Investigators examined whether dementia risk differs across individual angiotensin receptor blockers compared with lisinopril. The study used observational data to emulate randomized target trials, drawing on a linked network of 44 US health care organizations. The design focused on incident cognitive outcomes—chiefly all-cause dementia—and included several cerebrovascular and mortality endpoints as secondary measures. This framing positions individual ARB agents, not just the class as a whole, as the exposure of interest.
From an initial screen of 22,025,599 individuals, the authors identified 2,051,275 adults who met eligibility. Eligible participants were aged 55 to 80 years and were naive to ACEI/ARB therapy at initiation of either an ARB or lisinopril. The study team performed separate 1:1 propensity-score matching for each agent-level comparison; the abstract did not enumerate the covariates included in the propensity scores. Matched cohorts varied widely in size—from 10,113 up to 857,773 patients—and median follow-up across comparisons ranged from 4.7 to 7.2 years, providing multi-year observation for incident outcomes.
Cox proportional-hazards models estimated hazard ratios (HRs) for incident all-cause dementia. Pooled ARBs, taken together, were associated with a lower incidence of dementia compared with lisinopril (HR, 0.86; 95% CI, 0.85 to 0.87). The study team observed substantial heterogeneity when agents were disaggregated: individual ARB HRs versus lisinopril ranged from 0.66 to 0.93. Those agent-level differences indicate that not all ARBs carry the same association with dementia risk in this observational emulation. The source did not report specific covariate adjustment lists or proportional-hazards test results in the abstract.
The authors also compared non-losartan ARBs with losartan in within-class analyses. Within-class hazard ratios spanned 0.68 to 0.94, mirroring the between-class ordering by HR estimates and demonstrating consistent heterogeneity when the reference was another ARB. The most commonly prescribed agent, losartan, showed the least favorable association with dementia among the ARBs examined. The abstract does not provide agent-by-agent numerical details for every comparison, but it reports that the ordering of agents by HR was reproduced in both between-class and within-class analyses.
Secondary endpoints included Alzheimer disease, vascular dementia, cerebral infarction, death, and a combined outcome of dementia or death. The investigators reported considerable agent-level variation across these secondary outcomes as well. The source abstract did not include the full set of agent-specific effect estimates for each secondary endpoint; readers seeking exact secondary-outcome HRs must consult the full preprint or request aggregate results from the corresponding author.
Sensitivity analyses encompassed lagged exposure windows, treatment-persistence analyses, negative-control outcomes, and calculation of E-values. Negative-control associations were near null within class, and the authors stated that sensitivity analyses supported the robustness of the primary findings. The abstract did not publish the numerical outputs for these sensitivity checks; therefore the magnitude of E-values and detailed negative-control estimates are not reported here because the source did not provide them in the abstract.
This target-trial emulation used large observational datasets, but key limitations remain. The source abstract does not list the precise confounders adjusted for in propensity scoring, nor does it provide the full agent-level results for every secondary endpoint and sensitivity analysis. TriNetX individual-level data are not publicly available; aggregate results can be requested from the corresponding author. The study team noted that the observed agent-level heterogeneity in dementia associations requires confirmation in randomized trials. Ethical oversight was handled at Weill Cornell Medicine, which determined IRB review was not required because analyses used de-identified secondary data. Competing interests were disclosed: one author reported unrelated advisory and financial roles; other authors reported no relevant disclosures.
Limitations documented in the source include the observational design and reliance on electronic health record and claims coding for outcome ascertainment. Specifics on propensity-score covariates, diagnostic adjudication methods, and full numeric outputs for sensitivity analyses were not provided in the abstract and therefore could not be summarized here. These unanswered questions define the next clinical and methodological steps that the authors identified—chief among them is randomized confirmation of agent-level differences in dementia risk.