Beyond Binary Detection: A Multi-Dimensional Taxonomy of Cancer Misinformation on Reddit

This study addresses the limitations of binary misinformation detection by introducing a multi-dimensional taxonomy for characterizing cancer-related misinformation on Reddit. The researchers analyzed over 134,000 posts across 24 subreddits dedicated to breast, lung, colon, and prostate cancer. They developed a seven-dimensional framework—covering misinformation presence, information type, cancer stage, misinformation type, risk level, stance, and topical focus—to move beyond simple true/false labels. Using expert-annotated data from oncologists, the team evaluated the performance of various large language models (LLMs) in identifying and categorizing this content at scale.

The researchers found that roughly 6% of the analyzed cancer discussions contained misinformation. The content was not monolithic; it frequently involved narratives that promoted unsupported alternative treatments, expressed deep-seated distrust of pharmaceutical companies and medical institutions, or provided misleading advice on screening and diagnostic procedures. The study demonstrates that few-shot prompting significantly enhances the ability of LLMs to classify these nuanced dimensions, suggesting that automated, scalable annotation is a viable path forward for monitoring health misinformation in online communities.

Online platforms are critical spaces for patients to share experiences and seek support, but they also serve as vectors for harmful medical misinformation. By providing a structured, multi-dimensional taxonomy, this research offers a foundation for more sophisticated computational analysis. Moving beyond binary labels allows researchers and platform moderators to distinguish between benign uncertainty and high-risk misinformation, ultimately helping to mitigate the real-world harms—such as delayed care or the adoption of ineffective therapies—that arise from exposure to inaccurate health content.