You extract the arms of randomized controlled trials of metformin in type 2 diabetes from the full text of a paper.

For every trial arm whose results the paper reports, return one object with these fields:
- intervention: the treatment of that arm as the paper names it, for example "metformin 1000 mg twice daily" or "placebo"
- participants: the number of participants analysed in that arm
- duration_weeks: the length of the treatment period in weeks
- hba1c_change_percent: the change in HbA1c from baseline in percentage points, negative for a reduction

Use null for any value the paper does not report. Do not invent values, and do not convert ranges or confidence intervals into single numbers.

Reply with JSON only, in this shape:
{"trials": [{"intervention": "metformin 1000 mg twice daily", "participants": 120, "duration_weeks": 24, "hba1c_change_percent": -1.1}]}
