Risk of Alzheimer Dementia After High-Dose vs Standard-Dose Influenza Vaccination.

Accessible, low-cost preventive measures are needed to reduce risk of dementia. 1 Several antimicrobial vaccines have been linked with a reduced risk of dementia in older adults. 2-4 Our previous study found that receiving ≥1 vs zero inactivated influenza vaccines (IIVs) was associated with a 40% lower risk of Alzheimer dementia (AD) over a 4-year follow-up. 2 Similar to other studies of influenza vaccination and AD, however, our study did not differentiate between standard influenza vaccines and those with immunogenic enhancements (i.e., increased antigen dosage, adjuvanation, or antigen recombination). 5,6 Whether immunogenic enhancements modify the effect of influenza vaccination on AD risk remains unclear. Plausible mechanisms for such a modifying effect can be categorized as antimicrobial or nonantimicrobial. Compared with standard IIVs, enhanced IIVs confer greater protection against influenza infection, thereby decreasing risk of severe illness and the associated systemic inflammation that can promote neuroinflammation and neurodegeneration. 7 Possible nonantimicrobial mechanisms include induction of trained immunity, modulation of homeostatic imbalances associated with aging (e.g., inflammaging, immunosenescence), and modification of the neuroinflammatory response to AD pathologies leading to clearance and/or decreased injury to adjacent healthy parenchyma. 8 A recent study comparing dementia risk after recombinant herpes zoster (HZ) vaccination vs live-attenuated HZ vaccination found that dementia risk was lower after the recombinant vaccine and that this effect was stronger in women than in men. 9 Whether this sex-dependent modulation extends to other immunogenic enhancements, such as increased antigen dosage, is unknown. This study investigated whether antigen dose influences the effect of influenza vaccination on AD risk by comparing AD risk among adults ≥65 years old who received high-dose or standard-dose influenza vaccines. We hypothesized that the risk of AD among adults ≥65 will be lower after high-dose IIV (H-IIV) than after standard-dose IIV (S-IIV), with a greater difference among women than men.

This population-based retrospective cohort study used target trial emulation (TTE) to address the causal question: Among adults ≥65 years, what is the effect of vaccination with an H-IIV vs a standard IIV (standard dosage, nonadjuvanted, nonrecombinant) at baseline on the 3-year risk of incident AD, had no one been lost to follow-up? TTE begins by specifying the protocol of a hypothetical randomized trial and then uses observational data to emulate it, aligning eligibility, treatment assignment, and follow-up duration to mitigate immortal-time and related selection biases. 10,11 Full specifications of the target trial and emulation protocol are in eTable 1. The REporting of studies Conducted using Observational Routinely collected health Data guideline was followed. 12 Data were obtained from IQVIA PharMetrics Plus for Academics (PMPA), a US commercial claims database with adjudicated inpatient, outpatient, and pharmacy claims from all 50 states. Data available for this study spanned January 2014 through September 2022. The study period started August 1, 2014, and ended July 31, 2019. The study ended in 2019 to avoid the coronavirus disease 2019 pandemic (eFigure 1). Mortality data were not available. Eligible persons had ≥2 years of continuous enrollment in medical and pharmaceutical insurance reportable to the PMPA database; had data for age, sex, and geographic region; were 65 years or older; had no previous International Classification of Diseases (ICD) records (eTable 2) for mild cognitive impairment (MCI), encephalopathy, or dementia of any cause; and had no previous claims for any of the 4 medications that were US Food and Drug Administration approved for symptomatic AD treatment (donepezil, galantamine, rivastigmine, memantine; Figure 1 ). Similar to our previous comparison of influenza vaccinees and nonvaccinees, variables related to AD diagnosis, or to both AD diagnosis and influenza vaccine uptake or dosage, were included as covariates. 2 Variables strongly related to influenza vaccination but not AD were excluded. 13 Covariates included demographics, medical comorbidities, medications, routine vaccinations, and health care utilization proxies ( Table 1 ). For all covariates besides influenza vaccination history, baseline values were defined by the most recent measurement during the 2 years before enrollment. For history of influenza vaccinations, the cumulative doses of enhanced and standard-dose influenza vaccines over the past 2 influenza seasons were separate covariates. Detailed variable definitions are provided in eTable 2. Baseline Characteristics of Study Participants Abbreviations: COPD = chronic obstructive pulmonary disease; IIV = inactivated influenza vaccine; IPTW = inverse probability of treatment weighting; NSAID = nonsteroidal anti-inflammatory drug; SMD = standardized mean difference; Td = tetanus toxoid and reduced diphtheria toxoid; Tdap = tetanus toxoid, reduced diphtheria toxoid, and acellular pertussis. Frequency and percentage (of the column sample size) are provided for categorical variables; mean and SD are provided for continuous variables. See eTable 2 for complete variable definitions. The sample size (n) for each column equals the number of person-trials that were assigned to each treatment; each person can contribute up to 3 person-trials. SMDs are significant if ≥0.10, indicating inadequate covariate balance between the high-dose and standard-dose influenza vaccine groups. Number of outpatient or inpatient health care encounters in the 2 years before baseline. Composite of posttraumatic stress disorder, panic disorder, anxiety disorder not otherwise specified, obsessive compulsive disorder, social phobia, and generalized anxiety disorder. Composite of substance use disorders involving any of the following: opioids; cannabis; sedatives, hypnotics, or anxiolytics; cocaine; amphetamines or other stimulants; hallucinogens; inhalants; and/or other psychoactive substances, including polysubstance use. “Sustained use” was defined as ≥2 prescription claims in any 6-month period during the 2 years before baseline. At least 1 Zostavax or Shingrix vaccination during the 2 years before baseline. At least 1 PPSV-23 or PCV-13 vaccination during the 2 years before baseline. At least 1 Tdap or Td vaccination during the 2 years before baseline. Includes high-dose, adjuvanted, and recombinant inactivated influenza vaccine. Both per-protocol (PP) and intention-to-treat (ITT) effects were estimated. In PP, participants were censored if they received any influenza vaccination during follow-up, so estimates better reflect baseline dosage (H-IIV vs S-IIV) without later influenza vaccinations. In ITT, participants were not censored for subsequent influenza vaccination, better reflecting real-world annual revaccination but introducing potential exposure misclassification if there is mismatch between baseline and follow-up vaccinations. Because the primary causal question concerned the baseline dose independent of subsequent vaccinations, the PP design was emphasized ( Figure 2 , eTable 1). 11,14 This flowchart illustrates the stepwise selection of eligible participants for sequential emulated trials using the IQVIA PharMetrics Plus for Academics database. The study period spanned August 2014 through July 2019, although data from January 2014 through September 2022 were available for this study. The initial database population from January 2014 through September 2022 consists of 28,917,653 unique individuals. The exclusion criteria in each of the 2 exclusion-criteria boxes (upper right) are not mutually exclusive, so individuals may be excluded for more than 1 reason. The final sample of eligible participants is classified as either high-dose or standard-dose influenza vaccine initiators, with follow-up information and censoring reasons provided for each group. AD = Alzheimer dementia; MCI = mild cognitive impairment. Although a total of 21 sequential trials were emulated in the primary analysis, illustrative examples from only 2 of the sequential trials (the first and the ninth) are depicted in this figure. Each unique participant is represented by a different color. “START” indicates the participant's enrollment in a new trial. The syringe icon represents an influenza vaccination event. For each person, the syringe labeled “1st” represents the vaccination resulting in that person's first trial enrollment; that person's next vaccination event is labeled “2nd,” and so on. The dosage of each influenza vaccine event is indicated by the “HIGH” or “STANDARD” designators affixed to the syringe icon. Participants without a dosage indicator at enrollment (the brown, green, and gray participants) could represent a person enrolled in either the high-dose or the standard-dose group. During follow-up, a dosage indicator is not shown if the trajectory would be the same regardless of dosage; for example, regardless of the dose of the second influenza vaccine that the brown person receives, this person will be censored in the PP analysis but not in the ITT analysis. Censorship is indicated by a red X. Diagnosis of incident Alzheimer dementia is indicated by the symbol with a question mark inside of a head. Participants are enrolled in a trial if they meet all eligibility criteria and receive a high-dose or standard-dose influenza vaccine during the trial's enrollment month. The first emulated trial (“trial 1”) begins with enrollment in August 2016 and then has a follow-up period from September 1, 2016, through July 31, 2019. Each month during an influenza vaccine season (marked by the purple columns) constitutes the enrollment period for a sequential emulated trial. Participants enrolled in a trial that began during a previous influenza vaccine season can enroll in an additional trial during each subsequent influenza vaccine season as long as the person meets the eligibility criteria and receives a high- or standard-dose influenza vaccine during a future trial's enrollment period. Each person can, therefore, enroll in up to 3 trials across the 3 influenza vaccine seasons, and each separate trial for that participant is referred to as a “person-trial.” In the figure, dashed arcs are used to illustrate that each vaccination event for a participant can influence the trajectories of multiple person-trials. For example, the blue participant received a high-dose vaccine in August 2016 and is assumed to meet all eligibility criteria (this is assumed for all 6 example participants), so they are enrolled in the high-dose arm of Trial 1 (the blue person's first person-trial). The blue participant received another influenza vaccine in September 2017, resulting in censorship from Trial 1 in the PP design but not in the ITT design. In both the ITT and PP designs, the vaccine event with a standard-dose vaccine in September 2017 qualifies them for enrollment in a new person-trial in the standard-dose arm of Trial 9 (the blue person's second person-trial). The pink participant receives a high-dose influenza vaccine in August 2016, September 2017, and September 2018 and is, therefore, enrolled in the high-dose arms of Trials 1, 9, and 21 (3 total person-trials). Censorship of ongoing person-trials due to the vaccinations in September 2017 and September 2018 differs between the ITT design (no censorship) and the PP design (censorship of preceding trials). The brown participant receives either a high-dose or standard-dose vaccine in August 2016 and is enrolled in the corresponding treatment arm for Trial 1. The brown participant then receives a vaccine in either February 2017 or April 2017, but they are ineligible for enrollment in a new person-trial in both cases: they are not eligible in February 2017 because they already enrolled in a person-trial during the 2016–2017 flu vaccine season (Trial 1 in August 2016), and they are not eligible in April 2017 because April is outside of the August–February influenza vaccine season. The green participant receives either a high-dose or standard-dose vaccine in August 2016 and is enrolled in the corresponding treatment arm for Trial 1; they receive no additional influenza vaccines during the follow-up period of Trial 1 and are not diagnosed with AD or lost to follow-up, so they are followed until the end of the study period in both the ITT and PP designs. The gray participant receives either a high-dose or standard-dose vaccine in August 2016 and is enrolled in the corresponding treatment arm for Trial 1; they are then followed until diagnosis of incident AD in July 2017. AD = Alzheimer dementia; ITT = intention-to-treat; PP = per-protocol. Vaccinations with specific IIV formulations were identified using drug names and Current Procedural Terminology codes (eTable 2). The only H-IIV formulation was Fluzone High-Dose, which has been approved for US adults ≥65 since 2010. 15 All standard-dose, nonadjuvanted, nonrecombinant IIVs commercially approved in the United States for adults aged ≥50 during the study period were grouped as S-IIVs. Eligible participants could be enrolled and assigned to immediate receipt of either H-IIV or S-IIV during the 3 influenza vaccination seasons from 2016 to 2019. Vaccination for seasonal influenza in the United States typically begins in August, peaks October–November, and becomes negligible by the end of February, 16 a pattern reflected in the PMPA (eFigure 1). The influenza vaccination season was, therefore, defined as spanning August–February. A total of 21 sequential trials (the total number of months in the influenza vaccine seasons from August 2016 through July 2019) were emulated. Individuals meeting the eligibility criteria during multiple vaccine seasons were allowed to enroll in up to 1 trial per season. For each trial, follow-up began the first day of the month after trial enrollment and ended at the earliest of the following: incident AD; censoring due to lapsed medical or pharmaceutical coverage in the PMPA (loss to follow-up); or the end of the study period. In the PP design, participants were also censored for protocol nonadherence, defined as any type of influenza vaccination during follow-up ( Figure 2 ). If protocol nonadherence and incident AD occurred in the same month, this was counted as an AD case rather than censored. Incident AD was defined as ≥1 ICD record for AD, senile dementia, or dementia of unspecified cause, or a claim for donepezil, galantamine, rivastigmine, or memantine (eTable 2). Because AD-specific ICD codes are significantly under-recorded in claims while nonspecific/senile dementia codes are common, we included senile and unspecified dementia codes to minimize false-negative misclassification, at the expense of likely misclassifying some non-AD dementia cases as AD. 17,18 The implications of potential bias introduced by this operational definition were explored in the sensitivity analyses. The specific ICD codes used in the outcome definition and additional justification regarding the inclusion of senile/unspecified dementia are provided in eTable 2. The outcomes of all person-trials across the 21 emulated trials were pooled ( Figure 2 ). A propensity score for each person-trial was computed using a logistic regression model with treatment assignment as the outcome and baseline covariates as predictors. Propensity scores were then applied to inverse probability of treatment weighting (IPTW) to create a pseudo-population in which baseline covariates are balanced between treatment groups. IPTW weights were stabilized. Covariate balance was assessed using standardized mean difference (SMD), with SMD 0.1, the above process was repeated with the addition of a quadratic term for that covariate to the propensity-score model. Separate propensity-score models for the probability of censoring due to loss to follow-up (ITT and PP) or due to protocol nonadherence (PP only) were computed and applied to inverse probability of censoring weighting (IPCW). The censoring weights were stabilized and then multiplied by the stabilized treatment weights to generate composite weights. The composite weights were truncated at the 99th percentile. For mathematical representations, see eAppendix 1. Risk difference (RD), number needed to treat (NNT), and risk ratio (RR) were estimated at each month of follow-up via a pooled logistic regression model. Bootstrapping with 500 resamples with replacement was used to compute 95% CIs. Results were considered statistically significant if the CI excluded the null value (RD = 0, RR = 1). Data processing and statistical analyses were performed using Python version 3.8.5 and R version 4.3.2. Six secondary analyses were performed. First, we repeated the primary analysis after stratifying the database by sex. 9,19 Second, to better isolate the effect of vaccine dosage at enrollment from pre-enrollment vaccinations, we excluded individuals vaccinated against influenza in the past 2 years, mimicking a 2-year influenza-vaccine washout. Third, we replaced AD with incident MCI as the outcome and treated AD as a competing risk. To account for high MCI misclassification rates clinically and in administrative databases, 20-22 we evaluated 3 case-identification algorithms: 1 MCI ICD code (eTable 2); 1 MCI code or an AD-related medication claim; and a stricter definition requiring ≥2 MCI codes or 1 MCI code plus 1 AD-related medication claim within 12 months. Fourth, because the influenza vaccine is recommended annually, we assessed AD risk after sustained (seasonal) treatment with repeated H-IIV vs S-IIV over 3 years (eAppendix 2). Fifth, we compared AD risk after adjuvanted IIV (A-IIV) vs S-IIV (eAppendix 3). Assessment of the recombinant IIV (R-IIV) was precluded by claim sparsity, with only 1,599 R-IIV claims in the PMPA from August 2016 through July 2019. Sixth, we assessed the replicability of our previous study comparing AD risk after influenza vaccination vs nonvaccination in the Optum Clinformatics database by applying the previous study's methodology to the PMPA (eAppendix 4). 2 Eight sensitivity analyses were performed. To assess robustness to outcome misclassification, we evaluated 4 AD case-identification algorithms: a stricter definition requiring ≥2 AD-related ICD or medication records on different days within 12 months; a definition without senile/unspecified dementia ICD codes; a definition using medication claims only; and a definition including all AD and related dementias (ADRD) codes. Fifth, to assess sensitivity to follow-up duration, we shortened the eligibility-determination period to 1 year, resulting in 28 emulated trials over up to 4 years of follow-up. Sixth, to assess detection bias and reverse causality, we repeated the primary analysis with a 1-month lag period in which the start of follow-up was redefined as the first day of the second calendar month after enrollment. Participants were censored if diagnosed with incident AD during the first calendar month after enrollment. A lag period was not used in the primary analysis because of the absence of robust evidence establishing a minimum induction period between vaccine exposure and effect on cognitive decline in AD, and previous studies demonstrating that the effect of HZ vaccination on dementia risk was insensitive to lag period. 8,19,23 Moreover, a recent study of infections and risk of neurodegenerative disorders found that the risk of AD increased within 1, 5, 10, and 15 years after influenza infection, but the effect was greatest within 1 year of infection. 24 In a sensitivity analysis to evaluate potential reverse causality, that study assessed bidirectional hazard ratios between influenza and AD and found that the hazard ratio was 30.29 when influenza precedes AD diagnosis within 1 year but only 8.84 when AD precedes influenza, with nonoverlapping CIs. 24 This asymmetry suggests that influenza infection more strongly predicts subsequent AD than vice versa, arguing against significant reverse causality. Because H-IIV confers greater protection against influenza infection than S-IIV in older adults by 4 weeks postvaccination, 25 the stronger short-term (<1 year) association between influenza infection and AD risk may reasonably extend to AD risk after H-IIV vs S-IIV. Use of an active comparator further decreased risk of reverse causality. Although dementia is associated with lower IIV uptake, 26 an association between dementia and receipt of H-IIV vs S-IIV has not been established. Moreover, there is no clear biological, behavioral, or policy rationale for such an association. Seventh, to probe residual confounding, we restricted eligibility to antihyperlipidemic adherers, defined as those with antihyperlipidemic claims covering ≥80% days in the previous 12 months. Eighth, we tested a negative-control outcome composite of ICD codes for acute, typically monophasic, painful conditions, including acute pancreatitis, acute appendicitis, acute cholecystitis, and adhesive capsulitis of the shoulder (eTable 2). 9 Because evaluation of potential sex-dependent effects represented a key secondary objective, we repeated all 8 sensitivity analyses after stratifying the database by sex. Because this study uses only deidentified, retrospective claims data, it was deemed nonhuman subjects research by the UTHealth Institutional Review Board and was approved with waivers of Health Insurance Portability and Accountability Act and informed consent. Study codes are available upon request. The authors cannot make data available due to licensing restrictions; interested parties can license the data by contacting IQVIA.

Before application of IPTW, 185,183 H-IIV person-trials and 53,918 S-IIV person-trials were included in the pooled emulated trials ( Figure 1 ). After implementation of the initial IPTW including only linear terms, the SMD was <0.1 for all covariates except age and influenza vaccination history. After adding quadratic terms for these to the propensity-score regression model, all SMDs were <0.1 ( Table 1 ). The distributions of baseline covariates and their SMDs before and after IPTW are shown in Table 1 . There were 97,188 person-trials censored because of influenza vaccination during follow-up. The average follow-up duration for the H-IIV and S-IIV groups was 14.3 and 15.0 months in the PP design and 19.0 and 19.2 months in the ITT design. Compared with S-IIV, the cumulative risk of incident AD was significantly lower after H-IIV from months 1 to 25 in the PP design and 1 to 28 in the ITT design ( Tables 2 and 3 , Figure 3 ). The maximum statistically significant RD was 0.0054 (NNT = 185.2) at month 25 in the PP design and 0.0037 (NNT = 270.3) at month 28 in the ITT design ( Table 2 ). Alzheimer Dementia Risk After High- vs Standard-Dose Influenza Vaccination Abbreviations: AD = Alzheimer dementia; RD = risk difference; RR = risk ratio. RD <0 indicates a lower risk of AD after high-dose influenza vaccination compared with standard influenza vaccination. RR <1 indicates lower AD risk after high-dose vaccination compared with standard influenza vaccination. Parentheses contain 95% CIs. Values are significant if the 95% CI excludes the null value (RD = 0, RR = 1). AD After High-Dose vs Standard-Dose Influenza Vaccination Across Analyses Abbreviations: AD = Alzheimer dementia; ADRD = Alzheimer dementia and related dementias; Dx = diagnostic; Rx = medication. This table summarizes the periods with significant differences in risk of AD or ADRD after high-dose vs standard-dose influenza vaccination. “Protective” indicates that AD or ADRD risk was lower after high-dose influenza vaccination compared with standard-dose vaccination. A graphical representation of effect estimates and CIs is provided in eFigure 15. The cumulative incidence of AD (y-axis) after high-dose (blue) or standard-dose (red) influenza vaccination. The shaded region indicates the 95% CIs for the cumulative incidence in each treatment group. Note that this interval is different from the 95% CIs for the effect estimates (risk difference and risk ratio) at each time point, as shown in Table 2 . Overlap in the 95% CIs of cumulative incidence at a given month of follow-up does not indicate that the 95% CIs for RD and RR for that month include the null value (RD = 0, RR = 1). AD = Alzheimer dementia; RD = risk difference; RR = risk ratio. Women who received an H-IIV compared with an S-IIV had a significantly lower AD risk from months 1 to 13 in the PP design and 1 to 17 in the ITT design, with a maximum effect of NNT = 416.7 at month 13 in the PP design (eTables 3 and 4, eFigure 2). Among men, the PP results were not statistically significant, but in the ITT results, risk of incident AD was lower after H-IIV from months 17 to 24, with a minimum NNT = 232.6 at month 24. After excluding people with an influenza vaccination in the past 2 years, AD risk was significantly lower after H-IIV from months 1 to 4 in the ITT results, but there were no significant differences in the PP results (eTables 5 and 6, eFigure 3). In the 3 analyses with varying definitions of incident MCI as the outcome and AD as a competing risk, the ITT results showed an increased risk of incident MCI after H-IIV vs S-IIV in the ICD-only analysis and the ICD-or-medication analysis, but not in the stricter definition requiring 2 ICD or 1 ICD plus 1 medication record in a 12-month period. In the PP results for the ICD-or-medication definition, the risk of MCI was increased after H-IIV from months 13 to 24 of follow-up, although the lower confidence bound ranged from 0.0000 to 0.0002 during this period. The PP results of the 2 other MCI analyses, including the analysis with a stricter case definition, were nonsignificant (eTables 7–9). Analysis of sustained (seasonal) vaccinations on incident AD risk demonstrated a reduced risk after H-IIV from months 1 to 27 in the PP design, with a maximum effect of NNT = 294.1 at month 27 (eTables 10–13, eFigures 4 and 5). In the comparison of AD risk after A-IIV vs S-IIV, there was no significant effect in the PP results. In the ITT results, AD risk was significantly lower after A-IIV vs S-IIV from months 22 to 35, with a maximum effect of NNT = 62.1 at month 35 (eTables 14–16, eFigure 6). Risk of AD was significantly lower after influenza vaccination vs nonvaccination (NNT = 137.0, eTable 17), supporting the replicability of our previous work using a different administrative database (eAppendix 4). 2 In the sensitivity analysis requiring 2 AD-related events in a 12-month period, AD risk was significantly lower after H-IIV from months 29 to 35, with a maximum effect of NNT = 79.4 at month 35 (eTables 18 and 19, eFigure 7). When the ICD codes for unspecified and senile dementia were removed from the case-identification definition, the number of incident AD cases in each of the H-IIV and S-IIV groups was approximately 80% lower than in the primary analysis, and there were no significant differences in AD risk after H-IIV vs S-IIV (eTables 20 and 21, eFigure 8). Similarly, there was no significant difference in AD risk between H-IIV and S-IIV when incident AD was defined by medication claims only; however, the number of cases was 62% lower than in the primary analysis, with a corresponding decline in estimate precision (eTables 22 and 23, eFigure 9). With ADRD as the outcome, the number of cases in the H-IIV and S-IIV groups was 10%–15% greater than in the primary analysis. Risk of ADRD was significantly lower after H-IIV from months 1 to 17, with a maximum effect of NNT = 357.1 at month 17 (eTables 24 and 25, eFigure 10). In the analysis with a 1-year eligibility-determination period, AD risk was significantly lower after H-IIV from months 1 to 19, with a maximum effect of NNT = 400.0 at month 17 (eTables 26 and 27, eFigure 11). In the analysis with a 1-month lag period, AD risk was significantly lower after H-IIV from 8 to 28 months, with a maximum effect of NNT = 148.2 at month 28 (eTables 28 and 29, eFigure 12). In the analysis including only antihyperlipidemic adherers (eTables 30–32, eFigure 13), there was no significant difference in AD risk after H-IIV vs S-IIV. In the negative-control outcome analysis, there was no significant difference in risk of the negative-control outcome after H-IIV vs S-IIV (eTables 33 and 34, eFigure 14). The results of all secondary analyses and sensitivity analyses, except those stratified by sex, are summarized in Table 3 and eFigure 15. In the sensitivity analyses stratified by sex, the duration of the significant effect of H-IIV vs S-IIV on AD risk was greater among women, as was the robustness of this effect to changes in outcome definition, follow-up duration, and lag period (eTables 35–43). This study provides Class II evidence that treatment with H-IIV vs S-IIV was associated with decreased incident dementia in individuals ≥65 years of age captured in this US health care claims database.

This retrospective cohort study used a large US claims database to investigate whether AD risk differs between adults ≥65 years old who received a high-dose influenza vaccine compared with a standard-dose influenza vaccine over a follow-up period of up to 3 years. We found that the risk of incident AD was significantly lower after H-IIV compared with S-IIV for the first 25 months postvaccination, with a minimum NNT of 185.2 at month 25. Key strengths include an active-comparator design restricted to vaccinated individuals; compared with vaccinated-unvaccinated contrasts, the active-comparator design reduces bias arising from differences in underlying health status and behaviors, including confounding by indication, healthy-vaccinee bias, surveillance bias, and unmeasured confounding. 27 Target trial emulation prevented immortal-time bias. 11 IPTW and IPCW mitigated baseline and time-varying confounding and addressed informative censoring due to loss to follow-up and protocol nonadherence. The null negative-control outcome analysis further argues against substantial residual confounding. Previous studies have demonstrated that sex modulates immunologic responses to antimicrobial vaccinations and their effect on dementia risk. 19,28-30 In the present study, both men and women exhibited lower AD risk after H-IIV vs S-IIV; however, the duration and robustness of this finding in sensitivity analyses were stronger among women. Further research is needed to elucidate whether the sex-dependent effect of influenza vaccine dosage on dementia risk reflects sexual dimorphism in influenza-specific immunity or non–influenza-specific effects (e.g., innate training), both of which tend to be greater among older women than their male peers. 1,29-33 After excluding individuals who received an influenza vaccine in the past 2 years, there was no significant difference in incident AD risk after H-IIV vs S-IIV, except for a decreased risk with H-IIV in months 1–4 of the ITT analysis. These overall nonsignificant findings suggest that the duration of reduction in AD risk after H-IIV vs S-IIV may depend on repeated vaccination at least every 2 years. However, in the secondary analysis comparing AD risk during sustained exposure to H-IIV vs S-IIV over a 3-year period, AD risk was lower after H-IIV for months 1–27, only slightly longer than in the primary analysis. The MCI findings should be interpreted cautiously given high misclassification in routine practice and especially in claims data. 20-22 Outside of specialized centers, MCI is often diagnosed without in-depth neuropsychological testing and can be confounded by comorbid conditions. 34-36 Studies of Medicare claims have found very low MCI detection rates (as low as 0.08%) among US primary care clinicians, with >92% of expected MCI cases lacking an MCI diagnosis in claims data. 21,22 In an Alzheimer Disease Research Center cohort, 50% of expert-diagnosed MCI cases were coded as having dementia in a traditional Medicare claims database. 37 Taken together, these studies suggest that most MCI cases in this dataset are likely undetected or, less frequently, diagnosed with dementia. These limitations likely contribute to the inconsistent MCI results across case definitions. There were no significant differences in incident AD risk after A-IIV vs S-IIV in the PP results, but in the ITT results, the risk of incident AD was significantly lower after A-IIV from months 22 to 35 of follow-up. This delayed onset may reflect involvement of longer-lasting immunologic effects of A-IIVs, including more durable humoral immunity and memory-like innate responses. 6 See eAppendix 3 for additional discussion, including future directions for research on the effect of adjuvants and AD. As explained in the Outcome Ascertainment methods subsection, ICD codes for senile and unspecified dementia were included in the primary analysis definition of incident AD to account for high dementia misclassification rates in claims data. 17,18 When these codes were excluded, there was no significant difference in AD risk after H-IIV vs S-IIV. However, this finding likely reflects, at least in part, the high false-negative misclassification rate for AD in claims data. 17,18 Alternatively, the differential effect of H-IIV vs S-IIV on dementia risk may not be specific to AD. Future studies incorporating adjudicated diagnoses are needed to determine whether the findings can be replicated using consensus- and biomarker-confirmed AD cases. In both the ITT and PP analyses with a 1-month lag period, no significant differences were observed from months 1 to 7 of follow-up. From months 8 to 28, however, H-IIV was associated with a significantly lower risk of AD. These findings suggest that the apparent protective effect of H-IIV vs S-IIV in months 1–7 of the primary analysis may instead reflect preexisting, undiagnosed dementia (reverse causality) and/or differences in dementia surveillance between high-dose and standard-dose recipients (detection bias). Furthermore, when restricted to antihyperlipidemic adherers, no significant differences in AD risk were observed between H-IIV and S-IIV. Compared with the primary analysis, this subgroup included 63% fewer person-trials and 64% fewer cases ( Figure 1 , eTable 30). Although the effect sizes were consistent with those of the primary analysis, the CIs were considerably larger. These findings suggest that healthy-vaccinee bias may have influenced the primary analysis results. In the negative-control outcome analysis, the absence of a significant difference in risk between H-IIV vs S-IIV supports the attribution of the observed effect in the primary analysis, at least partially, to the exposure (i.e., vaccine dosage) instead of solely confounding factors. A limitation of this study is the lack of mortality data. Because H-IIVs provide better influenza protection than S-IIVs, S-IIV recipients may have higher mortality or infection-related morbidity. 38,39 If deaths were not accounted for as competing risks, AD risk among S-IIV recipients could be underestimated, biasing results away from finding lower AD risk after H-IIV. Because this study censored individuals with lapses in medical or pharmaceutical coverage reported to the PMPA, participants who died during follow-up were likely censored from the at-risk population shortly after death. However, we are unable to differentiate unenrollment after death from unenrollment due to other reasons (e.g., changes in insurance). IPCW was used to adjust for potential informative censoring due to nonrandom unenrollment and, in the PP design, for protocol nonadherence. Given the years-long preclinical and prodromal phases of AD, the follow-up duration of up to 3 years (or up to 4 years in the sensitivity analysis with a 1-year eligibility-determination period) represents another limitation, although this duration is similar to several other studies of vaccination and dementia risk. 40 Moreover, given the evidence for a significant increase in AD risk within a year after influenza infection, the greater protection against infection conferred by H-IIV vs S-IIV within 28 days postvaccination represents a plausible mechanism for an effect of vaccine dosage on AD risk over a 3-year follow-up. 24,25 Another limitation is this database's lack of socioeconomic and lifestyle data (e.g., income, education, exercise, and health literacy), a common issue among administrative databases. Higher socioeconomic status is associated with more access to specialists and frequent check-ups, which could bias toward greater AD detection rates. 41,42 Conversely, certain lifestyle and socioeconomic factors may be associated with routine vaccination practices and lower AD risk. 43-45 Our study design mitigated potential biases related to sociodemographics and lifestyle by requiring ≥2 years of medical and pharmaceutical coverage for eligibility, adjusting for comorbidities and health care utilization, using an active-comparator design, and censoring at lapse in coverage. This study is limited by the underrepresentation of patients ≥65 in the PMPA. Only 5% of the database's patients are ≥65, despite this age group representing approximately 16% of the US population in 2019. 46 In addition, most older adults in the database are enrolled in commercially managed Medicare supplemental plans. 47 Results may, therefore, not generalize to those in traditional Medicare or the uninsured. Furthermore, 1 study found that the capture rate for medication dispensing in PMPA was particularly poor for those aged ≥65 because patients in this group often have both commercial insurance and traditional Medicare coverage, and dispensing covered entirely by traditional Medicare plans would not be captured in PMPA. 48,49 This study demonstrated a decreased risk of incident Alzheimer dementia among adults ≥65 years who received a high-dose influenza vaccine vs a standard-dose influenza vaccine, with a longer and more robust effect among women than men. Longitudinal and ideally prospective studies with diverse populations, comprehensive assessments of cognition, sociodemographics, lifestyle, and biomarkers, and follow-up periods that are comparable to the decades-long preclinical phase of AD are needed to better understand the long-term effect of influenza vaccination on cognitive health and neurodegeneration. Future investigation should evaluate the potential mechanisms underlying the apparent dose-dependent effect, including the extent to which this effect is mediated by greater protection against influenza infection or by nonmicrobial pathways (e.g., trained immunity, inflammaging). 8,50 Studies should also examine whether influenza vaccination influences clinical progression after symptom onset (e.g., MCI to AD). Understanding the mechanisms through which influenza vaccines and immunogenic enhancements influence AD pathology and presentation could inform the targeted interventions and public health strategies to mitigate the growing population burden of AD.