Association between the use of anti-herpetic drugs and subsequent initiation of Alzheimer’s disease drug treatment: Dutch population-based inception cohort study Authors: Ali Alghamdi, Spyros Balafas, Jens H J Bos, Barbara C van Munster, Karim Rafie, Amalia M Dolga, Eelko Hak Published: 2026 ABSTRACT AbstractObjectivesTo examine whether exposure to anti-herpetic drugs (AHDs: acyclovir, valacyclovir, famciclovir) is associated with reduced risk of Alzheimer’s disease (AD) treatment initiation.DesignPopulation-based retrospective matched cohort study.SettingUniversity Groningen community pharmacy database IADB.nl, covering approximately 125 Dutch pharmacies (1994–2024).Participants262 757 adults aged 50–80 years without prior dementia or AD treatment. Exposed individuals with antiherpetic prescriptions (n=23 887) were matched 1:10 to unexposed controls (n=238 870) by age, sex and calendar time.InterventionAHDs: acyclovir, valacyclovir, famciclovir.Main outcome measuresInitiation of AD drug treatment, defined as at least two prescriptions for rivastigmine, donepezil, galantamine or memantine within 1 year. Cox proportional hazards models estimated HRs with 95% CIs, adjusted for comorbidities and medications. Analyses were stratified by period (1994–2018 vs 2019–2024) and drug type.ResultsDuring follow-up, 2495 participants initiated AD treatment. The age of the participants was 65 (SD 9), and 59% were female. Any AHD exposure was associated with 90% reduced hazard of AD treatment (HR 0.09, 95% CI 0.07 to 0.13, p<0.001). Similar association was found in both periods: HR 0.14 (95% CI 0.09 to 0.20) in period one and HR 0.05 (95% CI 0.03 to 0.10) in period 2. All three AHDs were associated with a lower likelihood of future AD drug prescription: valacyclovir HR 0.10, acyclovir HR 0.09, famciclovir HR 0.07. The incidence rate of AD treatment initiation was substantially lower among AHD users compared with unexposed individuals overall (0.69/1000 person-years (py) vs 4.96/1000 py, p<0.001), with this association evident in both period 1 (0.65/1000 py vs 3.74/1000 py, p<0.001) and period 2 (0.81/1000 py vs 8.22/1000 py, p<0.001).ConclusionsAHD exposure was consistently associated with markedly lower risk of AD treatment initiation, with similar findings observed in recent years. These findings support the hypothesis that herpesvirus reactivation may contribute to AD pathogenesis and suggest antiviral therapy could have preventive implications. Confirmation through prospective studies and randomised trials is needed. Strengths and limitations of this study Large, population-based retrospective cohort study spanning 30 years (1994–2024) with comprehensive matching (1:10 ratio based on age, sex and index date) and multivariable adjustment for demographic and clinical covariates, enhancing the validity of causal inference. Large, population-based retrospective cohort study spanning 30 years (1994–2024) with comprehensive matching (1:10 ratio based on age, sex and index date) and multivariable adjustment for demographic and clinical covariates, enhancing the validity of causal inference. Use of a validated community pharmacy prescription database covering approximately 125 Dutch pharmacies with high-quality medication records and no reliance on patient recall, eliminating information bias and enabling robust exposure and outcome assessment. Use of a validated community pharmacy prescription database covering approximately 125 Dutch pharmacies with high-quality medication records and no reliance on patient recall, eliminating information bias and enabling robust exposure and outcome assessment. The primary outcome was initiation of Alzheimer’s disease (AD) medication rather than a clinical diagnosis; while Dutch prescribing guidelines restrict these drugs primarily to AD, misclassification cannot be excluded, particularly since rivastigmine is also prescribed for Parkinson’s disease dementia. The primary outcome was initiation of Alzheimer’s disease (AD) medication rather than a clinical diagnosis; while Dutch prescribing guidelines restrict these drugs primarily to AD, misclassification cannot be excluded, particularly since rivastigmine is also prescribed for Parkinson’s disease dementia. Residual confounding by unmeasured variables—including education, socioeconomic status, lifestyle factors (diet, physical activity, cognitive engagement) and genetic factors such as APOE status—may bias estimates and limit causal inference. Residual confounding by unmeasured variables—including education, socioeconomic status, lifestyle factors (diet, physical activity, cognitive engagement) and genetic factors such as APOE status—may bias estimates and limit causal inference. The database lacks cognitive assessments, standardised dementia screening instruments, neuroimaging and biomarkers (amyloid-β, tau, neuroinflammatory markers), limiting characterisation of disease severity and pathology. The database lacks cognitive assessments, standardised dementia screening instruments, neuroimaging and biomarkers (amyloid-β, tau, neuroinflammatory markers), limiting characterisation of disease severity and pathology. Introduction Alzheimer’s disease (AD) is the most common cause of dementia worldwide, accounting for 60%–70% of cases and affecting an estimated 47 million people in 2015, with projections of over 130 million by 2050.15 In the Netherlands, approximately 290 000 people were living with dementia in 2021, and this number is expected to exceed half a million by 2040.6 Dementia ranks as the seventh leading cause of death globally.7 Current pharmacological treatments for AD, including cholinesterase inhibitors (donepezil, rivastigmine and galantamine) and memantine, have been available for approximately 20–30 years. These agents work through different mechanisms—cholinesterase inhibitors increase acetylcholine availability by inhibiting its enzymatic breakdown, while memantine acts as an N-Methyl-D-Aspartate (NMDA) receptor antagonist. However, despite their established use, these medications remain primarily symptomatic, slowing cognitive decline rather than halting or reversing the underlying pathology. Their clinical benefit is modest, typically delaying cognitive deterioration by 6–12 months, and their effectiveness varies considerably among patients. Furthermore, their therapeutic window is limited, often becoming less effective as the disease progresses to advanced stages.8 9 Given the limited effectiveness of current therapies, attention has shifted to drug repurposing strategies. The Drug Repurposing for Effective Alzheimer’s Medicines (DREAM) study highlighted several candidates, including immune-modulating agents, such as tumour necrosis factor (TNF) inhibitors.10 Because infectious triggers may contribute to AD, particularly herpesvirus reactivation, anti-herpetic drugs (AHDs) represent a biologically plausible repurposing opportunity. Herpesviruses establish lifelong latency and can reactivate in later life, triggering neuroinflammation, amyloid-beta deposition and tau hyperphosphorylation.11 12 Epidemiological studies have reported mixed associations between herpes zoster and dementia: some show an increased risk,13 while others report no significant relationship.1416 More recently, antiviral treatment has been associated with protective effects in Taiwanese,17 Italian18 and Swedish19 cohorts, though with heterogeneity in effect size possibly due to biases or lack of precision. In the Netherlands, herpes zoster vaccination was implemented as a population-wide programme after 2018. A joint memorandum by Zorginstituut Nederland and the Health Council (2016) concluded there was no justification for a national shingles vaccination programme under the Health Insurance Act, and broader introduction occurred only after 2018 for selected high-risk groups.20 The effect of AHDs on AD risk in this context is unknown. We therefore used the widely researched IADB.nl prescription database with a large population sample both before and after implementation of herpes zoster vaccination to examine whether exposure to acute acyclovir, valacyclovir, or famciclovir drug treatment was associated with reduced risks of initiating AD treatment among older adults. Methods Study design and data source We conducted a retrospective population-based inception cohort study using the University of Groningen IADB.nl community pharmacy database, covering the period 1994–2024. The database contains longitudinal dispensing records from about 125 community pharmacies and polyclinics in the northern Netherlands and includes approximately 1.3 million patients. It is broadly representative of the Dutch population, as registration is independent of age, sex or insurance status.21 Each patient has a unique anonymous identifier, allowing follow-up over time. The database records the dispensing date, dosage, quantity, prescriber and Anatomical Therapeutic Chemical (ATC) classification of each prescription. Data on over-the-counter medicines, hospital-dispensed drugs and clinical diagnoses are not included. Participants Eligible participants were adults aged 50–80 years with at least 1 year of registration before and after the index date. We excluded individuals with a prescription for anti-Alzheimer’s drugs (donepezil, rivastigmine, galantamine or memantine; ATC N06DA02, N06DA03, N06DA04, N06DX01) in the year before the index date, as well as those prescribed immunosuppressants or cancer therapies (ATC L01–L04). Participants with follow-up shorter than 1 year were also excluded from both groups. Exposure definition Exposed individuals were defined as those receiving at least one prescription for valacyclovir (J05AB11), acyclovir (J05AB01) or famciclovir (J05AB09).22 The index date was the date of the first prescription. Unexposed individuals were those without any prescription for AHDs during the study period. They were matched to exposed individuals in up to a 1:10 ratio by age, sex and calendar time at index date. Only initial treatment episodes were considered, to avoid misclassification from repeated or long-term antiviral use. Outcome definition In accordance with earlier studies from our group, the primary outcome was initiation of pharmacological treatment for AD, defined as at least two prescriptions within 1 year for donepezil, rivastigmine, galantamine or memantine.9 23 24 The case definition requiring at least two drug dispensings improves validity by excluding temporary or off-label use. In the Netherlands, these drugs are almost exclusively prescribed for AD and not for other indications, ensuring high specificity.25 Covariates We adjusted for demographic (age and sex) and clinical factors (diabetes mellitus, hypertension, hypercholesterolaemia, myocardial infarction, epilepsy, Parkinson’s disease, psychotic diseases) that could confound the association between AHD use and AD risk.26 Comorbidities were measured using prescription data as proxies, defined as at least two dispensings within the year before the index date. Classes considered were antidiabetics (A10), antihypertensives (C02–C09), cholesterol-lowering drugs (C10), anticoagulants (B01), antidepressants (N06A), antiepileptics (N03AE), anti-Parkinson’s drugs (N04), anticholinergics (A03AB, A03AA, A03E) and benzodiazepines (N05BA, N05CD, N05CF).27 28 Statistical analysis Baseline characteristics were compared using χ² tests for categorical variables and t tests for continuous variables. 3-year cumulative incidence and incidence rates of AD were calculated per 1000 person-years. Kaplan-Meier curves were generated to visualise cumulative incidence over 1000 days after index date, with group comparisons assessed by log-rank tests. Cox proportional hazards models estimated HRs and 95% CIs. Multivariable models adjusted for age, sex and covariates that differed significantly at baseline. To explore temporal trends, analyses were stratified by calendar period: 1994–2018 (period 1) and 2019–2024 (period 2). The cut-off year 2019 was chosen to reflect a shift in clinical and prescribing context.29 Statistical significance was assessed using two-sided tests with the significance level being α=0.05. All the analyses were performed using R (V.4.3.2) and software libraries therein. The analytical methods were selected to provide complementary perspectives on the association between AHD exposure and AD treatment initiation. Incidence rates were calculated as a descriptive measure to quantify absolute event occurrence over person-time independently of modelling assumptions, offering an intuitive sense of effect magnitude prior to confounding adjustment. Kaplan-Meier curves were used as a non-parametric tool to visualise temporal separation between exposure groups, with log-rank tests providing a global comparison of survival functions. The primary analysis used Cox proportional hazards regression to estimate adjusted HRs while accounting for measured confounding through multivariable adjustment. Where Schoenfeld residual diagnostics indicated violation of the proportional hazards assumption, time-varying Cox-GAM models were fitted, allowing the exposure effect to vary smoothly over follow-up time. Patient and public involvement This study used routinely collected, anonymised prescription data from the IADB.nl database; no primary data were collected from patients. Patients and the public were therefore not involved in developing the research question, study design, outcome measures or recruitment. The research question was informed by community-level priorities in AD research, as reflected in advocacy agendas such as Alzheimer Nederland and the DREAM initiative. Findings will be disseminated through open-access publication, with key results shared in plain-language formats with relevant patient organisations including Alzheimer Nederland. Results Baseline characteristics The overall cohort comprised 238 870 individuals without AHD treatment and 23 887 with AHD treatment (figure 1). Due to matching by age, sex and calendar time, demographic characteristics were well-balanced between exposed and unexposed groups. Mean age was 65 years (SD 9) in both groups and the proportion of women was 59% in both groups. Study flow chart of cohort selection. Flow of participants through the study, including inclusion criteria, exclusions, matching and final analytic sample. AD, Alzheimer’s disease; AHDs, anti-herpetic drugs. Among AHD users, there were small but statistically significant differences across drug types. Mean age was 65 years (SD 9) for valacyclovir and acyclovir users, and 66 years (SD 9) for famciclovir users. The proportion of women was 58% among valacyclovir users, 62% among acyclovir users and 61% among famciclovir users. Despite demographic matching, several clinical characteristics differed significantly between exposed and unexposed groups. Participants exposed to AHDs had higher prevalence of drug dispensings for high-risk comorbidities at baseline. AHD users more commonly received antidiabetics (12% vs 10%, p<0.001), antihypertensives (43% vs 37%, p<0.001), cholesterol-lowering drugs (26% vs 22%, p<0.001), anticoagulants (24% vs 20% among drug-specific users, p<0.001), antidepressants (10% vs 8.6%, p<0.001), anticholinergics (0.8% vs 0.4%, p<0.001), benzodiazepines (15% vs 11%, p<0.001) and antiepileptics (0.5% vs 0.4%, p=0.02). Anti-Parkinson’s drug use did not differ between groups overall (1.3% in both, p<0.001), though there were differences across specific AHD types (p=0.003) (table 1). Table 1Baseline characteristics of participants according to anti-herpetic drug exposureNo AHD(n=238 870)Any AHD(n=23 887)P valueValacyclovir(n=18 539)Acyclovir(n=4159)Famciclovir(n=1189)P valueAge at index date (mean±SD, years)65 (9)65 (9)ns65 (9)65 (9)66 (9) 0.002 Gender (n, %) Female141 230 (59)14 123 (59)ns10 844 (58)2558 (62)721 (61) 0.003 Drug prescription (n, %) Antidiabetics24 077 (10)2880 (12) <0.001 2279 (12)454 (11)147 (12) <0.001 Anticoagulants46 668 (20)46 668 (20) <0.001 4505 (24)957 (23)293 (25) <0.001 Antihypertensives88 437 (37)10 378 (43) <0.001 8117 (44)1736 (42)525 (44) <0.001 Cholesterol Lowering drugs52 001 (22)6330 (26) <0.001 5029 (27)5029 (27)990 (24) <0.001 Antidepressants20 628 (8.6)2491 (10) <0.001 1957 (11)400 (9.6)134 (11) <0.001 Antiepileptics874 (0.4)111 (0.5) 0.02 85 (0.5)20 (0.5)6 (0.5) <0.001 Anti-Parkinson’s drugs3014 (1.3)3014 (1.3) <0.001 186 (1.0)37 (0.9)14 (1.2) 0.003 Anticholinergics1053 (0.4)202 (0.8) <0.001 148 (0.8)42 (1.0)12 (1.0) <0.001 Benzodiazepines27 038 (11)3521 (15) <0.001 2632 (14)699 (17)190 (16) <0.001 Follow-up in days (median (IQR))946 (242)1064 (121) <0.001 1063 (123)1066 (116)1068 (110) <0.001 Comparison of demographic and clinical characteristics between exposed and unexposed participants.The bold entries indicate p<0.05.AHD, anti-herpetic drug; ns, not significant. Baseline characteristics of participants according to anti-herpetic drug exposure Comparison of demographic and clinical characteristics between exposed and unexposed participants. The bold entries indicate p<0.05. AHD, anti-herpetic drug; ns, not significant. Similar patterns of comorbidity prevalence were observed across the three AHD types, with all showing higher rates of concurrent medication use compared with unexposed individuals. Incidence of AD treatment initiation During the overall 3-year follow-up period, 2455 individuals (1.03%) in the unexposed group and 40 individuals (0.17%) in the AHD-exposed group initiated AD treatment (p<0.001). This corresponded to incidence rates of 4.96 and 0.69 per 1000 person-years, respectively (p<0.001). The relative risk at 3 years comparing any AHD use to no AHD use was 0.16 (95% CI 0.11 to 0.21, p<0.001), indicating an 84% reduction in cumulative incidence among AHD users (table 2). Table 2Incidence of AD treatment initiationNo AHD (n=238 870)Any AHD(n=23 887)P valueValacyclovir(n=18 539)P valueAcyclovir (n=4159)P valueFamciclovir(n=1189)P valueCases of AD within 3 years (n, %)2455 (1.03)40 (0.17) <0.001 33 (0.18) <0.001 6 (0.14) <0.001 1 (0.08) 0.002 Incidence rate AD (per 1000 py)4.960.69 <0.001 0.75 <0.001 0.55 <0.001 0.31 <0.001 Relative risk (95% CI)0.16 (0.11 to 0.21) <0.001 0.17 (0.12 to 0.24) <0.001 0.13 (0.06 to 0.28) <0.001 0.07 (0.01 to 0.51) 0.009 Incidence of AD treatment initiation among exposed and unexposed participants in overall, including absolute numbers, percentages and incidence rates.P values from χ² or log-rank tests.The bold entries indicate p<0.05.AD, Alzheimer’s disease; AHD, anti-herpetic drug; py, person years. Incidence of AD treatment initiation Incidence of AD treatment initiation among exposed and unexposed participants in overall, including absolute numbers, percentages and incidence rates. P values from χ² or log-rank tests. The bold entries indicate p<0.05. AD, Alzheimer’s disease; AHD, anti-herpetic drug; py, person years. Period 1 (1994–2018) In the earlier period, 1341 unexposed individuals (0.85%) and 28 AHD users (0.18%) initiated AD treatment within 3 years (p<0.001). Incidence rates were 3.74 per 1000 person-years for unexposed versus 0.65 per 1000 person-years for AHD users (p<0.001), corresponding to a relative risk of 0.21 for any AHD use. Drug-specific analyses showed consistent association across all agents. Valacyclovir users had 21 cases (0.19%) with an incidence rate of 0.68 per 1000 person-years (p<0.001) and a relative risk of 0.22. Acyclovir users had 6 cases (0.18%) with an incidence rate of 0.64 per 1000 person-years (p<0.001) and a relative risk of 0.21. Famciclovir users had only 1 case (0.10%) with an incidence rate of 0.35 per 1000 person-years (p=0.004) and a relative risk of 0.12 (online supplemental table S1). Period 2 (2019–2024) In the more recent period, the incidence of AD treatment initiation was higher in the unexposed group: 1114 cases (1.36%) compared with 12 cases (0.15%) among AHD users (p<0.001). Incidence rates were substantially higher than in period 1: 8.22 per 1000 person-years for unexposed versus 0.81 per 1000 person-years for AHD users (p<0.001). The relative risk for any AHD use was 0.11, representing an 89% reduction in AD treatment initiation. All 12 cases among AHD users occurred in the valacyclovir group (0.17%, incidence rate 0.92 per 1000 person-years, p<0.001), with a relative risk of 0.12. No AD cases occurred among acyclovir users (0 of 801, p=0.001) or famciclovir users (0 of 159, p=0.156) during this period, precluding calculation of relative risks for these agents (online supplemental table S2). Overall drug-specific three-year relative risks Across the entire study period, drug-specific relative risks demonstrated strong and consistent association. Valacyclovir showed a relative risk of 0.17 (95% CI 0.12 to 0.24, p<0.001), representing an 83% reduction in AD treatment initiation. Acyclovir demonstrated a relative risk of 0.13 (95% CI 0.06 to 0.28, p<0.001), indicating an 87% reduction. Famciclovir showed the strongest statistical significance with a relative risk of 0.07 (95% CI 0.01 to 0.51, p=0.009), representing a 93% reduction. The protective associations were consistent across both calendar periods, with notably stronger effects observed in more recent years (table 2). Kaplan-Meier analyses Kaplan-Meier survival curves demonstrated consistently lower cumulative incidence of AD treatment initiation among AHD users compared with unexposed individuals throughout follow-up. The curves diverged early in the observation period and the separation persisted, indicating a sustained association throughout the 3-year follow-up window. In period 1 (1994–2018), the cumulative incidence curves for any AHD use versus no use showed clear early divergence that remained stable over time. By 3 years, the unexposed group reached a cumulative incidence of approximately 0.9%, while AHD users remained below 0.2%. When stratified by drug type, valacyclovir and acyclovir users demonstrated similar protective patterns with consistently flat survival curves, while famciclovir showed minimal events, resulting in a nearly horizontal curve (figure 2). Kaplan-Meier curves for cumulative incidence of Alzheimer’s disease (AD) treatment initiation among anti-herpetic drug (AHD) users and unexposed individuals, Period 1 (1994–2018). Cumulative incidence of AD treatment initiation was estimated using the Kaplan-Meier method. Blue line: AHD users; red line: unexposed individuals. Shaded areas represent 95% CIs. Numbers at risk are displayed below the x-axis at each time interval. In period 2 (2019–2024), the divergence between exposed and unexposed groups was even more pronounced. The unexposed group showed a steeper rise in cumulative incidence, reaching approximately 2.0% by 3 years, while the AHD-exposed group remained below 0.3%. The binary exposure curves demonstrated a marked association, with the AHD group maintaining a consistently low and relatively flat trajectory throughout follow-up. When examined by individual drug type, valacyclovir users showed a modest gradual increase in cumulative incidence reaching approximately 0.3% by 3 years. Notably, both acyclovir and famciclovir users showed completely flat survival curves with no observed AD treatment initiations during the entire follow-up period, resulting in cumulative incidences remaining at 0% throughout the 3-year observation window (figure 3). Kaplan-Meier curves for cumulative incidence of Alzheimer’s disease (AD) treatment initiation among AHD users and unexposed individuals, period 2 (2019–2024). Cumulative incidence of AD treatment initiation was estimated using the Kaplan-Meier method. Blue line: AHD users; red line: unexposed individuals. Shaded areas represent 95% CIs. Numbers at risk are displayed below the x-axis at each time interval. AHD, anti-herpetic drug. The consistent separation of curves across both time periods and all drug types, combined with highly significant log-rank test results (p<0.001 for all comparisons except famciclovir in period 2), provide strong evidence for a protective association between AHD use and AD treatment initiation (online supplemental figures S1 and S2). Cox proportional hazards models In multivariable Cox regression analyses adjusted for age, sex, and baseline comorbidities, exposure to AHDs was associated with a substantially reduced hazard of AD treatment initiation. In the overall cohort, any AHD use was associated with a 90% reduction in AD treatment initiation compared with no use (HR 0.09, 95% CI 0.07 to 0.13, p<0.001) (table 3). Table 3Overall HRs for Alzheimer’s disease treatment initiationHR95% CIP valueAny AHD0.090.07 to 0.13 <0.001 Valacyclovir0.100.07 to 0.14 <0.001 Acyclovir0.090.04 to 0.21 <0.001 Famciclovir0.070.01 to 0.52 <0.01 Anti-diabetics0.970.85 to 1.110.734Anti-hypertensives0.980.88 to 1.080.697Cholesterol lowering1.080.96 to 1.200.179Anti-coagulants1.110.99 to 1.240.061Anti-depressants1.341.18 to 1.52 <0.001 Anti-epileptics0.720.54 to 0.95 <0.05 Anti-Parkinson’s1.841.65 to 2.05 <0.001 Anti-cholinergic1.110.69 to 1.790.651Benzodiazepines0.970.84 to 1.110.685Overall HRs for Alzheimer’s disease treatment initiation associated with AHD exposure. Multivariable Cox regression results showing adjusted HRs across the entire study period, adjusted for age, sex and relevant comorbidities.The bold entries indicate p<0.05.AHD, anti-herpetic drug. Overall HRs for Alzheimer’s disease treatment initiation Overall HRs for Alzheimer’s disease treatment initiation associated with AHD exposure. Multivariable Cox regression results showing adjusted HRs across the entire study period, adjusted for age, sex and relevant comorbidities. The bold entries indicate p<0.05. AHD, anti-herpetic drug. When stratified by calendar period, protective associations were observed in both periods. In period 1 (1994–2018), AHD exposure was associated with an 86% reduced hazard (HR 0.14, 95% CI 0.09 to 0.20, p<0.001) (online supplemental table S3). In period 2 (2019–2024), the association was even stronger, with a 94% reduction (HR 0.05, 95% CI 0.03 to 0.10, p<0.001) (online supplemental figure S4). Drug-specific analyses in the overall cohort showed consistent relation across agents. Valacyclovir demonstrated a 90% risk reduction (HR 0.10, 95% CI 0.07 to 0.14, p<0.001), acyclovir showed a 91% reduction (HR 0.09, 95% CI 0.04 to 0.21, p<0.001) and famciclovir showed a 93% reduction (HR 0.07, 95% CI 0.01 to 0.52, p<0.01), though with wide CIs reflecting small event numbers (table 3). When examined by period, valacyclovir showed protective associations in period 1 (HR 0.32, 95% CI 0.11 to 0.91, p<0.05), with a non-significant trend in period 2 (HR 0.21, 95% CI 0.02 to 2.23, p=0.200). Acyclovir showed a non-significant trend in period 1 (HR 0.35, 95% CI 0.09 to 1.23, p=0.104) (online supplemental table S3), while in period 2, both acyclovir and famciclovir had extremely low HRs with wide CIs precluding precise estimation (online supplemental table S4). Several covariates were independently associated with AD risk. Anti-Parkinson’s medication use showed the strongest associations across analyses (HR 1.84, 95% CI 1.65 to 2.05, p<0.001 overall; HR 2.26, 95% CI 1.94 to 2.64, p<0.001 in period 1; HR 1.50, 95% CI 1.28 to 1.76, p<0.001 in period 2). Antidepressant use was consistently associated with increased risk (HR 1.34, 95% CI 1.18 to 1.52, p<0.001 overall; HR 1.60, 95% CI 1.35 to 1.89, p<0.001 in period 1; HR 1.1, 95% CI 0.90 to 1.33, p=0.338 in period 2). Antiepileptic use was associated with reduced AD risk overall (HR 0.72, 95% CI 0.54 to 0.95, p<0.05). Other medications including antidiabetics, antihypertensives, cholesterol-lowering drugs, anticoagulants, anticholinergics and benzodiazepines showed no significant associations with AD risk across periods (table 3 and online supplemental tables S3 and S4). Time-varying HRs Time-varying HR analyses demonstrated that the association was strongest in the early follow-up period, with HRs decreasing sharply within the first year and then stabilising at low levels throughout the remainder of follow-up. In the overall cohort, the HR declined rapidly in the earliest follow-up period after 1 year, remaining stable thereafter (online supplemental figure S3A). When stratified by calendar period, the time-varying patterns differed substantially. In period 1 (1994–2018), the HR showed an initial steep decline followed by modest fluctuations between 1 and 2 years, eventually stabilising near 0.1 by 3 years. In period 2 (2019–2024), the association emerged more rapidly and remained consistently stronger throughout follow-up, with HRs declining steeply within the first few months and remaining close to 0.1 throughout the observation period (online supplemental figure S3B). Absolute risk reduction The absolute risk reduction (ARR) and number needed to prevent (NNP) provide clinically meaningful estimates of the magnitude of association between AHD use and reduced AD risk. Overall, across the 3-year follow-up period, any AHD use resulted in an ARR of 0.86%, corresponding to an NNP of 116, meaning that 116 individuals would need to receive AHD treatment to prevent one case of AD treatment initiation. Period 1 (1994–2018) In this period, any AHD use was associated with an ARR of 0.67% and an NNP of 149. Drug-specific analyses showed similar relationships: valacyclovir had an ARR of 0.66% (NNP=152), acyclovir had an ARR of 0.67% (NNP=149) and famciclovir demonstrated the largest ARR of 0.75% (NNP=133). Period 2 (2019–2024) In the more recent period, the ARRs were substantially larger than in period 1. Any AHD use resulted in an ARR of 1.21% with an NNP of 83, indicating that fewer individuals needed to receive treatment to prevent one AD case compared with the earlier period. Valacyclovir showed an ARR of 1.19% (NNP=84). Both acyclovir and famciclovir, with zero observed events, demonstrated ARRs of 1.36% (NNP=74), representing the strongest connection in this period. The lower NNP values in period 2 compared with period 1 reflect a larger absolute risk difference in more recent years, likely attributable to both higher baseline AD incidence rates in the unexposed population and a stronger observed association with AHD treatment. Discussion In this large, population-based cohort study, use of AHDs was associated with a substantially reduced risk of initiating treatment for AD. The association was consistent across two calendar periods (Period 1: 1994–2018; Period 2: 2019–2024) where in period 2 herpes zoster for high-risk persons was introduced in the Netherlands. All three AHDs showed protective associations, with valacyclovir demonstrating reduced incidence and hazard of AD treatment initiation, and remarkably, no AD cases occurring among acyclovir or famciclovir users in the later period. Comparison with other studies Our results align with findings from Taiwan, where antiviral treatment was associated with a ~90% reduced risk of dementia in patients with herpesvirus infections (adjusted HR 0.09, 95% CI 0.08 to 0.11).17 The magnitude of our overall effect estimate (HR 0.099, 95% CI 0.072 to 0.136) is remarkably consistent with these findings. Similarly, a nationwide Korean cohort reported a more modest but significant protective association, with adjusted HRs ranging from 0.76 to 0.79 depending on analytic method.15 In Sweden, registry-based analyses found a smaller yet consistent reduction in dementia risk among antiviral users (HR 0.89, 95% CI 0.86 to 0.92).19 The increase of the protective association observed in our study—from an 86% reduction in period 1 (1994–2018) to a 94% reduction in period 2 (2019–2024)—is particularly noteworthy. This temporal pattern may reflect several factors: increased prescribing practices with longer treatment durations, earlier diagnosis of dementia enabling detection of protective effects, better patient adherence to antiviral therapy, or changes in the characteristics of patients receiving AHDs. Differences between populations—such as genetic background, underlying risk factors and healthcare systems—could also explain variations in effect size across studies. Interpretation and possible mechanisms The protective association between AHD exposure and AD treatment initiation is biologically plausible. Herpesvirus reactivation has been linked to amyloid-beta deposition, tau hyperphosphorylation and neuroinflammation.11 12 Antiviral therapy could mitigate these processes by suppressing viral replication and reducing associated inflammatory cascades. It should be noted that many herpesvirus infections, particularly recurrent episodes of herpes simplex virus (HSV) and HZ, are self-limiting and often resolve spontaneously without antiviral treatment.30 HSV reactivations typically cause self-resolving mucocutaneous lesions that heal within 7–10 days without intervention, while herpes zoster is characterised as a painful but self-limited dermatomal rash.30 In clinical practice, antiviral treatment is often reserved for severe infections, immunocompromised patients, or to reduce the duration and severity of symptoms and post-herpetic neuralgia.31 32 All three anti-herpetic agents showed protective associations in our study, with overall HRs of 0.10 for valacyclovir, 0.09 for acyclovir and 0.07 for famciclovir. Pharmacological differences may also play a role: valacyclovir and famciclovir have higher bioavailability and central nervous system penetration than acyclovir.33 However, the CIs for these estimates overlap considerably, and the small number of events among famciclovir users limits precise estimation. Beyond antiviral activity, some agents may exert direct neuroprotective or anti-inflammatory effects.34 Several concomitant medications showed notable associations with AD risk. Anti-Parkinson’s drugs demonstrated the strongest association with increased risk of medication for dementia across all analyses (HR 1.84 overall, 95% CI 1.65 to 2.05), likely reflecting confounding by indication. Parkinson’s disease is itself a major risk factor for dementia, with 60%–85% of patients developing dementia over the disease course.35 The observed association with anti-Parkinson’s medications therefore likely reflects shared underlying neurodegenerative pathology rather than a causal effect of the medications themselves. Antidepressant use was also associated with increased AD risk (HR 1.34 overall, 95% CI 1.18 to 1.52). This finding should be interpreted with caution as depression itself is a well-established risk factor for dementia, with late-life depression associated with approximately a twofold increase in dementia risk.36 Additionally, depression is highly prevalent in individuals with dementia, creating a complex bidirectional relationship that may confound the observed associations. The association between antidepressants and AD may reflect confounding by indication,3740 whereby the underlying depression—rather than its treatment—contributes to dementia risk. Additionally, depressive symptoms often occur in dementia, potentially representing reverse causation. Interestingly, antiepileptic use was associated with reduced AD risk (HR 0.72, 95% CI 0.54 to 0.95). The relationship between antiepileptic drugs and dementia risk is complex and drug-specific. While older antiepileptic agents with cognitive adverse effects (such as valproate, phenytoin, carbamazepine) have been associated with increased dementia risk, newer agents like levetiracetam appear to have neutral or potentially protective effects.41 42 Other medication classes including antidiabetics, antihypertensives, cholesterol-lowering drugs, anticoagulants, anticholinergics and benzodiazepines showed no significant associations with AD risk in adjusted models. Clinical and policy implications Our results support the hypothesis that herpesvirus reactivation contributes to AD pathogenesis and suggest that antiviral therapy may delay the need for AD treatment in older adults. Although only 3% of participants used AHDs prophylactically, this low prevalence did not materially affect our study results, and our findings remained robust. Given the ARRs observed, even short-term antiviral use may yield meaningful population-level benefits. These findings are particularly relevant in the Netherlands, where herpes zoster vaccination was introduced only after 201820; the observed association therefore primarily reflects antiviral prescribing rather than vaccination. Future research should assess whether vaccination and antiviral treatment act synergistically to reduce dementia risk. Further studies with clinical diagnoses, biomarker data and stratification by patient subgroups are warranted to confirm these findings. Ultimately, randomised controlled trials of antiviral therapy for dementia prevention would provide the most definitive evidence, although logistical and ethical challenges may limit feasibility. Strengths and limitations Strengths of this study include the use of a large, representative prescription database with long-term follow-up and a robust exposure definition. Matching and multivariable adjustment reduced confounding, and consistency across incidence rates, Kaplan-Meier curves and Cox models strengthens the validity of the findings. Several limitations must also be considered. First, the outcome was initiation of AD treatment rather than a clinical diagnosis. While Dutch prescribing guidelines restrict these drugs almost exclusively to AD, some degree of misclassification cannot be excluded. Notably, rivastigmine, one of the three cholinesterase inhibitors commonly prescribed for AD, is also frequently used to treat dementia associated with Parkinson’s disease and may be prescribed for motor symptoms in non-demented Parkinson’s patients.35 43 This dual indication means that some rivastigmine prescriptions in our cohort may reflect Parkinson’s disease treatment rather than AD treatment, potentially biasing estimates toward the null if the protective association between AHDs and Parkinson’s disease differs from that with AD. Second, residual confounding by unmeasured variables, including education, lifestyle or socioeconomic status, is possible.3 Third, the database lacks information on cognitive assessments, dementia subtypes and biomarkers, limiting clinical interpretation. Fourth, the relatively small number of famciclovir and acyclovir users, particularly in the second period, resulted in very few events and consequently wide CIs that limit precise estimation of drug-specific effects for these agents. Finally, prescribing and health-seeking behaviours may differ between exposed and unexposed individuals, which could partly explain the observed associations. However, given the high effectiveness estimates with large precision, it is highly unlikely that bias or imprecision can explain these findings. Conclusions This large, population-based cohort study found that use of AHDs was consistently associated with a markedly reduced risk of initiating treatment for AD, with stronger protective associations in recent years. These findings support the hypothesis that herpesvirus reactivation contributes to AD pathogenesis and suggest that antiviral therapy may delay treatment onset in older adults. Confirmation through studies using clinical diagnoses and biomarkers, as well as exploration of potential synergy with herpes zoster vaccination, is warranted. Supplementary material