Assessing Open-Weight, Large Language Model for Symptom Extraction from Dialysis Notes: A Cost-Effective and Privacy-Preserving Approach
Bibliographic record
Abstract
Background: Hemodialysis often leads to intradialytic symptoms that affect quality of life. Traditional methods to study these rely on patient questionnaires or free-text dialysis notes, which are associated with cost and quality challenges. This project evaluates open-source large language models (LLMs) that can run offline, aiming to extract symptoms from nursing notes while maintaining data privacy. Methods: We designed 36 information extraction agents by combining four open-source LLMs with distinct prompting approaches, such as in-context learning, chain-of-thought, and rule-based guidance. Each agent identified the presence of four target symptoms across 100 dialysis nursing notes. Outputs were compared to expert annotations performed independently and without knowledge of the agents’ results. Evaluation metrics included accuracy, recall, precision, specificity, and balanced F1 score. Results: Among 100 nursing notes, chest pain was noted twice, dyspnea 14 times, dizziness 23 times, and nausea 16 times. The best-performing agent, based on Mixtral 8x7B paired with a simple prompt, achieved 93% specificity, 71% sensitivity, 63% precision. Performance remained strong across most symptoms except for nausea, and generally decreased with smaller models or more elaborate prompting strategies (Figure). Conclusion: This study provides the first assessment of open-source LLMs operating offline on consumer-level hardware for extracting symptoms from dialysis notes. It offers an accessible, low-cost method to support research into the quality of life of dialysis patients. Funding: Government Support – Non-U.S.Performance of the top 6 agents. All agents use Mixtral-8x7B-instruct-v0.1 with various prompting strategies.
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 | 0.005 | 0.025 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.002 | 0.003 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.002 |
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".