Enabling Drug-Induced Liver Injury Surveillance Through Automated Medication Extraction From Clinical Notes: A Medical Information Mart for Intensive Care IV Real-World Large Language Models Validation Study
Bibliographic record
Abstract
Background: Drug-induced liver injury (DILI) presents a significant diagnostic challenge, often leading to delayed detection. Unstructured clinical notes contain comprehensive medication data vital for DILI surveillance but are difficult to analyze systematically. Large language models (LLMs) show promise for automated extraction but require real-world clinical data validation to assess feasibility for clinical applications like DILI surveillance. Methods: We retrospectively validated an LLM system on 100 randomly sampled Medical Information Mart for Intensive Care IV (MIMIC-IV) discharge summaries. Gold standard unique medication lists were derived via manual annotation and manual deduplication based on normalized drug names. LLM outputs underwent identical deduplication. Performance was assessed using precision, recall, and F1-score comparing deduplicated lists. MIMIC-IV data use agreement (DUA) compliance was ensured. Results: Comparison yielded a precision of 0.85, recall of 1.00, and an F1-score of 0.92 for unique medication identification. The 174 false positives resulted from parsing or normalization errors; no medication hallucinations occurred. A subsequent DILI database lookup failed for approximately 6.2% of correctly identified unique medications, evaluated as a separate feasibility measure. Conclusions: The LLM demonstrates high accuracy and perfect recall for unique medication extraction and identification from complex clinical notes, establishing technical feasibility. This represents a feasible and possible integration of LLM towards developing automated tools for enhanced DILI surveillance and improved patient safety.
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 | 0.005 | 0.002 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 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".