# DeVisE: Behavioral Testing of Medical Large Language Models

arXiv: https://arxiv.org/abs/2506.15339 · HTML: https://mlsanity.com/paper/2506.15339/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 12 of its 35 references that are in the map.
- First posted on arXiv 18 Jun 2025.

## Builds on (most cited first, 12 of 12)

- [Domain-Specific Language Model Pretraining for Biomedical Natural Language Processing](https://mlsanity.com/paper/2007.15779/) (2020)
- [Beyond Accuracy: Behavioral Testing of NLP models with CheckList](https://mlsanity.com/paper/2005.04118/) (2020)
- [Adapted Large Language Models Can Outperform Medical Experts in Clinical Text Summarization](https://mlsanity.com/paper/2309.07430/) (2023)
- [VALSE: A Task-Independent Benchmark for Vision and Language Models Centered on Linguistic Phenomena](https://mlsanity.com/paper/2112.07566/) (2021)
- [Clinical Outcome Prediction from Admission Notes using Self-Supervised Knowledge Integration](https://mlsanity.com/paper/2102.04110/) (2021)
- [ViLMA: A Zero-Shot Benchmark for Linguistic and Temporal Grounding in Video-Language Models](https://mlsanity.com/paper/2311.07022/) (2023)
- [SemEval-2024 Task 2: Safe Biomedical Natural Language Inference for Clinical Trials](https://mlsanity.com/paper/2404.04963/) (2024)
- [Predicting Intensive Care Unit Length of Stay and Mortality Using Patient Vital Signs: Machine Learning Model Development and Validation](https://mlsanity.com/paper/2105.04414/) (2021)
- [Analysing zero-shot temporal relation extraction on clinical notes using temporal consistency](https://mlsanity.com/paper/2406.11486/) (2024)
- [What Do You See in this Patient? Behavioral Testing of Clinical NLP Models](https://mlsanity.com/paper/2111.15512/) (2021)
- [SEME at SemEval-2024 Task 2: Comparing Masked and Generative Language Models on Natural Language Inference for Clinical Trials](https://mlsanity.com/paper/2404.03977/) (2024)
- [Evaluating the Robustness of Adverse Drug Event Classification Models Using Templates](https://mlsanity.com/paper/2407.02432/) (2024)

## Built on by (most cited first, 2 of 2)

- [MEDEQUALQA: Evaluating Biases in LLMs with Counterfactual Reasoning](https://mlsanity.com/paper/2510.12818/) (2025)
- [How Well Do LLMs Understand Drug Mechanisms? A Knowledge + Reasoning Evaluation Dataset](https://mlsanity.com/paper/2511.06418/) (2025)

## Abstract (arXiv)

Large language models (LLMs) are increasingly applied in clinical decision support, yet current evaluations rarely reveal whether their outputs reflect genuine medical reasoning or superficial correlations. We introduce DeVisE (Demographics and Vital signs Evaluation), a behavioral testing framework that probes fine-grained clinical understanding through controlled counterfactuals. Using intensive care unit (ICU) discharge notes from MIMIC-IV, we construct both raw (real-world) and template-based (synthetic) variants with single-variable perturbations in demographic (age, gender, ethnicity) and vital sign attributes. We evaluate eight LLMs, spanning general-purpose and medical variants, under zero-shot setting. Model behavior is analyzed through (1) input-level sensitivity, capturing how counterfactuals alter perplexity, and (2) downstream reasoning, measuring their effect on predicted ICU length-of-stay and mortality. Overall, our results show that standard task metrics obscure clinically relevant differences in model behavior, with models differing substantially in how consistently and proportionally they adjust predictions to counterfactual perturbations.