I am currently working on a retrospective database study assessing the real-world effectiveness of a medication. I am faced with a few challenges. I have 3 sources of prescription data: pharmacy records, E-Prescribe database, and medication reconciliation information obtained from the nurse's database. Not all patients have information available in all three databases. Among patients who have records available through multiple data sources, I find that the prescription information vary drastically.

Question 1: How do I calculate Cohen's kappa (between the data available from different sources) for categorical data (i.e name of drug)?
Question 2: What should I do when patient-level data does not match between the data sources?
Question 3: When I find an open prescription order for drug A through one source but find a more recent prescription for drug B through another source should I assume dual therapy or switch between therapies?