If you have or will use APC modeling, see the critiques, and pay attention to them in your measurement plans.
The MTF data ask about grade of first use, rather than age at first use of a drug, rendering the measurement problem for APC modeling even more difficult than it is in the typical public use data context when birthdate is not disclosed.
No need to repeat the MTF or NSDUH mistakes in your measurements or construction of variables.
Is the NSDUH approach a mistake? Only when the goal requires more refined data than the PUF offer us. It could offer us other information such as number of days from a specified origin to an event (or the month of an event) -- i.e., a constructed variable that gives the time dimension a fine-grained rather than coarse-grained unit of measurement.
Switching variables also could be provided, such as a switch to say whether the event occurred before or after a specific birthday.
Is the MTF approach a mistake? Here, grade of first use rather than age of first use is even more useless than what is faced in NSDUH.
Why a change in measurement approach has not followed the papers from MTF that speak of APC modeling escapes me. "Grade" is a very fuzzy construct in demography and in epidemiology. Its utility in APC models is not at all clear, for reasons described in this paper , with some very nice graphical displays:
