Elamur Research⌕
☰
home / papers / 2023 / Self-Verification Improves Few-Shot Clinical Information Extraction

Self-Verification Improves Few-Shot Clinical Information Extraction

Paper · 2023 · Zelalem Gero, Chandan Singh, Hao Cheng et al.

Summary

Clinical information extraction often lacks labeled training data. Few-shot prompting of LLMs can stand in for supervised training, but its outputs are still error-prone and hard to inspect, which matters in a safety-critical field like medicine. The authors propose self-verification (SV), a four-call chain in which one LLM works through different prompts. It first produces an Original extraction, then searches for missed items (Omission), then attaches a supporting quote from the input to each item (Evidence), and finally discards items the evidence does not support (Prune). The idea rests on the observation that verifying an answer is usually easier than generating it. SV was tested on ICD code prediction (MIMIC-III, MIMIC-IV), clinical trial arm extraction (EBM-NLP) and medication status extraction (CASI) with GPT-3.5, ChatGPT and GPT-4. It raised F1 in every model–task combination. The quoted evidence spans also agree well with human annotations, so experts can audit the extracted items quickly. Code and prompts are public.

Key points

  • Results: Across models and tasks, SV adds 0.056 F1 on average. For GPT-4, clinical trial arm F1 rises from 0.419 to 0.530, and the authors report a gain of over 0.3 on medication status.
  • Ablations: Omission mainly boosts recall and Prune mainly boosts precision, and combining them gives the best F1. A single concatenated "large prompt" performs slightly worse than the plain baseline.
  • Evidence quality: For GPT-4, 93% of evidence spans overlap human-marked intervention spans in EBM-NLP (84% for GPT-3.5). Annotators mark fewer than 3.7% of tokens, so chance overlap would be low.
  • Data and models:
    • Data: 250 MIMIC-III reports, 250 MIMIC-IV discharge summaries (top-50 ICD codes), 100 EBM-NLP abstracts and 105 CASI snippets.
    • Models: text-davinci-003, gpt-3.5-turbo and gpt4-0314, run at temperature 0.1 via Azure OpenAI.
  • Best model by task: GPT-3.5 does best on the short-input tasks, while GPT-4 leads on the MIMIC ICD tasks. Code is at github.com/microsoft/clinical-self-verification.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksBiochemistry, Genetics and Molecular Biology · 234 worksMolecular Biology · 231 works

🧭 Topics 1Research subjects and languages the work is about.Information extraction · 167 works

Cite

Cite as Zelalem Gero, Chandan Singh, Hao Cheng, Tristan Naumann, Michel Galley, Jianfeng Gao and Hoifung Poon. 2023. Self-Verification Improves Few-Shot Clinical Information Extraction. arXiv:2306.00024 · IMLH 2023.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2306.00024 · arXiv 2306.00024

@misc{gero2023self,
  title = {{Self-Verification Improves Few-Shot Clinical Information Extraction}},
  author = {Zelalem Gero, Chandan Singh, Hao Cheng, Tristan Naumann, Michel Galley, Jianfeng Gao and Hoifung Poon},
  year = {2023},
  doi = {10.48550/arxiv.2306.00024},
  eprint = {2306.00024},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2306.00024}
}

References

46 references, 20 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Self-verification constitutes an important step towards unlocking the potential of LLMs in healthcare settings. As LLMs continue to generally improve in performance, clinical extraction with LLMs + SV seems likely to improve as well. Agrawal, M., Hegselmann, S., Lang, H., Kim, Y., and Sontag, D. Lar… — Large language models are few-shot clinical information extractors
  2. One limitation of SV is that it incurs a high computational cost as multiple LLM calls are chained together; however, these costs may continue to decrease as models become Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J. D., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et… — Language Models are Few-Shot Learners
  3. Chase, H. Langchain: Building applications with llms through composability. https://github.com/hwchase17/ langchain, 1 2023. ↗
  4. Madaan, A., Tandon, N., Gupta, P., Hallinan, S., Gao, L., Wiegreffe, S., Alon, U., Dziri, N., Prabhumoye, S., Yang, Y., et al. Self-refine: Iterative refinement with self-feedback. arXiv preprint arXiv:2303.17651, 2023. find
  5. Dao, T., Fu, D., Ermon, S., Rudra, A., and Ré, C. Flashattention: Fast and memory-efficient exact attention with io-awareness. Advances in Neural Information Processing Systems, 35:16344– 16359, 2022. find
  6. Moon, S., Pakhomov, S., and Melton, G. Clinical abbreviation sense inventory. 2012. find
  7. Nye, B., Li, J. J., Patel, R., Yang, Y., Marshall, I. J., Nenkova, A., and Wallace, B. C. A corpus with multi-level annotations of patients, interventions and outcomes to support language processing for medical literature. In Proceedings of the conference. Association for Computational Linguistics.…
  8. NIH Public Access, 2018. find
  9. Edin, J., Junge, A., Havtorn, J. D., Borgholt, L., Maistro, M., Ruotsalo, T., and Maaløe, L. Automated medical coding on mimic-iii and mimic-iv: A critical review and replicability study. arXiv preprint arXiv:2304.10909, 2023. — [arXiv:2304.10909](https://arxiv.org/abs/2304.10909) find
  10. Gao, L., Dai, Z., Pasupat, P., Chen, A., Chaganty, A. T., Fan, Y., Zhao, V. Y., Lao, N., Lee, H., Juan, D.-C., et al. Rarr: Researching and revising what language models say, using language models. arXiv preprint arXiv:2210.08726, 2022.
  11. OpenAI. Gpt-4 technical report, 2023.
  12. Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C. L., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., et al. Training language models to follow instructions with human feedback. arXiv preprint arXiv:2203.02155, 2022.
  13. Gao, L., Madaan, A., Zhou, S., Alon, U., Liu, P., Yang, Y., Callan, J., and Neubig, G. Pal: Program-aided language models, 2023. find
  14. Paranjape, B., Lundberg, S., Singh, S., Hajishirzi, H., Zettlemoyer, L., and Ribeiro, M. T. Art: Automatic multi-step reasoning and tool-use for large language models, 2023. find
  15. Gutiérrez, B. J., McNeal, N., Washington, C., Chen, Y., Li, L., Sun, H., and Su, Y. Thinking about gpt-3 in-context learning for biomedical ie? think again. arXiv preprint arXiv:2203.08410, 2022.
  16. Patil, S. G., Zhang, T., Wang, X., and Gonzalez, J. E. Gorilla: Large language model connected with massive apis. arXiv preprint arXiv:2305.15334, 2023.
  17. Johnson, A., Bulgarelli, L., Pollard, T., Celi, L. A., Mark, R., and Horng IV, S. Mimic-iv-ed. PhysioNet, 2021. find
  18. Peng, B., Galley, M., He, P., Cheng, H., Xie, Y., Hu, Y., Huang, Q., Liden, L., Yu, Z., Chen, W., et al. Check your facts and try again: Improving large language models with external knowledge and automated feedback. arXiv preprint arXiv:2302.12813, 2023.
  19. Johnson, A. E., Pollard, T. J., Shen, L., Lehman, L.-w. H., Feng, M., Ghassemi, M., Moody, B., Szolovits, P., Anthony Celi, L., and Mark, R. G. Mimic-iii, a freely accessible critical care database. Scientific data, 3(1):1–9, 2016.
  20. Press, O., Zhang, M., Min, S., Schmidt, L., Smith, N. A., and Lewis, M. Measuring and narrowing the compositionality gap in language models, 2022. find
  21. Jojic, A., Wang, Z., and Jojic, N. Gpt is becoming a turing machine: Here are some ways to program it, 2023. find
  22. Rajani, N. F., McCann, B., Xiong, C., and Socher, R. Explain yourself! leveraging language models for commonsense reasoning. arXiv preprint arXiv:1906.02361, 2019.
  23. Kornblith, A. E., Singh, C., Devlin, G., Addo, N., Streck, C. J., Holmes, J. F., Kuppermann, N., Grupp-Phelan, J., Fineman, J., Butte, A. J., et al. Predictability and stability testing to assess clinical decision instrument performance for children after blunt torso trauma. PLOS Digital Health, 1(8… find
  24. Rashkin, H., Nikolaev, V., Lamm, M., Aroyo, L., Collins, M., Das, D., Petrov, S., Tomar, G. S., Turc, I., and Reitter, D. Measuring attribution in natural language generation models. arXiv preprint arXiv:2112.12870, 2021.
  25. Laursen, M., Pedersen, J., Hansen, R., Savarimuthu, T. R., and Vinholt, P. Danish clinical named entity recognition and relation extraction. In Proceedings of the 24th Nordic Conference on Computational Linguistics (NoDaLiDa), pp. 655–666, 2023. find
  26. Ribeiro, M. T., Singh, S., and Guestrin, C. ” why should i trust you?” explaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, pp. 1135–1144, 2016.
  27. Liu, S., Wright, A. P., Patterson, B. L., Wanderer, J. P., Turer, R. W., Nelson, S. D., McCoy, A. B., Sittig, D. F., and Wright, A. Assessing the value of chatgpt for clinical decision support optimization. MedRxiv, pp. 2023–02, 2023. find
  28. Rudin, C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature machine intelligence, 1(5):206–215, 2019.
  29. Lundberg, S. M. and Lee, S.-I. A unified approach to interpreting model predictions. Advances in neural information processing systems, 30, 2017.
  30. Scheurer, J., Campos, J. A., Korbak, T., Chan, J. S., Chen, A., Cho, K., and Perez, E. Training language models with language feedback at scale. arXiv preprint arXiv:2303.16755, 2023. find
  31. Ma, Y., Cao, Y., Hong, Y., and Sun, A. Large language model is not a good few-shot information extractor, but a good reranker for hard samples! arXiv preprint arXiv:2303.08559, 2023.
  32. Schick, T., Dwivedi-Yu, J., Dessı̀, R., Raileanu, R., Lomeli, M., Zettlemoyer, L., Cancedda, N., and Scialom, T. Toolformer: Language models can teach themselves to use tools, 2023.
  33. MacNeil, S., Tran, A., Mogil, D., Bernstein, S., Ross, E., and Huang, Z. Generating diverse code explanations using the gpt-3 large language model. In Proceedings of the 2022 ACM Conference on International Computing Education Research-Volume 2, pp. 37–39, 2022. find
  34. Shin, T., Razeghi, Y., Logan IV, R. L., Wallace, E., and Singh, S. Autoprompt: Eliciting knowledge from language models with automatically generated prompts. arXiv preprint arXiv:2010.15980, 2020.
  35. Singh, C., Askari, A., Caruana, R., and Gao, J. Augmenting interpretable models with llms during training. arXiv preprint arXiv:2209.11799, 2022a. — [arXiv:2209.11799](https://arxiv.org/abs/2209.11799) find
  36. Singh, C., Morris, J. X., Aneja, J., Rush, A. M., and Gao, J. Explaining patterns in data with language models via interpretable autoprompting. arXiv preprint arXiv:2210.01848, 2022b. — [arXiv:2210.01848](https://arxiv.org/abs/2210.01848) find
  37. Singh, C., Hsu, A. R., Antonello, R., Jain, S., Huth, A. G., Yu, B., and Gao, J. Explaining black box text modules in natural language with language models. arXiv preprint arXiv:2305.09863, 2023. — [arXiv:2305.09863](https://arxiv.org/abs/2305.09863) find
  38. Toma, A., Lawler, P. R., Ba, J., Krishnan, R. G., Rubin, B. B., and Wang, B. Clinical camel: An open-source expert-level medical language model with dialogue-based knowledge encoding. arXiv preprint arXiv:2305.12031, 2023. find
  39. Wang, B., Deng, X., and Sun, H. Iteratively prompt pre-trained language models for chain of thought. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pp. 2714–2730, Abu Dhabi, United Arab Emirates, December 2022. Association for Computational Linguistics. UR… find
  40. Wang, Y., Wang, L., Rastegar-Mojarad, M., Moon, S., Shen, F., Afzal, N., Liu, S., Zeng, Y., Mehrabi, S., Sohn, S., et al. Clinical information extraction applications: a literature review. Journal of biomedical informatics, 77:34–49, 2018. find
  41. Wiegreffe, S. and Pinter, Y. Attention is not not explanation. arXiv preprint arXiv:1908.04626, 2019.
  42. Wu, C., Zhang, X., Zhang, Y., Wang, Y., and Xie, W. Pmc-llama: Further finetuning llama on medical papers. arXiv preprint arXiv:2304.14454, 2023.
  43. Wu, T., Terry, M., and Cai, C. J. Ai chains: Transparent and controllable human-ai interaction by chaining large language model prompts. In CHI Conference on Human Factors in Computing Systems, pp. 1–22, 2022. find
  44. Xu, B., Wang, Q., Mao, Z., Lyu, Y., She, Q., and Zhang, Y. k nn prompting: Beyond-context learning with calibration-free nearest neighbor inference. arXiv preprint arXiv:2303.13824, 2023. — [arXiv:2303.13824](https://arxiv.org/abs/2303.13824) find
  45. Zhang, H., Liu, X., and Zhang, J. Summit: Iterative text summarization via chatgpt. arXiv preprint arXiv:2305.14835, 2023. find
  46. Zweigenbaum, P., Demner-Fushman, D., Yu, H., and Cohen, K. B. Frontiers of biomedical text mining: current progress. Briefings in bioinformatics, 8(5):358–375, 2007. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/gero-2023-self-verification-improves-few-shot/