Timing Errors May Skew GLP-1RA Cardiovascular Risk Studies, Letter Warns

A new letter to the editor published in the Journal of Clinical Sleep Medicine is drawing attention to a subtle but consequential methodological problem that may distort real-world studies of glucagon-like peptide receptor agonists, the blockbuster class of drugs that includes semaglutide and tirzepatide. Written by Güney Sarıoğlu, a cardiologist at Battalgazi State Hospital in Malatya, Turkey, the letter argues that the timing of GLP-1RA exposure in observational analyses may introduce a well-known but frequently overlooked source of error called immortal time bias, potentially inflating or deflating estimates of the drugs’ cardiovascular effects in patients with obstructive sleep apnea and obesity.

The letter is a critical appraisal of a real-world study by Ahn and colleagues that examined whether GLP-1RAs act as cardiovascular risk modifiers in people with obstructive sleep apnea and obesity. Real-world studies of this kind have become enormously influential because they mine large clinical databases to answer questions that randomized trials either have not yet addressed or cannot practically address. With millions of patients now prescribed GLP-1RAs for type 2 diabetes, obesity, and increasingly for their demonstrated cardiovascular benefits, the stakes for getting these observational analyses right could hardly be higher. Policy decisions, prescribing patterns, and reimbursement frameworks increasingly rest on the kind of database evidence that Sarıoğlu’s letter scrutinizes.

At the heart of the critique lies a technical concept that has shaped pharmaco-epidemiology for nearly two decades. Immortal time bias arises when a period of time during which the outcome of interest cannot occur is improperly included in one group’s follow-up, typically the treated group. The classic formulation comes from epidemiologist Samy Suissa, whose 2008 paper in the American Journal of Epidemiology laid out how this bias operates: if researchers define the exposed group by a prescription that occurs sometime after cohort entry, but count that patient’s follow-up from the moment of entry, the patient must survive long enough to receive the prescription. That guaranteed survival window, the ‘immortal time,’ makes the treated group appear artificially protected, generating spuriously favorable results for the drug.

Sarıoğlu points out that this structure is particularly easy to fall into when studying GLP-1RAs, because these drugs are often initiated months or even years after a patient first enters the health system with obesity, sleep apnea, or diabetes. When investigators anchor their analysis at the date of an obstructive sleep apnea diagnosis or at a baseline clinic visit, but classify patients as GLP-1RA users only once a prescription appears later in their record, the exposed group has, by construction, accumulated event-free time before treatment ever began. Unless the analysis explicitly accounts for that window—through techniques such as time-dependent exposure modeling, matching on the time to treatment, or active-comparator new-user designs—the resulting hazard ratios can suggest cardiovascular protection that reflects study design rather than pharmacology.

The letter also situates its argument within a broader and ongoing refinement of how epidemiologists understand these biases. A 2025 paper by Miguel Hernán and colleagues in the journal Epidemiology provided a structural description of the family of biases that generate immortal time, framing them through the lens of causal diagrams and target trial emulation. That work emphasized that immortal time bias is not a single mistake but a constellation of design choices—how cohorts are defined, how exposure is classified, how follow-up begins and ends—that collectively manufacture a comparison between people who could not yet have experienced an event and those who could. Sarıoğlu’s letter applies this modern framework to the specific case of GLP-1RAs in sleep apnea populations, effectively asking whether the original study emulated the randomized trial it intended to mimic.

The target trial framework, as it is known, asks investigators to specify, before touching the data, the randomized trial they would ideally conduct: who would be eligible, how treatment would be assigned, when follow-up would start, and what outcome would be measured. In a well-executed emulation, the moment of cohort entry and the moment treatment is assigned coincide, or the analysis explicitly handles the gap between them. When they diverge—as they do whenever a prescription recorded at an arbitrary later date defines the exposed group—the emulation drifts away from the trial it was meant to mirror, and the divergence is precisely where bias enters. Sarıoğlu’s central claim is that the timing of exposure classification in real-world cardiovascular analyses of GLP-1RAs represents exactly such a divergence, and that readers should interpret effect estimates from such studies with corresponding caution.

Why does this matter so much for this particular drug class and this particular patient population? Obstructive sleep apnea affects roughly a billion people worldwide and is strongly associated with obesity, hypertension, arrhythmias, and increased cardiovascular mortality. GLP-1RAs have generated intense excitement because randomized trials in other populations, notably patients with type 2 diabetes and established cardiovascular disease, showed meaningful reductions in major adverse cardiovascular events. Translating those findings to sleep apnea populations through observational data is an attractive and legitimate research strategy. But it is also a strategy in which the exposure is highly patterned by the very health trajectories under study: patients who remain well enough, engaged enough with care, and clinically stable enough to receive a GLP-1RA prescription are systematically different from those who deteriorate, drop out, or die before such a prescription is written. Any analysis that does not neutralize this selection can convert healthier-patient dynamics into apparent drug benefit.

The letter does not claim that GLP-1RAs lack cardiovascular benefits, nor does it assert that the original study’s conclusions are necessarily wrong. Its point is narrower and, in a sense, more important: the direction and magnitude of any bias introduced by exposure timing cannot be determined from the published results alone, and the credibility of real-world evidence for this drug class depends on design features that must be transparently reported. Sarıoğlu, writing as the sole author of the letter, conceived the commentary, reviewed the relevant literature, and drafted and revised the manuscript, drawing on no external funding and declaring no competing interests. The letter is based exclusively on critical appraisal of previously published work and involves no new data collection, which means its contribution is methodological rather than empirical—it is a lens, not a dataset.

The wider lesson extends well beyond sleep medicine. As GLP-1RAs are studied for an ever-expanding list of outcomes—from kidney disease and heart failure to dementia and addiction—real-world database studies will continue to proliferate, and each carries the same vulnerability if exposure timing is mishandled. The epidemiological community has developed reliable remedies: defining cohort entry at the moment of treatment eligibility, modeling exposure as a time-varying covariate, using new-user designs that exclude prevalent users, and emulating target trials with explicit cloning, censoring, and weighting strategies. Sarıoğlu’s letter serves as a reminder that applying these tools is not pedantic hair-splitting but the difference between evidence that can guide patient care and evidence that merely reflects who managed to stay alive and in care long enough to fill a prescription.

Published on 9 September 2026 as a letter to the editor in the Journal of Clinical Sleep Medicine, the commentary adds a careful methodological voice to one of the most consequential drug-evidence debates of the decade. Whether future real-world analyses of GLP-1RAs in obstructive sleep apnea and obesity will confirm, revise, or overturn the cardiovascular signals reported to date remains an open question. What the letter makes clear is that answering it responsibly requires paying close attention not just to whether patients took these drugs, but to precisely when the clock on their follow-up started—and whether that clock was fair to both the treated and the untreated.

Subject of Research: Methodological bias in real-world observational studies of GLP-1 receptor agonist exposure timing and cardiovascular risk in obstructive sleep apnea and obesity

Article Title: Timing of GLP-1RA exposure in real-world cardiovascular risk analyses

Article References: Sarıoğlu, G. (2026). Timing of GLP-1RA exposure in real-world cardiovascular risk analyses. Journal of Clinical Sleep Medicine, 22(1), Article 161. https://doi.org/10.1007/s44470-026-00188-3

Image Credits: AI Generated

DOI: 10.1007/s44470-026-00188-3

Keywords: GLP-1 receptor agonists, immortal time bias, cardiovascular risk, obstructive sleep apnea, obesity, pharmaco-epidemiology, real-world evidence, observational studies, target trial emulation, type 2 diabetes, methodology, Journal of Clinical Sleep Medicine