Dementia is one of the major public health threats that individuals face as they age. Epidemiological studies suggest that cardiovascular disease, hypercholesterolemia, hypertension and diabetes are important risk factors for the development of dementia [1–4]. Some studies suggest that medications used to treat these risk factors are also associated with a reduced incidence of dementia [5–10]. Statins are a class of lipid-lowering agents that act by inhibiting the second enzyme in the cholesterol synthetic cascade, 3-hydroxy-3-methylglutaryl-coenzyme A (HMG-CoA) reductase. Many studies have investigated whether statin treatment might be of benefit for subjects with dementia or at risk for dementia, but the conclusion remains ambiguous because of conflicting results [9–15].

Initial studies by our group and that of Jick et al suggested that statins might be beneficial as a therapy for dementia [9, 10]. Subsequent studies using a cross-sectional analytic method consistently observed a reduced prevalence and incidence of dementia among statin users (reviewed by Sjogren et al[16]). However, the cross-sectional nature of these studies raises concerns based on the potential for unanticipated biases including confounding by indication. These concerns have prompted investigators to pursue other strategies to test whether statins might protect against dementia.

Two-wave epidemiological studies examined the effects of statins on incident Alzheimer's disease (AD), but failed to show a statistical benefit associated with statin use [11, 17]. The number of subjects on statins who developed incident AD in these two-wave studies was only in the single digits, however, which raises concerns about whether they had sufficient power to detect any putative statin effects.

Prospective incidence trials have also been examined. Cognitive arms were added onto ongoing cardiology studies of which the original goal was to test the efficacy of statins in preventing cardiovascular incidents [18, 19]. These studies also failed to show a reduction in incident dementia associated with statin use, but concerns can be raised about the use of an add-on design and whether the studies were sufficiently powered to detect small effects of statins on incident dementia.

Two studies were also performed to investigate whether statins might delay the progression of cognitive decline in subjects with mild to moderate AD [12, 14]. Both of these studies were quite promising because they showed reduced progression of measures of cognitive function, but they were limited by small cohort sizes.

The cumulative review of these various studies leads to a mixed picture, with multiple studies both suggesting and refuting that statins might reduce the incidence or progression of AD.

Powering studies sufficiently to detect modulation of incident dementia by medication presents a significant challenge for investigators in the field because only a small fraction of subjects in any database are using any particular medication, and only 1–2% of those subjects develop dementia over the course of short-term follow-up. Large population databases provide a useful mechanism to address this problem. For instance, the Decision Support System (DSS) database of the United States Veterans Affairs (VA) medical system is a population database that obtains records from medical centers throughout the USA, which contains diagnostic, pharmaceutical and demographic information on approximately 4.5 million subjects [20]. In this study, we used existing information within this database to obtain prospective data that allowed us to test the hypothesis that use of statins is associated with a reduced incidence of dementia. We found that simvastatin is associated with a significant reduction in incident dementia and atorvastatin is associated with a more modest reduction that is of borderline significance.

We previously reported that statins are associated with a reduced prevalence of dementia. Subsequent studies on this subject produced mixed results. We hypothesized that the ambiguity might arise from differences in efficacy among different statins, variation in responses in the population and inadequate sample size to power the studies sufficiently to detect such modifying factors. To address these issues, we used a very large population database to perform prospective studies examining the effects of statins on the incidence of dementia. The prospective format used in the current study avoids effects arising from changes in statin use associated with the diagnosis of dementia, and could account any differences observed between our previous study and this one [9]. We found that simvastatin is associated with a reduced incidence of both dementia and PD.

Studies using population databases have strengths and weaknesses that must be considered in interpreting results. The large numbers of subjects provides enormous power for analyses and the structure of the databases allows for prospective studies of incidence. The strong statistical power associated with the large number of subjects also enables detailed subcategorization of the cohorts, which can provide novel insights. For instance, the ability to subdivide statin users by each statin allowed us to examine the effects of three different statins: atorvastatin, lovastatin and simvastatin. The ability to track subjects over a period of time allowed us to examine incident dementia cases, which represents a significant difference from our previous study on statins, which examined prevalent cases [9]. The focus on incident cases in the current study avoided problems associated with some of the biases that can occur in studies of prevalence, and could explain the differences observed between the two studies [9]. In addition, the covariates provide some adjustment for confounding by indication.

The nature of the prospective study, however, is quite different from that in a randomized clinical trial. Subjects on medications in population databases necessarily have comorbid illnesses, whereas random clinical trials can control the type or degree of comorbid illness or even exclude subjects with comorbid illnesses. Assessing the nature of the sample and the comparators correctly becomes a very important issue in studies examining population databases. We used two comparators, a CV comparator and warfarin. We feel that the CV comparator is the more robust, because the strong link between hypercholesterolemia and cardiovascular disease suggests that the statin group should resemble the CV comparator group. Warfarin was chosen as a comparator to provide a specific medication as a reference, although any particular medication will always have particular concerns relevant to that medication. We chose warfarin because it had not been shown to modify the course of dementia. Warfarin is used to treat many types of vascular disease, including deep vein thrombosis, stroke and atrial fibrillation [23]. Despite use of warfarin to treat vascular disease, referencing simvastatin to warfarin produced results that were substantially similar to those observed when referencing simvastatin to the CV comparator.

Lovastatin users also provided a useful reference, although this group was not designed as a reference at the outset of the study. The incidence of dementia among subjects taking lovastatin was similar to that of the CV comparator, and well above that of the simvastatin group. The case mix of the simvastatin and lovastatin groups might resemble each other more than the case mixes for the other comparators, yet simvastatin showed a greater efficacy for reducing dementia, although it is important to be cognizant that simvastatin is a newer medication than lovastatin, which could lead to a selection bias. The strength of reduction in dementia for simvastatin compared with lovastatin is consistent with many studies showing that simvastatin reduces measures of plasma cholesterol more effectively than lovastatin [24, 25]. The presence of a strong and significant difference in HRs for dementia between these two groups (lovastatin and simvastatin) that are congruent for statin use supports the hypothesis that the efficacy of simvastatin arises from the action of simvastatin rather than from a prescription bias.

Analysis of the frequency of hospitalizations and the Charlson Index were used to compare the different groups in this study and used to control for comorbid disease. The CV comparator and each of the statins exhibited similar hospitalization rates and Charlson indices, which suggests that these groups exhibited similar rates of comorbid illnesses. The warfarin group showed elevated Charlson indices and elevated hospitalization rates, which is consistent with a hypothesis that these groups had a higher general rate of morbidity than most of the other groups. The potential differences in comorbid illnesses that might exist among the groups did not exert strong effects on the resulting HRs. When we used models that adjusted for potential interactions between Alzheimer's disease and other chronic illnesses, we did not observe a strong change in the size of the HR describing the relationship between use of simvastatin and incident dementia.

Quantitative indices of subject health, such as laboratory values or imaging studies, are not available in the DSS database, in contrast to databases prepared by academic investigators and in clinical trials [20]. The absence of quantifiable measures means that we were not able to determine the degree of cholesterol reduction associated with each statin, nor were we able to quantify cognition. Although we used the ICD9 diagnostic code for senile dementia of the Alzheimer type, 331.0, misdiagnosis occurs, thus we used the generic term, dementia. Diagnoses of AD in population databases are also not rigorously controlled, and the diagnostic measures often do not meet the NINDS-ARDA criteria for AD. Previous studies indicate that the diagnoses of AD in the VA databases are 70–95% accurate [26]. The presence of ≥70% of cases that probably do have AD suggests that the reduction in incidence of dementia associated with statin use observed in our study also applies to AD, although the exact degree is risk reduction for AD might differ from that observed in our study.

The strength of reduction of incidence of dementia observed with simvastatin is striking. Further studies are required to determine whether this effect represents a biological action of simvastatin or a statistical bias skewing results obtained from the DSS database. If the reduction in incident dementia derives from biological actions of simvastatin, studies in the literature suggest potential biological mechanisms that might contribute to this action. Prior studies indicate that simvastatin is more effective than pravastatin or lovastatin at modifying some measures of lipid metabolism, such as reductions in cholesterol and LDL, and increases in HDL. Simvastatin has a similar efficacy as atorvastatin with respect to reducing measures of lipid metabolism [25, 27]. Simvastatin is better that atorvastatin on some measures (e.g., raising HDL), but atorvastatin is better than simvastatin on other measures (e.g., lowering LDL) [25, 28–30]. The size of the difference in HR that we observed for simvastatin compared with the other statins appears larger than that observed in studies examining vascular lipid parameters. This raises the possibility that the putative benefit of simvastatin arises from another contributing factor. One factor could be the ability to penetrate the blood-brain barrier. The statins differ in their lipophilicity and ability to cross the blood-brain barrier, with the rank order of permeabilities being lovastatin > simvastatin > atorvastatin [31–33]. Simvastatin is atypical, because of its strong efficacy but intermediate permeability. The combination of lipophilicity and efficacy gives simvastatin a unique pharmacological profile compared with the other statins. These factors might lead simvastatin to be more potent than lovastatin. Atorvastatin has strong anti-inflammatory properties, but the inability of atorvastatin to penetrate the blood-brain barrier might reduce its efficacy in preventing neurodegenerative disease [35]. Further studies are needed to clarify this issue.

The ability to examine multiple disorders is a salient strength of population databases such as the DSS database. In this study, we also examined the efficacy of statins towards PD. The association of statins with reduced incidence dementia and PD raises the possibility that the action of statins against these diseases shares a common mechanism. Many investigators have noted that statins reduce inflammation, osteoporosis, and fractures, as well as diseases directly caused by heart disease [40–42]. Inflammation contributes to the pathophysiology of AD and PD. We recently observed that subjects with severe AD pathology who were taking statins show less inflammation than those not taking statins [43]. It is possible that the putative ability of statins to reduce inflammation contributes to the reduction in incidence of degenerative disease associated with simvastatin.

Simvastatin is associated with a significant reduction in incident dementia and incident Parkinson's disease. Atorvastatin is associated with a modest reduction in incident dementia that is of borderline significance when age is included as a covariate, but insignificant when other comorbid diseases are included as covariates. Further studies are required to determine whether this effect represents a biological action of simvastatin or an unanticipated statistical bias present in the DSS database.