Free Text Analysis: Identification of adverse drug events in clinical notes
Bibliographic record
Abstract
ObjectiveTo explore the prevalence of adverse drug events (ADE) leading to drug changes using Artificial Intelligence (AI) based free text analysis of primary care encounter notes . ApproachWe used electronic medical record l encounter notes linked to population-based drug dispensation records. Dispensation records were used to identify drug switching to another agent in the same therapeutic class, which could suggest an undesirable side-effect requiring discontinuation of the drug. We annotated clinical notes by identifying ADEs after retrieving the last encounter note written for that patient on or before the identified date. We ran different AI models, including BERT and Large Language Models (LLMs), to identify ADEs. ResultsWe annotated 1085 notes with 362 ADEs. These were used to train BERT models (precision or PPV: 0.584 ) and prompt-based LLM (0.563) but with improved sensitivity (0.725) and specificity (0.739) . ConclusionsIdentifying ADEs requires deep medical knowledge and simple Natural Language Processing models trying to identify ADEs only on the basis of language of encounter notes was not successful based on the BERT model performance. Prompt based models have deep knowledge but were trained with the objective of producing grammatically correct, coherent sentences, not with the medical knowledge required. LLMs also require significant computational power. ImplicationsWhile AI does have significant potential to advance our capacity for complex analytic tasks the current LLMs are not yet able to adequately identify ADEs in a corpus of electronic clinical notes.
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 | 0.037 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.009 | 0.004 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.004 | 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".