Association between statin use and Alzheimer’s disease with dose response relationship

This study aimed to determine the dose–response relationship between the levels of statin exposure and the incidence of Alzheimer’s disease (AD). We included 119,013 Korean adults (≥ 60 years old) using a database from the Korean National Health Insurance Service (2002–2013). Statin exposure was treated as a time-varying variable. Incidence of AD was defined by the first claim code for AD with anti-Alzheimer drugs. AD occurred in 9467 cases during a median 7.2 years of follow-up. Overall, statin use was not associated with an increased risk of AD incidence [adjusted hazard ratio (aHR) = 1.04; 95% confidence interval (CI) = 0.99–1.10]. When examined by level of statin exposure, statin prescription < 540 days during a 2-year window time was associated with a higher risk for incidence of AD compared to statin non-use. However, days of prescription ≥ 540 and cumulative defined daily dose ≥ 540 of statin were associated with decreased risk of AD [aHR (95% CI) = 0.87 (0.80–0.95) and 0.79 (0.68–0.92), respectively]. Our findings indicate that less persistent statin use is associated with increased risk of AD, whereas persistent and adherent statin use is associated with decreased risk of AD.

www.nature.com/scientificreports/ However, in the real world, the status of statin use can change considerably. Many patients discontinue statin use, and there is variability in drug adherence. Conversely, statin use dramatically increased during the last two decades 19 , and many statin non-users turned into statin users, (2) The dose-response effect was not examined mainly because statin use was determined by a questionnaire [12][13][14][15][16] , and the number of study participants was too small (110-582 cases among 8 prospective cohort studies) [12][13][14][15][16][17] , (3) Most of the study durations were too short (mean duration of follow-up ≤ 5 years) 12,13,15,16 , and (4) a Taiwanese study with the largest sample size suffered from lack of potential confounder information, such as baseline cholesterol level 18 . In this study, we aimed to evaluate the association of statin use with incidence of AD considering the time-varying status of statin use along with the dose-response relationship.

Results
Baseline characteristics. The mean age of the study population was 66.0 [standard deviation (SD), 4.9] years, and 48.1% of the subjects were male (Supplementary Table 1  www.nature.com/scientificreports/ association of statin use with risk of AD, and no significant difference by type of statin (e.g., fungus-derived and synthetic) was observed for risk of AD (Supplementary Table 4).

Discussion
We investigated the dose-response relationship between statin use and the incidence of AD while considering time-varying exposure. The strengths of our study include (1) a large population-based representative sample; (2) the use of pharmacy claims data that enabled us to collect accurate information on statin use including dosage and statin intensity; and (3)  This trend is similar in other countries and suggests that previous studies that do not consider statin use as a time-varying covariate suffer from misclassification bias.  Statin use (≥ 540 days or cDDD) was associated with reduced risk of AD, while statin use (< 540 days or cDDD) was associated with increased risk of AD compared to statin non-use. Statin exposure by cDDD and incidence of AD showed an inverse J-shaped association with a significant P for trend among statin users (P-trend < 0.001). www.nature.com/scientificreports/ Our study showed that statin users with poor persistence and poor adherence had an elevated risk for AD compared to statin non-users. While this seems to indicate that statin use can increase AD risk, we should consider indication bias. Statin treatment is indicated in patients with high cholesterol levels or at high risk of developing cardiovascular events. Conversely, statin non-users in our study were not at such risk. As hypercholesterolemia or comorbid cardiovascular risk factors are associated with higher AD risk 22 , an increased risk in the poor persistence and poor adherence group may be a reflection of such risk. In our study, both high and low cholesterol levels were also associated with higher risk of developing AD. Hypercholesterolemia has received attention as a risk factor for cognitive decline or AD 23,24 . High cholesterol levels in serum have been positively correlated with accumulation of amyloid β in the brains of patients with AD 25 .
On the contrary, patients who maintained their statin prescription with good adherence had an even lower risk of developing AD than statin non-users, suggesting a protective effect of statin use. The more evident dose-response relationship by cDDD compared to days of statin prescription suggests that this protective effect is also dependent upon statin dosage and potency. Our finding is consistent with previous studies that showed a stronger protective effect with a longer cumulative duration of statin use and use of a higher potency statin 18,26 .
There are multiple possible explanations for our findings, which may suggest either a causal relationship, confounding effect, or even reverse causal relationship. First, a direct protective effect of statin use on AD can be suggested. As statins are usually prescribed when hypercholesterolemia is detected, statin users with hypercholesterolemia are possibly at a higher risk of AD 27 . Our study findings suggest that persistent use with good compliance can reverse the AD risk, while an elevated risk cannot be reduced by non-persistent use. Our epidemiological finding is also supported by preclinical evidence. Experimental studies have suggested that simvastatin reduced the level of Aβ in the brain of guinea pigs 28 and reduced levels of AD markers in cerebrospinal fluid in humans 29 . A brain autopsy study showed that patients taking statins were less likely to have amyloid deposits, although there was no relationship between statins and amyloid load 30 . Second, there is a possibility that nonpersistent use of and poor adherence to statin are markers of a less healthy lifestyle or lower socioeconomic status, which are also AD risk factors. Heavy alcohol consumption 31 , and lower educational attainment 32 are both reported to be associated with the occurrence of AD. On the contrary, good persistence and adherence to statin therapy could be a marker of a healthier lifestyle and socioeconomic status, rather than suggesting a direct effect of statin use on AD. Third, there is the possibility of an inverse association between statin use and AD. Poor drug adherence might be a sign of the prodromal stage of AD. Executive function and working memory are shown to be important factors for medication adherence 33 . Even very mild cognitive impairment can contribute to inadequate medication adherence 34 . However, the more evident association between cDDD and AD risk is not well addressed by inverse causality, suggesting that the first theory is a more plausible explanation of our findings.
Our study has important clinical and public health implications. Some people hesitate to take statins or discontinue their use due to concerns about long-term adverse effects such as dementia; however, individuals who are at high risk of cardiovascular disease should not be discouraged from taking statins due to concerns about cognitive impairment. In addition, our findings suggest that persistent statin use should be encouraged among patients who have begun taking statins because of the potential for decreased risk of both cardiovascular disease and AD.
There are several limitations that need to be considered when interpreting our study. First, there could be concerns about indication bias as people who are and are not treated with statins might differ in terms of cardiovascular condition, functional status, and health behaviors, which are often not measured. A randomized controlled trial would be the only solution to overcome this bias, but is not feasible due to practical and ethical concerns. In this observational study, we tried to overcome this by investigating dose-response relationship by adjustment of all potential confounders available in our data. In addition, we conducted sensitivity analyses with a subgroup of participants who were indicated for statin treatment (Supplementary Table 5), using propensity score matching or adjustment (Supplementary Table 6), and the results were consistent with the main analyses. Second, as the national data were not intentionally collected solely for this study, we could not obtain information on ADrelated factors, such as education level and APOE genotype. Third, our outcome determination is based on claims data from clinics and hospitals based on a diagnosis of AD and prescription of an antidementia drug. Underuse of antidementia drugs might cause selection bias in the definition of AD incidence. As a periodic screening test for AD was not conducted as in a prospective cohort study, there is the possibility of underdiagnosis, especially in individuals with poor medical care access. However, this suggests that the protective effect of statin use on AD could be even stronger. Fourth, the subtypes of dementia cannot be clearly distinguished using claims data, and accurate determination of subtype is not always performed in a typical primary care practice where brain magnetic resonance imaging is often not available 35 . In addition, AD and vascular dementia frequently co-exist, which accounts for the protective effect of statin therapy on AD 36 . However, after excluding AD cases with vascular dementia, the results were not changed (Supplementary Table 7). Fifth, we did not consider concomitant drugs as potential confounders in our analysis. Currently, several medications, such as metformin 37 , aspirin 38 , benzodiazepines 39 , and zolpidem 40 are suspected to affect cognitive function and dementia incidence. However, the evidence is too conflicting 41,42 to be included as definite confounders. Sixth, reverse causality could exist. Preclinical or early dementia (before it is formally diagnosed) increases the risk of treatment noncompliance. To complement this possibility, we additionally performed 2-year, 4-year, 6-year, and 8-year lag time analysis, excluding dementia cases that occurred in the time lags. The results at 2-year and 4-year lags were consistent with the main findings, although the confidence interval became wider and the significance level was attenuated (Supplementary Table 8). However, this association was not evident with longer lag times (6 or 8 years) probably due to the relatively short follow up in our study. Finally, although factors for adherence include age, socioeconomic status, social support, depression, and co-payments and could confound our results, these factors were not considered fully in our study due to lack of information. Our findings indicate that, although hypercholesterolemia can increase the risk of AD, persistent and adherent use of statins can reduce and even reverse www.nature.com/scientificreports/ the risk. Statin use itself is not associated with AD risk, and persistence of and good adherence to statin therapy should be emphasized in clinical practice and public health education.

Methods
Study setting. The National Health Insurance (NHI) in Korea covers approximately 97% of the Korean population, while the remaining 3% are covered by the Medical Aid program. All medical utilization and prescription information covered by NHI are collected with International Classification of Diseases, 10th revision (ICD-10) diagnosis codes for each outpatient visit or hospitalization, since medical facilities and pharmacies submit medical and pharmacy claims for reimbursement. A biennial national health screening program (NHSP) is provided to all NHI members over 40 years of age, which contains a questionnaire on health behavior (e.g., past medical history, smoking, and alcohol consumption), anthropometric measurements (e.g., weight, height, and blood pressure), and a blood test (e.g., fasting glucose and lipid levels) to screen for cardiovascular risk factors. In addition, NHI has established a national database linked to the results of the NHSP, including information on utilization of medical facilities and a death registry database 43 . This national database is widely used in epidemiologic studies 44 45 . To reflect cumulative exposure to statins, we used two definitions for statin exposure: days of statin prescription and cumulative defined daily dose (cDDD) during each 2-year time window. Days of statin prescription was defined as the total number of days of statin prescription regardless of dosage or intensity. The cDDD was defined as the summation of total DDD and was considered not only the number of prescription days, but also the dosage and therapeutic intensity of the drug.
To examine the dose-response relationship, we categorized days of statin prescription and cDDD into five groups: < 90, 90-180, 180-365, 365-540, and ≥ 540 days and cDDD considering the distribution of statin use. Based on this definition, a larger number of days of statin prescription and cDDD can be regarded as markers of persistent and adherent use of statin therapy (which means 'not discontinuing' and 'not skipping' the medication, respectively).
Outcome: incidence of dementia. The incidence of AD was defined using the prescription records for anti-Alzheimer drugs (donepezil, galantamine, rivastigmine, or memantine) with ICD-10 codes related to AD (F00 and G30) in medical expense claims submitted to the NHI service 44 . In Korea, there is a relatively strict requirement to be reimbursed for prescriptions for anti-Alzheimer drugs if a patient has a Mini Mental State Examination score ≤ 26 and either a Clinical Dementia Rating ≥ 1 or a Global Deterioration Scale score ≥ 3. Physicians who prescribe these anti-Alzheimer drugs need to document the evidence in patients' medical records in accordance with NHI's strict reimbursement policy. Statistical analysis. We used the t-test and chi-square tests for continuous and categorical variables to examine the differences between statin non-users and statin users. Time-dependent Cox proportional hazards regression analyses were performed to evaluate the association between statin use and the incidence of AD. The time-dependent model has strengths over the conventional Cox model in that it is able to reflect the timevarying factors over time rather than as fixed baseline factors 46 . The follow-up time for each subject was divided into a time window of 2 years, and the cumulative use of each medication during the previous time window (i.e., previous 2 years) was calculated as exposure ( Supplementary Fig. 1). This was also used in previous studies examining the effect of statin use on dementia 15,17 or medication adherence on mortality 47 . The statin use, days of statin prescription, and cDDD were modeled as time-dependent variables. We adjusted for age and sex in model 1. In addition, we adjusted for lifestyle variables (BMI, smoking status, and alcohol consumption), socioeconomic factors (income status), and medical information (history of hypertension, diabetes, and serum cholesterol levels) in model 2.
In sensitivity analyses, (1) we restricted inclusion to a cohort for whom statins were indicated (n = 34,526), including patients with diabetes, those who had high total cholesterol ≥ 240 mg/dL, and inpatient or outpatient diagnosis history of cardiovascular events (I21, I22, I60, I61, I62, I63, I64), (2) we performed a propensity score matching (1:3) analysis with adjustment for propensity scores to determine the independence of association