Fill drug columns with "skipped" to preserve the semantic meaning (drug not given) and use NaN only for true missing or unknown values in other columns. This separation helps downstream analysis or models distinguish between absence-by-design and missing data.