EXSCEL: The Large Neutral Result
Last verified May 2026 · Phase 3/4 · Finished; published September 2017 · NCT01144338 ↗
Size does not guarantee a positive result, and EXSCEL is the clearest demonstration of that here. It enrolled more than twice SELECT's participants for a comparable duration and found no significant reduction. Exenatide is exendin-based rather than a human GLP-1 analog, and its once-weekly formulation produces a different exposure profile from semaglutide or dulaglutide — the usual explanations for why it behaved differently. Open-label adherence issues in a pragmatic trial design have also been proposed.
- Enrollment
- 14,752
- Duration
- A median 3.2 years, with the trial running until enough events had accumulated rather than to a fixed date
- Drug
- Exenatide once-weekly (Bydureon)
- Population
- Adults 18 and over whose type 2 diabetes ran between an HbA1c of 6.5% and 10.0%, with or without heart disease — 73.1% arrived with it, 26.9% without. Average age 62.0, 38.0% women, average HbA1c 8.0%, average BMI 31.8, about 12 years since diagnosis. ★ The background treatment is what makes this trial unusual: 77% on metformin, 47% on sulfonylureas and 46% on insulin, with participants free to add other drugs during the trial. That is a real clinic population rather than a curated one, and it cuts both ways.
Primary endpoint
Cardiac death, a nonfatal heart attack, or a nonfatal stroke — whichever came first
Treatment arm
839 of 7,356 (11.4%); 3.7 events per 100 person-years
Comparator
905 of 7,396 (12.2%); 4.0 events per 100 person-years
Treatment difference: Hazard ratio 0.91 (95% CI 0.83 to 1.00); P<0.001 for non-inferiority, P=0.06 for superiority
⚠ A near miss rather than a flat line: the reduction came out at 9%, with the confidence interval touching 1.00 and a p-value of 0.06. It cleared the safety bar the FDA required and failed the conventional test for benefit. ★ Sitting between the neutral lixisenatide trial and the clearly positive liraglutide and semaglutide ones, it is the single result that most complicates any tidy story about this drug class.
Secondary endpoints
| Endpoint | Treatment | Comparator | Difference |
|---|---|---|---|
| Death from a cardiac cause ⚠ Fewer, but it does not count. Because the main test failed, everything below it is descriptive by the trial's own pre-set rules. | 340 of 7,356 (4.6%) | 383 of 7,396 (5.2%) | Hazard ratio 0.88 (95% CI 0.76 to 1.02) |
| Heart attacks that were not fatal A small numerical difference and nothing more. | 354 of 7,356 (4.8%) | 383 of 7,396 (5.2%) | Hazard ratio 0.95 (95% CI 0.82 to 1.10) |
| Strokes that were not fatal Slightly fewer, again not formally testable. | 187 of 7,356 (2.5%) | 218 of 7,396 (2.9%) | Hazard ratio 0.86 (95% CI 0.70 to 1.05) |
| Death from any cause at all ⛔ This is the most misquoted number in the trial. Deaths were 6.9% against 7.9%, which on its own would look convincing — but the trial's pre-set testing order meant that once the main endpoint failed, this could not be declared significant. Anyone citing this trial as evidence the drug saves lives is reading past the rule the trial set for itself in advance. | 507 of 7,356 (6.9%) | 584 of 7,396 (7.9%) | Hazard ratio 0.86 (95% CI 0.77 to 0.97) |
| Admitted to hospital for heart failure No effect, matching the pattern across this whole class and standing in contrast to SGLT2 inhibitors, which reduce these admissions substantially. | 219 of 7,356 (3.0%) | 231 of 7,396 (3.1%) | Hazard ratio 0.94 (95% CI 0.78 to 1.13) |
| Admitted with an acute coronary event Reported in support of the main analysis, with nothing to separate the arms. | Reported as supportive endpoint | Reported as supportive endpoint | Hazard ratio 0.97 (95% CI 0.87 to 1.07) |
| Three-month average blood sugar ⚠ A modest gain that understates the drug, because both arms could freely add other diabetes medication — including, later in the trial, drugs from this same class. | Reduced | Reference | Between-group difference −0.53 percentage points (95% CI −0.57 to −0.50) |
| Weight change ⛔ Roughly 1 to 2 kg. Nothing here is relevant to the weight drugs sold today, and it is part of why this product could not compete once semaglutide arrived. | Reduced | Reference | Between-group difference −1.27 kg (95% CI −1.40 to −1.13) |
Adverse events
| Event | Treatment rate | Comparator rate |
|---|---|---|
| Reactions where the injection went in ★ A problem specific to this product rather than the class. It used a slow-release microsphere depot, and a meaningful minority of patients developed visible lumps under the skin — something the liquid pens do not cause, and a real reason people stopped. | Higher with exenatide | Lower with placebo |
| Pancreatitis, confirmed No increase, consistent with every other large trial in this class. | 31 of 7,344 (0.4%) | 39 of 7,389 (0.5%) |
| Pancreatic cancer No imbalance, in a trial large enough for the question to mean something — 14,752 people over three years. | 20 of 7,344 (0.3%) | 24 of 7,389 (0.3%) |
| Medullary thyroid cancer ★ Not one case in 14,752 people. This is the cancer behind the boxed warning on this class, and it comes from rodent studies. | 0 of 7,344 | 0 of 7,389 |
| Severe low blood sugar needing help from someone else ⚠ Slightly higher on the drug — and the reason is what people were already taking. Nearly half this trial was on insulin and 47% on a sulfonylurea, which are what cause lows. The GLP-1 does not do it alone. | 247 of 7,344 (3.4%) | 219 of 7,389 (3.0%) |
| Stopped taking the assigned treatment ⛔ 43.0% against 45.2% — nearly half of everyone in both arms stopped. That is a striking figure and it matters for interpretation: a trial where half the participants are no longer taking what they were assigned has a real problem detecting a modest benefit. | 43.0% | 45.2% |
Subgroup analyses
- People who had already had cardiovascular disease — 73.1% of the trial: 10% fewer events, with the confidence interval just clearing 1
⚠ Slightly better than the trial overall, and it was a pre-planned subgroup rather than the primary test — so it is a hint about where any benefit lives, not a result.
- People with no previous cardiovascular disease — 26.9%: No signal at all
★ The same split that shows up in the liraglutide trial: whatever benefit exists in this class concentrates in people who have already been damaged.
- People whose kidney function was already reduced: 12% fewer events, not statistically distinguishable from the rest
Directionally consistent with the pattern of more benefit where there is more damage, without reaching the point where it can be claimed.
Clinical significance
EXSCEL's value is corrective. It is the trial to remember when a marketing page compresses this literature into 'GLP-1s are cardioprotective', because the largest trial in the class says that is not true of every drug in it. ⚠ For a weight-management reader, exenatide is not a realistic option in any case; the relevance is entirely to how confidently class-wide claims should be made.
Frequently Asked Questions
References
- 1.Holman RR, Bethel MA, Mentz RJ, Thompson VP, Lokhnygina Y, Buse JB, Chan JC, Choi J, Gustavson SM, Iqbal N, Maggioni AP, Marso SP, Öhman P, Pagidipati NJ, Poulter N, Ramachandran A, Zinman B, Hernandez AF; EXSCEL Study Group. Effects of Once-Weekly Exenatide on Cardiovascular Outcomes in Type 2 Diabetes. N Engl J Med. 2017. PMID: 28910237.
- 2.Amylin Pharmaceuticals, LLC; AstraZeneca. Exenatide Study of Cardiovascular Event Lowering Trial (EXSCEL): A Trial to Evaluate Cardiovascular Outcomes After Treatment With Exenatide Once Weekly in Patients With Type 2 Diabetes Mellitus. ClinicalTrials.gov, NCT01144338. 2017. https://clinicaltrials.gov/study/NCT01144338
- 3.AstraZeneca. Bydureon (exenatide extended-release) product portfolio communication — U.S. market discontinuation, 2024. AstraZeneca corporate disclosure. 2024. https://www.astrazeneca.com/our-therapy-areas/cardiovascular-renal-and-metabolism.html