Are Medium-Sized Transformers Models still Relevant for Medical Records Processing?
Bibliographic record
Abstract
As large language models (LLMs) become the standard in many NLP applications, we explore the potential of medium-sized pretrained transformer models as a viable alternative for medical record processing. Medical records generated by healthcare professionals during patient admissions remain underutilized due to challenges such as complex medical terminology, the limited ability of pretrained models to interpret numerical data, and the scarcity of annotated training datasets. Objective: This study aims to classify numerical values extracted from medical records into seven distinct physiological categories using CamemBERT-bio. Previous research has suggested that transformer-based models may underperform compared to traditional NLP approaches in this context. Methods: To enhance the performance of CamemBERT-bio, we propose two key innovations: (1) incorporating keyword embeddings to refine the model's attention mechanisms and (2) adopting a number-agnostic strategy by removing numerical values from the text to encourage context-driven learning. Additionally, we assess the criticality of extracted numerical data by verifying whether values fall within established standard ranges. Results: Our findings demonstrate significant performance improvements, with CamemBERT-bio achieving an F1 score of 0.89 - an increase of over 20% compared to the 0.73 F1 score of traditional methods and only 0.06 units lower than GPT-4. These results were obtained despite the use of small and imbalanced training datasets.
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 distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.003 | 0.002 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
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 teacher head, 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".