Validation of Chain-of-Thought prompt structuring to improve a language model’s capability to extract new clinical concepts from clinical text messages