The recombinant shingles vaccine is associated with lower risk of dementia

Dementia prevention remains a major public health priority. Varicella-zoster virus causes varicella and herpes zoster (shingles), and shingles vaccination is recommended for older adults due to the risk of complications. Prior observational studies suggested shingles vaccination might reduce dementia risk, but most compared vaccinated versus unvaccinated populations—designs vulnerable to selection biases such as the healthy-vaccinee effect. A recent natural experiment around an age-eligibility cutoff indicated the live zoster vaccine may protect against dementia, particularly in women, but the live vaccine has been discontinued in favor of a recombinant vaccine. Whether the recombinant vaccine confers similar or greater protection against dementia was unknown. This study leverages a natural experiment in the U.S. healthcare system—the rapid uptake of the recombinant vaccine and concurrent disuse of the live vaccine after October 2017—to estimate the association between recombinant vaccination and subsequent dementia risk while mitigating major sources of selection bias.

Multiple cohort studies have reported associations between shingles vaccination and reduced dementia incidence, but many relied on vaccinated versus unvaccinated comparisons susceptible to confounding and healthy-vaccinee bias. A notable exception was a natural experiment in Wales comparing individuals just above and below an eligibility age cutoff, which found protective effects of the live zoster vaccine limited to women. Additional studies explored links between herpesvirus infections and dementia risk, a hypothesis debated for decades, with some evidence implicating herpes infections or reactivations in neurodegenerative processes. Given the replacement of the live vaccine by the recombinant vaccine (which provides superior protection against shingles and includes adjuvants), determining whether the recombinant vaccine also reduces dementia risk addresses both translational and mechanistic questions raised by prior literature.

Data source and design: Retrospective cohort analyses used de-identified electronic health records from the TriNetX US Collaborative Network (62 healthcare organizations; >100 million patients). The natural experiment exploited the rapid transition from live to recombinant shingles vaccines after October 2017. Cohorts and exposures: Two cohorts included patients who received their first shingles vaccine dose at age ≥65: (1) primary cohort vaccinated between 1 Nov 2017 and 31 Oct 2020 (predominantly recombinant; 95% recombinant; median follow-up 4.15 years, IQR 3.16–4.99) and (2) comparator cohort vaccinated between 1 Oct 2014 and 30 Sep 2017 (predominantly live; 98% live; median follow-up 6.0 years, IQR 5.2–6.0). Individuals vaccinated in Oct 2017 were excluded (transition month). Exclusions applied for prior or within-1-month post-vaccination diagnoses of dementia (ICD-10 F01, F02, F03), Parkinson’s disease (G20), and other degenerative nervous system diseases (G30–G32), to limit reverse causation. Covariates and matching: Sixty covariates were used for propensity-score matching (1:1, caliper 0.1) including sociodemographic factors (age matched in 2-year bins from 65 up to ≥95, sex, ethnicity, race, marital status), broad ICD-10 comorbidity categories (with further granularity for key categories such as neoplasms, cardiovascular, psychiatric, endocrine/metabolic), prior herpes infections, history of influenza vaccination, and selected Z-codes related to healthcare use/health factors showing imbalances (SMD >0.15). Matching aimed to achieve SMD <0.1 for all covariates. The estimand is best interpreted as the average treatment effect in controls. Outcomes: The primary outcome was first dementia diagnosis from 3 months to 6 years post-vaccination (ICD-10 F01, F02, F03, G30, G31.0, G31.83). Secondary outcomes included all-cause mortality, composite dementia or death, individual dementia subtypes, herpes zoster infection (B02), and a composite negative control outcome of acutely painful, dementia-unrelated conditions. Statistical analysis: Incidence curves used Kaplan–Meier estimators. Proportional hazards were tested via Schoenfeld residuals (cox.zph); due to violation in the primary analysis (P<0.0001), Cox models were not used. Instead, restricted mean time lost (RMTL) over 6 years was estimated (survRM2), reporting RMTL ratios and absolute differences translated into additional diagnosis-free days among those affected. Time-varying hazard ratios (HRs) used generalized survival models with natural cubic splines (rstpm2), selecting degrees of freedom by AIC. Sex moderation was tested via permutation (1,000 permutations). Missingness for sex, race, ethnicity used explicit categories and was included in matching. Two-sided P<0.05 was considered significant. Analyses used Python (propensity scores) and R 4.2.1. Secondary/robustness analyses: Stratification by sex; restriction to patients confirmed to have received the predominant vaccine in each window; limiting exposure windows to 6 months pre/post October 2017; limiting follow-up to 18 months fully pre-COVID-19; excluding individuals who received both vaccines; adjusting for socioeconomic deprivation (ICD-10 Z59). Coarsened exact matching (CEM) on key covariates (age, sex, race, neurological comorbidities) was used with parametric and bootstrap variance. Follow-up horizons were aligned at the cohort or matched-pair levels. Additional analyses compared shingles vaccines (recombinant and live) against influenza and Tdap vaccines given in the same exposure windows. Data/code availability: Aggregate outputs are available at https://osf.io/9frxm/; code is available at https://osf.io/9frxm/. TriNetX data access requires permission and potential costs.

Primary analysis (propensity-score matched, n=103,837 per cohort): The recombinant-predominant cohort had a lower risk of dementia over 6 years than the live-predominant cohort, with an RMTL ratio of 0.83 (95% CI 0.79–0.87; P=2.9×10^-15), corresponding to 17% more diagnosis-free time and +164 days (95% CI 124–205) lived without dementia diagnosis among those affected. Time-varying HRs were <1 within the first year and approached 1 toward year 6. Sex stratification: Protective associations were observed in both sexes, larger in women: females RMTL ratio 0.78 (95% CI 0.73–0.83; P=2.3×10^-15), +222 days (168–276); males RMTL ratio 0.87 (95% CI 0.81–0.94; P=0.00028), +122 days (56–187). The effect difference by sex was significant (permutation test P=0.017). Other outcomes: No significant difference in all-cause mortality (RMTL ratio 0.98; 95% CI 0.95–1.01; P=0.22). The composite of dementia or death favored the recombinant cohort (RMTL ratio 0.93; 95% CI 0.91–0.96; P=3.8×10^-7; +64 days, 39–89). Negative control outcome showed no difference (RMTL ratio 0.97; 95% CI 0.91–1.03; P=0.29). Herpes zoster infection: Lower incidence after recombinant vaccination (RMTL ratio 0.65; 95% CI 0.61–0.69; P=4.3×10^-28), adding +381 days (326–435) without zoster among those affected; similar magnitude in women (0.64) and men (0.65), with no sex moderation (P=0.87). Robustness analyses: Results were consistent when: aligning follow-up horizons (cohort-wise RMTL ratio 0.83; 95% CI 0.79–0.87; P=4.3×10^-15; +165 days 121–209); restricting to predominant vaccine recipients (0.82; 0.79–0.86; P=7.5×10^-14; +173 days 131–214); adjusting for socioeconomic deprivation (0.84; 0.80–0.88; P=1.4×10^-14; +157 days 117–196); excluding individuals who received both vaccines (0.79; 0.74–0.83; P=1.5×10^-17; +214 days 165–263); restricting exposure windows to 6 months on either side of the step change (0.83; 0.76–0.92; P=0.00025; +160 days 74–246). Coarsened exact matching yielded similar estimates (parametric variance 0.82; 0.77–0.87; P=1.6×10^-11; +192 days 137–248; bootstrap variance 0.82; 0.79–0.86; P<0.001; +192 days 151–235). A pre-pandemic limited follow-up analysis also showed lower risk (HR 0.74; 95% CI 0.62–0.90; log-rank P=0.0019), with no evidence of PH violation (P=0.56). Comparisons with other vaccines: Recombinant zoster vs influenza: RMTL ratio 0.77 (n=209,031; P=1.4×10^-10), +213 days; recombinant vs Tdap: 0.73 (n=98,353; P=2.6×10^-5), +271 days. Live zoster vs influenza: 0.86 (n=41,466; P=1.2×10^-6), +144 days; live vs Tdap: 0.86 (n=64,035; P=2.1×10^-6), +142 days.

The natural experiment comparing individuals vaccinated before (live vaccine predominant) versus after (recombinant vaccine predominant) October 2017 suggests that recombinant shingles vaccination is associated with a clinically meaningful reduction in dementia risk over 6 years, with 17% more diagnosis-free time and about 5.5 months of delayed diagnosis among those who develop dementia. The protective association is consistent across sexes, larger in women by approximately 9 percentage points, and robust to multiple analytic choices, including alternative matching strategies, exposure windows, and adjustment for socioeconomic deprivation. The lack of association with all-cause mortality and with a negative control outcome reduces concern about residual confounding from general health status or healthcare-seeking behavior, while the significant composite outcome (dementia or death) argues against survivorship bias. Mechanistically, protection could be mediated by reducing herpes zoster infections, in line with the observed reduction in zoster risk and longstanding hypotheses linking herpesvirus infections with dementia. Superior anti-zoster efficacy of the recombinant vaccine could account for its greater apparent protective association relative to the live vaccine. The attenuation of the time-varying HR toward the end of follow-up—mirroring waning protection against zoster—suggests a possible delay rather than permanent prevention of dementia onset, though this pattern was not consistently present across all analyses and requires replication. As an observational study, causality cannot be established. Nevertheless, the rapid vaccine transition mitigates key selection biases inherent in vaccinated versus unvaccinated comparisons. Consistency across secondary analyses, similar mortality, and null findings for negative controls support the validity of the association. These findings provide motivation for randomized controlled trials to confirm causality and to inform health-economic evaluations of recombinant shingles vaccination as a potential dementia risk-reduction strategy.

Receiving the recombinant shingles vaccine is associated with a lower 6-year risk of dementia compared with receiving the live shingles vaccine, translating to 17% more time lived without a dementia diagnosis and substantial additional diagnosis-free days among those who develop dementia. The association is robust across analyses, present in both sexes (greater in women), and extends to comparisons versus influenza and Tdap vaccines. These results warrant mechanistic studies into antiviral and adjuvant-mediated pathways and support the design of large-scale randomized controlled trials to confirm causality, determine durability of protection, and evaluate cost-effectiveness and optimal vaccination strategies in older adults.

Key limitations include the observational design and reliance on EHR-coded diagnoses without independent validation. Socioeconomic and lifestyle data are sparse, though adjustment for socioeconomic deprivation was performed in a secondary analysis. Diagnosis-free time may not equate to disease-free time due to diagnostic delays; if such delays are similar across cohorts, differences should parallel disease-free time. The study did not examine effects of multiple vaccine doses. Differences in vaccination uptake and population characteristics over time could introduce confounding, although restricting exposure windows to 6 months on either side of the 2017 step change and extensive matching reduce this risk. In the primary analysis, confidence intervals did not account for the paired nature of matched data; however, coarsened exact matching with paired bootstrap variance produced similar estimates. As with all studies using a live, continuously updated EHR network, exact counts may vary over time.