Reasoning Models for Text Mining in Oncology - a Comparison Between o1 Preview and GPT-4o

PurposeChain-of-thought prompting is a method to make a Large Language Model (LLM) generate intermediate reasoning steps when solving a complex problem to increase its performance. OpenAIs o1 preview is an LLM that has been trained with reinforcement learning to create such a chain-of-thought internally, prior to giving a response and has been claimed to surpass various benchmarks requiring complex reasoning. The purpose of this study was to evaluate its performance for text mining in oncology. MethodsSix hundred trials from high-impact medical journals were classified depending on whether they allowed for the inclusion of patients with localized and/or metastatic disease. GPT-4o and o1 preview were instructed to do the same classification based on the publications abstracts. ResultsFor predicting whether patients with localized disease were enrolled, GPT-4o and o1 preview achieved F1 scores of 0.80 (0.76 - 0.83) and 0.91 (0.89 - 0.94), respectively. For predicting whether patients with metastatic disease were enrolled, GPT-4o and o1 preview achieved F1 scores of 0.97 (0.95 - 0.98) and 0.99 (0.99 - 1.00), respectively. Conclusiono1 preview outperformed GPT-4o for extracting if people with localized and or metastatic disease were eligible for a trial from its abstract. o1 previewss performance was close to human annotation but could still be improved when dealing with cancer screening and prevention trials as well as by adhering to the desired output format. While research on additional tasks is necessary, it is likely that reasoning models could become the new state of the art for text mining in oncology and various other tasks in medicine.