The study, published in Frontiers in Endocrinology, identified adults aged 18 to 80 with type 2 diabetes who had been hospitalized with severe kidney dysfunction requiring dialysis or renal replacement therapy around the time of discharge. Patients with documented delirium or altered consciousness during the qualifying hospitalization were excluded, so the analysis focused on new, incident neuropsychiatric events arising after the acute episode. The researchers compared two mutually exclusive exposure groups: patients taking a GLP-1RA without an SGLT2 inhibitor, and patients taking an SGLT2 inhibitor without a GLP-1RA. Before statistical balancing, the cohorts comprised 3,822 GLP-1RA users and 5,937 SGLT2 inhibitor users; after 1:1 propensity score matching, 3,398 patients remained in each group.

The matching was necessary because the two groups started out strikingly different. GLP-1RA users were younger on average (55.5 versus 59.5 years), less often male, and had higher body mass index, while SGLT2 inhibitor users carried a heavier burden of cardiovascular disease: ischemic heart disease was present in 52.2 percent of SGLT2 inhibitor users versus 30.1 percent of GLP-1RA users, and beta-blocker and diuretic use were also far more common in the SGLT2 inhibitor group. Propensity matching substantially improved balance across age, sex, kidney function, lipid parameters and most medications, but residual differences persisted for ischemic heart disease, chronic obstructive pulmonary disease, glycated hemoglobin, beta-blocker use and diuretic use—imbalances the investigators explicitly carried into subsequent sensitivity analyses.

The primary analysis examined the window from 90 to 365 days after the index hospitalization, a landmark approach designed to capture post-acute neurological events rather than complications of the hospital stay itself. In this window, GLP-1RA use was associated with better delirium-free survival than SGLT2 inhibitor use: 118 delirium events among 3,032 GLP-1RA patients versus 155 among 2,982 SGLT2 inhibitor patients, a difference that reached statistical significance on log-rank testing (P = 0.009). After multivariable adjustment, GLP-1RA exposure remained independently associated with a lower hazard of incident delirium, with an adjusted hazard ratio of 0.668 (95 percent confidence interval, 0.488–0.914; P = 0.012). By contrast, depression and a composite neurological outcome showed no significant between-group differences during the same primary window.

The temporal pattern proved to be the study’s most striking feature. During the early 90-to-180-day interval, the direction of association reversed for some outcomes: GLP-1RA users had a higher risk of depression than SGLT2 inhibitor users (hazard ratio, 1.84; 95 percent CI, 1.07–3.15; P = 0.023), and a similar early excess appeared for the composite neurological outcome (hazard ratio, 1.71; 95 percent CI, 1.00–2.92; P = 0.045). Delirium, however, showed no significant difference in that early period. By extended follow-up out to 730 days, the early depression difference had vanished entirely (hazard ratio, 0.937; 95 percent CI, 0.748–1.175), while the lower delirium risk associated with GLP-1RA use persisted (hazard ratio, 0.729; 95 percent CI, 0.574–0.926; P = 0.009).

Yet the authors themselves flag that this apparent durability does not survive stricter scrutiny. In exploratory sensitivity analyses using three mutually exclusive, non-overlapping intervals—90–180, 181–365 and 366–730 days—the adjusted association between GLP-1RA use and delirium was nonsignificant in every single interval, with hazard ratios of 0.778, 0.814 and 0.926 respectively. Meanwhile, after adjustment for age, sex, ischemic heart disease, diuretic use and beta-blocker use, GLP-1RA treatment was associated with a higher hazard of depression and overall neurological complications across all three intervals. The divergence between crude and adjusted estimates suggests that residual covariate imbalance after propensity matching may have shaped some of the secondary signals, and the authors emphasize that these interval-specific comparisons are hypothesis-generating rather than formal tests of a treatment-by-time interaction.

Beyond the drug-class comparison, the analysis surfaced a second, independent signal: statin use was associated with a lower risk of incident delirium during the primary window (adjusted hazard ratio, 0.674; 95 percent CI, 0.492–0.924; P = 0.014). The investigators also observed an exploratory association between low-density lipoprotein cholesterol in the range of 100 to 129.9 mg/dL and a higher delirium hazard, with LDL-C ranges evaluated as separate indicator variables rather than a single reference category. The authors caution that this intermediate-range finding should not be read as defining an optimal LDL-C target, but they note it is intriguing in light of prior work in type 2 diabetes describing an inverted U-shaped relationship between LDL-C and cognitive performance, with an inflection point near 104 mg/dL.

The biological plausibility for a delirium-specific GLP-1RA effect rests on converging mechanisms. Delirium is an acute brain disorder highly sensitive to systemic inflammation, and it has been mechanistically linked to microglial activation, endothelial dysfunction, blood–brain barrier disruption, oxidative stress and impaired cerebral energy metabolism—all disturbances that persist after severe kidney injury. Experimental studies suggest GLP-1 receptor activation can shift microglia from pro-inflammatory toward anti-inflammatory phenotypes, and clinical work in type 2 diabetes has associated GLP-1RA therapy with reduced oxidative stress, improved mitochondrial function, and, in the case of liraglutide in Alzheimer’s disease, preserved cerebral glucose metabolism and improved blood–brain glucose transfer. Larger real-world studies add context: a longitudinal study of 109,778 patients with type 2 diabetes reported lower dementia incidence among GLP-1RA users (adjusted hazard ratio, 0.90), and a target-trial emulation of more than one million patients linked semaglutide to a lower risk of a first Alzheimer’s disease diagnosis. Depression, by contrast, involves monoaminergic and glutamatergic neurotransmission, stress-axis dysregulation and psychosocial factors that inflammatory modulation may not reach, which could explain why the delirium signal did not extend to mood outcomes.

The early depression advantage seen with SGLT2 inhibitors has a possible mechanistic footing as well. These drugs exert rapid metabolic and hemodynamic effects after initiation, including early improvements in glycemic dynamics, which could matter most during the vulnerable early recovery phase after acute kidney disease. The authors note, however, that glycated hemoglobin remained slightly higher among GLP-1RA users after matching (9.0 versus 8.6 percent), and because longitudinal HbA1c changes were not assessed, the contribution of differential glycemic improvement remains uncertain. An agent-specific sensitivity analysis comparing empagliflozin—the most common SGLT2 inhibitor, accounting for 83.7 percent of that cohort—with semaglutide, which accounted for 53.3 percent of the GLP-1RA cohort, found no significant differences in depression, delirium or composite neurological outcomes after matching, further underscoring the fragility of the class-level signals.

The study’s limitations are substantial and the authors are unusually explicit about them. As a retrospective electronic health record analysis, it remains vulnerable to residual and unmeasured confounding, and because clinical guidelines steer clinicians toward GLP-1RAs or SGLT2 inhibitors based on distinct metabolic, cardiovascular and kidney profiles, confounding by indication cannot be excluded. Death was censored rather than modeled as a competing event, although mortality was low and similar between groups (3.5 versus 3.8 percent, P = 0.510). Delirium and depression were identified through diagnostic codes rather than standardized assessments, a method with high specificity but low sensitivity that likely undercounts events; medication records do not guarantee adherence; and laboratory data were incomplete. The authors conclude that both the lower delirium risk associated with GLP-1RA use in the primary analysis and the outcome-specific, time-varying patterns across follow-up windows should be considered exploratory, and that prospective studies are needed to determine whether these associations hold—and whether treatment effects on the recovering brain truly change over time.

Subject of Research: Comparative neuropsychiatric outcomes of GLP-1 receptor agonists versus SGLT2 inhibitors after dialysis-requiring acute kidney disease in type 2 diabetes

Article Title: Distinct temporal neuropsychiatric outcomes associated with GLP-1 receptor agonists versus SGLT2 inhibitors after acute kidney disease in type 2 diabetes

Article References: Cheng, Y.-C., Huang, Y.-S., Chen, J.-Y., Wu, V.-C., Chen, R.-M., Lu, K.-C., Kuo, K.-L., Lu, C.-L., Zheng, C.-M., & Hou, Y.-C. (2026). Distinct temporal neuropsychiatric outcomes associated with GLP-1 receptor agonists versus SGLT2 inhibitors after acute kidney disease in type 2 diabetes. Frontiers in Endocrinology, 17, 1963944. https://doi.org/10.3389/fendo.2026.1963944

Image Credits: AI Generated

DOI: 10.3389/fendo.2026.1963944

Keywords: GLP-1 receptor agonists, SGLT2 inhibitors, acute kidney disease, type 2 diabetes, delirium, depression, statins, LDL cholesterol, neuroinflammation, propensity score matching, TriNetX, kidney-brain axis