MedPT: A Massive Medical Question Answering Dataset for Brazilian-Portuguese Speakers

TL;DR

MedPT addresses the critical lack of native Brazilian Portuguese medical data by introducing a large-scale, real-world question-answer corpus collected from patient-doctor interactions. The authors implement a rigorous multi-stage curation pipeline and LLM-assisted annotation to produce a high-signal resource spanning 3,200 topics and 104 specialties, with 384,095 QA pairs and rich metadata. They validate the dataset’s utility through a subspecialty-routing task using a 1.7B-parameter model, achieving a peak $0.94$ F1 on a 20-class setup and demonstrating strong few-shot learning gains, while qualitative analysis reveals clinically meaningful ambiguities in misclassifications. Publicly releasing MedPT aims to foster equitable, culturally aware medical NLP tools for the Portuguese-speaking world and beyond, enabling more accurate patient care and domain-adapted AI systems.

Abstract

While large language models (LLMs) show transformative potential in healthcare, their development remains focused on high-resource languages, creating a critical barrier for others as simple translation fails to capture unique clinical and cultural nuances, such as endemic diseases. To address this, we introduce MedPT, the first large-scale, real-world corpus for Brazilian Portuguese, comprising 384,095 authentic question-answer pairs from patient-doctor interactions. The dataset underwent a meticulous multi-stage curation protocol, using a hybrid quantitative-qualitative analysis to filter noise and contextually enrich thousands of ambiguous queries. We further augmented the corpus via LLM-driven annotation, classifying questions into seven semantic types to capture user intent. Our analysis reveals its thematic breadth (3,200 topics) and unique linguistic properties, like the natural asymmetry in patient-doctor communication. To validate its utility, we benchmark a medical specialty routing task: fine-tuning a 1.7B parameter model achieves an outstanding 94\% F1-score on a 20-class setup. Furthermore, our qualitative error analysis shows misclassifications are not random but reflect genuine clinical ambiguities (e.g., between comorbid conditions), proving the dataset's deep semantic richness. We publicly release MedPT to foster the development of more equitable, accurate, and culturally-aware medical technologies for the Portuguese-speaking world.

Figures (2)

• Figure 1: Distribution of medical questions by type.
• Figure 2: Distribution of the top five most represented medical professions by number of answers.