No trial has tested it, and the observational evidence now runs to a few thousand people. A matched cohort at a large dialysis organization followed 2,492 patients who started in-center hemodialysis while on one of these drugs, against 2,492 who did not. Hospitalization ran 9% lower and mortality 17% lower [2]. Before that cohort, the entire world literature was 388 patients across 19 studies [1]. The groups in the larger study differed substantially before anyone was treated, so the honest summary is encouraging rather than established.
Everything known about these drugs comes from people whose kidneys work well enough to be in a trial. This is what the evidence looks like for people whose kidneys have failed, and the shape of it is the point.
Who is missing from the trials
The large cardiovascular and weight-loss trials exclude end-stage kidney disease, and the kidney trial that made news enrolled people with chronic kidney disease rather than people on dialysis. That is a distinction the coverage rarely draws. So a patient on dialysis, who may need to lose weight to qualify for a transplant, is asking a question no randomized trial has answered.
This review gathered what exists: 19 studies, 388 patients, 91.5% on renal replacement therapy [1]. Semaglutide accounted for 266 of them, liraglutide 66, dulaglutide 47, and tirzepatide nine. Nine patients, in the world literature.
What the pooled studies found
On semaglutide, weight fell 3.09 kg at three months, 95% CI −5.95 to −0.24, and 3.68 kg at six months, 95% CI −5.71 to −1.65. At twelve months it was 7.00 kg, 95% CI −11.38 to −2.61, with heterogeneity of 79.2%. Glycated hemoglobin fell 0.75 percentage points at twelve months, 95% CI −1.07 to −0.43.
Some estimates did not exclude no effect: BMI at three months, 95% CI −2.92 to 0.09, and HbA1c at six months, 95% CI −1.20 to 0.20. Both liraglutide weight estimates crossed zero as well.
The estimates that are usable
Not all of it is like that. Any gastrointestinal event came in at 39%, 95% CI 25% to 55%, and discontinuation because of side effects at 9%, 95% CI 4% to 20%. Those are wide but they say something: side effects are common in this population and roughly one person in eleven stopped because of them.
The twelve-month weight figure is similar: a real effect with an interval running from about two and a half to eleven kilograms. Heterogeneity was high enough that the studies disagreed substantially. It is enough to say the drug does something and not enough to say how much, which a high heterogeneity statistic always implies.
The cohort that multiplied the evidence base
A matched cohort then followed 2,492 patients who began in-center hemodialysis while on one of these drugs, against an equal number who did not [2]. Hospitalization ran 9% lower (incidence rate ratio 0.91, 95% CI 0.87 to 0.96) and mortality 17% lower (IRR 0.83, 95% CI 0.73 to 0.95). Both are incidence rate ratios rather than hazard ratios, because hospitalization recurs and the model counts episodes.
The baseline table is the part to read first. Among patients on a GLP-1 drug at dialysis initiation, 98% had diagnosed diabetes against 73% of the comparison group. Predialysis nephrology care had reached 76% against 67%, and mean BMI was 33 against 30. Predialysis nephrology care is one of the best-established predictors of doing well on dialysis and has nothing to do with any drug. The full reading is in the dialysis cohort.
What the trial evidence one stage earlier says
FLOW enrolled people with type 2 diabetes and chronic kidney disease, at semaglutide 1.0 mg, and its subgroup analyses publish numbers needed to treat [3]. To prevent one primary kidney outcome over three years you would treat 22 people with established atherosclerotic disease, 13 with heart failure, or 17 at high cardiovascular risk. Every interaction test came back null.
Those numbers describe kidney protection in diabetic kidney disease at a diabetes dose. They do not transfer to someone taking a weight-management dose without either condition, and they do not transfer to dialysis. The arithmetic is in the FLOW numbers needed to treat.
A measurement problem that runs through all of it
Kidney function is usually reported as an estimated glomerular filtration rate, which is calculated from creatinine, and creatinine is a waste product of muscle. A drug that takes muscle mass off lowers creatinine and raises the calculated rate with the kidney doing exactly what it was doing before.
A matched study in polycystic kidney disease shows the pattern cleanly. Both eGFR and a stage threshold defined off eGFR improved, while reaching end-stage disease, which does not depend on creatinine, did not [4]. That is what a measurement artifact looks like, and also what an underpowered hard endpoint looks like in 320 pairs. The check is set out in the eGFR and muscle analysis.
What the authors conclude
That semaglutide may be considered in end-stage kidney disease, that the results are primarily hypothesis-generating, and that prospective randomized trials are urgently needed. All three clauses belong together. A drug can be reasonable to try in a population where nothing else works while the evidence for it remains this thin.
Why it belongs on a site about buying
Because nobody on this roster screens for kidney failure, and a person on dialysis filling out a telehealth intake form is in a category the whole evidence base excludes. Dosing, clearance and hypoglycemia risk are all different when kidneys have stopped working, and none of that is a question an online questionnaire is built to catch. The census of what goes unasked counts how much.
More generally, this is the clearest example available of what the gap between trial populations and real ones actually costs. Exclusion criteria are written to protect a trial’s result. They also decide, years later, which patients have evidence and which have nineteen small studies.