Whether these drugs reduce drinking has been asked several times, including in the trials covered in two trials that asked different questions. What had not been asked is what happens when people stop taking them, and a Swedish register study built specifically to answer that [1].
The design deserves a moment. Rather than matching people on a drug to different people not on it, this study compared each person with themselves, contrasting periods when they were taking a GLP-1 drug against periods when they were not. Everything stable about a person — their genetics, their social circumstances, their baseline relationship with alcohol — cancels out by construction, which removes the confounding that troubles most of this literature. Across 167,026 residents of Stockholm County with type 2 diabetes, 41,109 received a GLP-1 drug at some point during the study period.
While people were taking one, alcohol-related hospitalizations ran at a rate ratio of 0.55 (95% CI 0.43–0.70) against their own unexposed periods, and substance-related hospitalizations at 0.61 (95% CI 0.50–0.74). DPP-4 inhibitors, included as a comparison treatment, produced no consistent reductions, which suggests what was measured is not simply the effect of being under active medical care. How much that conclusion depends on the comparison chosen is the subject of the overdose study with five comparators.
The discontinuation windows are the new information. In the first 182 days after stopping, the alcohol association was still there but weaker, at 0.70 (95% CI 0.50–0.98). In days 183 to 364 it had gone, at 1.07 (95% CI 0.71–1.63), an interval sitting squarely across no effect. For substance use disorder more broadly, neither post-discontinuation window reached significance at all.
The practical reading the authors themselves draw is that substance use deserves clinical monitoring when one of these drugs is stopped. That matters more than it might sound, because stopping is the normal outcome rather than the exception, as other register studies of this kind and the discontinuation literature both show. If a benefit exists and fades within a year of stopping, then the population most exposed to that fade is large, and largely unwatched. The weight side of the same stopping problem is covered in what happens to muscle.