Cancer Reasoning Traces

Paper: Reasoning with LLMs for Cancer Treatment Outcome Prediction
Authors: Geetha Krishna Guruju, Raghu Vamsi Hemadri et al.
License: CC BY 4.0
Dataset size: 24,856 samples
Modality: Text
Task: Clinical reasoning generation (Chain-of-Thought)
Code: OncoReason GitHub Repository

Dataset Overview

The Cancer Reasoning Traces dataset contains structured chain-of-thought (CoT) reasoning and commentary derived from oncology patient summaries in the MSK-CHORD dataset.
Each record corresponds to a single anonymized cancer patient and captures the step-by-step clinical reasoning process behind survival and treatment outcome prediction — without including the final prediction itself.

This dataset enables the training and evaluation of reasoning-aligned large language models (LLMs) that can articulate clinically grounded, interpretable reasoning for oncology tasks.

Dataset Structure

Data Preparation

• Derived from structured patient summaries built using the MSK-CHORD oncology database, which contains rich clinical, biomarker, and treatment information.
• Each summary was provided as input to the DeepSeek-R1 reasoning model, which generated multi-step chain-of-thought explanations using a standardized oncology-specific prompt.
• The responses were parsed into two structured components:
  1. chain_of_thought: ordered reasoning steps
  2. comments: concise notes about uncertainty or missing information
• Outputs were validated for schema consistency and formatted into JSON for reproducible downstream use.

Use Cases

This dataset can be used for:

• Training reasoning-aligned LLMs for clinical and biomedical applications
• Evaluating interpretability and consistency in medical CoT generation
• Studying how LLMs reason about prognostic and treatment-related factors
• Fine-tuning models for structured CoT distillation or reward-based reasoning alignment (e.g., GRPO)

License

This dataset is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0).

Citation

If you use this dataset in your research or project, please cite the following paper:

OncoReason: Structuring Clinical Reasoning in LLMs for Robust and Interpretable Survival Prediction
Raghu Vamshi Hemadri, Geetha Krishna Guruju, Kristi Topollai, Anna Ewa Choromanska
arXiv preprint arXiv:2510.17532
https://arxiv.org/abs/2510.17532

@misc{hemadri2025oncoreasonstructuringclinicalreasoning,
      title={OncoReason: Structuring Clinical Reasoning in LLMs for Robust and Interpretable Survival Prediction}, 
      author={Raghu Vamshi Hemadri and Geetha Krishna Guruju and Kristi Topollai and Anna Ewa Choromanska},
      year={2025},
      eprint={2510.17532},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2510.17532}, 
}