Language Models Lag Behind: Inconsistent Alignment with Recent Removal of Race from Clinical Algorithms
Bibliographic record
Abstract
<bold>BACKGROUND:</bold> Several medical societies have revised commonly used clinical algorithms to remove the social construct of race as a predictor. Given the rise in the adoption of large language models (LLMs), we assessed whether LLMs trained since October 2023 are aware of the recent removal of race from lung function reference equations and models used to estimate Glomerular Filtration Rate (eGFR). <bold>METHODS:</bold> We evaluated a combination of 12 state-of-the-art and lightweight language models by OpenAI, Meta, Microsoft, and DeepSeek. We used various prompts to ask LLMs to identify the latest FEV<sub>1</sub> and eGFR reference models and their input parameters, as recommended by the ERS/American Thoracic Society (ATS) and the American Society of Nephrology (ASN), respectively. We ran each prompt-model combination five times with default parameters and compared model outputs with the Medical Society's recommendations. <bold>RESULTS:</bold> Regardless of model size and reasoning capability, less than 3% of model outputs correctly identified the removal of race from lung function equations. In contrast, 100% of outputs from state-of-the-art models and 25% from lightweight models correctly identified the removal of race from eGFR (Figure). No statistically significant difference was observed between open-source and proprietary models. <bold>CONCLUSIONS:</bold> Evaluated LLMs consistently failed to correctly identify the latest recommendations of ERS and ATS. <fig><object-id>erj;66/suppl_69/PA2023/F1</object-id><object-id>F1</object-id><object-id>F1</object-id><graphic></graphic></fig>
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 | 0.002 | 0.000 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| 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 | 0.002 | 0.002 |
| Research integrity | 0.000 | 0.001 |
| 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".