Associations of Anti-Hypertensive Treatments with Alzheimer's Disease, Vascular Dementia, and Other Dementias: "Patients with AD were around half as likely to be prescribed ARBs"
Download the PDF here
Journal of Alzheimer's Disease
Neil M. Davies1, 2, Patrick G Kehoe3, Yoav Ben-Shlomo1, Richard M. Martin1, 2
1School of Social and Community Medicine, Faculty of Medicine and Dentistry, University of Bristol, Canynge Hall, Bristol, UK
2MRC Centre for Causal Analysis in Translational Epidemiology (CAiTE), University of Bristol, Oakfield House, Bristol, UK
3Dementia Research Group, School of Clinical Sciences, Faculty of Medicine and Dentistry, University of Bristol, John James Laboratories, Frenchay Hospital, Bristol, UK
"Analyses using conditional logistic regression gave similar results. When we restricted the analyses to patients exposed either to ARBs or ACE-Is as sole-therapy (Supplementary Table 4), there was little evidence of an association of ACE-I sole-therapy (OR; 1.01, 95%CI, 0.91-1.12), but there was an inverse association of ARB sole-therapy (OR; 0.63, 95%CI, 0.45-0.88), with probable or possible AD"
We investigated whether angiotensin II receptor blockers (ARBs) and angiotensin converting enzyme inhibitors (ACE-Is) are more strongly associated with Alzheimer's disease (AD), vascular dementia (VaD), and other dementias, than other anti-hypertensive drugs. We conducted a nested case-control analysis within the UK general practice research database, with prospectively recorded anti-hypertensive prescribing data. We sampled cases aged ≥60 years and diagnosed between 1997-2008 (5,797 with AD, 2,186 with VaD, 1,214 with unspecified/other dementia) which were matched to up to four controls by age, general practice and gender. We computed odds-ratios and dose response effects for AD, vascular and unspecified/other dementia, comparing those prescribed ARBs or ACE-Is for at least six months with patients prescribed other anti-hypertensives. We controlled for matching factors, co-morbidities, smoking status, an area measure of socioeconomic status, consultation rate and blood pressure and accounted for reverse causality by introducing time-lags of up to eight years prior to diagnosis/index date. Patients diagnosed with AD, vascular and unspecified/other dementia had fewer prescriptions for ARBs and ACE-Is. Inverse associations with AD were strongest for ARBs (odds-ratio; 0.47, 95%CI, 0.37-0.58) compared with ACE-Is (odds-ratio; 0.76, 95%CI, 0.69-0.84) (pdifference < 0.001). Associations of ARBs with AD were stronger than for vascular dementia (pdifference = 0.01) and unspecified/other dementia (pdifference = 0.23). There were inverse dose-response relationships between ARBs and ACE-Is with AD (both ptrend < 0.01). The inverse association of ACE-Is with AD diminished when using longer time lags but the ARB-AD association persisted. Patients with AD were around half as likely to be prescribed ARBs. Further randomized controlled trial evidence is required to rigorously test these findings.
Alzheimer's disease (AD) and vascular dementia (VaD) are common forms of dementia. AD has characteristic neuropathological hallmarks of cortical and cerebrovascular deposits comprised mainly of amyloid- (A ) peptides and intracellular neurofibrillary tangles comprised of hyperphosphorylated forms of microtubule-associated tau protein . VaD pathology is more heterogeneous and results from various forms of cerebral ischemic disease and damage and diminished cerebral blood flow due to cardiac function abnormalities . AD and VaD often co-exist as a mixed dementia , thereby complicating diagnosis . Hypertension is a risk factor for dementia. While anti-hypertensive drugs reduce VaD risk
, the association between hypertension and AD appears more complicated [5, 6]. Observed reductions in cerebral blood flow [7, 8] may be secondary to AD , microvascular abnormalities may precede neuropathology [1, 10], and intra-arterial infusion of A raises blood pressure in normotensive or naturally hypertensive rats . Additionally, a number of reported A -degrading enzymes, also important in vasoregulation, have altered behavior in postmortem brain tissue [5, 12].
Angiotensin-1 converting enzyme-inhibitors (ACE-Is) reduce blood pressure by reducing both the formation of the vasoconstrictor angiotensin II and the degradation of the vasodilator bradykinin. Reported associations with dementia has been inconsistent; some studies suggest modest cognitive benefits in mild cognitive impairment [5, 12] or AD  while oth- ers show an increased risk of dementia and disability . The latter adverse associations are consistent with genetic, biochemical, and neuropathological evidence that ACE-Is may contribute to A -related pathology in AD because ACE reputedly degrades A . In contrast, the newer angiotensin II receptor blockers (ARBs), which block angiotensin II signaling rather than its production, are therefore 'ACE sparing' and theoretically preserve ACE's proposed A -degrading function while still being anti-hypertensive. ARBs protect against both cognitive deficits and A related pathology in animal models of AD . In one small trial, elderly hypertensive patients treated with losartan (an ARB) had better cognitive outcomes than atenolol- treated patients , while more recent observational data suggested ARBs were associated with lower rates of dementia incidence and slower progression of AD compared to lisinopril (an ACE-I) and other antihypertensive treatments .
We conducted a case-control study nested with UK primary care data to examine associations of ARBs and ACE-Is with AD and VaD. Our study improves on previous observational studies, by controlling for possible confounders including blood pressure, which may determine the type of anti-hypertensive prescribed, and protopathic bias . We hypothesized
a priori that AD would be more strongly associated with ARBs than ACE-Is, compared to other anti-hypertensives,
and the associations would be strongest for diagnoses of AD compared to non-specific dementia or VaD.
There were a total of 20,021 cases and 77,475 controls, of which 9,197 (46%) cases and 39,166 (51%) controls were on anti-hypertensive treatment. Amongst those on anti-hypertensive treatment, 2,227 patients were categorized as probable AD; 3,570 possible AD; 2,186 probable VaD; and 1,214 as unspecified or other dementia group (Table 1). Coronary heart disease, stroke, and diabetes co-morbidity were more common in patients with probable VaD compared to those with probable AD or controls (all p values < 0.001). There was little difference in mean blood pressure across the different groups. Treatment with ACE-Is was more common in patients with probable VaD (38%) and controls (38%) than probable AD (33%). Treatment with ARBs was more common in controls (7%) than those who developed probable AD (4%).
In fully-adjusted models, patients ever prescribed either ARBs or ACE-Is were less likely to develop probable AD, possible AD, VaD, and unspecified/other dementia than patients ever prescribed other antihypertensive drugs (Table 2). Fewer patients ever prescribed ARBs developed AD (fully adjusted odds ratio (OR); 0.47, 95%CI: 0.37-0.58) compared with other anti-hypertensive drugs. The association was weaker with ACE-Is (OR; 0.76, 95%CI: 0.69-0.84) (test for difference between ARB and ACE-I associations p < 0.001) . Associations of ARBs with AD were stronger than for VaD (p difference = 0.01) and unspecified/other dementia (p difference = 0.12); in contrast there was little statistical evidence that associations of ACE-Is with AD were stronger than for unspecified/other dementia (p difference = 0.15) and VaD (p difference = 0.22). The fully adjusted model hardly differed from the basic adjusted associations for probable and possible AD. However, after adjustment for confounders, inverse associations of ARBs and ACE-Is with VaD and unspecified/other dementia increased in magnitude beyond that observable by chance. Further adjustment for the index of multiple deprivation or smoking status had negligible impact on observed associations (Supplementary Tables 2 and 3). As there was little evidence of differences in associations for probable and possible AD outcomes (p = 0.28), we combined probable and possible AD into a single category in subsequent tables to enhance statistical power. We found evidence of a dose response relationship between DDDs of ACE-I and probable/possible AD (p < 0.001) (Table 3). Similarly, we found evidence of a dose response relationship between ARBs and probable/possible AD (p = 0.009) (Table 3); there was little evidence that these associations were non-linear (both p > 0.26). We also found evidence of a dose response relationship between DDDs of ACE-Is and VaD (p = 0.01), but this was not evident for ARBs (p = 0.63). We found no evidence of any interactions between exposure to either ARBs or ACE-Is and age
and probable or possible AD (both LR-tests p > 0.10).
Tables 4 and 5 show the fully adjusted associations of ARBs and ACE-Is with probable or possible AD and VaD after the introduction of sequentially increasing time lag periods. The inverse association of ACE-Is with probable AD was increasingly attenuated with increased lags. In contrast, even after the introduction of lag times of eight years, the 50% reduction in risk of probable and possible AD associated with ARBs with remained (Table 5). There was little evidence of
an association of ACE-Is or ARBs with VaD after the introduction of a lag time of at least two years.
Analyses using conditional logistic regression gave similar results. When we restricted the analyses to patients exposed either to ARBs or ACE-Is as sole-therapy (Supplementary Table 4), there was little evidence of an association of ACE-I sole-therapy (OR; 1.01, 95%CI, 0.91-1.12), but there was an inverse association of ARB sole-therapy (OR; 0.63, 95%CI, 0.45-0.88), with probable or possible AD. Results were similar when we classified exposure based on initial therapy received (OR for ACE-I as initial therapy = 1.04; 95%CI: 0.96-1.12; OR for ARB as initial therapy = 0.66; 95%CI: 0.50, 0.86).
This is the first large UK-based investigation of the association of anti-hypertensive medication exposure with risk of AD and other dementias in primary care. We tested whether ARBs had a stronger association with AD than ACE-Is or other anti-hypertensives and whether the association was specific to AD rather than VaD. This hypothesis was generated because of ARBs influence on two biological pathways. While both ARBs and ACE-Is reduce angiotensin II signaling, now believed to be involved in the pathobiology of AD , ARBs are unlikely to interrupt ACE-mediated A degradation (unlike ACE-Is) . These mechanisms of action suggest that ARBs may have benefits over ACE-Is in the etiology of AD .
We found that fewer patients prescribed ARBs and ACE-Is went on to develop probable AD. Patients taking ARBs and ACE-Is had 53% and 24% lower risks of AD respectively, compared to patients taking other anti-hypertensive medications in basic adjusted analyses. These associations did not differ by age, co-morbidities, and blood pressure, suggesting little confounding by observed co-morbidities. The inverse association was not specific, as those ever prescribed ACE-Is or ARBs were also less likely to develop VaD and adjustment for confounders strengthened these associations. However, associations of ARBs (but not ACE-Is) with AD were stronger than for VaD and unspecified/other dementia; this finding was supported by our time-lag models, where associations of ACE-Is with AD and VaD and ARBs with VaD, substan- tially reduced with longer lags while ARBs remained inversely associated with AD.
Our findings that ACE-Is are inversely associated with dementia are consistent with some but not all previous studies . Some studies suggest that ACE-Is or specific types of ACE-Is may be positively associated with risk of dementia . Very recently, in a largely male study population from the United States, patients prescribed ARBs were reported to have lower inci-
dence and rate of progression of AD , and stronger associations were found for ARBs than either ACE-Is or other non-ACE-I/ARB anti-hypertensives, which is consistent with our results and other publications [4, 5, 13, 25-29]. Our results are consistent with some randomized controlled trial (RCT) evidence where patients treated with losartan for six months had bet- ter word recall than those prescribed the beta-blocker, atenolol . Another RCT showed that valsartan improved word recall from baseline assessments, while there was no benefit for enalapril-treated patients . That specific angiotensin II targeting drugs demonstrate protective effects on AD risk [13, 29] and general cognition supports the 'angiotensin II' hypothesis of AD  whereby ARBs and ACE-Is are protective by reducing angiotensin II-mediated inhibition of acetylcholine release [30-32]. However, recent publication of secondary outcome measures on cognitive impairment and cognitive decline in the much larger
ONTARGET and TRANSCEND trials provide less convincing evidence . For ONTARGET, there was weak evidence that the ARB telmisartan outperformed the ACE-I ramipril (OR of cognitive impairment 0.90, 95%CI: 0.80-1.01, p = 0.06). However, this was not replicated in the TRANSCEND study, a placebo-controlled RCT of telmisartan in ACE-I intolerant subjects (OR of cognitive impairment 0.97, 95%CI: 0.81-1.17, p = 0.76); a greater benefit might have been expected as the comparator was placebo. An accompanying commentary , highlighted that neither trial differentiated type of pathology; hence specific benefits for AD pathology would be attenuated, especially given that patient populations were selected for
high cardiovascular disease risk which would increase the ratio of vascular to AD pathology. In addition, the mean age of patients in these trials was around 16 years younger than patients in our study, and the trials had a shorter follow-up period. AD pathology increases with age; hence our population should have lower cognitive reserve and greater AD pathology.
Similar arguments apply to meta-analyses which suggest that ACEI/ARB regimes do not reduce dementia risk [35, 36]. Whether these methodological differences are sufficient to explain our discrepant findings and/or whether the results from the observational studies reflect uncontrolled confounding or other types of bias is uncertain (see below).
Our study has several strengths. It is the first large nationally representative study of primary care subjects with full prescribing histories and measurement of co-morbidities such as blood pressure for both males and females. We used patients' long-term prospective prescription histories, defining exposures using 5.92 million eligible prescriptions over 24 years. We included exposure time lags of up to eight years prior to diagnosis. We tested whether the associationsbetween ACE-Is and ARBs were specific to AD, or whether fewer patients exposed to ARBs or ACE-Is also developed VaD. We were able to adjust for blood pressure prior to index date, which may be a strong determinant of prescribing decisions for type of anti-hypertensive. We estimated dose response associations between exposures and outcomes and assessed the specificity of the dose-response association with AD. Validation exercises have shown 80-90% agreement between a diagnosis of dementia in the GPRD database and diagnoses confirmed by correspondence with the GPs  or blinded medical record review .
There are several possible alternative explanations for our observed associations. Chance can be excluded given the precision of the results and the consistency with previous studies . The prospective nature of exposure recording excludes recall bias. Confounding may explain discrepant findings between our observational results and the trial results. We controlled for
several potential confounding factors. Critically, we adjusted for blood pressure history, but this adjustment did not meaningfully change the associations for AD. The sample was restricted to patients treated with anti- hypertensives; hence all patients should have similar indications for anti-hypertensive treatment. Adjusting for IMD (a proxy measure of individual level education and socioeconomic characteristics) and smoking behavior made little difference to the results. However, while GPRD has been shown to be relatively good in the recording of current smoking behavior, past smoking is less well recorded and so our results may be subject to some residual confounding due to smoking .
Our results may suffer from protopathic bias, in which pre-clinical manifestation of disease may determine exposure. We used two strategies to address this. First, examining time lags between exposure and disease assumed that persistent associations with
incrementally larger lags were less likely to reflect pre-morbid disease. Our lagged sensitivity analyses suggested that the inverse association of ARBs with AD was robust, in contrast to the associations with VaD and between ACE-Is and AD. Second, we calculated associations for patients exposed to ARB or ACE-I sole-therapy to avoid any confounding due to
drug tolerance and switching. There was evidence of an inverse association of ARB sole-therapy, but not ACE-I sole-therapy. Prescribing habits may have changed over time. If these changes are related to the likelihood of diagnosis or type of dementia diagnosis then this could bias the results. However, the robustness of the time lag associations between ARBs and AD makes
this less likely.
Similarly, our other sensitivity analyses confirmed the consistency of association for ARBs even when we reclassified exposure based on sole or initial therapy only suggesting that immortal-time bias is unlikely to explain our findings. We are uncertain why associations between ACE-I and AD diminish with latency period though one potential explanation could be different prescribing patterns as regards centrally or peripherally acting ACE-Is. One previous study failed to find association between ACE-Is (as a 'class') with incidence of AD but on sub-group analyses revealed lower risk of cognitive decline for centrally-acting ACE-Is while peripherally-acting ACE-Is were associated with increased risk of dementia and higher levels of disability . We did not conduct a similar post-hoc analysis as we regard the classification of ACE-Is into centrally and peripherally-acting compounds as controversial. This is exemplified by current conflicting evidence around the central action of lisinopril [40-42] and ramipril [40, 43].
It is likely that there will be some under-diagnosis and/or under-recording of AD and VaD in the GPRD . While the validity of a GPRD diagnosis of AD and dementia is relatively good, we are unaware of comparative data for a VaD diagnosis . We believe that these potential misclassifications of the dementia outcomes would be non-differential in relation to the type of antihypertensive prescribed and therefore more likely to attenuate, rather than generate, observed associations.
The observational and biological evidence in favor of ARBs protecting against dementia, presented here and by others , strengthens the need for them to be studied more rigorously in the future. We show that ARBs are associated with a reduced risk of AD which could provide significant patient and socio-economic benefits if these data reflect a true causal effect. Further
evidence from randomized trials with detailed phenotypic assessment of dementia type are required to verify whether these associations are causal or reflect some form of bias. Such studies should examine both secondary prevention of mild cognitive impairment to dementia as well as tertiary prevention for patients with newly diagnosed AD.