The effectiveness of amphetamine (e.g. Adderall) and methylphenidate (e.g. Ritalin) for the symptoms of ADHD, that is, lack of executive function, has been demonstrated beyond reasonable doubt in many short-term studies. (Mandatory reference) Does it last though? Well, the literature is shockingly sparse, but I was able to extract three main arguments:

1. (against tolerance, strong): Longer studies (RCTs) don't have worse results than shorter ones
2. (against tolerance, medium): People seem to take the same dosages for many years (i.e. no dose escalation)
3. (for tolerance, strong): In a large, long RCT (the MTA study), effects disappeared after three years in the observational followup. Yes, they've considered selection effects.

Argument I: RCT Length vs. Effect Size

If tolerance exists, then longer studies should find smaller effects than shorter studies. This is the best kind of design: Splitting people into two groups for a time T, comparing the difference between amphetamine and sugar-pill people. In studies with longer T, this is not smaller than in studies with smaller T.

So if tolerance exists at all, it must happen very slowly, on the scale of months to years. (subtle foreshadowing)

However, the authors claim, with reference to this study, that while development of tolerance to CNS Drugs is a priori likely, is rarely takes more than 28 weeks.

Argument II: The Stable Swedish Doses

The dose of Swedish People generally seems to stabilize over time. If there was tolerance, we would expect the dose to creep up endlessly. Notice the dropout (people don't want to take Adderall that long) and, in general, observational nature of the data however. Lastly, the doses are already pretty close to the max "doctor won't look at you weird" dose, so maybe people just give up trying to get higher doses prescribed, even though the effect is declining.

Counterargument I: The MTA Study

Scott Alexander writes that in the large and long MTA study, after three years the original stimulant effect seems to have disappeared. One might claim that people being allowed to switch treatment groups "drowned out any original effect". The authors of the MTA study seem not convinced, but why?

For any selection to occur, underlying disease severity would have to correlate with switching groups, e.g. people with light ADHD drop medication, and people with severe ADHD start it, thus the underlying effect is drowned out. Using the many other participant variables the MTA study collected, we can try to estimate "how bad is the ADHD of this person". Then, we split people into five groups with similar estimated severity, and among these subgroups, there is still no effect.

However, this argument is always only as good as the prognostic power of our measured "standard things you should control for" variables, therefore selection effects cannot be ruled out with total certainty.

The Rest of the Literature Isn't Good

The other arguments don't seem that good! The only systematic review of the subject published to date seems very much overconfident, and the popular science picked up on the "no evidence for tolerance" claim, calling it the "tolerance myth"

For example, they write:

Allen monitored the behavioural effects of methylphenidate in 108 children treated for 3–10 years. They observed that the dose of methylphenidate, once adjusted for growth, did not significantly change during the follow-up. Nevertheless, they warned that using [milligrams per kilogram doses] may overcorrect for growth with increasing age.".

Another way to read that is "as the children grew up, the dose crept up". That is, because the effect of stimulants isn't strongly mediated by weight or height, uncontrolling for it would reveal a diminishing effect, which is evidence for tolerance. To be fair, you can also read that in the next sentence.

" Retrospective chart review of 166 children and adolescents attending a specialist service between 1976 and 1990 noted that 60% of those receiving off-label high doses of methylphenidate developed tolerance (defined as a failure to maintain a clinical response to the same dose), but none of those receiving less than 60 mg/day did. Unfortunately, the authors did not explore the potential reasons underlying the apparent loss of efficacy. Nevertheless, participants who developed tolerance responded well to d-amphetamine or tolerance resolved after a month’s medication break "

This is almost tautologically true: If someone were to develop tolerance, they would escalate the dose above 60mg. Likewise, almost by definition, the children who haven't climbed there haven't developed tolerance yet.

" In line with these findings, two randomised withdrawal studies have shown that stimulants remain effective for most individuals after 6 or 24 months of treatment, although a minority were able to stop the medication without relapsing".

Randomized withdrawal trials work by collecting a bunch of people with adderall bottles, and then replacing the pills of one group with sugar ones, to see how they do in comparison to the still-on-adderall group. Obviously the sugar-pill-people could also do as bad as they do because their receptors have been downregulated, meaning the observed evidence doesn't disprove tolerance.

"Alternative explanations, such as natural fluctuations of symptoms, limited compliance, life events and co-occurrent mental health conditions."

First of all, if there is an explanation, there is an explanandum, that is, there seems to be a need to explain the apparent loss of stimulant-effect over time. Some patients seem to be complaining, better write an overconfident review to calm their worries.

Natural fluctuations and life events are not a good candidate here, because we know from both RCT/short-term study evidence (i.e. in the placebo group) and declining severity of ADHD with age that symptoms slowly but surely improve over time, so it ought to cut in the other direction!

There is also evidence I haven't covered here. First, the general prior of CNS Drugs inducing tolerance, the brain building more dopamine-sucking things in response to Ritalin, and whatever stuff you can learn from lab animals. Lastly, since recreational stimulant use induces tolerance (e.g. smoking meth), and the drug-liking scores (it's a "how was it?" Likert scale) in studies seem to decline quickly, one could carefully review if this is generalizable.

Stimulants are the only thing that move the needle much (and oh is it much) on executive function, which is highly heritable (I mean even nature.com says it), and doesn't respond well to CBT or other non-medication options. Insofar as any tolerance exists, it's probably more on the scale of months to years, courtesy of Argument I. In conclusion, despite much effort, I can't get myself to say more than "consider medication breaks once in a while" here.

(and ask your doctor or whatever; not medical advice)