News · Heart & Metabolic
A diabetes drug tracked a third fewer deaths in over-75s, alongside a stroke signal nobody can explain
Trials of SGLT2 inhibitors barely recruited the very old. A Japanese target trial emulation of 8,486 people aged 75 and over found lower death and dialysis rates than a rival drug class, and more strokes.
- In people over 75, this diabetes drug class tracked about a third fewer deaths.
- It also tracked far less kidney failure than the drug class it was compared with.
- Strokes went the other way and were more common, which nobody can yet explain.
- Heart failure admissions and heart attacks showed no clear difference either way.
- 8,486 Japanese patients, compared in records rather than randomized, so cause is not settled.
The trials that made SGLT2 inhibitors a standard diabetes drug were largely run in people who were not very old.
That is not an oversight so much as how trials work: recruit people likely to survive the follow-up, avoid those on eight other medicines, keep the protocol clean. The result is a drug class prescribed heavily to people in their eighties on evidence gathered mostly from people in their sixties.
Writing in Age and Ageing, researchers went looking for that missing evidence in Japanese health records. Among 8,486 people aged 75 and over with type 2 diabetes, SGLT2i initiation was associated with lower all-cause mortality than starting a different diabetes drug, and with a much lower risk of reaching dialysis.
They also found more strokes, and they do not know why.
What the comparison actually was
This is not a drug against nothing. Both groups were treated.
Individuals aged 75 or over with type 2 diabetes who newly initiated an SGLT2 inhibitor or a DPP-4 inhibitor were compared, which puts two active diabetes treatments side by side. DPP-4 inhibitors are the mild, well-tolerated option that Japanese prescribing favors in older patients, so this is close to the choice a doctor actually faces.
Type 2 diabetes is a disease in which your blood glucose, or blood sugar, levels are too high, and both drugs lower it. What separates them is everything else they do.
The two results that point the same way
Death and dialysis both moved, and both moved substantially.
All-cause mortality came in about a third lower, and the risk of end-stage renal disease or dialysis roughly two-thirds lower. The kidney result fits what is already known: this drug class was designed around the kidney, and its protective effect there is one of the better-established findings in diabetes medicine.
Chronic kidney disease means that your kidneys are damaged and can’t filter blood as they should, and diabetes and high blood pressure are the most common causes. Treatments cannot cure kidney disease, but they may slow it. A drug that keeps an 80-year-old off dialysis is doing something a patient would feel.
The confidence range on the dialysis figure is very wide, which is what happens when few people in a cohort reach dialysis at all. The direction is clear; the size is not.
The finding that does not fit
Stroke occurred more frequently among SGLT2i users, at roughly 40% higher.
This is where the article has to slow down. Randomized trials of this drug class in younger patients have not found a stroke excess, and a result that contradicts trial evidence in a database study is usually the database rather than the drug.
The authors are careful about it. They write that the higher observed stroke risk requires cautious interpretation because residual confounding, competing mortality and heterogeneity in vulnerable older adults may contribute to this signal.
Competing mortality is the subtle one. If a drug keeps people alive longer, those people are around to have strokes that the comparison group did not live long enough to have. A treatment that reduces death can look as though it increases every non-fatal event, purely by arithmetic.
That does not prove the stroke finding is an artifact. It means it cannot be read as a drug effect on this evidence.
Why the death result is the most fragile part
Doctors choose. That is the problem with every study of this shape.
An 80-year-old prescribed a drug that increases urine output, lowers blood pressure and risks dehydration is, on average, an 80-year-old who is walking well, eating properly and not falling. A frail one gets the gentler drug. Fit 80-year-olds outlive frail ones regardless of prescription.
A target trial emulation reduces this by fixing eligibility and follow-up in advance, and inverse probability weighting adjusts for what was recorded. Neither can adjust for the impression a doctor formed in a consulting room and never wrote down, and a one-third mortality difference is exactly the size that kind of unrecorded judgement can produce.
What this changes
For guidelines, something. For an individual, almost nothing yet.
The useful contribution here is that the very old were finally studied at all, with a design better than the usual database comparison, and the kidney and mortality results point the same way as the trial evidence in younger patients rather than against it.
The stroke signal is the reason not to go further than that. It is either a real hazard specific to frail older people, an artifact of people living longer, or confounding nobody caught, and this study cannot separate the three.
Anyone over 75 taking a diabetes drug already has a prescriber weighing kidney function, falls, dehydration and how many other tablets are in the box. That conversation is where this belongs, and one database study does not settle it.
People also ask
What did the study find?
SGLT2i initiation was associated with lower all-cause mortality (hazard ratio 0.677, 95% CI 0.500-0.916) and lower risk of end-stage renal disease or dialysis (HR 0.374, 95% CI 0.157-0.890). No clear differences were observed for heart failure hospitalisation (HR 1.103) or myocardial infarction (HR 1.015). Stroke occurred more frequently among SGLT2i users (HR 1.420, 95% CI 1.165-1.828).
What is an SGLT2 inhibitor?
A diabetes drug that makes the kidneys excrete glucose in urine rather than reabsorbing it. Large trials have shown benefits for heart failure and kidney disease well beyond blood sugar control, which is why the class matters more than its glucose effect alone.
Why does age 75 need its own study?
Because the landmark trials of this drug class recruited relatively few people that old, and the very old differ in ways that matter: more competing causes of death, more frailty, higher risk of dehydration and low blood pressure from a drug that increases urine output.
What is a target trial emulation?
A method that takes routine health records and analyses them as if they were the randomized trial nobody ran, by specifying eligibility, treatment start and follow-up in advance. It removes some biases of ordinary database studies. It cannot remove all of them.
How seriously should the stroke finding be taken?
Cautiously, and the authors say so. Residual confounding, competing mortality and heterogeneity in vulnerable older adults may contribute to the signal, and it runs against a large body of randomized evidence in younger patients. It is a reason to look, not a reason to conclude.
Why might the comparison be flattering to SGLT2 inhibitors?
Because doctors do not prescribe at random. A drug that causes dehydration in frail patients tends to be given to the fitter ones, and fitter 80-year-olds live longer whatever they take. Matching on recorded characteristics cannot fully undo that.
What should an older person with diabetes do?
Nothing on the strength of one database study. The choice between diabetes drugs in your late seventies involves kidney function, frailty, falls and dehydration risk, and belongs with the doctor who knows all four. This is general information rather than medical advice.