Fine-Tune, Don't Prompt, Your Language Model to Identify Biased Language in Clinical Notes

• URL: http://arxiv.org/abs/2603.10004v1
• Date: Mon, 16 Feb 2026 22:39:28 GMT
• Title: Fine-Tune, Don't Prompt, Your Language Model to Identify Biased Language in Clinical Notes
• Abstract summary: Clinical documentation can contain emotionally charged language with stigmatizing or privileging valences.<n>We present a framework for detecting and classifying such language.<n>Fine-tuning with lexically primed inputs consistently outperformed prompting approaches.
• Score: 31.925112343488976
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Clinical documentation can contain emotionally charged language with stigmatizing or privileging valences. We present a framework for detecting and classifying such language as stigmatizing, privileging, or neutral. We constructed a curated lexicon of biased terms scored for emotional valence. We then used lexicon-based matching to extract text chunks from OB-GYN delivery notes (Mount Sinai Hospital, NY) and MIMIC-IV discharge summaries across multiple specialties. Three clinicians annotated all chunks, enabling characterization of valence patterns across specialties and healthcare systems. We benchmarked multiple classification strategies (zero-shot prompting, in-context learning, and supervised fine-tuning) across encoder-only models (GatorTron) and generative large language models (Llama). Fine-tuning with lexically primed inputs consistently outperformed prompting approaches. GatorTron achieved an F1 score of 0.96 on the OB-GYN test set, outperforming larger generative models while requiring minimal prompt engineering and fewer computational resources. External validation on MIMIC-IV revealed limited cross-domain generalizability (F1 < 0.70, 44% drop). Training on the broader MIMIC-IV dataset improved generalizability when testing on OB-GYN (F1 = 0.71, 11% drop), but at the cost of reduced precision. Our findings demonstrate that fine-tuning outperforms prompting for emotional valence classification and that models must be adapted to specific medical specialties to achieve clinically appropriate performance. The same terms can carry different emotional valences across specialties: words with clinical meaning in one context may be stigmatizing in another. For bias detection, where misclassification risks undermining clinician trust or perpetuating patient harm, specialty-specific fine-tuning is essential to capture these semantic shifts. * Equal contribution.

Related papers

• Toward Better Assessment of LLMs' Performance in Clinical Error Detection [13.451007655769827]
  We evaluate 15 large language models (LLMs) for automated detection of errors in clinical documentation.<n>We find that 13 of 15 models fall below the level of random pairwise discrimination, even while achieving F1 scores that standard practice would read as moderate.<n>We show that F1 and pairwise accuracy are driven in opposite directions by the same underlying bias, so that ranking models by F1 may systematically promote the weakest discriminators.
  arXiv  Detail & Related papers  (2026-08-17T14:41:53Z)
• Beyond Scalar Scores: Exploring LLM-based Metrics for Clinical Significance Evaluation in Radiology Reports [49.5225801722164]
  Existing metrics obscure this requirement by reducing report quality to a medically ungrounded scalar.<n>We study this boundary using ReEvalMed benchmark as testbed and evaluate metric-level clinical significance.
  arXiv  Detail & Related papers  (2026-06-17T08:10:30Z)
• Measuring What VLMs Don't Say: Validation Metrics Hide Clinical Terminology Erasure in Radiology Report Generation [10.15221228043609]
  This paper investigates the use of decoding strategies that lead to high aggregate token-overlap scores despite template collapse.<n>We introduce Clinical Association Displacement (CAD), a vocabulary-level framework that quantifies shifts in demographic-based word associations in generated reports.<n>We show that deterministic decoding produces high levels of semantic erasure, while sampling generates diverse outputs but risks introducing new bias.
  arXiv  Detail & Related papers  (2026-03-02T08:59:39Z)
• A Federated and Parameter-Efficient Framework for Large Language Model Training in Medicine [59.78991974851707]
  Large language models (LLMs) have demonstrated strong performance on medical benchmarks, including question answering and diagnosis.<n>Most medical LLMs are trained on data from a single institution, which faces limitations in generalizability and safety in heterogeneous systems.<n>We introduce the model-agnostic and parameter-efficient federated learning framework for adapting LLMs to medical applications.
  arXiv  Detail & Related papers  (2026-01-29T18:48:21Z)
• From Generative Modeling to Clinical Classification: A GPT-Based Architecture for EHR Notes [0.0]
  This study presents a GPT-based architecture for clinical text classification.<n>Rather than updating all model parameters, the majority of the GPT-2 backbone is frozen.<n>The proposed method is evaluated on radiology reports from the MIMIC-IV-Note dataset.
  arXiv  Detail & Related papers  (2026-01-29T16:33:47Z)
• Specialised or Generic? Tokenization Choices for Radiology Language Models [2.081299660192454]
  vocabulary used by language models (LM) plays a key role in text generation quality.<n>General, medical, and domain-specific tokenizers on the task of radiology report summarisation across three imaging modalities.<n>Our findings demonstrate that medical and domain-specific vocabularies outperformed widely used natural language alternatives when models are trained from scratch.
  arXiv  Detail & Related papers  (2025-08-13T17:13:56Z)
• Contrastive Learning with Counterfactual Explanations for Radiology Report Generation [83.30609465252441]
  We propose a textbfCountertextbfFactual textbfExplanations-based framework (CoFE) for radiology report generation. Counterfactual explanations serve as a potent tool for understanding how decisions made by algorithms can be changed by asking what if'' scenarios. Experiments on two benchmarks demonstrate that leveraging the counterfactual explanations enables CoFE to generate semantically coherent and factually complete reports.
  arXiv  Detail & Related papers  (2024-07-19T17:24:25Z)
• Making the Most Out of the Limited Context Length: Predictive Power Varies with Clinical Note Type and Note Section [70.37720062263176]
  We propose a framework to analyze the sections with high predictive power. Using MIMIC-III, we show that: 1) predictive power distribution is different between nursing notes and discharge notes and 2) combining different types of notes could improve performance when the context length is large.
  arXiv  Detail & Related papers  (2023-07-13T20:04:05Z)
• Language Model Classifier Aligns Better with Physician Word Sensitivity than XGBoost on Readmission Prediction [86.15787587540132]
  We introduce sensitivity score, a metric that scrutinizes models' behaviors at the vocabulary level. Our experiments compare the decision-making logic of clinicians and classifiers based on rank correlations of sensitivity scores.
  arXiv  Detail & Related papers  (2022-11-13T23:59:11Z)
• Few-Shot Cross-lingual Transfer for Coarse-grained De-identification of Code-Mixed Clinical Texts [56.72488923420374]
  Pre-trained language models (LMs) have shown great potential for cross-lingual transfer in low-resource settings. We show the few-shot cross-lingual transfer property of LMs for named recognition (NER) and apply it to solve a low-resource and real-world challenge of code-mixed (Spanish-Catalan) clinical notes de-identification in the stroke.
  arXiv  Detail & Related papers  (2022-04-10T21:46:52Z)
• Hurtful Words: Quantifying Biases in Clinical Contextual Word Embeddings [16.136832979324467]
  We pretrain deep embedding models (BERT) on medical notes from the MIMIC-III hospital dataset. We identify dangerous latent relationships that are captured by the contextual word embeddings. We evaluate performance gaps across different definitions of fairness on over 50 downstream clinical prediction tasks.
  arXiv  Detail & Related papers  (2020-03-11T23:21:14Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.