Scholarlink

This study evaluates the ability of fine-tuned medium-sized large language models (7–9B parameters) to interpret and categorize open-ended patient reports of the most bothersome problems and their functional impact, as reported by people with Parkinson’s disease (PD). Using a dataset of 323,263 reports from 30,841 patients, we compare the performance of Llama, Gemma, and Mistral against a deep learning baseline across 14 medical symptom domains. The results indicate that carefully prompted and fine-tuned LLMs achieve strong classification performance in identifying symptoms from patient reports. These findings highlight the potential of well-designed LLMs as scalable alternatives to specialized machine learning models in healthcare NLP, which usually require thorough preprocessing of the data and are tailored to one specific use case.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Leveraging Fine-Tuned State-of-the-Art LLMs for Symptom Classification of Patient-Reported Problems in Parkinson’s Disease

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Vanessa Richter,
  • Abhishek Hosamath,
  • Lakshmi Arbatti,
  • Michael Neumann,
  • Vikram Ramanarayanan

摘要

This study evaluates the ability of fine-tuned medium-sized large language models (7–9B parameters) to interpret and categorize open-ended patient reports of the most bothersome problems and their functional impact, as reported by people with Parkinson’s disease (PD). Using a dataset of 323,263 reports from 30,841 patients, we compare the performance of Llama, Gemma, and Mistral against a deep learning baseline across 14 medical symptom domains. The results indicate that carefully prompted and fine-tuned LLMs achieve strong classification performance in identifying symptoms from patient reports. These findings highlight the potential of well-designed LLMs as scalable alternatives to specialized machine learning models in healthcare NLP, which usually require thorough preprocessing of the data and are tailored to one specific use case.