Drugs and the harm they cause to an individual and others is highly subjective and contextualized within a culture/society. There is not going to be some math formula to give you the answer.
If you wanted to know the best web application framework to build your next REST API, would you build a mathematical theory for deducing it, or find out what the relevant experts in the field think?
I disagree. I am a psychiatrist and participated in some clinical research. Step 0 is quantifying. It's not always perfect, but if you don't know what you are measuring, your measurements are useless.
The idea that any one set of metrics and measurements have more intrinsic truth-value than simply aggregating the internal mental weights of all the relevant experts in the field is just like, your opinion man.
If you want to present an alternate, mathematical model for drug harm and explain and argue for the validity of the choices you make for how to measure harm and how the data was gathered over what time period, etc etc - no one's stopping you.
If you wanted to make a legitimate measure of the harm of various drugs despite the inability to build a mathematical model, you could do a survey asking people about how often they use various drugs and their opinion of the harm they've been caused by various drugs (as user or non-user). You'd get systemic bias from peoples imperfect ability to judge harm but at least you'd have something instead of nothing.
IIRC they asked psychiatrists and not pharmacologists.
Well, see, that's the tricky bit right there.
What I don't understand about these studies is what mathematical basis they use for comparing the relative impact of these various factors. I would assume that these would all be completely different, and as a consequence, incommensurable. However, studies like this seem to have found appropriate relative weights for these factors, and were able to add them all together. I would be interested in knowing if it's really possible to do this without producing a figure that's totally bogus, and if so, how.