CheckMate 9ER was the pivotal study that established the cabozantinib-nivolumab combination, often abbreviated CaboNivo, as a first-line option for patients with advanced renal cell carcinoma, a disease that arises in the kidneys and, once metastatic, carries a historically poor prognosis. The original trial demonstrated a significant overall survival benefit for the immunotherapy-targeted therapy duo compared with sunitinib, a tyrosine kinase inhibitor that had long served as a benchmark treatment. Patients who received CaboNivo lived longer, and the finding helped reshape treatment guidelines around the world. But the trial’s overall survival results, like those of many modern oncology studies, carried a hidden complication that statisticians have increasingly sought to address.

That complication is subsequent therapy. In advanced renal cell carcinoma, patients whose initial treatment stops working almost always move on to further lines of treatment, and the sequence of therapies a patient receives can profoundly influence how long they ultimately live. When patients in the control arm of a trial later cross over to newer, more effective treatments, or when imbalances arise in the quality and intensity of post-progression care between the two arms, the measured overall survival benefit of the experimental regimen can be diluted or distorted. The observed survival difference then reflects not only the effect of the initial treatment but also a tangled web of downstream decisions made after the trial’s assigned therapy ended.

To untangle this web, the research team, led by Marc-Oliver Grimm of Jena University Hospital in Germany together with colleagues from industry and health economics consultancies and the University of Bari in Italy, turned to two-stage estimation, a counterfactual modeling approach designed to isolate the effect of the initial randomized treatment from the effects of everything that follows. The method proceeds in two distinct analytical steps, each addressing a different piece of the puzzle. It is a technique that has gained traction in oncology health technology assessment precisely because it offers a way to model what survival would have been in a hypothetical world where subsequent therapy exerted no influence.

In the first stage, the investigators used multivariate modeling to estimate how much subsequent therapy extended overall survival among patients who discontinued their original assigned study treatment. This step required careful characterization of the patients who moved on to later lines of therapy, including the types of treatments they received and the clinical characteristics that predicted both the receipt of subsequent therapy and the survival benefit it conferred. By quantifying this extension, the model could then estimate what those patients’ survival would have been in the absence of the additional treatments, constructing the counterfactual scenario that lies at the heart of the analysis.

In the second stage, the researchers computed an adjusted hazard ratio for overall survival comparing CaboNivo with sunitinib, now freed from the confounding effects of subsequent treatment. The hazard ratio is the workhorse statistic of survival analysis, expressing the relative risk of death over time between two groups, and an adjusted value below one indicates a survival advantage for the experimental arm. By feeding the stage-one counterfactual estimates into this calculation, the team produced a picture of the true relative efficacy of the two first-line regimens, unclouded by the downstream therapeutic landscape.

The results were revealing. Under the two-stage estimation, the adjusted counterfactual median overall survival was 42.7 months for patients treated with cabozantinib plus nivolumab, compared with just 21.3 months for those treated with sunitinib, a difference of more than twenty-one months. The corresponding adjusted hazard ratio was 0.53, with a 95 percent confidence interval of 0.43 to 0.65 and a p value below 0.001, indicating that the combination cut the hazard of death nearly in half in this counterfactual scenario. By contrast, the unadjusted estimates told a more modest story: median overall survival of 46.5 months for CaboNivo versus 35.5 months for sunitinib, with a hazard ratio of 0.79 and a p value of 0.0196.

The comparison between the adjusted and unadjusted figures is where the analytical insight becomes vivid. Counterintuitively, the counterfactual median survival for the CaboNivo arm was actually lower than the observed median, 42.7 months versus 46.5 months, which reflects the removal of the survival extension that subsequent therapies contributed to patients in that arm. Yet the counterfactual median for the sunitinib arm fell even more sharply, from 35.5 months to 21.3 months, indicating that patients in the control arm derived a larger relative boost from the therapies they received after discontinuing their assigned treatment. When those downstream effects were removed from both arms, the underlying advantage of the initial CaboNivo regimen emerged as considerably larger than the raw trial comparison had shown.

For clinicians, the finding offers reassurance that the survival benefit of first-line cabozantinib plus nivolumab is not an artifact of favorable downstream care patterns but a genuine property of the regimen itself. For patients with advanced renal cell carcinoma, a disease in which treatment sequences have grown increasingly complex with the arrival of immune checkpoint inhibitors and multiple targeted agents, the implication is that starting with the combination may confer a deeper and more durable survival advantage than conventional analysis suggests. The authors conclude that removing the effect of subsequent therapy increased the estimated overall survival benefit of CaboNivo versus sunitinib, reinforcing the combination as a first-line standard of care in advanced renal cell carcinoma.

The analysis also carries broader methodological significance for the field of oncology evidence generation. Two-stage estimation is increasingly relevant to health technology assessment bodies, which must judge the value of new cancer therapies using trial data that are inevitably contaminated by subsequent treatment effects, and the approach offers a structured, transparent way to model counterfactual scenarios that no randomized trial can directly observe. As a post hoc analysis, however, the study relies on modeling assumptions rather than prospective randomization to the counterfactual condition, and its findings should be interpreted alongside the original trial results rather than as a replacement for them. The CheckMate 9ER trial is registered as NCT03141177, and the original research underlying this summary was published in Oncologic Therapy in November 2025, with the summary article appearing in Advances in Therapy in October 2026 under open access, funded by Ipsen, which was involved in the design, execution, analysis, and writing of the summary.

Subject of Research: Two-stage estimation of overall survival adjusting for subsequent therapy in the phase 3 CheckMate 9ER trial of cabozantinib plus nivolumab versus sunitinib in advanced renal cell carcinoma

Article Title: Summary of Research: Two-Stage Estimation of Overall Survival in the Phase 3 CheckMate 9ER Trial, Adjusting for the Impact of Subsequent Therapy

Article References: Grimm, M.-O., Karim, E., Kapetanakis, V., Lothgren, M., Ogareva, A., Shukla, P., Truscott, J., & Porta, C. (2026). Summary of Research: Two-Stage Estimation of Overall Survival in the Phase 3 CheckMate 9ER Trial, Adjusting for the Impact of Subsequent Therapy. Advances in Therapy. https://doi.org/10.1007/s12325-026-03773-3

Image Credits: AI Generated

DOI: 10.1007/s12325-026-03773-3

Keywords: advanced renal cell carcinoma, cabozantinib, nivolumab, sunitinib, CheckMate 9ER, overall survival, two-stage estimation, subsequent therapy, hazard ratio, phase 3 trial, immunotherapy, post hoc analysis

The original SURMOUNT-OSA program comprised two Phase 3, randomized, double-blind, placebo-controlled trials spanning 52 weeks. Study 1 enrolled 234 adults with moderate-to-severe sleep apnea, defined as an apnea-hypopnea index of at least 15 events per hour, and obesity, defined as a body mass index of at least 30 kilograms per square meter, who were not using PAP therapy. Study 2 enrolled 235 similar participants who were on PAP therapy, and asked them to withhold the device for roughly seven days before laboratory polysomnography so that their underlying disease severity could be measured. Participants received either the maximum tolerated dose of tirzepatide, 10 or 15 milligrams, or placebo. The primary results, published previously, showed clinically meaningful reductions in the apnea-hypopnea index, in sleep apnea-specific hypoxic burden, in body weight, and in systolic blood pressure, along with improvements in high-sensitivity C-reactive protein and patient-reported outcomes.

The new analyses asked a deceptively simple question: did the drug work equally well in everyone, or did its effectiveness depend on who the patient was at the start of the trial? The researchers stratified participants by baseline age, sex, apnea-hypopnea index severity, body mass index, and neck circumference, then estimated changes from baseline to week 52 using mixed models for repeated measures, adjusting for baseline values, geographic region, sex, and treatment. Because these exploratory analyses were descriptive rather than powered for formal statistical testing of heterogeneity, the authors caution that apparent differences between subgroups may reflect chance, residual confounding, differing baseline distributions, or regression to the mean. Still, the consistency of the direction of benefit across nearly every slice of the population is striking.

The headline finding is that tirzepatide outperformed placebo across all baseline subgroups. Participants on the drug experienced apnea-hypopnea index reductions ranging from roughly 20 to more than 50 events per hour depending on subgroup, compared with far smaller changes on placebo. When stratified by age, reductions ranged from 27.7 to 34.1 events per hour; by sex, from 19.8 to 32.6 events per hour; by baseline apnea-hypopnea index, from 12.1 to 52.2 events per hour; by body mass index, from 25.2 to 34.4 events per hour; and by neck circumference, from 23.9 to 30.8 events per hour. Sleep apnea-specific hypoxic burden, an emerging measure of the oxygen deprivation burden imposed by the disease, and systolic blood pressure also improved consistently, with blood pressure falling by roughly 6.7 to 10.2 millimeters of mercury across subgroups and studies—a magnitude that rivals or exceeds the modest reductions typically achieved with PAP therapy itself.

Some of the numerical patterns tracked physiology in intuitive ways. Men, who had higher baseline apnea-hypopnea indices than women in both trials, posted larger absolute reductions, from 30.6 events per hour in men versus 19.8 in women in Study 1, and 32.6 versus 24.8 in Study 2. Participants with the most severe disease at baseline, an apnea-hypopnea index of at least 70 events per hour, saw the largest absolute drops, at 43.3 events per hour in Study 1 and 52.2 events per hour in Study 2, although their relative improvement was smaller than in patients with milder disease. Younger participants on PAP therapy in Study 2 showed numerically greater reductions in the apnea-hypopnea index, hypoxic burden, and body weight than those aged 50 and older, hinting that obesity may play a stronger causal role in sleep apnea among younger patients, while age-related factors such as reduced lung function and weakened upper airway protective reflexes may dilute the benefit of weight loss in older adults.

Perhaps the most clinically resonant results concern categorical shifts in disease severity. Among tirzepatide-treated participants, 67.7 percent in Study 1 and 79.0 percent in Study 2 moved to a less severe apnea category by week 52, whereas most placebo participants, 64 to 70 percent, saw no clinically relevant change. Improvement crossed every baseline characteristic examined: by age, 62.4 to 77.8 percent of tirzepatide-treated participants improved depending on subgroup and study; by sex, 60.5 to 86.1 percent; and by body mass index, the proportion improving ranged from 57.0 to 80.5 percent. Even in the very severe subgroup with a baseline apnea-hypopnea index of at least 70 events per hour, where average starting values hovered around 86 to 91 events per hour, over half of tirzepatide-treated participants in both studies achieved reductions of at least 50 percent, and 31.8 to 64.6 percent improved in severity class.

The neck circumference findings add a physiologically interesting wrinkle. Neck circumference has been proposed as a better proxy than body mass index for the fat deposition around the upper airway that narrows breathing passages during sleep. Participants with larger necks, who tended to be younger, male, and heavier, showed numerically greater absolute reductions in apnea-hypopnea index and body weight than those with smaller necks, a pattern consistent across both studies. Meanwhile, weight reductions increased with higher baseline body mass index, particularly in Study 2, and apnea-hypopnea index improvements were broadly similar across body mass index categories, suggesting that patients with even the highest degrees of obesity stand to benefit. The authors note that these results contradict the assumption that heavier patients would need to lose more weight to achieve comparable airway improvement.

Underlying the results is a central scientific question: how much of tirzepatide’s effect on sleep apnea flows through weight loss, and how much is independent? A separate analysis of the SURMOUNT-OSA data found a linear association between the magnitude of weight reduction and improvements in the apnea-hypopnea index and hypoxic burden, pointing to weight loss as the dominant driver. But the authors caution that weight-independent effects cannot be excluded. Tirzepatide may influence several of the pathogenic pathways that produce sleep apnea, including insulin resistance and systemic inflammation, both of which have been identified as antecedent risk factors for the disorder and may affect airway muscle function and ventilatory control. Sex differences in these pathways, and evidence from other studies that women tend to lose more weight on incretin therapies than men, may help explain why weight changes by sex were inconsistent between the two trials even as apnea measures improved in both.

The authors are careful about the limitations. The 52-week duration of the trials leaves the long-term effects unknown, particularly since longer studies of tirzepatide have shown additional weight loss over time. Subgroup analyses were not powered for statistical significance, the models did not mutually adjust for all baseline characteristics, and factors such as socioeconomic status, prior apnea surgery, PAP adherence, and apnea endotypes were not captured. Discontinuation rates were also higher in the placebo arm, which may have inflated the apparent treatment difference. Still, the consistency of benefit across two independent trials, nearly five hundred participants, and every baseline characteristic examined represents an unusually robust signal for an exploratory analysis.

For clinicians and patients, the practical takeaway is considerable. Severity improvements on tirzepatide may open the door to alternative therapies such as oral appliances or positional therapy for patients shifted into milder disease categories, and the substantial blood pressure reductions may carry independent cardiovascular significance given the known links between sleep apnea, resistant hypertension, and cardiometabolic disease. Whether tirzepatide and PAP therapy together produce additive benefits, and whether the improvements endure over years, remain open questions that the authors say warrant further study. What is already clear is that the era of treating sleep apnea with machines alone is ending, and that a single medication can meaningfully reshape the disease across virtually the full spectrum of patients who have it.

Subject of Research: Effects of tirzepatide on obstructive sleep apnea severity and cardiometabolic measures across baseline patient subgroups in the SURMOUNT-OSA Phase 3 trials

Article Title: Association of tirzepatide with changes in OSA-related measures based on baseline characteristics – post hoc analyses of SURMOUNT-OSA

Article References: Falcon, B., Xie, C. C., Redline, S., Grunstein, R., Turnbull, C. D., Rapoport, D. M., Wang, H., Chakladar, S., Dimitriadis, G. K., Lau, E., Bednarik, J., Liao, B., & Malhotra, A. (2026). Association of tirzepatide with changes in OSA-related measures based on baseline characteristics – post hoc analyses of SURMOUNT-OSA. Journal of Clinical Sleep Medicine, 22(1), Article 155. https://doi.org/10.1007/s44470-026-00162-z

Image Credits: AI Generated

DOI: 10.1007/s44470-026-00162-z

Keywords: tirzepatide, obstructive sleep apnea, SURMOUNT-OSA, obesity, GLP-1 receptor agonist, apnea-hypopnea index, hypoxic burden, blood pressure, weight loss, sleep medicine, Phase 3 trials, post hoc analysis