Introduction

Dementia is a significant global health concern, impacting more than 50 million individuals today, with projections indicating that the figure will surpass 150 million by 2050 [1]. In Australia, around 433,300 individuals are living with dementia, and this figure is expected to almost double by 2054 [2]. Rather than arising abruptly in late life, dementia can be viewed as a clinical manifestation of cumulative biological ageing processes, shaped over decades by vascular injury, neuroinflammation, metabolic stress, and declining brain resilience [3,4,5]. Hypertension, affecting about one-third of the global population, is a well-established modifiable risk factor for dementia and a key driver of these chronic vascular and inflammatory trajectories [6, 7]. While controlling blood pressure is a key strategy for reducing dementia risk [7, 8], increasing attention has focused on whether specific antihypertensive medications (AHMs) may have neuroprotective benefits beyond their blood pressure-lowering effect [9].

Angiotensin receptor blockers (ARBs) and angiotensin-converting enzyme inhibitors (ACEIs), which modulate the renin-angiotensin system (RAS), have attracted particular interest because RAS signalling in the brain and cerebral vasculature is implicated in neuroinflammation, oxidative stress, and vascular stiffness, key pathways in ageing-related cognitive decline [3, 9]. These agents may exert pleiotropic effects, including improved cerebral blood flow, reduced neuroinflammation, and enhanced microvascular resilience, positioning them as potential modulators of chronic vascular and inflammatory ageing trajectories [9]. However, clinical data on the dementia-preventive effects of RAS-targeting AHMs remain limited and mixed [9,10,11,12,13]. Some evidence suggests that ARBs may offer superior cognitive protection compared to ACEIs, potentially through stimulation of angiotensin II (Ang-II) type 2 and type 4 receptors (AT2/AT4), which are associated with reduced oxidative stress and improved cerebral perfusion [14]. For instance, a retrospective cohort analysis based on Optum’s de-identified Clinformatics® Data Mart, which included more than six million hypertensive patients, reported that ARB users had a 20% lower adjusted risk of developing Alzheimer’s disease and related dementias (ADRD) compared with ACEI users (hazard ratio [HR] = 0.80; 95% CI [confidence interval]: 0.79–0.80) over an average follow-up of 5.6 years [13]. Similarly, a multinational cohort study conducted in Hong Kong, the UK, Sweden, and Australia, including nearly two million participants, showed that initiating ARB therapy was linked to a lower risk of developing all-cause dementia (HR = 0.92; 95% CI: 0.89–0.94) over a median follow-up of 5.6 to 8.4 years [11].

However, data from clinical trials are limited, and only two trials with short follow-up (less than five years) have compared the effects of ARBs with ACEIs. The ONTARGET and TRANSCEND randomised controlled trials (RCTs) found no significant difference in dementia or cognitive decline between ACEI and ARB treatment groups (odds ratio (OR) = 0.97, 95% CI: 0.89–1.06) over a median follow-up of 4.9 years [12]. Likewise, a 2025 target trial emulation using the SPRINT cohort reported no significant difference in adjudicated mild cognitive impairment/dementia risk between ARB and ACEI users after accounting for medication adherence (risk ratio (RR) = 0.94, 95% CI: 0.66–1.29) during 4 years of follow-up [10].

These conflicting findings and the short follow-up of clinical trials highlight ongoing uncertainty about the extent to which ARBs and ACEIs influence the risk of developing dementia and whether observed differences reflect true pharmacological effects on ageing-related vascular and inflammatory pathways or residual confounding in real-world prescribing. Prior studies often lacked adjustment for key lifestyle factors, such as diet and physical activity, which are established dementia risk modifiers and important determinants of vascular and inflammatory ageing [15]. To date, no large-scale, real-world study with rigorous exposure definition (≥ 80% adherence) has simultaneously examined head-to-head comparisons of individual ARBs and ACEIs while adjusting for physical activity and diet, as existing evidence is mostly limited to class-level analyses that preclude precise drug-level comparisons.

Using linked data from the 45 and Up Study through the Chronic Conditions Umbrella Programme Linkage (CUPL) [16], we addressed these gaps by defining exposure rigorously, examining individual drug effects, and exploring effect modification by sex, diet, physical activity, and comorbidities, with a view to understanding how specific RAS-targeting AHMs may modulate long-term vascular and inflammatory processes that culminate in dementia.

Methods

Data source, study participants and study design

This study leveraged the Sax Institute’s 45 and Up Study, a large, ongoing cohort of 267,357 adults aged 45 years and older in New South Wales (NSW), Australia [17]. Study participants were recruited from the general population through random selection using the Services Australia Medicare enrolment database, which provides near-complete coverage of the Australian population. Initial recruitment occurred between 2005 and 2009, with follow-up data collected through subsequent survey waves: Wave 2 (2012–2015), Wave 3 (2018–2020), and Wave 4 (2023–ongoing). People aged 80 years or above, along with those living in rural or remote regions, were intentionally oversampled. Approximately 19% of those invited participated, representing ~ 11% of the NSW population aged 45 years and over. All participants provided consent for ongoing follow-up and linkage to routinely collected health data using participant project numbers [18]. Data analysis was performed through a secure linkage platform developed for the 45 and Up Study to integrate various administrative and health datasets [16]. The data sources that were linked and used in this study included: the 45 and Up Study baseline survey, along with data from waves 2, 3, and 4 (https://www.saxinstitute.org.au/solutions/45-and-up-study/use-the-45-and-up-study/data-and-technical-information), which was linked by the Centre for Health Record Linkage (CHeReL) to the NSW Admitted Patient Data Collection (APDC), the NSW Emergency Department Data Collection (EDDC), and the NSW Mental Health Ambulatory Data (MHAMB). CHeReL employed a probabilistic linkage method, in which records with uncertain match probabilities were manually reviewed to ensure accuracy (www.cherel.org.au). The NSW APDC contains records of all hospital admissions in public and private hospitals in NSW from 2001 to 2023, while the NSW EDDC includes emergency department presentations from 2005 to 2023, and the NSW MHAMB comprises outpatient mental health service data from 2001 to 2023. These datasets were provided by the NSW Ministry of Health [19,20,21], which we gratefully acknowledge. In addition, the Pharmaceutical Benefits Scheme (PBS) data, provided by Services Australia, which includes information on subsidised prescription medications dispensed in Australia from 2004 to 2023. Linkage of the 45 and Up Study cohort to PBS data was facilitated by the Sax Institute using a unique identifier and deterministic matching. Finally, the 45UpDeaths dataset, which was prepared by the Sax Institute using data from the National Death Index, provided by the Australian Institute of Health and Welfare, and covered the period 2005 to 2024.

Hypertension cohorts were identified using multiple data sources. A primary diagnosis of hypertension was found using the NSW APDC for hospitalisations based on specific codes from the International Classification of Diseases, 10th revision, Australian modification (ICD-10-AM) and the EDDC, which included codes from ICD-10-AM, ICD, 9th edition, clinical modification (ICD-9-CM), and Systematised Nomenclature of Medicine, Clinical Terms (SNOMED CT), as shown in Supplementary Table 1 (Table S1). Self-reported hypertension was also identified from responses to the question “Has a doctor EVER told you that you have hypertension?” in the 45 and Up Study baseline survey or waves 2, 3, and 4 (https://www.saxinstitute.org.au/solutions/45-and-up-study/use-the-45-and-up-study/data-and-technical-information). Additionally, hypertension was identified through AHM use in PBS records using Anatomical Therapeutic Chemical (ATC) codes in Table S1. Participants were classified as hypertensive if hypertension was identified using any of the data sources (APDC/EDDC diagnoses, self-report, or AHM dispensing). The first date of hypertension diagnosis was determined as the earliest recorded date from all available data sources. In cases of disagreement between sources (e.g., no self-reported hypertension but an ICD-coded diagnosis present), the individual was still classified as hypertensive. This strategy prioritised sensitivity to capture the full spectrum of treated and diagnosed cases. As mentioned below, we also performed a sensitivity analysis by excluding patients whose hypertension diagnosis was based only on AHM use, given the other possible clinical indications for most of these drugs.

A prospective cohort analysis was conducted among patients with hypertension who initiated AHMs between 2004 and June 30, 2022, with follow-up data available through June 30, 2023. This design ensured that all participants had at least one year of continuous AHM exposure. Patients were excluded if they had no recorded use of ARBs or ACEIs during the study period, had less than 80% exposure to either medication, had a recorded diagnosis of dementia prior to their hypertension diagnosis, or had less than 12 months of follow-up after the index date. The index date was defined as the date of the first ARB or ACEI prescription during the eligibility period (between 2004 and June 30, 2022). To ensure sufficient exposure to the medication and reduce the likelihood of including individuals with undiagnosed or pre-existing dementia, follow-up for dementia outcomes began one year after treatment initiation and continued until the earliest of dementia diagnosis, death, or end of follow-up (30 June 2023). The selection flowchart is presented in Fig. 1.

Exposure to ARBs and ACEIs

AHM claims in the PBS database were identified using ATC codes (Table S1) [22]. The main exposure variable was as-treated use of either an ARB or an ACEI. Exposure to ARBs was defined as the use of at least one drug from the ARB class, as listed in the Australian Medicines Handbook (AMH), including candesartan, irbesartan, losartan, olmesartan, telmisartan, and valsartan [23], and exposure to ACEIs was defined as exposure to at least one of the ACEI class, such as captopril, enalapril, fosinopril, lisinopril, perindopril, quinapril, ramipril, and trandolapril [23].

The level of medication exposure was assessed using the proportion of days covered (PDC), a common metric for adherence in chronic disease [24]. PDC was calculated as the total supply days covered by medications divided by the total follow-up days, multiplied by 100 [25]. Our PDC calculation accounted for treatment switching and discontinuation, such that days covered by any drug within the same class contributed to cumulative exposure. Participants exposed to both classes were classified according to the class accounting for ≥ 80% of their total follow-up days covered, reflecting predominant long-term as-treated exposure [25, 26].

Participants were categorised into ARB or ACEI groups based on predominant predominant as-treated exposure over follow-up. The ARBs group consisted of participants with ≥ 80% PDC for ARBs and < 80% PDC for ACEIs, while the ACEIs group included those with ≥ 80% PDC for ACEIs and < 80% PDC for ARBs. We used an 80% PDC cutoff in this study, as it is the most commonly used cutoff in previously published studies [25, 27]. Within each group, exposure to individual drugs was calculated by dividing the number of days covered by a specific drug by the total number of days the participant was exposed to the ARB or ACEI group. To categorise exposure to individual drugs within each group, we required a PDC of ≥ 80% for the specific drug.

Dementia outcomes were assessed after participants were categorised according to their predominant as-treated exposure, reflecting sustained long-term treatment patterns. This approach captures real-world prescribing and adherence behaviours while minimising misclassification due to short-term switching or discontinuation [25, 26], which is particularly relevant given the long prodromal phase of dementia.

Main outcome

The study’s primary endpoint was the development of dementia. Dementia was identified using ICD-10-AM codes in APDC; through ICD-10-AM, ICD-9-CM, and SNOMED CT codes in EDDC; through ICD-10-AM codes in MHAMB and death data; and through self-reported confirmation of a dementia or Alzheimer’s diagnosis in Wave 4 of the 45 and Up Study, responding to the question, “Has a doctor EVER told you that you have dementia or Alzheimer’s disease?” (https://www.saxinstitute.org.au/solutions/45-and-up-study/use-the-45-and-up-study/data-and-technical-information) and through the dispensing of anti-dementia medications in the PBS (Table S1). A primary incident dementia event was defined as the earliest recorded date of a dementia diagnosis in any of the linked data sources, with no hierarchical prioritisation of data sources was applie. Identification of dementia using multiple linked data sources in Australian settings has been shown to improve sensitivity while maintaining high specificity [28, 29]. Although no formal validation has been conducted specifically within the 45 and Up Study, prior research in this cohort suggests that this multi-source approach captures approximately 77–92% of the expected age-specific dementia incidence [29]. A national validation study comparing Australian administrative datasets with adjudicated dementia diagnoses in the ASPREE trial reported a positive predictive value of 72.6%, specificity of 98.2%, and sensitivity of 70.7% [28], figures comparable to or exceeding those reported in other international linked data studies [30], supporting the reliability of multi-source dementia ascertainment in Australia. The secondary outcomes of the study included all-cause mortality and dementia-related mortality. Both outcomes were identified using linked death records from the 45 and Up Study, which include date and cause of death information coded using ICD-10-AM. Due to inconsistent recording of dementia subtypes across datasets and their absence in some sources (e.g., self-reports, PBS), reliable classification was not possible; therefore, a composite dementia outcome was used.

Covariates

Covariates were selected based on their potential roles as confounders, specifically those that could influence both the use of ARBs or ACEIs and the risk of dementia [14, 31, 32]. Comorbidities, including atrial fibrillation, heart failure, depression, diabetes, coronary heart disease, dyslipidaemia, schizophrenia, and stroke, were captured using ICD-10-AM codes from the APDC, through ICD-10-AM, ICD-9-CM, and SNOMED CT codes from the EDDC, and through self-reported responses to the question “Has a doctor EVER told you that you have [condition]?” in the baseline or follow-up waves (2, 3, and 4) of the 45 and Up Study surveys (https://www.saxinstitute.org.au/solutions/45-and-up-study/use-the-45-and-up-study/data-and-technical-information) (Table S1). Additionally, diagnoses of atrial fibrillation and dyslipidaemia were identified through the dispensing of condition-specific medications listed in the PBS (Table S1). A comorbidity was considered present if it was documented at or prior to the index date [31, 33]. Concurrent baseline use of medications of interest — non-steroidal anti-inflammatory drugs (NSAIDs), statins, anticoagulants, and anticholinergics with a cognitive burden score of 2 or 3 [34] — was identified from PBS data using ATC codes (Table S1). Concurrent medication use at baseline was defined as having prescriptions within six months before or after the index date.

Age (calculated from birth year to index date), sex, smoking history, physical activity, and dietary habits were extracted from the baseline 45 and Up Study survey (https://www.saxinstitute.org.au/solutions/45-and-up-study/use-the-45-and-up-study/data-and-technical-information). The total physical activity was calculated by combining minutes spent walking, in moderate-intensity activity, and in vigorous-intensity activity (weighted by a factor of two) [35]. Participants performing 150 min or more of total physical activity per week were considered physically active, in line with the World Health Organisation’s (WHO) global guidelines for health [36]. Dietary habits were evaluated based on self-reported frequency of consumption of red meat (beef, lamb, or pork), poultry (chicken, turkey, or duck), processed meats (such as bacon, sausages, salami, devon, or burgers), and fish or seafood (https://www.saxinstitute.org.au/solutions/45-and-up-study/use-the-45-and-up-study/data-and-technical-information). Participants were classified into two categories: vegetarians, who reported no consumption of any of these foods, and non-vegetarians, which included regular meat eaters (consumed red meat, poultry, or processed meat more than once per week), semi-vegetarians (consumed these meats once per week or less), and pesco-vegetarians (consumed fish or seafood at least once per week but no red or processed meat or poultry) [37].

Statistical analysis

To ensure comparability between ARB and ACEI users, a 1:1 propensity score matching (PSM) approach was applied, accounting for key baseline confounders: age, sex, smoking history, diabetes, atrial fibrillation, congestive heart failure, coronary heart disease, chronic kidney disease, dyslipidaemia, dihydropyridine calcium channel blockers [DHP CCBs], thiazide diuretics, beta-blockers, non-DHP CCBs, antiplatelet agents, anticoagulants, and statins. Matching was conducted without replacement, using a calliper width of 0.05 standard deviations of the logit of the PS. The balance of the matched variables was evaluated using absolute standardised mean differences (ASMD), with values less than 0.10 indicating acceptable alignment [38].

Baseline characteristics of both the unmatched and matched groups were compared using descriptive statistics. Cox proportional hazards regression was employed to estimate the adjusted HRs for dementia incidence, all-cause mortality, and dementia-related mortality. Visual inspection of Schoenfeld residuals was used to assess the proportional hazards assumption [39]. The number of events per predictor was assessed to prevent violations of the proportional hazards assumption [40]. To precise the model estimates the model was adjusted for relevant lifestyle factors (physical activity and diet), baseline use of concurrent medications (other AHM use, NSAIDs, and anticholinergics), as well as key baseline comorbidities recorded (Parkinson’s disease, depression, and schizophrenia). Except for age, matching variables were not included as covariates in the Cox proportional hazards model to preserve the matched design and prevent overadjustment; age was included because a minimal residual imbalance remained after matching. Dementia-free survival between exposure groups was also compared using Kaplan–Meier survival curves and log-rank tests.

Three subgroup analyses were performed. First, exploratory, hypothesis-generating comparisons used lisinopril as the reference agent to examine potential heterogeneity in dementia risk across individual ARBs and ACEIs. Second, exploratory within-class analyses compared irbesartan with other ARBs to assess possible variation in dementia risk among agents within the same drug class. These choices were guided by prior evidence suggesting that lisinopril and irbesartan have relatively minimal or neutral effects on dementia risk within their respective classes, making them appropriate comparators for detecting potential variations across and within drug classes [41, 42]. Third, subgroup analyses by sex were performed, as previous studies suggested potential variation in drug response and dementia susceptibility between men and women [43].

Three sensitivity analyses were conducted. First, to minimise potential misclassification due to the multiple clinical indications of AHMs (e.g., heart failure, proteinuria) and to reduce confounding by indication, patients whose hypertension diagnosis was identified based solely on AHM use were excluded. Second, to address the potential influence of death as a competing event, given that death can preclude the onset of dementia and is common in older populations, we estimated dementia’s cumulative incidence via a Fine and Gray model while accounting for death as a competing risk [44]. Third, as an alternative approach to confounding adjustment, we conducted an inverse probability of treatment weighting (IPTW) analysis based on the propensity score in the full eligible cohort. Stabilised weights were applied, covariate balance before and after weighting was assessed using ASMD, and weighted Cox proportional hazards models with robust variance estimation were used to estimate HRs for incident dementia. Statistical data management and analyses were conducted using SAS software (version 9.4, SAS Institute Inc., Cary, NC, USA).

Ethics approval

The conduct of the 45 and Up Study was approved by the University of NSW Human Research Ethics Committee. Ethical approval for this study was granted by the NSW Population & Health Services Research Ethics Committee (Reference Number: 2021/ETH12383) following review and approval by the Sax Institute Scientific Review Panel. Project access was authorised by the Sax Institute, with data availability commencing in March 2025. Researchers accessed the data securely through the Sax Institute’s Secure Unified Research Environment (SURE). All datasets were de-identified before analysis, with participant-level linkage across databases enabled using unique Sax Institute Project Person Numbers (PPN).

Results

Flowchart of cohort selection and exclusions

Figure 1 shows the selection process from the full 45 and Up Study cohort (N = 267,357) to the 266,411 individuals available for this research, after accounting for participant withdrawals. A total of 64,403 participants met eligibility criteria and were included in the final analytic cohort. Of these, 36,507 were exposed to ARBs and 27,896 to ACEIs. After matching, 25,787 patients remained in each exposure group, with a mean age of 66.3 (9.0) years, 48.2% women, and a mean follow-up of 11.3 (5.2) years (Fig. 1).

Fig. 1
figure 1

Participant selection process for the final cohort (ACEIs, Angiotensin-Converting Enzyme Inhibitors; AHM, Antihypertensive Medication; ARBs, Angiotensin II Receptor Blockers)

Full size image

Baseline characteristics

Table 1 shows demographic and clinical features of ARB and ACEI users before and after PSM. After matching, both groups were well-balanced on baseline key confounders (all ASMDs < 0.1). Almost all covariates demonstrated acceptable balance, with ASMDs below 0.1, except for age, which showed marginal imbalance (ASMD = 0.1; Table 1 and Fig. S1). In the IPTW sensitivity analysis conducted in the full eligible cohort, baseline characteristics were likewise well balanced after weighting, with ASMDs for all measured covariates reduced to below 0.1, indicating adequate balance (Table S2).

Table 1 Comparison of baseline characteristics between ARB- and ACEI-treated patients with hypertension, before and after propensity score matching
Full size table

Dementia incidence rates among unmatched groups

Table S3 presents the crude incidence rates of dementia among unmatched patients treated with ARBs and ACEIs. The ARB group had a dementia incidence of 6.2 per 1000 person-years (95% CI: 6.0–6.4), compared with 8.7 per 1000 person-years (95% CI: 8.4–9.0) in the ACEI group, reflecting a significantly reduced incidence in the ARB group (p < 0.001; Table S3).

Adjusted dementia risk in matched groups

ARB use was linked to a markedly reduced risk of developing dementia compared to ACEI use (HR = 0.72; 95% CI: 0.65–0.80, p < 0.001), after controlling for diet, physical activity, comorbidities and concomitant medications (Table 2).

Table 2 Hazard ratios with 95% confidence intervals estimated using multivariable Cox regression
Full size table

Exploratory analyses comparing individual ARBs and ACEIs suggested potential differences in dementia risk (Table 2). Among ARBs, olmesartan (n = 2088; 62 events) showed the lowest risk (HR = 0.32; 95% CI: 0.16–0.62), followed by candesartan (n = 6815; 506 events, HR = 0.41; 95% CI: 0.24–0.69), telmisartan (n = 6832; 439 events, HR = 0.42; 95% CI: 0.25–0.71), and irbesartan (n = 9670; 965 events, HR = 0.45; 95% CI: 0.27–0.75), with all showing statistically significant benefits relative to lisinopril. Estimates for less frequently used drugs, such as captopril (n = 116; 36 events), were based on small numbers and thus less precise (HR = 4.9; 95% CI: 1.04–23.4) (Table 2).

Exploratory comparative analysis among individual ARBs

In exploratory, hypothesis-generating analyses examining heterogeneity within the ARB class, olmesartan had the lowest risk of dementia (HR = 0.55; 95% CI: 0.43–0.71) compared with irbesartan. All other ARBs did not show statistically significant benefits relative to irbesartan (Table S4).

Sex-based subgroup analysis

Figure 2 provides HRs for dementia risk associated with ARB versus ACEI use, stratified by sex. The apparent protective association of ARBs was consistent in females (HR = 0.73; 95% CI: 0.61–0.88) and males (HR = 0.74; 95% CI: 0.64–0.87), without significant sex-based interaction (p for interaction = 0.153) (Fig. 2, Table S5).

Fig. 2
figure 2

Sex-specific subgroup analysis of dementia risk (ARB = angiotensin II receptor blocker; ACEI = angiotensin-converting enzyme inhibitor; CI = confidence interval; HR = hazard ratio; ref = reference)* *Adjusted for physical activity, diet, baseline use of comedications (DHP CCBs, thiazide diuretics, beta-blockers, non-DHP CCBs, other antihypertensives, antiplatelet agents, NSAIDs, anticoagulants, anticholinergics, and statins), and comorbidities during follow-up (atrial fibrillation, congestive heart failure, depression, schizophrenia, coronary heart disease, stroke, and dyslipidaemia)

Full size image

Sensitivity analysis: excluding hypertension defined by medication use

Table S6 displays the results of a sensitivity analysis excluding 11,447 participants whose hypertension diagnosis was based solely on AHM dispensing records. The findings were consistent with the main analysis, showing that ARB users had a significantly lower risk of developing dementia compared with ACEI users (HR = 0.76; 95% CI: 0.67–0.86) (Table S6).

Sensitivity analysis: adjusted for competing risk of death

After adjusting for the competing risk of death with the Fine-Gray subdistribution hazard model, the findings remained consistent. ARB use was linked to a 29% lower subdistribution hazard of dementia compared with ACEI use (Table S7).

Sensitivity analysis: Inverse probability of treatment weighting

After applying IPTW based on the propensity score, the findings remained consistent with the primary analysis. In the IPTW-weighted cohort, ARB use was associated with a significantly lower risk of incident dementia compared with ACEI use (adjusted hazard ratio 0.72, 95% CI 0.68–0.76) (Table S8).

All-cause mortality risk

Results indicated a significant reduction in the risk of death for ARB users compared to ACEI users (HR = 0.77; 95% CI: 0.73–0.82, p < 0.001). In exploratory agent-level analyses, olmesartan showed the greatest risk reduction (HR = 0.64; 95% CI: 0.56–0.74), followed by telmisartan (HR = 0.91; 95% CI: 0.85–0.97) and candesartan (HR = 0.92; 95% CI: 0.86–0.98) compared with irbesartan (Table S9).

ARBs showed varied mortality risk compared to lisinopril. Olmesartan was associated with the greatest reduction in mortality risk (HR = 0.34; 95% CI, 0.24–0.48) (Table S9). Perindopril was linked to the lowest risk of death among ACEIs (HR = 0.70; 95% CI: 0.50–0.86) compared with lisinopril (Table S9). After excluding patients whose hypertension diagnosis was based solely on AHM records, ARBs were significantly associated with the reduction of all-cause mortality compared with users of ACEIs (HR = 0.80; 95% CI: 0.75–0.85) (Table S9).

Dementia-related mortality

Table S10 presents the analysis of dementia-related mortality among users of RAS-acting AHMs. Compared with ACEIs, ARBs were associated with a significant reduction in dementia-related mortality (HR = 0.81; 95% CI: 0.74–0.89; p < 0.001). In exploratory within-class analyses, olmesartan showed the greatest reduction in dementia-related mortality (HR = 0.52; 95% CI: 0.33–0.82) compared to irbesartan (Table S10). After excluding patients whose hypertension was identified solely through AHM records, dementia-specific mortality did not differ significantly between ARB and ACEI users (HR = 0.84; 95% CI: 0.66–1.06) (Table S10).

Kaplan–Meier survival curves and log-rank tests

Figure S2 shows dementia-free survival after matching. ARB users (red) had a slightly higher survival probability than ACEI users (blue), with a significant difference by log-rank test (p < 0.001), suggesting a potential protective effect of ARBs (Fig. S2).

Discussion

In this large PSM cohort of 51,574 patients, ARB use was associated with a 28% lower risk of developing dementia compared to ACEI use (HR = 0.72; 95% CI: 0.65–0.80) over an average follow-up period of 11 years in a real-world clinical setting. This association was consistent across three subgroup analyses (based on sex and head-to-head individual ARBs and ACEIs comparison) and three sensitivity analyses (excluding hypertension defined only by AHM use, adjustment for the competing risk of death and IPTW analyses). Fine-Gray analyses account for death as a competing event and reflect cumulative incidence rather than the instantaneous hazard. The similarity of the Fine-Gray (HR = 0.71) and cause-specific Cox results (HR = 0.72) indicates that differences in mortality did not drive the lower dementia risk associated with ARB use.

To our knowledge, this is the first study to examine the risk of dementia in people with hypertension who received ARB compared with ACEI, with adjustment for two key modifiable lifestyle risk factors for dementia, diet and physical activity, which are increasingly recognised as critical determinants of cognitive health, vascular ageing, and dementia risk [15]. Prior observational studies and randomised trials have largely overlooked these variables [11, 13, 41, 45, 46], potentially confounding the observed associations and their links to underlying ageing biology. By accounting for these factors, our study enhances causal inference. It provides a more accurate estimate of how RAS-acting AHMs modulate long-term vascular and inflammatory ageing processes that drive dementia.

While prior trials, such as ONTARGET and TRANSCEND, have reported no significant difference in the risk of dementia between ARBs and ACEIs [10, 12], their relatively short follow-up periods (median 5 years) may have limited their ability to detect effects on the protracted ageing trajectories leading to dementia [10]. Our findings are consistent with several large observational studies showing lower dementia risk with ARB use compared to ACEIs [13, 41, 45, 46], and with longer follow-up better capturing cumulative impacts on brain ageing pathways. A cohort of over 800,000 patients reported a 19% lower risk of dementia with ARBs versus lisinopril, with an average follow-up of 2.5 years [41]. A multinational study of nearly 2 million individuals found that ARB initiation was associated with an 8% reduction in all-cause dementia risk, with median follow-up ranging from 5.6 to 8.4 years [11]. Another large cohort study reported a 20% lower adjusted risk of developing AD and related dementias (ADRD) among ARB users compared to ACEI users, with a follow-up of 5.6 years [13].

Exploratory analyses of individual ARBs and ACEIs suggested some variation in their associations with dementia risk; however, several comparisons involved small numbers of participants and events, leading to wide confidence intervals. For example, the HR for captopril was 4.9 (95% CI: 1.04–23.4) based on only 36 events. These findings should be interpreted cautiously and considered hypothesis-generating rather than definitive. The primary inference remains at the class level, with ARBs overall associated with lower dementia risk than ACEIs, potentially through superior modulation of brain RAS signalling, neuroinflammation, and vascular resilience during ageing. The potential heterogeneity among individual drugs warrants further investigation in larger datasets, but no strong conclusions can be drawn from the present study.

Both ARBs and ACEIs modulate the RAS but through different mechanisms implicated in brain ageing [10]. ARBs block AT1 receptors directly, reducing oxidative stress, inflammation, and amyloid-beta accumulation, while preserving AT2 receptor stimulation, which may confer neuroprotection against vascular stiffness and chronic inflammatory trajectories [14, 47, 48]. Some ARBs, such as telmisartan and candesartan, also have partial agonist activity at peroxisome proliferator-activated receptor gamma (PPAR-γ), offering additional anti-inflammatory benefits that may enhance brain resilience to ageing [49, 50]. By contrast, ACEIs reduce Ang-II production but do not completely suppress its effects, and by reducing amyloid-β degradation, they may have a more limited impact on the vascular and inflammatory ageing processes culminating in dementia [51, 52]. The pharmacological distinctions provide plausible biological explanations for the observed differences in dementia risk. Recent guidelines indicate that each drug class can be used interchangeably for treating high blood pressure and cardiovascular disorders, as they offer similar therapeutic benefits [53].

Beyond cognitive outcomes, ARB use was also associated with a statistically significant reduction in all-cause mortality compared to ACEI use (HR = 0.77; 95% CI: 0.73–0.82, p < 0.001), consistent with previous observational studies [11]. Among individual agents, olmesartan again demonstrated the greatest reduction in mortality risk, while perindopril was the most favourable among ACEIs. In addition, ARB use was associated with a significantly lower risk of dementia-related mortality (HR = 0.81; 95% CI: 0.74–0.89; p < 0.001), supporting the possibility that modulation of RAS signalling may influence not only cognitive outcomes but also broader survival trajectories in ageing populations.

Strengths and limitations

A major strength of our study is the use of the 45 and Up Study, Australia’s largest ongoing study of health and ageing, linked to administrative health datasets. The extended follow-up period (mean 11 years) allowed us to assess long-term outcomes and reduce reverse causation. Several covariates, including baseline diet and physical activity, were used to adjust the Cox proportional hazards model. We also applied a ≥ 80% PDC threshold to define exposure, thereby improving the accuracy of capturing consistent medication use.

However, the study has limitations. As an observational study, causal inference is limited, and potential residual confounding from unmeasured factors (e.g., genetic susceptibility, undiagnosed cognitive impairment, frailty, renal function stage, or clinician treatment preference) may persist. Although the ≥ 80% PDC threshold is a validated measure of adherence [25, 27], it does not guarantee actual medication ingestion and may not fully capture complex medication-taking behaviours. We accounted for switching between drug classes, but we lack data on titration, reasons for switching, or dose changes. We followed as-treated exposure to ARBs or ACEIs, which captured the predominant medication class prescribed during the follow-up. While this approach captures real-world treatment patterns, exposure classification relied on cumulative post-baseline information and may be subject to adherence-related selection and potential immortal-time bias [54]. Therefore, the results should be interpreted as reflecting the effect of sustained exposure among adherent users. Confounding by indication may persist despite matching and adjustment. Low event counts for some ARBs (losartan and eprosartan) limited subgroup analyses. We were unable to examine dementia subtypes because they were not generally recorded. While the cohort is broadly representative of older adults in NSW, the findings may not generalise to younger populations [17]. Comorbidities and comedication use were  assessed only at baseline, potentially missing changes over time. Atrial fibrillation and dyslipidaemia were identified via medication records, which may lead to misclassification. We did not have a placebo or neutral comparator group, which is difficult to establish in observational studies due to confounding by indication. Finally, the lack of blood pressure (BP) measurements and longitudinal data prevented adjustment for differences in baseline BP, achieved targets, or treatment intensity. As BP management is central to vascular and brain ageing, class- or agent-specific dementia protection reported in this study should be interpreted cautiously without such data. Nevertheless, BP-lowering effects are generally comparable across RAS-acting AHMs [55], and studies adjusting for BP have reported similar associations with dementia risk [56].

Clinical and research implications

Prevention is crucial for dementia, as there is currently no cure. This study found that ARBs were associated with significantly reduced dementia risk, relative to ACEIs, offering a preventive, inexpensive, and easily administered option for those already on AHMs. Given no known risk in prescribing ARBs over ACEIs for hypertension, clinicians could consider this change while awaiting further validation. The observed differences within drug classes highlight the potential need for individualised RAS-acting AHMs based on both cardiovascular and cognitive outcomes. Future research should clarify the mechanisms underlying these possible effects, particularly regarding blood–brain barrier penetration, and validate the findings across diverse populations to ensure broad applicability.

Conclusion

ARB use was associated with a significantly lower risk of incident dementia and a modest reduction in all-cause mortality compared to ACEI use among patients with hypertension, independent of key modifiable risk factors such as physical activity and diet. The benefits appeared to be agent-specific, with olmesartan, candesartan, and telmisartan showing the most pronounced protective effects. While causality cannot be established from this observational study, the findings highlight the potential role of ARBs, particularly certain agents, in reducing dementia risk, warranting further confirmation in randomised controlled trials.