Can open source large language models be used for tumor documentation in Germany? : An evaluation on urological doctors’ notes

Abstract

Background Tumor documentation in Germany is currently a largely manual process. It involves reading the textual patient documentation and filling in forms in dedicated databases to obtain structured data. Advances in information extraction techniques that build on large language models (LLMs) could have the potential for enhancing the efficiency and reliability of this process. Evaluating LLMs in the German medical domain, especially their ability to interpret specialized language, is essential to determine their suitability for the use in clinical documentation. Due to data protection regulations, only locally deployed open source LLMs are generally suitable for this application. Methods The evaluation employs eleven different open source LLMs with sizes ranging from 1 to 70 billion model parameters. Three basic tasks were selected as representative examples for the tumor documentation process: identifying tumor diagnoses, assigning ICD-10 codes, and extracting the date of first diagnosis. For evaluating the LLMs on these tasks, a dataset of annotated text snippets based on anonymized doctors’ notes from urology was prepared. Different prompting strategies were used to investigate the effect of the number of examples in few-shot prompting and to explore the capabilities of the LLMs in general. Results The models Llama 3.1 8B, Mistral 7B, and Mistral NeMo 12 B performed comparably well in the tasks. Models with less extensive training data or having fewer than 7 billion parameters showed notably lower performance, while larger models did not display performance gains. Examples from a different medical domain than urology could also improve the outcome in few-shot prompting, which demonstrates the ability of LLMs to handle tasks needed for tumor documentation. Conclusions Open source LLMs show a strong potential for automating tumor documentation. Models from 7–12 billion parameters could offer an optimal balance between performance and resource efficiency. With tailored fine-tuning and well-designed prompting, these models might become important tools for clinical documentation in the future. The code for the evaluation is available from https://github.com/stefan-m-lenz/UroLlmEval. We also release the data set under https://huggingface.co/datasets/stefan-m-lenz/UroLlmEvalSet providing a valuable resource that addresses the shortage of authentic and easily accessible benchmarks in German-language medical NLP.