Paper

MedHELM: Holistic Evaluation of Large Language Models for Medical Tasks

Lee, Jennifer, Nateghi, Fateme, Kakkar, Vikas, Vedak, Shivam, Jain, Sneha S, Lungren, Matthew P, Revri, Anurang, Bedi, Suhana, Alekseyev, Alex, Unell, Alyssa, Chia-Chun Chiang, Reis, Eduardo, Koyejo, Sanmi, Kashyap, Mehr, Oez, Mert, Lin, Steven, Fuentes, Miguel, Keyes, Timothy, Aali, Asad, Fries, Jason Alan, Chen, Jonathan, Liang, Percy, Chung, Philip, Aydin Zahedivash, Goh, Ethan, Soetikno, Brian, Patel, Birju, Kotecha, Nikesh, Capasso, Robson, Sharma, Aditya, Saralkar, Rachna, Morse, Keith, Divi, Vasu, Daneshjou, Roxana, Qiu, Hao, Pham, Tho, Mawji, Bilal, Ravi, Nirmal, Mai, Yifan, Bethel Mieso, Lugtu, Carlene, Swaminathan, Akshay, Ghoddusi, Faraz, Alsentzer, Emily, Nayak, Ashwin, Patel, Hinesh, Aghaeepour, Nima, Kennedy, Vanessa, Shah, Nigam H, Gensheimer, Michael F, Dash, Dev, Gatidis, Sergios, Sharma, Harshita, Jain, Shrey, Pfeffer, Mike, Wornow, Michael, Jindal, Jenelle, Zhou, Vicky, Horvitz, Eric, Hong, Christy, Schulman, Kevin, Wang, Thomas, Black, Kameron, Chiou, Albert S, Mohana, Roy, Bannett, Yair, Fayanju, Oluseyi, Shah, Shreya, Cui, Hejie, Dong-han, Yao, Grolleau, Francois, Ambers, Nerissa, Chaudhari, Akshay, Helzer,

2025

Overview

While large language models (LLMs) achieve near-perfect scores on medical licensing exams, these evaluations inadequately reflect the complexity and diversity of real-world clinical practice. We introduce MedHELM, an extensible evaluation framework for assessing LLM performance for medical tasks with three key contributions. First, a clinician-validated taxonomy spanning 5 categories, 22 subcategories, and 121 tasks developed with 29 clinicians. Second, a comprehensive benchmark suite comprising 35 benchmarks (17 existing, 18 newly formulated) providing complete coverage of all categories and subcategories in the taxonomy. Third, a systematic comparison of LLMs with improved evaluation methods (using an LLM-jury) and a cost-performance analysis. Evaluation of 9 frontier LLMs, using the 35 benchmarks, revealed significant performance variation. Advanced reasoning models (DeepSeek R1: 66% win-rate; o3-mini: 64% win-rate) demonstrated superior performance, though Claude 3.5 Sonnet achieved comparable results at 40% lower estimated computational cost. On a normalized accuracy scale (0-1), most models performed strongly in Clinical Note Generation (0.73-0.85) and Patient Communication & Education (0.78-0.83), moderately in Medical Research Assistance (0.65-0.75), and generally lower in Clinical Decision Support (0.56-0.72) and Administration & Workflow (0.53-0.63). Our LLM-jury evaluation method achieved good agreement with clinician ratings (ICC = 0.47), surpassing both average clinician-clinician agreement (ICC = 0.43) and automated baselines including ROUGE-L (0.36) and BERTScore-F1 (0.44). Claude 3.5 Sonnet achieved comparable performance to top models at lower estimated cost. These findings highlight the importance of real-world, task-specific evaluation for medical use of LLMs and provides an open source framework to enable this.

• Publisher: Cornell University Library, arXiv.org
• Subject: Benchmarks / Computing costs / Large language models / Medical research / Performance evaluation / Taxonomy