1. do randomized control against *inert* placebo 2. measure net health benefit over 10 years

If #1 and #2 look good, by all means roll this out.