MedCalc-Bench: Evaluating Large Language Models for Medical Calculations

TL;DR

MedCalc-Bench introduces the first dedicated benchmark for evaluating LLMs on medical calculations, addressing a gap where clinicians rely on rule-based and equation-based calculators. The authors curate 55 MDCalc-based tasks, assemble 1047 instances with ground-truth answers and explanations from real and synthesized notes, and evaluate a wide range of models under zero-shot and few-shot prompting. Results show GPT-4 leads with about 50.9% accuracy in a one-shot CoT setting, while open and smaller models lag, highlighting persistent gaps in calculator knowledge, parameter extraction, and arithmetic. Through detailed error analysis, the work identifies primary bottlenecks and offers concrete directions (enhanced domain knowledge, robust extraction, and reliable arithmetic) for advancing medical calculation capabilities in LLMs, underscoring the need for continued development before clinical deployment.

Abstract

As opposed to evaluating computation and logic-based reasoning, current benchmarks for evaluating large language models (LLMs) in medicine are primarily focused on question-answering involving domain knowledge and descriptive reasoning. While such qualitative capabilities are vital to medical diagnosis, in real-world scenarios, doctors frequently use clinical calculators that follow quantitative equations and rule-based reasoning paradigms for evidence-based decision support. To this end, we propose MedCalc-Bench, a first-of-its-kind dataset focused on evaluating the medical calculation capability of LLMs. MedCalc-Bench contains an evaluation set of over 1000 manually reviewed instances from 55 different medical calculation tasks. Each instance in MedCalc-Bench consists of a patient note, a question requesting to compute a specific medical value, a ground truth answer, and a step-by-step explanation showing how the answer is obtained. While our evaluation results show the potential of LLMs in this area, none of them are effective enough for clinical settings. Common issues include extracting the incorrect entities, not using the correct equation or rules for a calculation task, or incorrectly performing the arithmetic for the computation. We hope our study highlights the quantitative knowledge and reasoning gaps in LLMs within medical settings, encouraging future improvements of LLMs for various clinical calculation tasks.

Figures (3)

• Figure 1: Example instances of the MedCalc-Bench dataset.
• Figure 2: Accuracy vs. Number of Parameters for Equation-Based Calculators
• Figure 3: Accuracy vs. Number of Parameters for Rule-Based Calculators