This paper compares the performance of BERT and GPT in Named Entity Recognition (NER) for Spanish electronic health records (EHRs), focusing on breast cancer data. The study evaluates two approaches: fine-tuning BERT with annotated data and using GPT with few-shot learning and external knowledge. The results show that both methods achieve comparable performance in terms of precision, recall, and F-score. However, GPT demonstrates slightly better overall effectiveness, particularly in F-score, and requires minimal data annotation. The study highlights the potential of GPT in medical data processing due to its ability to handle complex medical terminologies and contextual nuances without extensive training. The findings suggest that GPT, especially with few-shot learning and external knowledge, can be a viable alternative to traditional NER methods like BERT for structuring Spanish EHRs. The research also emphasizes the importance of prompt design in leveraging GPT's capabilities for accurate entity extraction. The study concludes that while BERT offers high precision, GPT's adaptability and efficiency make it a promising tool for medical NER tasks, particularly in resource-constrained environments. The results indicate that GPT can significantly reduce the need for extensive data annotation, making it more efficient for real-time applications in healthcare.This paper compares the performance of BERT and GPT in Named Entity Recognition (NER) for Spanish electronic health records (EHRs), focusing on breast cancer data. The study evaluates two approaches: fine-tuning BERT with annotated data and using GPT with few-shot learning and external knowledge. The results show that both methods achieve comparable performance in terms of precision, recall, and F-score. However, GPT demonstrates slightly better overall effectiveness, particularly in F-score, and requires minimal data annotation. The study highlights the potential of GPT in medical data processing due to its ability to handle complex medical terminologies and contextual nuances without extensive training. The findings suggest that GPT, especially with few-shot learning and external knowledge, can be a viable alternative to traditional NER methods like BERT for structuring Spanish EHRs. The research also emphasizes the importance of prompt design in leveraging GPT's capabilities for accurate entity extraction. The study concludes that while BERT offers high precision, GPT's adaptability and efficiency make it a promising tool for medical NER tasks, particularly in resource-constrained environments. The results indicate that GPT can significantly reduce the need for extensive data annotation, making it more efficient for real-time applications in healthcare.