Prediabetes is not a disease with a treatment. It is a threshold — a fasting glucose or an HbA1c above normal and below the line that defines type 2 diabetes — and the thresholds have moved. The American Diabetes Association dropped the impaired-fasting-glucose cutoff to 100 mg/dL in 2003 and the HbA1c cutoff to 5.7% in 2010, each change enlarging the population the word describes.[1]
Against that threshold, GLP-1 drugs produce the largest categorical effect in this literature. In the one phase 3 trial run specifically in adults with obesity and prediabetes, 103 of 127 participants on semaglutide 2.4 mg had returned to normoglycemia at week 52, against nine of 64 on placebo — 81% against 14%, odds ratio 19.8 (95% CI 8.7 to 45.2, P < 0.0001).[2] The non-drug comparator everyone reaches for is the Diabetes Prevention Program, which randomized adults at risk to lifestyle intervention, metformin or placebo and reported its incidence reduction in 2002;[10] that arithmetic is set out in the metformin comparison.
Three facts sit underneath the 19.8 and none of them appear on a checkout page. A third to three-fifths of prediabetes reverts without any drug. Progression to diabetes explains under a quarter of the harm prediabetes predicts. And the effect is contingent on the drug still being in the body.
What the word actually names
A Cochrane review pooled 103 prospective cohort studies of people with intermediate hyperglycemia, with follow-up running from one to 24 years. In the 47 studies that reported restitution of normoglycemia, regression ran 33% to 59% within one to five years of follow-up and 17% to 42% across six to eleven years, graded moderate certainty.[1] That is a spontaneous reversion rate, in cohorts on no treatment at all.
The trajectory is similarly loose. In 10,310 adults aged 46 to 70 in the Atherosclerosis Risk in Communities study, 60% met the prediabetes definition at baseline and roughly 30% progressed to diabetes over a maximum of about 30 years, at a median seven years to diagnosis.[3] Most people carrying the label for three decades did not cross the line.
The dedicated trial, and exactly what it measured
STEP 10 enrolled adults with a body-mass index of 30 kg/m² or higher and prediabetes by NICE criteria at 30 sites in five countries, randomizing 138 to semaglutide 2.4 mg and 69 to placebo for 52 weeks with diet and activity counseling in both arms. Mean baseline weight was 111.6 kg and mean HbA1c 5.9%. Body weight fell 13.9% against 2.7%, an estimated treatment difference of −11.2% (95% CI −13.0 to −9.4), which sits alongside the broader obesity figures in the semaglutide weight-loss article. Adverse events leading to discontinuation occurred in 6% of the semaglutide arm against 1% of placebo.[2]
The reversion endpoint was defined strictly — HbA1c below 6.0% and fasting plasma glucose below 5.5 mmol/L, both at once. The trial’s two primary endpoints were percentage weight change and the proportion reverting.[2] Neither is a clinical event. Nobody counted a heart attack, a retinopathy diagnosis or a death, and the trial was not sized to.
The 14% placebo rate is the figure worth holding. It is not directly comparable to the cohort range above — one year against one to five years, a dual laboratory criterion against looser ones — but it establishes that the denominator moves on its own, and a reader told only “81%” has been handed a number without one.
The endpoint the trials chose covers a minority of the risk
Every prediabetes trial in this class measures progression to type 2 diabetes or reversion away from it. The ARIC analysis asked how much of the harm associated with prediabetes that endpoint actually accounts for. Across about 30 years and 7,069 events — 1,937 atherosclerotic cardiovascular events, 2,109 heart failure, 3,288 chronic kidney disease and 4,785 deaths — prediabetes was associated with any complication at a hazard ratio of 1.21 (95% CI 1.15 to 1.27) against normoglycemia. After accounting for intervening diabetes as a time-varying term, the association was 1.18 (95% CI 1.12 to 1.24), leaving 85% (95% CI 75 to 94) of the excess risk intact.[3]
Two things follow. The association itself is modest — a hazard ratio of 1.21 is not the disease the word implies. And progression to diabetes explained less than a quarter of it, so a drug that prevents progression is acting on the smaller share of what the label marks.
Reversion earns its keep in a study with no GLP-1 in it
The strongest defense of the STEP 10 endpoint comes from the Diabetes Prevention Program Outcomes Study. Among 1,990 participants, those who returned to normal glucose regulation at least once during the original trial had a 56% lower later diabetes risk than those who consistently met prediabetes criteria: hazard ratio 0.44 (95% CI 0.37 to 0.55, P < 0.0001). Critically, the benefit was unaffected by which arm produced the reversion — the interaction tests returned P = 0.1722 for lifestyle and P = 0.3304 for metformin.[4] Reversion looks like it matters regardless of how it was achieved, which is the argument for treating it as a legitimate target.
The same paper carries the result that complicates it. Among participants who did not return to normal glucose regulation, those assigned to the intensive lifestyle intervention had a higher subsequent diabetes risk than the placebo group — hazard ratio 1.31 (95% CI 1.03 to 1.68, P = 0.0304) — and a lower chance of reaching normoglycemia later (odds ratio 0.59, 95% CI 0.42 to 0.82).[4] Failing to revert despite an effective intervention identified a group at worse risk than never having been treated. Nothing equivalent has been reported for a GLP-1 non-responder.
On hard outcomes, the ranking inverts
The follow-up analysis asked whether reversion lowered cardiovascular risk. Across 2,775 participants and ten years, estimated 10-year Framingham risk was highest in those who remained prediabetic at 16.2%, intermediate in those who reverted at 15.5%, and lowest at 14.4% in the people who developed diabetes — all pairwise comparisons P < 0.05. The authors attribute the pattern in part to higher or increasing use of lipid and blood-pressure medication in the diagnosed group, and the prediabetes group’s own score fell over the decade, from 18.6% in year one to 15.9% in year ten.[5]
Crossing into diabetes triggered treatment of the risk factors that actually kill people. Staying just below the line did not. The gap between reverting and not reverting was 0.7 percentage points of estimated risk, and no trial in this drug class has measured a cardiovascular event in a prediabetes population at all.
What happened when the drug stopped
The three-year tirzepatide data show diabetes diagnoses accumulating in the treated group within four months of withdrawal while the placebo rate barely moved; those figures and their washout belong to the SURMOUNT-1 write-up. The semaglutide equivalent is the STEP 1 extension, which followed 327 completers for a year after treatment and lifestyle support both ended, and reported that the cardiometabolic improvements gained over 68 weeks reverted toward baseline at week 120 for most variables.[6] The weight half of that story is in what the withdrawal trials found and the STEP 1 explainer.
The gaps are as informative as the findings. STEP 10 built in a 28-week off-treatment period, and its published report states both primary endpoints at week 52 only.[2] The three-year liraglutide prediabetes trial reached 50% completion and names its own limitation plainly: withdrawn individuals were not followed up after discontinuation, so its diabetes counts describe people still taking the drug.[7] The 56-week trial that preceded it enrolled 3,731 adults of whom 61.2% had prediabetes at baseline.[8] What none of these designs produced is a durable off-drug glycemic outcome.
No label offers this, and the economics assume decades
Read on DailyMed on September 15, 2026, the Wegovy label does not contain the string “prediabetes” anywhere, and its indications cover cardiovascular risk reduction, weight reduction and noncirrhotic MASH. The Zepbound label contains “pre-diabetes” exactly once, in a baseline-characteristics table for the sleep-apnea trials, and its indications cover weight reduction and obstructive sleep apnea. Neither carries a diabetes-prevention indication, which means prescribing for prediabetes is off-label everywhere — a distinction explored in the two-indication article.
The health-economic case makes the duration explicit. A cost-effectiveness model built on STEP 10 used a base-case treatment duration of 20 years, returning an incremental cost-effectiveness ratio of £17,547 per quality-adjusted life-year; varying duration between one year, as the trial actually ran, and lifelong gave a range of £7,152 to £17,695.[9] A 52-week reversion figure was extrapolated across two decades of continuous dispensing, which is the commitment the prevention argument quietly contains. What that costs at cash-pay rates is what the cost calculator is for.
One more thing separates a trial from a telehealth order. STEP 10 and the three-year tirzepatide analysis used branded product manufactured to the approved specification. Most cash-pay sellers dispense compounded semaglutide or tirzepatide, which is not FDA-approved and which the FDA does not review for safety, efficacy or quality before it reaches a patient — see what a compounded vial contains and the sellers listed on the compounded semaglutide board.
What none of this settles
No trial has shown that reverting to normoglycemia on a GLP-1 reduces a clinical event. The evidence that reversion predicts a better outcome comes from a cohort in which it was achieved by lifestyle, metformin or chance, and the one analysis that looked at cardiovascular risk found the reverters only 0.7 percentage points better off than the people who stayed prediabetic — and worse off than the people who got a diabetes diagnosis and the prescriptions that came with it.[4][5] Whether a laboratory category restored by a drug carries the same prognosis as one restored otherwise is an open question, not a settled one.
What is established is narrow and real. Semaglutide 2.4 mg moves adults with obesity and prediabetes out of the prediabetes range far more often than placebo over a year, and the effect size is large. Whether that is worth twenty years of injections, at a price the trials never had to clear, is a question about the person and the price rather than about the pharmacology — and the prescribing decision belongs to a clinician who has seen the labs, not to a checkout.