See my parallel comment... are they reporting 1/4 of a household eats 25% individually (for a 5% reduction across a family of 4). Or, does one person eating 5% less cause the other 3 to also eat 5%... if it's the latter, this potentially has health impacts across people who aren't taking the drug.
If the dataset includes household size you could limit the analysis to that subset. Alternatively you could focus on households with lower average basket sizes and hypothesise that those households have fewer adults. Obviously those measure aren't going to be anywhere near perfect but might be informative.
