Association between timing of statin treatment after diabetes diagnosis and risk of dementia: a nationwide observational study

Summary

Background

The burden of dementia is projected to rise with aging populations, particularly in those with diabetes; however, there are no effective treatments, and attention has turned to prevention through treatment of modifiable risk factors such as low-density lipoprotein cholesterol. We tested the hypothesis that early initiation of statin treatment following type 2 diabetes diagnosis is associated with lower risk of all-cause dementia than delayed or no initiation.

Methods

Using nationwide registers, we identified 132,585 statin-naïve individuals in Denmark developing type 2 diabetes from 2006 through 2019. Follow-up was until dementia, death, emigration, or December 31, 2021. Using a clone-censor weighting design emulating a target trial we compared: i) no statin within 5 years after diagnosis (reference), ii) late statin (start at 1–5 years), and iii) early statin (start at ≤ 1 year). The primary outcome was 10-year risk of all-cause dementia based on Aalen-Johansen survival curves with death treated as competing risk.

Findings

During median 7.1 years of follow-up (interquartile range: 3.8–9.9 years), 3592/132,585 = 2.7% developed dementia. Estimated 10-year absolute risks of dementia from the weighted Aalen-Johansen analysis were 3.9% in those with no statin initiation, 3.3% in those with early statin initiation, and 3.5% in those with late statin initiation. Following diabetes diagnosis and compared to no statin initiation, early statin treatment was associated with 15% lower relative risk of dementia at 10 years (95% confidence interval: 9%–21%) while the corresponding relative risk for late statin treatment was 10% lower (3%–15%). Corresponding absolute risk differences were 0.59% (0.34%–0.86%) and 0.38% (0.12%–0.61%) lower, respectively.

Interpretation

In statin-naïve individuals developing type 2 diabetes, statin therapy was associated with lower relative and absolute risks of dementia, most pronounced in those with early statin initiation. There is a need for trial evidence to confirm the findings.

Funding

The Danish Cardiovascular Academy and the Danish Heart Foundation.

Research in context.

Evidence before this study

We searched PubMed from conception to February 2026 with a general purposed search combining terms “statin”, “dementia” and “diabetes” and a Medical Subject Headings (MeSH) based search combining terms “Hydroxymethylglutaryl-CoA Reductase Inhibitors” and “Dementia”. Searches were not limited to English-language publications in order to capture the full body of literature on statin use and risk of dementia, and inclusion criteria were kept broad accordingly. Other relevant articles referenced to were also assessed.

Based on accumulated evidence, the 2024 Lancet Commission on dementia prevention, intervention, and care put forward elevated levels of low-density lipoprotein cholesterol from midlife as a newly recognized modifiable risk factor to prevent dementia, while retaining diabetes as one of the other 13 modifiable risk factors. Reduced dementia risk associated with statin use has to date only been reported in observational studies. In an updated systematic review and meta-analysis by Westphal Filho et al. of 55 observational studies, statin use at baseline was associated with lower risk of all-cause dementia (hazard ratio 0.86, 95% CI 0.82 to 0.91), Alzheimer’s dementia (0.82, 0.74 to 0.90), and vascular dementia (0.89, 0.77 to 1.02) when compared to non-users. The included studies were assessed at moderate- to low risk of bias in the report. Completed randomized trials exploring the effects of statins on dementia have largely been negative, but are limited in number and duration and thus in statistical power, which may affect results. No trials have been conducted specifically in individuals with type 2 diabetes. We found no studies that examined the importance of starting statin treatment at different stages following a diabetes diagnosis on risk of dementia.

Added value of this study

Using nationwide Danish registers and a clone-censor weighting design to reduce risk of biases, in 132,585 statin-naïve individuals who developed type 2 diabetes, this study found that early statin treatment, within one year of diabetes diagnosis, was associated with the lowest risk of developing all-cause dementia. Late statin treatment, started within 1–5 years of diagnosis, was associated with lower risk of all-cause dementia than not initiating statin treatment, which conferred the highest risk of dementia.

Implications of all the available evidence

There is currently no cure for dementia, so attention has turned to targeting modifiable risk factors in midlife to prevent or delay development of dementia. Our results, added to currently available evidence, highlight a potential for timely intervention targeting low-density lipoprotein cholesterol lowering with statins when a diagnosis of type 2 diabetes is established. Only randomized trials can finally establish if early statin therapy after a diagnosis of diabetes will reduce the risk of dementia.

Introduction

Advances in prevention and treatment of prevalent non-communicable diseases such as ischemic heart disease, stroke, chronic obstructive pulmonary disease, and cancers, and of infectious diseases have improved life expectancy globally and people increasingly reach old age.¹ Consequently, the global prevalence of dementia is increasing, with an estimated 57 million people living with Alzheimer’s or vascular dementia in 2019 and a projected 153 million affected by 2050.² There is currently no cure for dementia, and cognitive decline can unfold over decades prior to receiving a diagnosis, so attention has turned to targeting modifiable risk factors in midlife with the aim of preventing or delaying dementia. The 2024 Lancet Commission on dementia prevention, intervention, and care identified 14 modifiable risk factors, collectively estimated to have the potential to prevent or delay nearly half of all dementia cases.³ In the report, diabetes remained a recognized risk factor, and notably, elevated low-density lipoprotein (LDL) cholesterol was added for the first time. Due to the high prevalence of hypercholesterolemia, the Commission estimated that treatment of elevated LDL cholesterol has potential to prevent the highest number of dementia cases.

With the global increase in obesity, more people are living with type 2 diabetes.⁴ As hypercholesterolemia is more prevalent in those with diabetes compared to those without, these patients will often have multiple risk factors for developing dementia. Thus, early intervention with statins before any symptoms of cognitive decline may be essential to slow or prevent development of dementia. LDL cholesterol lowering through statin treatment is already a central treatment in type 2 diabetes with efficacy primarily established for prevention of cardiovascular diseases⁵; however, the importance of early statin treatment following diabetes diagnosis on the long-term prospect of cognitive decline and dementia is unknown.

Using nationwide data from Danish health registers, we tested the hypothesis that early initiation of statin treatment following diabetes diagnosis (within 1 year) in statin-naïve individuals is associated with lower risk of all-cause dementia than late initiation (within 1–5 years) or no initiation (none within 5 years). For this purpose, we used a clone-censor weighting design emulating a target trial.⁶ This is a new method aimed at reducing time-related biases in observational research (Supplementary Fig. S1).

Methods

Study design

In this nationwide cohort study of individuals with type 2 diabetes, we leveraged several Danish health and demographic registers. Danish residents get a unique civil registration number at birth or immigration that is used for administrative contacts with the Danish healthcare system. The Danish healthcare system provides free healthcare for all Danish residents and through national registers the civil registration number can be linked to provide comprehensive information on general demographics, health care utilization, including registration of diseases and biochemical measurements at hospitals, socioeconomic factors, and vitality, in essence making it possible to follow the entire Danish population as a cohort with virtually complete follow-up.⁷

For outcomes and medical covariates, we extracted diagnosis codes registered by Danish local codes or International Classification of Diseases (ICD) codes in the Danish Patient Register covering hospital admissions (since 1977) and outpatient contacts (since 1995).⁸ Education attainment and income statistics were from the Danish Student Register and the Danish Income Statistics Register.^9,10 The national Danish Registry of Medicinal Products Statistics holds information on all drugs dispensed in Danish pharmacies since 1995 allowing for individual level information on pharmacotherapy use identified by Anatomical Therapeutic Chemical (ATC) codes.¹¹ LDL cholesterol measurements were from the Danish Register of Laboratory Results for Research with routine laboratory measurements from Danish hospital laboratories.¹² Analytes are identified by Nomenclature, Properties, and Units (NPU) codes. Data became available at different timepoints depending on geographical location/region in Denmark and all regions were covered from 2015. Danish hospital laboratories generally adhere to international standards for internal and external quality assurance.¹²

Access to the data registers was granted after institutional review by Statistics Denmark (project 708406). Ethics approval is not required for such anonymous register-based research in Denmark.

Participants

From January 1, 2006 through December 31, 2019 statin-naïve individuals above the age of 40 years living in Denmark with incident type 2 diabetes were followed from the date of diabetes diagnosis. Type 2 diabetes was defined by either first registration of a type 2 diabetes diagnosis code or first prescription dispensed for a glucose-lowering drug. For type 2 diabetes diagnosis after age 40, glucose-lowering drugs, and registered diagnoses have reported high validity with sensitivity and positive predictive values of >90%.¹³ Individuals registered with dementia before diabetes diagnosis (ICD-8 codes: 2900–2901 and ICD-10 codes: F00, F01, F03, G30) and individuals registered with type 1 diabetes (ICD-10 code E10) or gestational diabetes (ICD-10 code O24) were excluded. In 2003 a joint task force including the European Society of Cardiology, the European Association for the Study of Diabetes, and the International Diabetes Federation Europe for the first time specified the same treatment goals for total cholesterol and LDL cholesterol in individuals with diabetes as for individuals with established cardiovascular disease.¹⁴ January 2006 was chosen as the first possible point of entry into our cohort to allow for a period of guideline recognition and uptake. Individuals were followed until the date of dementia, death, censoring (emigration from Denmark), 10 years after diagnosis, or end of follow-up December 31, 2021, whichever came first.

Exposures

Three treatment groups were defined by timing of first statin initiation following type 2 diabetes diagnosis: i) no statin within 5 years after diagnosis (reference), ii) late statin initiation (within 1–5 years), and iii) early statin initiation (≤1 year after diagnosis).

Outcomes

The primary outcome was 10-year risk of all-cause dementia (ICD-10 codes: F00, F01, F03, and G30). Previous validation studies have shown that dementia diagnoses have a positive predictive value of 86% in Danish Registers, meaning that when registered diagnoses are reconfirmed through detailed reading of all patient files 86% are correct.¹⁵ Secondary outcomes were subgroups of dementia by Alzheimer’s dementia (ICD-10 codes F00 and G30) and a group of vascular-related dementias, consisting of vascular (ICD-code F01) or unspecified (ICD-10 code F03) dementia diagnosis; Alzheimer’s disease functioned as an internal “negative control” outcome in our analysis: while genetic inhibition of HMG-CoA reductase (the target of statins) reduces the risk of vascular-related dementias and ischemic heart disease, the corresponding effect on Alzheimer’s disease is negligible.¹⁶

Statistical analysis

A challenge for observational studies assessing the effect of pharmacotherapies, and particularly when evaluating different times of treatment initiation, is aligning new use of a drug (dispensing a prescription) with start of follow-up (diagnosis of diabetes). Defining treatment groups based on whether a prescription was dispensed before or after cohort entry risks introducing immortal-time bias.¹⁷ For example, if individuals are followed from diabetes diagnosis and classified according to whether they fill a statin prescription within one year, those who eventually initiate treatment must survive and remain event-free until they receive the prescription (Supplementary Fig. S1, upper panel). This “immortal” period, during which the outcome cannot occur, can bias estimates in favor of treatment. Clone-censor weighting is a methodological approach designed to minimize immortal-time bias, enabling the emulation of a hypothetical randomized trial using real-world data (Supplementary Fig. S1, lower panel, and Supplementary Table S1)⁶; however, no amount of adjustment in an observational dataset can fully remove inherent biases in these data sources. In brief, each individual was triplicated at the start of follow-up (at diabetes diagnosis), with one clone assigned to each treatment strategy: “no statin”, “late statin”, and “early statin”. During follow-up, clones were censored when they deviated from their assigned strategy. For example, if an individual initiated treatment 3 months after diagnosis, the clones assigned to “no statin” and “late statin” were censored at that time, and only the clone aligned with the actual treatment strategy, “early statin”, remained under follow-up. Because this approach introduces informative censoring driven by treatment initiation that may depend on covariates, inverse probability weighting is used to adjust for covariate imbalances. We fit pooled logistic regression models (biweekly intervals) to estimate each individual’s probability of initiating treatment given covariate history; pooled logistic regression approximates hazard ratios under short intervals and rare outcomes. These predicted initiation probabilities were then used to compute stabilized inverse probability weights to account for informative censoring, which were applied to the clones. Both the numerator and denominator models included follow-up time modelled as a quadratic polynomial (in 14-day intervals since diabetes diagnosis) to flexibly capture the non-linear pattern of statin initiation over time. The denominator model was further adjusted for age, sex (as designated at birth), descent (Danish descent or other descent, meaning individuals who immigrated to Denmark or descendants of these individuals), year of entry into observation, educational attainment (“Basic education”: International Classification for Education (ISCED 0–2),¹⁸ “High school or vocational education” (ISCED 3–4), or “higher education” (ISCED ≥ 5)), income (“Low”: lowest quartile, “Middle”: interquartile range, and “High”: highest quartile), use of drugs including calcium channel blockers, renin-angiotensin system inhibitors, beta blockers, thiazides or thiazide-like drugs, antiplatelets, antidepressives, insulin, glucose-lowering drugs with effect on atherosclerotic cardiovascular diseases (sodium-glucose cotransporter-2 inhibitor or glucagon-like peptide-1 receptor agonist), number of glucose-lowering drugs (below two or two and above), history of hypertension, cardiovascular disease (myocardial infarction, coronary revascularization, stroke, or peripheral arterial disease), or chronic obstructive pulmonary disease, and LDL cholesterol levels. Because LDL cholesterol is a key predictor of statin initiation, we included it as a time-varying variable. It was modeled categorically (LDL cholesterol: <1.4, 1.4–1.8, 1.8–2.6, 2.6–4.9, and >4.9 mmol/L; <55, 55–70, 70–100, 100–190, and >190 mg/dl), as treatment decisions may be non-linearly influenced by guideline-based thresholds.¹⁹ The value reflected the most recent LDL cholesterol measurement (NPU01568, NPU10171, or DNK35308) within the prior 365 days and was updated biweekly when a new measurement became available; otherwise, the most recent value was carried forward. As lipid profiles were not available in all individuals, a missing indicator category was also included in the main model. While multiple imputation is often the gold standard for dealing with missing values in outcome models, in this model predicting treatment initiation with statin, we maintained the “Unknown” category in the main model for LDL cholesterol, as forcing imputed continuous values into the model would assume clinicians based their prescribing decisions on LDL values they did not possess. Medical history variables were similarly time-updated if events occurred after diabetes diagnosis, and use of pharmacotherapies was time-updated when prescriptions were dispensed. Other than missing values for LDL cholesterol as described above, no education was registered in 5137 individuals (4% which was kept as “Unknown”) and no income was registered in 56 individuals (0.04% which was imputed to median). All other covariates were complete as presented in the baseline tables. ATC and ICD codes used to identify covariates are detailed in Supplementary Table S2.

A weight-adjusted Aalen–Johansen estimator was used to calculate cumulative incidence of primary and secondary outcomes, treating deaths as competing events. Relative and absolute risks, also accounting for competing risk of death, are reported at 3, 5, 8, and 10 years. To test the differences over 10 years for the primary outcome, treatment curve differences were compared by calculating a p-value for restricted mean time lost.

In sensitivity analyses, all-cause dementia was analyzed stratified by sex (men and women), age above median age at diagnosis, and descent (Danish descent and other descent). In a further sensitivity analysis multiple imputation of all missing covariates was performed, and the resulting cumulative incidences were averaged across imputed datasets. Also, a sensitivity analysis was conducted with a one-year lag in statin exposure compared to the main analysis, with groups defined as: i) no statin within 6 years after diagnosis (reference), ii) late statin initiation (within 2–6 years), and iii) early statin initiation (≤2 years after diagnosis), to assess the robustness of the findings. Finally, since people with undiagnosed, developing dementia may be less likely to initiate or continue preventive treatments such as statins, we conducted a sensitivity analysis to assess whether reverse causation due to delayed diagnosis was influencing our findings. In this analysis, dementia events occurring within one year of statin initiation were recoded as a competing risk (alongside death) rather than dementia.

Non-parametric bootstrapping for 1000 iterations with the percentile method (0.025 and 0.975 quantiles) was used to estimate 95% confidence intervals (CI) and p-values. As computation times were prohibitively long with multiple imputation and results were similar to the main analyses, bootstrapping was not performed in sensitivity analyses using imputed values.

To illustrate the cumulative incidence of statin initiation in the full cohort and first LDL cholesterol measurements among individuals without measurement before diabetes diagnosis, marginal (unweighted) Aalen–Johansen estimators were used, treating emigration and death as competing risks.

Continuous variables are reported by mean with standard deviation or median with interquartile range, categorical variables by count and percentage, risk differences by percent and 95% CI, and relative risks by ratio and 95% CI. R version 4.4.1 was used for analyses. STROBE guidelines for reporting of observational studies were adhered to in the reporting of this study.

Role of funding source

The organizations funding this study were not involved in designing the study; in the collection, analysis, and interpretation of data; in the writing of the report; or in the decision to submit the paper for publication.

Results

Participant characteristics

From 370,386 individuals who were diagnosed with diabetes from 2006 through 2019, 132,585 individuals met the eligibility criteria (Supplementary Fig. S2). During median 7.1 years (interquartile range: 3.8–9.9 years) of follow-up, 3592 (2.7%) were diagnosed with all-cause dementia, 2532 (1.9%) with vascular-related dementia, and 1571 (1.2%) with Alzheimer’s dementia, while 511 (0.4%) received diagnoses of both vascular-related and Alzheimer’s dementia during follow-up. In individuals who did not start statins within five years after diagnosis (no statin), baseline age was higher at diabetes diagnosis than in individuals with early or late statin treatment (Table 1). In individuals who had LDL cholesterol measured in the year prior to a diabetes diagnosis, levels were lowest in the no statin group, higher in the late statin group, and highest for the early statin group. In individuals with no LDL cholesterol measurements in the year prior to diabetes diagnosis (62%), ∼65% had a measurement done within five years and ∼80% within ten years after diagnosis (Supplementary Fig. S3). Within one year after diabetes diagnosis many initiated statins, while the rate of new users leveled off after five years and women appeared less likely to initiate statins (Supplementary Fig. S4). After applying cloning and weighting covariates were perfectly balanced at baseline and remained balanced across five years (Supplementary Figs. S5–S9).

Table 1.

Baseline characteristics at type 2 diabetes diagnosis.

	No statin within 5 years (N = 56,693)	Statin <1 year (N = 52,496)	Statin 1–5 years (N = 23,396)
Sex
Women	26,523 (46.8%)	21,245 (40.5%)	9910 (42.4%)
Men	30,170 (53.2%)	31,251 (59.5%)	13,486 (57.6%)
Age, years	63.5 (14.0)	59.8 (10.6)	59.5 (10.9)
Descent
Danish	49,997 (88.2%)	45,778 (87.2%)	20,020 (85.6%)
Non-Danish	6696 (11.8%)	6718 (12.8%)	3376 (14.4%)
Year of entry into observation	2010 (4.11)	2010 (3.98)	2010 (4.01)
Education
Basic	21,354 (37.7%)	19,642 (37.4%)	8732 (37.3%)
High school or vocational	23,362 (41.2%)	24,125 (46.0%)	10,738 (45.9%)
Higher	8933 (15.8%)	7276 (13.9%)	3286 (14.0%)
Unknown	3044 (5.4%)	1453 (2.8%)	640 (2.7%)
Income
Low	15,163 (26.7%)	12,081 (23.0%)	5688 (24.3%)
Middle	27,804 (49.0%)	26,852 (51.2%)	11,747 (50.2%)
High	13,726 (24.2%)	13,563 (25.8%)	5961 (25.5%)
Medical history
Cardiovascular disease	4159 (7.3%)	3828 (7.3%)	970 (4.1%)
Chronic obstructive pulmonary disease	4763 (8.4%)	2520 (4.8%)	1176 (5.0%)
Hypertension	11,520 (20.3%)	8500 (16.2%)	3995 (17.1%)
Pharmacotherapy
Beta-blockers	9755 (17.2%)	7340 (14.0%)	3666 (15.7%)
Thiazides, thiazide-like drugs	15,494 (27.3%)	13,912 (26.5%)	6573 (28.1%)
Renin-angiotensin system inhibitors	13,534 (23.9%)	14,417 (27.5%)	5993 (25.6%)
Calcium channel blockers	9866 (17.4%)	8562 (16.3%)	4178 (17.9%)
Antiplatelets	7611 (13.4%)	6690 (12.7%)	2416 (10.3%)
Antidepressants	5553 (9.8%)	4017 (7.7%)	1929 (8.2%)
LDL cholesterol (mmol/L)a
Mean	2.8 (0.83)	3.6 (0.87)	3.2 (0.81)
<1.4	742 (1.3%)	102 (0.2%)	75 (0.3%)
1.4–1.8	1331 (2.3%)	188 (0.4%)	163 (0.7%)
1.8–2.6	6550 (11.6%)	1827 (3.5%)	1518 (6.5%)
2.6–4.9	12,025 (21.2%)	16,972 (32.3%)	6819 (29.1%)
>4.9	252 (0.4%)	1502 (2.9%)	235 (1.0%)
Missing at diabetes diagnosisb	35,793 (63.1%)	31,920 (60.8%)	14,586 (62.3%)

Continuous variables are reported by mean (standard deviation) and categorical variables by count (percentage).

LDL: low density lipoprotein cholesterol.

^aIn mg/dl the LDL cholesterol categories correspond to <55, 55–70, 70–100, 100–190, and >190 mg/dl.

^bIn the individuals with missing LDL measurements before diabetes diagnosis, a ∼65% had a measurement done within five years and ∼80% within ten years after diagnosis.

All-cause dementia

Ten-year absolute risks of dementia were 3.9% in those with no statin initiation, 3.3% in those with early statin initiation, and 3.5% in those with late statin initiation (Fig. 1). Following diabetes diagnosis and compared to no statin initiation, early statin treatment was associated with 15% lower relative risk of dementia at 10 years (95% confidence interval: 9%–21%) while the corresponding relative risk for late statin treatment was 10% lower (3%–15%) (Figs. 1 and 2). Corresponding absolute risk differences were 0.59% (0.34%–0.86%) and 0.38% (0.12%–0.61%) lower. At 3, 5, and 8 years, relative and absolute risks were similarly highest with no statin, stepwise lower with late statin, and lowest with early statin (Figs. 1 and 2). For all comparisons of the curves over the 10 years, the p-value was <0.001. If multiple imputation was used for missing variables, results were similar to those reported in the main analysis (Supplementary Fig. S10). When statin exposure was instead defined as i) no statin within 6 years after diagnosis (reference), ii) late statin (within 2–6 years), and iii) early statin (≤2 years after diagnosis), the stepwise decrease in risk from no to late to early statin was consistent with the main analysis (Supplementary Fig. S11). When only including dementia events that occurred one year after statin initiation, results were also similar to the main analysis (Supplementary Fig. S12).

Fig. 1.

Timing of statin treatment following diagnosis of diabetes on risk of all-cause dementia. Cumulative incidence and absolute risk differences for all-cause dementia with early, late and no statin treatment following diagnosis of type 2 diabetes. The cumulative incidence curves are based on a weight-adjusted Aalen-Johansen estimator treating death as competing risk. All covariates as listed in methods were included for adjustments. Abbreviations: CI: Confidence interval.

Fig. 2.

Timing of statin treatment following diagnosis of diabetes on relative risk of all-cause dementia. Relative risk of all-cause dementia with early, late and no statin treatment 3, 5, 8, and 10 years following diagnosis of type 2 diabetes. Risks are based on a weight-adjusted Aalen-Johansen estimator treating death as competing risk. All covariates listed in the methods section were included for adjustments. At baseline (diabetes diagnosis) the individuals were triplicated in the clone-censor weighting design, thus number at risk in each arm was initially 132,585. Abbreviations: RR: Relative Risk, CI: Confidence interval.

Timing of statin treatment according to sex, age, and descent and all-cause dementia

In both women and men, early statin initiation was associated with the lowest risk at all time points, as in the main analysis (Fig. 3). Restricting analyses to individuals above the median age of 60 years to define a more relevant at-risk group yielded similar results as the main analysis (Supplementary Fig. S13), as did stratifying analyses by Danish and other descent (Supplementary Fig. S14).

Fig. 3.

Timing of statin treatment following diagnosis of diabetes on risk of all-cause dementia in women and men. Cumulative incidence and absolute risk differences for all-cause dementia with early, late, and no statin treatment following diagnosis of type 2 diabetes stratified by women, upper panel, and men, lower panel. The cumulative incidence curves are based on a weight-adjusted Aalen-Johansen estimator treating death as competing risk. All covariates listed in the methods section were included for adjustments. Abbreviations: CI: Confidence interval.

Vascular-related and Alzheimer’s dementia

For vascular-related dementia, risk was lower for early and late statin initiation compared to no initiation, and lower with early versus late statin initiation (Fig. 4, upper panel). Ten-year absolute risks of vascular-related dementia were 2.9% in those with no statin initiation, 2.3% in those with early statin initiation, and 2.4% in those with late statin initiation. This stepwise decrease in risk from no to late to early statin initiation for vascular-related dementia was also reflected in absolute risk differences at 3, 5, and 8 years (Fig. 4, upper panel). However, no clear pattern was seen for Alzheimer’s dementia (Fig. 4, lower panel), which could be considered as a “negative control”. Ten-year absolute risks of Alzheimer’s dementia were 1.7% in those with no statin initiation, 1.6% in those with early statin initiation, and 1.5% in those with late statin initiation (Fig. 4, lower panel).

Fig. 4.

Timing of statin treatment following diagnosis of diabetes on risk of vascular-related dementia and Alzheimer’s dementia. Cumulative incidence and absolute risk differences for vascular-related dementia, upper panel, and Alzheimer’s dementia, lower panel, with early, late, and no statin treatment following type 2 diabetes diagnosis. The cumulative incidence curves are based on a weight-adjusted Aalen-Johansen estimator treating death as competing risk. All covariates listed in the methods section were included for adjustments. Abbreviations: CI: Confidence interval.

Discussion

In a nationwide study in Denmark of 132,585 statin-naïve individuals developing type 2 diabetes, statin therapy was associated with lower relative and absolute risks of dementia, most pronounced in those with early statin initiation. These findings are novel.

Mechanistically, cumulative exposure to elevated LDL cholesterol throughout life is considered a risk factor for cognitive decline and dementia.³ In a meta-analysis of cohort studies, LDL cholesterol per 1 mmol/L increase in midlife was associated with 8% increased risk of developing all-cause dementia.²⁰ Statins are effective interventions to lower LDL cholesterol and protect against atherosclerotic cardiovascular disease,^21,22 which in turn may protect against cerebrovascular atherosclerosis, ischemic stroke, hypoperfusion, and tissue injury, offering a plausible mechanistic link to explain the benefits associated with earlier statin initiation. Other actions of statins, like lowering of triglyceride-rich remnant cholesterol particles, reduced inflammation, and local brain specific actions, including protection against protein dysfunction, altered neurotransmission capacity, and decreased neurodegeneration,23, 24, 25, 26 may also contribute to lower risk of dementia. Statins have also been suggested to reduce accumulation of Amyloid-β species,²⁴ possibly explaining findings of lower risk of Alzheimer’s dementia associated with use of statins²⁷; which however, was not documented in the present study.

Our report appears to be the first to evaluate the importance of timely intervention with statins on risk of dementia following a diabetes diagnosis using a trial emulation design on nationwide data. In support of our findings, a 2025 meta-analysis of 55 observational studies by Westphal Filo et al. indicated that statin use at baseline was associated with lower risks of dementia compared to non-use, including among individuals with diabetes.²⁷ Studies with longer exposure to statin at baseline were progressively associated with more protective estimates for risk of dementia. This meta-analysis was consistent with a previous meta-analysis from 2022 by Olmastroni et al. of observational studies, similarly indicating beneficial association of statin use with risk of dementia, including Alzheimer’s dementia.²⁸ Though, a call for randomized intervention trials have been part of both meta-analyses.

Clinical trials have not documented that use of statins prevent cognitive decline and dementia.^29,30 The Heart Protection study³⁰ found no difference in cognitive impairment assessed as a tertiary outcome by telephone questionnaire at the final visit of 20,536 high-risk individuals randomized to simvastatin or placebo with mean 5 years of follow-up; the endpoint of this study can be considered as a soft endpoint relative to the hospital registered diagnoses of dementia used in our study. The Prospective Study of Pravastatin in the Elderly at Risk (PROSPER) trial²⁹ had cognitive decline as a major prespecified endpoint, but found no differences in cognitive tests conducted at six different time points of mean 42 months of follow-up for 5804 individuals randomized to Pravastatin or placebo. Comparatively short follow-ups, limited statistical power, selection of individuals with advanced age, already established cardiovascular disease or accumulated risk-factors for cardiovascular disease, and for the PROSPER trial the inability of hydrophilic pravastatin to cross the blood-brain barrier,²⁴ may explain some of the differences from our results, which are based on an observational design emulating a trial with longer follow-up and substantially more statistical power. One record-linkage extended follow-up study of the PROSPER and West of Scotland Coronary Prevention Study (WOSCOPS) randomized clinical trials found no association between historical statin versus placebo allocation and subsequent admission to long-term care, although dementia was not assessed as an outcome.³¹ The newest meta-analysis by the Cholesterol Treatment Trialists’ Collaboration, collating results on adverse events from placebo-controlled randomized clinical trials, reported no statistically significant difference in the development of dementia including Alzheimer’s disease with statin therapy (rate ratio 0.93, 95% CI: 0.77–1.11).³² A more heterogeneous population with only 18% with diabetes at baseline, and very low dementia event rates of <0.1% per annum across trials with 4.5 years of follow-up, may explain the wide confidence interval. No trials have studied risk of dementia with statin use specifically in individuals with type 2 diabetes, which itself is a risk factor. Ongoing randomized trials including the pragmatic evaluation of events and benefits of lipid-lowering in older adults (PREVENTABLE)³³ and statins for extension of disability-free survival and primary prevention of cardiovascular events among older people (STAREE)³⁴ are aimed at estimating the effect of starting statin treatment in people aged > 70 years without dementia and without established cardiometabolic diseases. They may provide important evidence to the efficacy of statins on dementia in otherwise healthy people in old age. However, these trials are not aimed at quantifying the potential of timely intervention in high-risk individuals, such as those with diabetes, in the early stages of the disease to prevent long-term dementia.

The natural history of dementia typically evolves over 10–20 years and is longer than the median 7 years of follow-up in which we observed statin treatment to be associated with reduced risk of dementia.³ However, this aligns with the fact that statin therapy reduces the risk of atherosclerotic cardiovascular disease, in trials of typically 5 years of statin treatment, even if atherosclerosis leading to atherosclerotic cardiovascular disease develops over decades.³⁵ It is also worthwhile noting that the present study included only individuals with diabetes often with faster development of both dementia and atherosclerotic cardiovascular disease than those without diabetes. That said, as it is acknowledged that the latent period of dementia can be long, there is a need for randomized clinical trial evidence preferably with longer than ten years of follow-up, to evaluate whether the benefits observed in our study, found in individuals with a mean age of approximately 60 years with diabetes, are robust with respect to timing.

Strengths of our study include the use of comprehensive registers with nationwide non-selective coverage and individual level data on general demographic, socioeconomic, health, laboratory, and pharmacotherapy information, and use of a design that is robust to immortal-time bias, which is challenging to address with traditional observational methods. Another strength is our use of both time static and time-varying covariates for confounder control. Finally, long follow-up with essentially no losses to follow-up is a strength when studying the consequences of early intervention aimed at preventing dementia at advanced age.

There are naturally also potential limitations to the study. First, observational data offer challenges not faced in clinical trials and while the Danish health registers are comprehensive, well-studied, and internationally recognized for their quality, the data collection is not specifically purposed towards any study protocol. Thus, direct information on potential confounders such as blood pressure and smoking are not available. In their place, antihypertensive pharmacotherapy and history of hypertension or chronic obstructive pulmonary disease (to capture extensive smoking) were included in the models. Second, dementia outcomes were based solely on diagnostic ICD codes and, considering that some individuals had registered codes within both Alzheimer’s and vascular-related dementia, there may be crossover, rendering the interpretation of subgroup outcomes less robust. Third, a challenge with dementia is that it is often not diagnosed at hospitals but is classified by general practitioners as age-related cognitive decline. Thus, the registry-based approach only sees the tip of the iceberg. Fourth, dementia has a long pre-clinical phase, during which individuals may have reduced healthcare engagement and are less likely to initiate or continue preventive therapies such as statins. This could lead to lower observed statin exposure among those already on a trajectory towards dementia and bias associations. Fifth, it is worth pointing out that dementia rates in our cohort were relatively low at 0.5% per year, which makes it difficult to draw definitive conclusions. Sixth, statin use was not modelled as time-varying, meaning that important assumptions are made regarding assumed continued exposure despite potential discontinuation or deprescribing, particularly in the context of pre-clinical dementia; however, time-varying statin use is challenging to model in the used design. Ultimately, there is a question of residual bias due to confounders (both measured and unmeasured), which cannot be ruled out in observational studies, since they lack a randomizing procedure. Hence, even though results were robust across multiple analyses, we cannot completely exclude that the apparent differences observed in the present study could be partly explained by confounding due to e.g., unmeasured confounders. Only randomized trials can control for such confounding.

The main limitation of this study (and most other pharmaco-epidemiological studies) is confounding by indication. It is important to emphasize that a clone-censor weighting design emulating a target trial does not remove confounding by indication; thus, the benefit of statins may in part be explained by confounding factors. For example, the baseline characteristics in Table 1 show that “No statin within 5 years” was associated with four years older mean age, low income, and lower LDL cholesterol; factors that likely influenced the indication statin use. It is reassuring though, that the “negative control” of Alzheimer’s dementia was not associated with the same lower risk of disease as vascular-related dementia, which is in line with studies on genetic inhibition of the target for statins (HMG-CoA reductase).¹⁶

Clinically, our findings highlight a potential for prevention of delay of dementia through timely intervention targeting high LDL cholesterol when a diagnosis of type 2 diabetes is established. Importantly, neuroprotective effects of statins on dementia risk have to date only been documented in observational studies.^27,28 We should be cautious about accepting observational evidence alone, without confirmation in randomized clinical trials with long follow-up, to guide clinical practice, as doing so risks lowering the threshold of proof required for a hypothesis that is plausible. Trials aimed at quantifying the potential of timely intervention in the early stages of diabetes to prevent or delay dementia are, however, unlikely to be conducted, as it would be unethical to withhold statin therapy in those with diabetes due to the high cardiovascular risk. Nevertheless, similar randomized studies of statin therapy in individuals without diabetes with long-term follow-up should be conducted to answer this question reliably. More broadly, a meta-analysis of long-term follow-up of randomized trials of statin therapy with linkage to primary care records for 10–20 years after cessation of trial treatment could address the hypothesis perhaps more reliably than the present study. Death registry and hospitalization records alone may not be sufficient, while access to primary care records with linkage to statin allocation in the trials perhaps could be able to address this question even better.

Conclusion

In statin-naïve individuals developing type 2 diabetes, statin therapy was associated with lower relative and absolute risks of dementia, most pronounced in those with early statin initiation. There is a need for trial evidence to confirm the findings.

The recent diabetes guidelines recommend the use of SCORE2-Diabetes to guide prophylactic initiatives.⁵ Statins have been underused for many years, and the implementation of the SCORE2-Diabetes risk stratification will classify most patients with diabetes as high- or very high-risk patients. The current study may help to emphasize the importance of statin use in this high-risk group and reduce the current treatment inertia.

Contributors

All authors were involved in the conceptualization of the study and interpretation of the results. TT and SA conducted data preparation and statistical analyses. TT wrote the first draft of the manuscript. SA supervised the study. MØJ, BGN, and SA evaluated and edited the final manuscript. All authors had access to the data and agreed to submit for publication. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.

Data sharing statement

Data from the national registers are available on secure servers managed by third-party Statistics Denmark (https://www.dst.dk/) and cannot be made publicly available due to privacy laws. Authorizations with access through secure connections are granted to researchers affiliated to publicly funded Danish research and analysis institutions conditional on having an application reviewed and accepted by Statistics Denmark.

Declaration of interests

TT and MØJ report no conflicts of interest. SA reports support for meeting attendance from Abbott. BGN reports receiving grants for other research projects from AstraZeneca and Merck, and consulting fees used for research from Amarin, Amgen, Arrowhead, AstraZeneca, Esperion, Lilly, Marea Therapeutics, Merck, Novartis, Novo Nordisk, and Sanofi. BGN further reports speaker honoraria from Novartis, Amgen, and Sanofi, participation on advisory boards on two randomized trials (Horizon Lp(a) and CORE), and society leadership role as president of the European Atherosclerosis Society.