Viability of Open Large Language Models for Clinical Documentation in German Health Care: Real-World Model Evaluation Study
Bibliographic record
Abstract
Background: The use of large language models (LLMs) as writing assistance for medical professionals is a promising approach to reduce the time required for documentation, but there may be practical, ethical, and legal challenges in many jurisdictions complicating the use of the most powerful commercial LLM solutions. Objective: In this study, we assessed the feasibility of using nonproprietary LLMs of the GPT variety as writing assistance for medical professionals in an on-premise setting with restricted compute resources, generating German medical text. Methods: We trained four 7-billion-parameter models with 3 different architectures for our task and evaluated their performance using a powerful commercial LLM, namely Anthropic's Claude-v2, as a rater. Based on this, we selected the best-performing model and evaluated its practical usability with 2 independent human raters on real-world data. Results: In the automated evaluation with Claude-v2, BLOOM-CLP-German, a model trained from scratch on the German text, achieved the best results. In the manual evaluation by human experts, 95 (93.1%) of the 102 reports generated by that model were evaluated as usable as is or with only minor changes by both human raters. Conclusions: The results show that even with restricted compute resources, it is possible to generate medical texts that are suitable for documentation in routine clinical practice. However, the target language should be considered in the model selection when processing non-English text.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.022 | 0.048 |
| Meta-epidemiology (narrow) | 0.003 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".