Your math is bad. 1% of recipients is nowhere near the same as the 1% of smokers you used.
Therefore, to generally reduce the population of recipients you’d have to reduce the population of smokers.
Unless you have data showing a specific population skew or sampling issue I’d say it is perfectly good math, albeit with an assumption.
Now, of course, you could simply deny service to 1% of smokers who would be recipients—or other extremely skewed sampling attacks—but that is somewhat unlikely and in most cases directly against the goals of the program.