Tirzepatide, Semaglutide, and SGLT2 Inhibitors: Dementia Risk in Type 2 Diabetes

Abstract

Study at a Glance

Study Snapshot

HR = hazard ratio; NS = not statistically significant; MACE = major adverse cardiovascular events. All HRs derived from propensity-score-matched Kaplan-Meier survival analysis.

Study Facts Table

What Researchers Actually Did

Younis and colleagues used the TriNetX global federated network — a real-world database containing electronic health records from more than 150 million patients across more than 140 health care organizations, predominantly in North America and Western Europe — to emulate three separate target trials. Adults with type 2 diabetes who initiated tirzepatide, semaglutide, or an SGLT2 inhibitor between May 2022 and May 2023 were enrolled; the May 2022 lower bound was chosen because that is when tirzepatide received FDA approval for type 2 diabetes, ensuring all three drugs were concurrently available and that time zero was synchronized. Patients were required to have no history of dementia, no prior co-prescription of the study drugs, and no glucose-lowering therapy within the preceding six months — the last criterion operationalizes a new-user design intended to reduce carryover confounding and depletion-of-susceptibles bias. Contraindicated conditions (pancreatitis, thyroid cancer, gallstones, gastroparesis, ketoacidosis) were excluded.

Each treatment arm was propensity-score matched 1:1 to its comparator on 32 baseline covariates spanning demographics, socioeconomic status, cardiometabolic comorbidities, anthropometrics, biochemistry (HbA1c, eGFR), and concomitant medications. Balancing was confirmed using standardized mean differences below 0.10 in all matched pairs. Kaplan-Meier survival curves and Cox proportional hazard models were applied, with follow-up capped at 2 years. Bell’s palsy was pre-specified as a falsification outcome — a condition with no plausible biological link to any of the study drugs — to probe for residual confounding or systematic coding artifacts. E-values were calculated to quantify the minimum strength of unmeasured confounding that could nullify each observed association.

Key Findings: Primary Outcomes

• Tirzepatide vs. SGLT2 inhibitors (TTE1, n=14,462 per arm): 55 dementia events in the tirzepatide arm vs. 79 in the SGLT2-i arm; HR 0.66 (95% CI 0.47–0.93), p=0.02. E-value for the point estimate: 2.40; E-value for the CI limit closest to null: 1.36.
• Semaglutide vs. SGLT2 inhibitors (TTE2, n=57,959 per arm): 518 dementia events vs. 696; HR 0.71 (95% CI 0.63–0.78), p<0.01. E-value for the point estimate: 2.17; CI E-value: 1.88.
• Tirzepatide vs. semaglutide (TTE3, n=12,246 per arm): 51 dementia events vs. 73; HR 0.69 (95% CI 0.48–0.99), p=0.04. E-value for the point estimate: 2.26; CI E-value: 1.11.
• Bell’s palsy falsification endpoint showed no statistically significant difference across all three TTEs (TTE1: HR 0.69, p=0.16; TTE2: HR 1.03, p=0.85; TTE3: HR 0.66, p=0.19), providing at least partial reassurance against gross systematic bias.

Key Findings: Secondary Outcomes

• MACE, tirzepatide vs. SGLT2-i: HR 0.80 (95% CI 0.73–0.90), p<0.01 — tirzepatide favored.
• MACE, semaglutide vs. SGLT2-i: HR 0.84 (95% CI 0.80–0.88), p<0.01 — semaglutide favored.
• MACE, tirzepatide vs. semaglutide: HR 0.95 (95% CI 0.85–1.06), p=0.34 — no statistically significant difference. Authors note limited statistical power in this smaller comparison.
• All-cause mortality, tirzepatide vs. SGLT2-i: HR 0.29 (95% CI 0.23–0.37), p<0.01. Authors explicitly acknowledge this magnitude is likely not fully causal; E-value for the point estimate was 6.35 and for the CI limit was 4.85, but the authors themselves flag that such confounding remains plausible.
• All-cause mortality, semaglutide vs. SGLT2-i: HR 0.53 (95% CI 0.49–0.57), p<0.01.
• All-cause mortality, tirzepatide vs. semaglutide: HR 0.72 (95% CI 0.58–0.90), p<0.01 — tirzepatide favored.
• No formal subgroup analyses for incident dementia were conducted due to the low absolute event count.

Adverse Events and Safety Profile

This study was not designed to assess treatment-emergent adverse events, and no dedicated safety data are reported. Conditions constituting contraindications to the study drugs — pancreatitis, thyroid cancer, gallstones, gastroparesis, and diabetic ketoacidosis — were exclusion criteria rather than monitored outcomes. Tolerability differences between agents may exist in practice and could influence real-world adherence, but TriNetX does not capture refill behavior, dose titration, or explicit stop dates; therefore, differential discontinuation rates between arms remain unquantifiable in this dataset.

Statistical Approach and Rigor

The target trial emulation framework, combined with an active comparator new-user design and a six-month washout, represents a methodologically sound approach to minimizing prevalent-user bias, carryover confounding, and immortal time bias. Propensity score matching on 32 covariates achieved acceptable balance (SMD <0.10). The use of E-values to quantify sensitivity to unmeasured confounding is appropriate and reported transparently. However, several concerns temper confidence: the tirzepatide-versus-semaglutide comparison yields a CI E-value of only 1.11, meaning a confounder with an association as modest as 1.11-fold with both exposure and outcome could shift the confidence interval to include the null. Absolute event counts are low, particularly in TTE1 (55 vs. 79 events) and TTE3 (51 vs. 73 events), constraining precision and limiting subgroup analyses. Dementia ascertainment relied on ICD-10 codes, which may undercount true incident cases — especially with short follow-up. The intention-to-treat framework, while appropriate for emulating a trial, may attenuate true pharmacological differences when adherence is differential. Missing data are excluded rather than imputed by TriNetX, and completeness for HbA1c and BMI was below 85% in several arms.

Clinical Takeaway

For clinicians managing adults with type 2 diabetes who carry elevated dementia risk — an overlapping population given the 60% excess dementia risk conferred by type 2 diabetes — this study provides the first direct comparative signal that tirzepatide may reduce incident dementia more than semaglutide or SGLT2 inhibitors. The absolute event numbers are small, the follow-up is short relative to the decades-long neurodegenerative trajectory of dementia, and residual confounding from channeling bias (SGLT2 inhibitors preferentially prescribed in heart failure and chronic kidney disease; incretin therapies in obesity) cannot be excluded despite matching. The Bell’s palsy falsification endpoint is reassuring, and the effect direction is consistent across all three comparisons. This is a signal worth monitoring closely, not a practice-changing finding on its own.

Why This Matters Clinically

Dementia affects an estimated 57.4 million people globally and is projected to reach 152.8 million by 2050, with type 2 diabetes independently conferring a roughly 60% excess risk. No disease-modifying pharmacotherapy for dementia has passed phase 3 trials. The EVOKE and EVOKE+ trials of oral semaglutide in early Alzheimer’s disease failed to meet their primary endpoint, and the ELAD trial of liraglutide did not achieve significance on cerebral glucose metabolism. Against that backdrop, observational data suggesting a meaningful dementia risk reduction with tirzepatide — the most efficacious incretin-based therapy currently available — deserves scientific attention. If confirmed in randomized trials, the implication would be that selecting tirzepatide over semaglutide in a patient with type 2 diabetes and dementia risk factors serves dual metabolic and neuroprotective goals. The cardiovascular and mortality associations must be interpreted with particular caution given the magnitude and the plausibility of residual confounding.

CED Clinical Relevance

At CED Clinic, patients presenting with cannabis use and concurrent metabolic conditions — including type 2 diabetes, obesity, and metabolic syndrome — frequently carry the same cardiometabolic risk profile studied here. Many of these patients are already engaged in conversations about GLP-1 receptor agonist therapy for weight management and glycemic control. This study adds a preliminary, hypothesis-generating layer to those discussions: the agent chosen for metabolic reasons may also carry differential neurocognitive implications. That said, no prescription decision at CED Clinic should be altered on the basis of this single retrospective signal. The appropriate response is to document dementia risk factors formally, revisit the literature as RCT data emerge, and ensure patients understand the speculative nature of neuroprotection claims for any current diabetes medication.

Fits What We Already Know

This study situates within a growing body of observational and trial-derived evidence. A Korean nationwide cohort study cited by the authors found that SGLT2 inhibitors reduced all-cause dementia risk by 21% versus other oral glucose-lowering agents. Wang et al., using a similar TriNetX-based target trial emulation, found semaglutide associated with a 40–70% lower dementia incidence compared to insulin or other GLP-1 receptor agonists — a larger effect size than the semaglutide-versus-SGLT2-i comparison in the current study, which the authors attribute to the choice of a more active comparator. Tang et al. demonstrated GLP-1 receptor agonists associated with 31% lower dementia risk versus sulfonylureas and 23% lower risk versus DPP-4 inhibitors in older Swedish patients with type 2 diabetes. Pooled randomized controlled trial data from Nørgaard et al. showed GLP-1 receptor agonist treatment associated with a 53% lower dementia rate compared to placebo. In animal models, GLP-1/GIP dual agonism has demonstrated superior neuroprotective effects relative to GLP-1 monotherapy in models of Alzheimer’s disease and Parkinson’s disease, and tirzepatide specifically has shown a protective effect on spatial learning and memory in diabetic rats via modulation of insulin resistance and amyloid-beta accumulation — though two mouse model studies found neither semaglutide nor tirzepatide altered neurodegeneration pathology. The current study is the first to pit tirzepatide directly against semaglutide and SGLT2 inhibitors on a dementia endpoint in a large human cohort.

What This Study Teaches Us

In plain terms: among adults with type 2 diabetes followed for two years, those who started tirzepatide were diagnosed with dementia less frequently than those who started either semaglutide or an SGLT2 inhibitor — even after adjusting statistically for dozens of background differences between groups. The size of the difference was modest in absolute terms (55 vs. 79 events in the tirzepatide/SGLT2-i comparison; 51 vs.

Physician-Led, Whole-Person Care

A doctor who takes the time to truly understand you.

Personal care that starts with listening and is guided by experience and ingenuity.

Health, Longevity, Wellness

One-on-One Cannabis Guidance

Metabolic Balance

Leave a Message
Metabolic Care
Medical Consulting
Cannabis Care