Why Diabetes Drugs Began Running Cardiovascular Outcomes Trials

For much of diabetes drug development, lowering hemoglobin A1C was the main efficacy target, and that makes biological and clinical sense: chronic hyperglycemia contributes to microvascular complications, and A1C summarizes average glycemia. But a drug can improve a surrogate marker while causing harm through another pathway.

Concern about possible cardiovascular harm from a glucose-lowering medicine led FDA in December 2008 to issue special guidance for new type 2 diabetes therapies. Sponsors had to rule out an unacceptable increase in major cardiovascular events. In practice, this created a generation of large, randomized, event-driven cardiovascular outcomes trials, commonly called CVOTs.

The policy did more than answer a safety question. Some trials found cardiovascular, heart-failure, or kidney benefits, producing outcome-based indications and changing clinical guidelines. FDA withdrew the 2008 guidance in March 2020 and proposed a broader safety framework, so a dedicated CVOT is no longer a blanket U.S. rule for every new glucose-lowering drug. The evidence model it created still shapes development.

Why A1C was not enough#

A1C is a validated measure of glycemia and can be assessed over months in trials of manageable size. Waiting years for heart attacks, strokes, kidney failure, blindness, and death would make every initial diabetes efficacy trial much longer and larger.

Surrogates create efficiency, but they narrow the claim. A treatment can lower glucose while affecting fluid balance, blood pressure, lipids, weight, heart rhythm, thrombosis, hypoglycemia, or other pathways, so the net cardiovascular effect is the sum of those changes, not the A1C change alone.

History in several therapeutic areas shows that improving a biomarker does not guarantee better clinical outcomes. The regulatory question after 2008 was therefore not whether glucose fell. It was whether the drug caused an unacceptable increase in major cardiovascular events among people with type 2 diabetes.

The 2008 FDA confidence-bound framework#

The 2008 guidance recommended a prespecified meta-analysis of cardiovascular events across phase 2 and phase 3 trials, with events prospectively defined and adjudicated by a committee unaware of treatment assignment. The primary composite generally included cardiovascular death, nonfatal myocardial infarction, and nonfatal stroke, with other events sometimes included in broader analyses.

Before approval, the upper bound of the two-sided 95% confidence interval for the estimated risk ratio needed to rule out 1.8. If the preapproval evidence did not already rule out 1.3, a postapproval trial generally had to do so. These were safety margins, not treatment-benefit targets.

Suppose an estimated hazard ratio was 1.02 with a 95% confidence interval from 0.86 to 1.21. The point estimate is near no difference, and the upper bound rules out both 1.8 and 1.3. A point estimate of 0.95 with a much wider interval ending at 1.55 would rule out 1.8 but not 1.3, leaving a postapproval obligation under the historical framework.

Why a dedicated trial became the practical answer#

Ordinary glycemic trials were often short, enrolled relatively low-risk participants, and recorded few cardiovascular events. Pooling them could not generate a narrow confidence interval. Event definitions and follow-up also varied.

A dedicated CVOT enrolled thousands of adults at high cardiovascular risk and followed them until a target number of events occurred. Higher baseline risk produced more events in less calendar time, improving statistical information without requiring an impossibly large sample.

The trial randomized participants to the new therapy or control on top of contemporary diabetes and cardiovascular care. An independent clinical-events committee adjudicated suspected events using prespecified definitions. The design was usually intended first to show noninferiority, meaning the upper confidence bound stayed below the safety margin.

Event-driven design is about information, not a fixed anniversary#

For a time-to-event endpoint, precision depends heavily on the number of events, not simply the number enrolled, and a trial might plan to analyze after a specified number of cardiovascular deaths, heart attacks, or strokes accumulates.

If event rates are lower than expected, follow-up lasts longer or more participants are needed. If rates are higher, the target can arrive sooner. This is why two trials of similar enrollment can have different durations. Event-driven does not mean investigators wait blindly without safety review: data-monitoring committees evaluate accumulating unblinded data under a charter and can recommend action for harm, overwhelming benefit, or futility when prespecified boundaries apply.

Noninferiority does not prove protection#

A CVOT that rules out an unacceptable increase supports cardiovascular safety within its margin and studied population. To claim cardiovascular benefit, a trial needs a superiority analysis with appropriate control of type I error and persuasive results.

The distinction matters in headlines. When you read that a trial "met the primary cardiovascular endpoint," that can mean only that the upper confidence bound stayed below a noninferiority margin; the result may be compatible with modest benefit, no effect, or some increase smaller than the margin. So read the estimand, the point estimate, the confidence interval, the margin, the component outcomes, and the analysis hierarchy. A significant secondary component after a neutral primary composite may be hypothesis-generating rather than a definitive claim, depending on multiplicity control.

What counts as MACE#

The common three-component major adverse cardiovascular event composite includes cardiovascular death, nonfatal myocardial infarction, and nonfatal stroke, though some programs use a broader composite that adds hospitalization for unstable angina or another event.

Composite endpoints increase event counts, but the components differ in severity, frequency, and response to treatment, and a favorable composite may be driven mainly by one of them. Look at each component estimate, while recognizing that individual components often have less statistical power. Heart-failure hospitalization is frequently evaluated separately because its mechanism and time course differ from atherosclerotic events, and kidney outcomes also require carefully defined composites and attention to competing risk and changes in estimated filtration.

The trials found more than regulators expected#

The initial regulatory goal was to exclude unacceptable harm. Across the CVOT era many therapies turned out to be cardiovascularly neutral on the primary safety question, some trials showed reductions in major cardiovascular events, and others found pronounced reductions in heart-failure hospitalization or slower kidney-disease progression.

Those results moved diabetes treatment beyond a glucose-first comparison. If you have established atherosclerotic disease, heart failure, or chronic kidney disease, outcome evidence can matter more than a small difference in A1C. Regulatory labels and clinical guidelines began reflecting specific demonstrated benefits.

Benefits cannot be assigned automatically to every member of a drug class. Molecules, doses, study populations, endpoints, and trial results differ. A class pattern can support biological reasoning but does not replace product-specific evidence and current labeling.

The high-risk population creates a tradeoff#

Enrolling people with established cardiovascular disease or multiple risk factors makes trials feasible because events accrue. It also narrows direct applicability to younger or lower-risk adults, people with newly diagnosed diabetes, and those without cardiovascular disease.

Relative effects may transfer, but absolute benefit depends on baseline risk. A similar hazard ratio produces a larger absolute reduction when events are common. Safety may also differ with kidney function, frailty, concomitant medicines, or disease duration.

Subgroup analyses can explore consistency but are often underpowered for interaction. Absence of a significant interaction is not proof that every subgroup has an identical effect, and an isolated subgroup p-value can be misleading among many comparisons.

The 2020 U.S. policy change#

After reviewing many completed CVOTs, FDA noted that none conducted to satisfy the 2008 policy had identified an increased ischemic cardiovascular risk, while some demonstrated benefit. The agency withdrew the 2008 guidance on March 9, 2020.

FDA issued draft guidance proposing a broader preapproval safety evaluation rather than a narrow mandatory cardiovascular framework. It emphasized enough participants with relevant comorbidities, including older adults and people with chronic kidney disease, and adequate duration to characterize a range of organ-system risks.

The 2020 document remained draft guidance on FDA's site at the July 2026 research date. Label it accurately when you cite it: draft recommendations do not have the status of final guidance and are not legally binding. Product-specific discussions with FDA can still lead to additional outcomes work.

Europe and global programs can differ#

The European Medicines Agency's diabetes guideline Revision 2 became effective in January 2024. It expects a development program to support lack of excess cardiovascular risk using a population that includes relevant high-risk patients and sufficient data. The design can draw on the whole program, and the need for dedicated work depends on the evidence and product.

A global sponsor must design for the jurisdictions in which it seeks authorization. One region's withdrawal of a blanket policy does not erase another region's guidance, a safety signal, or the evidence needed for a desired outcomes indication. That is another reason a program can look larger than a single U.S. approval calculation would predict: the development question is global benefit and risk, not compliance with one historical document.

What CVOTs still cannot answer well#

Even trials with thousands of participants may be too small for rare harms, and follow-up of a few years may not characterize decades of use. Run-in periods, adherence, frequent visits, and eligibility criteria can make the people in the trial different from the people who will take the drug.

CVOTs also allow background treatment to evolve. That reflects ethical care but can dilute glucose differences and complicate interpretation of mechanisms. Trial sites may be concentrated in regions with different care systems from future users. So postmarketing surveillance, pragmatic trials, registries, mechanistic studies, and longer follow-up all remain important, because a neutral primary composite does not certify safety in every organ system.

How to read a CVOT result#

First work out whether the primary goal was noninferiority, superiority, or a hierarchy of both. Read the margin and the whole confidence interval, not just the point estimate you were shown. Then inspect the composite definition, its components, heart-failure and kidney outcomes, all-cause mortality, and important adverse events.

Check who actually enrolled, and how close they are to the person you are treating: established disease, risk factors only, kidney function, age, sex, diabetes duration, and background therapies. Review discontinuation, missing outcome follow-up, rescue treatment, and whether analyses preserved randomization.

Finally, separate trial evidence from an approved indication. A persuasive publication can inform practice, but regulatory labeling defines the claim reviewed for a specific product, population, and use.