Large Language Models (LLMs) have shown potential in medical data interpretation, but accurately analyzing clinical indicators like blood pressure, fasting glucose, total cholesterol, and low-density lipoprotein cholesterol remains challenging due to their complexity. This study evaluates LLM performance in interpreting these indicators and assesses the effect of providing explicit diagnostic criteria. We used 1,000 health checkup records from the Korean National Health Insurance Service and tested four LLMs: Claude 3.5 Sonnet, Gemini 1.5 Pro, GPT-4o, and LLaMA 3.1-70B. We analyzed basic prompt and constraint prompt with clear criteria. Providing criteria significantly improved accuracy, especially for fasting glucose and cholesterol, where most models achieved over 0.95 accuracy. Blood pressure interpretation showed less improvement due to its complexity. This study confirms that structured prompts enhance LLM performance, especially for complex clinical data.