For many people living with type 2 diabetes, basal insulin is a daily fixture. Researchers have long wondered whether adding a glucagon-like peptide-1 receptor agonist, or GLP-1RA, to a person's regimen might reduce insulin requirements enough to allow them to stop insulin altogether. A study published in the Annals of Internal Medicine set out to answer exactly that question, using a method called target trial emulation to mimic the rigor of a randomized controlled trial without actually running one.
The study analyzed electronic health record data from nearly 9,000 veterans with type 2 diabetes who were already on basal insulin and then started one of three drug classes: a GLP-1 receptor agonist, a sodium-glucose cotransporter-2 inhibitor (SGLT-2i), or a dipeptidyl peptidase-4 inhibitor (DPP-4i). All three are used to manage blood sugar in type 2 diabetes, but they work through different biological pathways. The researchers tracked each group for up to three years, looking for a specific outcome: whether participants went at least 12 consecutive months without filling an insulin prescription.
The results challenged a common assumption. Despite the growing reputation of GLP-1 receptor agonists as potent metabolic tools, adding one to basal insulin did not meaningfully increase the odds of stopping insulin when compared to either of the other two drug classes.
How the study was designed
Target trial emulation is a relatively recent analytical approach in epidemiology. The idea is to use real-world data, in this case electronic health records, and structure the analysis to resemble what a randomized clinical trial would look like. This helps reduce some of the biases that normally plague observational research, where patients who receive different treatments may differ from one another in important ways.
The researchers drew data from the U.S. Veterans Health Administration, one of the largest integrated healthcare systems in the country. They identified veterans with type 2 diabetes who were on basal insulin and who started one of the three drug classes between 2020 and 2022. To make the comparison fair, they used a technique called matching, creating nearly 9,000 sets of patients who were similar to one another across key characteristics before being assigned to different drug classes.
The primary outcome was insulin discontinuation, defined strictly as a gap of 12 months or more in insulin prescription fills over a three-year follow-up window. The researchers ran both an intention-to-treat analysis, which counts everyone regardless of whether they stayed on the drug, and a modified per-protocol analysis, which more closely tracks only those who continued taking the assigned medication.
Profile of the study population
The study population reflected the demographics of the Veterans Health Administration. About 63 percent of participants were 65 years or older, 93 percent were male, and 70 percent identified as White. Nearly half, 48 percent, had a hemoglobin A1c of 9 percent or higher at the start of the study, indicating that a substantial portion had poorly controlled blood sugar despite already being on insulin.
Within the GLP-1RA group, the majority, about 76.6 percent, were using semaglutide. The rest used dulaglutide, liraglutide, or exenatide. In the SGLT-2i group, nearly all participants used empagliflozin, and in the DPP-4i group, the overwhelming majority used alogliptin. This concentration in specific agents within each class is worth noting, because results might look different if the class were represented more evenly.
What the numbers showed
Over three years, the rates of insulin discontinuation were remarkably similar across all three groups. About 16.7 percent of those in the GLP-1RA group stopped insulin, compared to 17.9 percent in the SGLT-2i group and 17.1 percent in the DPP-4i group. In statistical terms, the GLP-1RA group actually had a slightly lower rate of discontinuation than the SGLT-2i group, though the difference was not statistically significant.
The risk ratio for GLP-1RA compared to SGLT-2i was 0.93, with a confidence interval ranging from 0.86 to 1.01. For GLP-1RA compared to DPP-4i, the risk ratio was 0.98, with a confidence interval of 0.87 to 1.09. Neither result crossed the threshold for statistical significance, meaning the study could not confirm a real difference between groups. The per-protocol analysis, which some researchers argue is a stricter test, produced similar findings.
The researchers also looked at subgroups, including people with higher or lower baseline blood sugar, different ages, and other characteristics. None of these subgroup analyses revealed a population where GLP-1RAs showed a clear advantage over the other drug classes for the specific goal of stopping insulin.
Why this finding matters for research
These results are notable because they run counter to a narrative that has built up around GLP-1 receptor agonists. The literature has consistently shown that GLP-1RAs can lower blood sugar meaningfully and reduce insulin doses in many patients. Some smaller studies and clinical observations have suggested that discontinuation might be achievable for some individuals. This large study, however, suggests that at a population level, across thousands of real-world patients, GLP-1RAs did not outperform other commonly used drug classes when the goal was stopping insulin entirely.
One possible explanation is that insulin discontinuation is a high bar. Reducing a dose is different from stopping altogether, and many patients on basal insulin may have underlying beta-cell function limitations that no oral or injectable non-insulin agent can fully compensate for. The study did not measure whether insulin doses were reduced, only whether insulin was stopped for at least 12 months.
Another consideration is that the study population was older, predominantly male, and had longstanding diabetes with suboptimal control. Findings in this demographic may not apply to younger patients, those earlier in their disease course, or populations with better baseline glycemic control.
Limitations the authors acknowledged
The researchers were transparent about several limitations. The use of electronic health records introduces the possibility that some prescriptions were filled but not taken, or that some patients obtained insulin through other means not captured in the records. This kind of misclassification could push the results toward finding no difference even if one exists.
There is also the possibility of residual confounding. Even with careful matching, patients who are prescribed different drug classes may differ in ways that the data cannot fully capture, such as motivation to reduce insulin use, access to diabetes education, or dietary habits. Target trial emulation reduces but does not eliminate this concern.
The study was funded by the U.S. Department of Veterans Affairs, and its population reflects the specific demographics of that system. Replication in more diverse populations, including more women and younger patients, would strengthen the conclusions.
The broader picture in GLP-1 peptide research
GLP-1 receptor agonists work by mimicking a natural gut hormone that stimulates insulin release in response to meals, suppresses glucagon, and slows gastric emptying. The literature suggests these mechanisms combine to lower blood sugar in ways that are meal-sensitive and less prone to causing dangerous drops in blood glucose compared to insulin itself.
Research into GLP-1 signaling continues to expand. Scientists are studying how different peptide structures affect the duration of action, the degree of glucose lowering, and the range of tissues influenced. Some newer investigational compounds are designed to activate not only the GLP-1 receptor but also related receptors for other gut hormones, with the goal of achieving broader metabolic effects. The study covered here does not speak to those emerging compounds, as it focused on agents that were in widespread clinical use between 2020 and 2022.
For researchers interested in the pharmacology of GLP-1 signaling and related metabolic peptides, this study adds an important data point: the ability of GLP-1 receptor agonists to lower blood sugar does not automatically translate into insulin independence at the population level, at least not more so than alternative drug classes. Understanding where and why this gap exists remains an active area of scientific inquiry.
