of both accuracy as well as reproducibility within its own responses, indicating a more reliable and consistent performance. The area most lacking in accuracy for both models was lung cancer staging, indicating a need for further refinement in this domain. Another key observation was the models' tendency to incorporate empathetic language, often beginning responses with expressions of sympathy and consistently advising confirmation with a medical professional. Our study underscores the potential and limitations of current AI models in patient education and support, highlighting areas for improvement and the importance of empathetic communication in AI interactions with patients. As the model continues to be trained on a larger and more comprehensive set of data, it is reasonable to anticipate further improvements in its ability to provide precise, detailed, and contextually appropriate responses. Citation Format: Asiyah Allibhai, Ahmed Allibhai, Anthony Brade, Zishan Allibhai. Evaluating the accuracy and reproducibility of ChatGPT models in answering lung cancer patient queries [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 1296.
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 machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.040 | 0.201 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.002 | 0.003 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
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 source (direct Gemma or distilled Codex), 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".