Enhancing Precision in Detecting Severe Immune-Related Adverse Events: Comparative Analysis of Large Language Models and International Classification of Disease Codes in Patient Records
Bibliographic record
Abstract
PURPOSE Current approaches to accurately identify immune-related adverse events (irAEs) in large retrospective studies are limited. Large language models (LLMs) offer a potential solution to this challenge, given their high performance in natural language comprehension tasks. Therefore, we investigated the use of an LLM to identify irAEs among hospitalized patients, comparing its performance with manual adjudication and International Classification of Disease (ICD) codes. METHODS Hospital admissions of patients receiving immune checkpoint inhibitor (ICI) therapy at a single institution from February 5, 2011, to September 5, 2023, were individually reviewed and adjudicated for the presence of irAEs. ICD codes and an LLM with retrieval-augmented generation were applied to detect frequent irAEs (ICI-induced colitis, hepatitis, and pneumonitis) and the most fatal irAE (ICI-myocarditis) from electronic health records. The performance between ICD codes and LLM was compared via sensitivity and specificity with an α = .05, relative to the gold standard of manual adjudication. External validation was performed using a data set of hospital admissions from June 1, 2018, to May 31, 2019, from a second institution. RESULTS Of the 7,555 admissions for patients on ICI therapy in the initial cohort, 2.0% were adjudicated to be due to ICI-colitis, 1.1% ICI-hepatitis, 0.7% ICI-pneumonitis, and 0.8% ICI-myocarditis. The LLM demonstrated higher sensitivity than ICD codes (94.7% v 68.7%), achieving significance for ICI-hepatitis ( P < .001), myocarditis ( P < .001), and pneumonitis ( P = .003) while yielding similar specificities (93.7% v 92.4%). The LLM spent an average of 9.53 seconds/chart in comparison with an estimated 15 minutes for adjudication. In the validation cohort (N = 1,270), the mean LLM sensitivity and specificity were 98.1% and 95.7%, respectively. CONCLUSION LLMs are a useful tool for the detection of irAEs, outperforming ICD codes in sensitivity and adjudication in efficiency.
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 | 0.040 | 0.117 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.002 | 0.003 |
| Bibliometrics | 0.005 | 0.002 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.003 | 0.002 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 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 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".