Disproportionality analysis of adverse event signals associated with seven Alzheimer’s drugs: insights from the FAERS database Kai Liu, Xiaohu Yang, Lingti Kong Abstract AbstractObjectiveTo compare adverse event reporting patterns for seven Alzheimer’s disease (AD) drugs (aducanumab, donanemab, lecanemab, donepezil, galantamine, rivastigmine, and memantine) using the FDA Adverse Event Reporting System (FAERS), and to explore potential reporting signals.MethodsFAERS reports from the first quarter of 2004 to the second quarter of 2026 were retrospectively analyzed. Adverse events were coded using MedDRA version 28.1 preferred terms (PTs) and system organ classes (SOCs). Disproportionality signals were detected using the reporting odds ratio (ROR), proportional reporting ratio (PRR), Bayesian confidence propagation neural network (BCPNN), and multi-item gamma-Poisson shrinker (MGPS). PT-level signals were additionally controlled using Benjamini–Hochberg false discovery rate correction. Drug-level comparisons, sex- and age-stratified analyses, time-to-onset (TTO) analysis, and Weibull distribution fitting were performed.ResultsA total of 39,301 valid reports involving the seven AD drugs were included. Reports were more frequent among women (52.3%, n = 20,537) and patients aged ≥65 years. Anti-Aβ monoclonal antibodies showed concentrated reporting signals in nervous system disorders, with amyloid related imaging abnormalities (ARIA) related events as the core PT-level signal. Donanemab and lecanemab were also associated with prominent infusion-related reaction signals. Conventional symptomatic drugs demonstrated broader multi-system reporting patterns: donepezil and galantamine were mainly associated with cholinergic cardiovascular and fall-related signals; rivastigmine with administration-site reactions and fall-related events; and memantine with confusional state, agitation, dizziness, and somnolence. Several PT-level signals were not listed in the reviewed prescribing information and were annotated as potential reporting signals. Sex- and age-stratified analyses showed broadly consistent PT-level spectra. TTO analyses revealed heterogeneous onset patterns among different drug and adverse events.ConclusionThis study utilized FAERS data to compare adverse event reporting signals for seven anti-AD drugs. Distinct reporting profiles were observed across drugs, and several potential ADE signals, including potential unlabeled reporting signals, were identified. These findings may support pharmacovigilance surveillance and hypothesis generation, but should be interpreted as reporting associations rather than evidence of causality.Keywords: adverse drug events, alzheimer’s disease drugs, disproportionality analysis, FDA adverse event reporting system (FAERS), signal Introduction Alzheimer’s disease (AD) is an insidious and progressive neurodegenerative disorder (; ; ). With population aging worldwide, the number of individuals at risk for AD and related dementias is increasing substantially: global dementia cases are projected to increase from 57.4 million in 2019 to 152.8 million by 2050 (). Pharmacological intervention remains a key approach for delaying disease progression and managing both cognitive decline and neuropsychiatric symptoms. For several decades, systemic anti-AD treatment has primarily included acetylcholinesterase inhibitors (AChEIs), such as donepezil, rivastigmine, and galantamine, and memantine, an N-methyl-D-aspartate (NMDA) receptor antagonist. These conventional agents provide symptomatic benefit at specific disease stages (). As disease-modifying therapies have emerged, clinical treatment guidelines for AD continue to evolve. New-generation biologicals, represented by amyloid-β (Aβ)-targeting monoclonal antibodies, including aducanumab, lecanemab, and donanemab, are intended for intervention during the very early stages of disease to reduce cerebral Aβ plaques and slow clinical decline (; ; ; ). Although these drugs may improve cognition or slow decline, they are also associated with clinically relevant safety concerns. AChEIs are known to increase central and peripheral cholinergic tone and can cause gastrointestinal cramping and nausea, as well as bradycardia, atrioventricular block, syncope, and a risk of falling (; ). Anti-Aβ monoclonal antibodies are known to reduce amyloid plaques, although amyloid-related imaging abnormalities (ARIA) have limited their broader clinical application (; ; ). Existing pre-marketing studies are limited by sample size, follow-up duration, and participant homogeneity, thus reducing the detection of delayed or rare adverse events. Therefore, there is a critical need for post-marketing safety evaluation including large pharmacovigilance databases (; ). FAERS, the World Health Organization global database of individual case safety reports (VigiBase), and the European Union pharmacovigilance system (EudraVigilance) provide key data sources for the mining of signals on a large scale. Based on reports of suspected adverse drug reactions, signal detection can identify emerging medication safety risks and support the safer use of medications (; ; ). Although individual FAERS-based studies have described adverse event signals for selected anti-Aβ monoclonal antibodies, direct comparative characterization across both disease-modifying monoclonal antibodies and conventional symptomatic therapies remains limited (; ; ; ; ). Moreover, comprehensive comparisons integrating subgroup analyses, PT-level temporal characteristics, and multiple signal detection algorithms remain scarce. On this basis, the present study used real-world data from the FAERS database to comparatively characterize adverse event reporting signals associated with seven current AD therapies, including three anti-Aβ monoclonal antibodies and four conventional symptomatic drugs. In particular, we aimed to provide a unified comparison of safety signals across mechanistically distinct drugs and to comprehensively evaluate reporting patterns through drug- and PT-level analyses, sex- and age-stratified analyses, time-to-onset characteristics, and multiple disproportionality algorithms. Our findings may contribute to post-marketing safety surveillance and support the rational selection of treatments and risk management (; ; ). Materials and methods Data source Data were extracted from the FDA Adverse Event Reporting System (FAERS) database covering the first quarter of 2004 through the second quarter of 2026. The FAERS data files were downloaded on 19 August 2026, which was defined as the database lock date for this analysis. The core FAERS modules were used, including DEMO for demographic and administrative information, DRUG for suspect and concomitant drug information, REAC for adverse event preferred terms, OUTC for patient outcomes, RPSR for report sources, THER for drug therapy start and end dates, and INDI for indication information. During data cleaning, a sequential de-duplication strategy was applied at the case level. For records sharing the same CASEID, the most recent report was retained according to FDA_DT; when multiple records had the same FDA_DT, the record with the largest PRIMARYID was retained. After case-level de-duplication, remaining duplicate PRIMARYIDs were further removed to ensure that each report was counted only once. The complete workflow for data extraction, cleaning, and de-duplication is shown in . Data filtering Using the FAERS database, we screened for the generic names of three anti-Aβ monoclonal antibodies (aducanumab, lecanemab, and donanemab) and four conventional agents (donepezil, rivastigmine, galantamine, and memantine), together with their corresponding brand names (Aricept, Exelon, Razadyne, Namenda, Aduhelm, Leqembi, and Kisunla). Drug names were manually harmonized across generic names, brand names, and common spelling variants. We extracted records in which the drug was coded as the primary suspect (PS) drug in the ROLE field of the database. Adverse events were described and classified according to MedDRA version 28.1 preferred terms (PTs) and system organ classes (SOCs). Statistical analysis Signal detection was performed using four disproportionality analysis algorithms: reporting odds ratio (ROR), proportional reporting ratio (PRR), Bayesian confidence propagation neural network (BCPNN), and multi-item gamma-Poisson shrinker (MGPS). For each drug–PT pair, a 2 × 2 contingency table was constructed. Because these estimators are derived from the same table and are not independent, their predefined thresholds were jointly applied as a conservative signal filter. Signals were defined as positive only when all four algorithm-specific criteria were met (). PT-level analyses were exploratory. Benjamini–Hochberg false discovery rate (FDR) correction was applied within each drug across all tested PTs, and main PT-level signals were required to meet both the four-algorithm criteria and the FDR-adjusted threshold. For sex-stratified analyses, only PTs positive in both female and male reports for the same drug were compared. Female-to-male reporting odds ratios with 95% confidence intervals were calculated from 2 × 2 tables using a 0.5 continuity correction. Fisher’s exact test was applied when any expected count was <5; Pearson’s chi-square test without continuity correction was applied when all expected counts were ≥5. P values were adjusted within each drug using the Benjamini–Hochberg method. Time to onset (TTO) was defined as the interval between the first recorded therapy start date (drug first exposure) and the first reported adverse event occurrence. Reports with valid non-negative TTO values were included in descriptive and non-parametric analyses. TTO was summarized as median and interquartile range (IQR). Between-drug differences were assessed using the Kruskal–Wallis rank-sum test, followed by pairwise Wilcoxon rank-sum tests with Benjamini–Hochberg correction. Weibull fitting was performed by maximum likelihood estimation among reports with positive TTO values. The shape parameter β was used to classify onset patterns as early-failure (β < 1), random (β = 1), or wear-out (β > 1). Analyses were performed using R version 4.5.1. Data processing used readxl, dplyr, purrr, tibble, stringr, writexl, and openxlsx; Weibull fitting used MASS; and visualization used ggplot2 and pheatmap. Categorical variables were summarized as frequencies and percentages, and continuous variables as mean ± standard deviation or median (IQR), as appropriate. All results were interpreted as reporting patterns rather than incidence or causality. Results Demographic characteristics A total of 39,301 adverse event reports related to the seven target drugs were included in the final analysis. summarizes demographic characteristics and report sources by drug. Conventional drugs accounted for higher report volumes, with rivastigmine having the largest number of reports (N = 13,261), followed by donepezil (N = 8,951) and memantine (N = 8,297). As clinical use expanded, lecanemab accumulated 3,552 reports, whereas fewer reports were available for donanemab (N = 2,780) and aducanumab (N = 515). In total, 52.3% (n = 20,537) of subjects were women, while 36.7% were men; 11.0% of all reports were missing sex information. The number of adverse event reports was higher among women than among men; this was consistent with the higher prevalence of AD among women. The reporting population was predominantly aged 65 years or older. By reporting country, the United States accounted for 35.5% of all reports, followed by Japan (11.6%) and the United Kingdom (9.5%). Signal detection at the SOC level presents SOC-level signals that met the predefined positivity criteria for all four disproportionality algorithms. The complete SOC-level results were provided in . To visualize differences in signal strength across SOCs, a heatmap of SOC-level signal intensity based on ROR values was generated (). Analysis revealed substantial differences in adverse event profiles across the seven anti-AD drugs; however, all drugs exhibited core positive signals in nervous system disorders. Monoclonal antibodies showed the most concentrated signal pattern. For aducanumab, donanemab, and lecanemab, nervous system disorders showed consistently strong signals across all four algorithms. For aducanumab, nervous system disorders showed ROR = 12.58 (95% CI: 11.18–14.16), PRR = 6.40 (χ2 = 2930.25), EBGM = 6.40 (EBGM05 = 5.68), and IC = 2.68 (IC025 = 2.52). The corresponding values were ROR = 7.25 (95% CI: 6.89–7.63), PRR = 4.77 (χ2 = 8033.55), EBGM = 4.76 (EBGM05 = 4.53), and IC = 2.25 (IC025 = 2.18) for donanemab and ROR = 7.19 (95% CI: 6.87–7.54), PRR = 4.74 (χ2 = 9419.82), EBGM = 4.74 (EBGM05 = 4.52), and IC = 2.24 (IC025 = 2.18) for lecanemab. In contrast, conventional symptomatic drugs showed broader multi-system distributions. Donepezil showed relatively strong signals for cardiac, psychiatric, and nervous system disorders. For cardiac disorders, ROR = 3.82 (95% CI: 3.66–3.98), PRR = 3.56 (χ2 = 4559.09), EBGM = 3.56 (EBGM05 = 3.41), and IC = 1.83 (IC025 = 1.77). Galantamine also showed a prominent signal for cardiac disorders, with ROR = 2.77 (95% CI: 2.49–3.09), PRR = 2.65 (χ2 = 373.56), EBGM = 2.65 (EBGM05 = 2.38), and IC = 1.41 (IC025 = 1.24). Memantine showed prominent signals for psychiatric and nervous system disorders. For psychiatric disorders, ROR = 3.65 (95% CI: 3.53–3.78), PRR = 3.19 (χ2 = 6152.60), EBGM = 3.19 (EBGM05 = 3.08), and IC = 1.67 (IC025 = 1.62). Although rivastigmine had the highest numbers of reports for general disorders and administration site conditions (8,912 reports) and nervous system disorders (7,751 reports), the SOC-level profile of this drug reflected a high-frequency distribution rather than highly concentrated disproportionality enrichment. Overall, adverse event signals for monoclonal antibodies were concentrated in the nervous system, whereas conventional drugs involved multiple systems, including cardiac, psychiatric, and gastrointestinal disorders. Signal detection at the PT level presents the overlap of PT-level positive signals identified by the four disproportionality algorithms (ROR, PRR, BCPNN, and MGPS) for each of the seven Alzheimer’s drugs. PTs located in the central intersection met the predefined positivity criteria of all four algorithms and were included in subsequent PT-level analyses. lists the top 30 most frequently reported positive signals and their signal strengths at the PT level for each of the seven drugs following the exclusion of false-positive signals related to indication-related confounding and product packaging. The complete positive PT-level results were provided in . PTs marked with an asterisk in were positive reporting signals that were not listed as adverse reactions in the reviewed prescribing information for the corresponding drug. In addition, ROR values for signals positive by all four detection methods were transformed as lg (ROR + 1), and a heatmap of PT-level positive signal strength based on ROR values was generated (; ). Monoclonal antibodies showed a concentrated pattern of target-related adverse event signals. ARIA-related PTs, including Amyloid related imaging abnormalities, Amyloid-related imaging abnormalities-edema/effusion (ARIA-E) and Amyloid-related imaging abnormalities-hemorrhage (ARIA-H) showed high disproportionality. For aducanumab, ARIA-E showed ROR = 9731.43 (95% CI: 8072.48–11731.30), PRR = 8472.20 (χ2 = 1,051,765.86), EBGM = 7356.74 (EBGM05 = 6102.62), and IC = 12.84 (IC025 = 6.87). For donanemab, the corresponding values were ROR = 7621.95 (95% CI: 6737.22–8622.86), PRR = 7068.68 (χ2 = 1,864,750.10), EBGM = 4127.06 (EBGM05 = 3648.01), and IC = 12.01 (IC025 = 8.51). Donanemab and lecanemab also showed strong infusion-related reaction signals, suggesting that cerebral microvascular injury and immunogenic responses to exogenous proteins may be important safety features of this drug class. Conventional symptomatic drugs exhibited broader peripheral cholinergic and non-specific central nervous system signals. Donepezil showed strong signals for bradycardia [ROR = 34.76 (95% CI: 32.31–37.39), PRR = 33.80 (χ2 = 23674.73), EBGM = 33.33 (EBGM05 = 30.98), and IC = 5.06 (IC025 = 4.89)], with syncope and falls also being prominent. Galantamine similarly showed a strong bradycardia signal [ROR = 13.84 (95% CI: 10.75–17.82), PRR = 13.69 (χ2 = 717.60), EBGM = 13.68 (EBGM05 = 10.63), and IC = 3.77 (IC025 = 3.14)], along with syncope. Rivastigmine showed pronounced disproportionality for application site erythema [ROR = 22.65 (95% CI: 20.59–24.92), PRR = 22.45 (χ2 = 8730.27), EBGM = 22.09 (EBGM05 = 20.08), and IC = 4.47 (IC025 = 4.26)]. Memantine was characterized by confusional state [ROR = 15.91 (95% CI: 14.87–17.03), PRR = 15.32 (χ2 = 11720.44), EBGM = 15.24 (EBGM05 = 14.25), and IC = 3.93 (IC025 = 3.81)], with dizziness and somnolence as other typical signals. In the label-comparison analysis, several asterisk-marked PTs in were identified as potential unlabeled reporting signals. These signals were detected by all four disproportionality algorithms and remained significant after FDR adjustment, but they were not described as adverse reactions in the reviewed prescribing information. Therefore, these findings may provide additional pharmacovigilance clues beyond known label information. However, because FAERS is a spontaneous reporting system, these asterisk-marked PTs should be interpreted as exploratory reporting signals. Sex-stratified PT-level signal analysis Because female reports accounted for a greater proportion of the total reports than male reports, PT-level ROR heatmaps were calculated separately for female and male reports to evaluate sex-stratified differences (). Sex-stratified PT-level analyses showed that the core signal spectra were broadly consistent with the overall PT-level results. Exploratory comparisons among PTs shared by female and male positive-signal sets identified selected sex-related reporting enrichments after False Discovery Rate (FDR) correction (). However, these differences were limited to a subset of PTs and did not materially alter the main drug-specific signal patterns observed in the overall analysis. For both female and male reports, ARIA-related events were the dominant signals for the three anti-Aβ monoclonal antibodies. Of the conventional symptomatic drugs, reports from males exhibited more prominent signals related to cholinergic syndrome and conduction abnormalities. For example, the ROR values for cholinergic syndrome were 423.58 for galantamine and 288.81 for donepezil. When considering reports from females, arrhythmia-related signals for donepezil were more prominent, including sinus bradycardia, torsade de pointes, and bradycardia (ROR = 65.27, 57.35, and 39.07, respectively). For both males and females, rivastigmine was characterized predominantly by reactions at the application site, whereas memantine was characterized predominantly by consciousness, behavioral, and neuropsychiatric symptoms. Age-stratified PT-level analysis Age-stratified analyses were performed using the ten most frequently reported PTs identified in the primary analysis for each drug (). Overall, the PT spectra remained largely consistent across adult age groups, with most differences limited to changes in signal ranking rather than the emergence of age-specific reporting patterns. Among the three anti-Aβ monoclonal antibodies, ARIA-related events, cerebral haemorrhage, headache, and infusion-related reactions remained the dominant reporting signals across adult age groups, with the 65–85 years subgroup showing the most complete and stable signal pattern. Likewise, the conventional symptomatic drugs exhibited broadly preserved PT spectra across age groups. Donepezil was characterized primarily by cardiovascular and neuropsychiatric PTs, whereas memantine remained dominated by confusional state, agitation, dizziness, and somnolence. Only modest variations in signal ranking were observed for rivastigmine and memantine, while donepezil and galantamine showed nearly identical PT profiles across adult age groups. Interpretation of the <18 years subgroup should be made with caution because of the limited number of reports and substantial missing age information. TTO analysis Analysis revealed differences in TTO between the seven anti-AD drugs (; ). Overall, memantine-associated adverse events occurred earliest, with a median TTO of 18 days. The median TTOs for donanemab, galantamine, donepezil, and lecanemab were observed during the early to mid-treatment period (36, 41, 45, and 50 days, respectively). The median TTO for rivastigmine was 76 days, while that of aducanumab was 147.5 days. TTO distributions differed significantly across the seven drugs according to the Kruskal–Wallis rank-sum test (H = 480.22, df = 6, P < 0.001; ). Pairwise Wilcoxon rank-sum tests with Benjamini–Hochberg correction were further performed to compare TTO distributions between individual drugs, and the complete pairwise results are provided in . Weibull analysis showed that most drugs had β values below 1, indicating early-failure onset patterns. Aducanumab showed a delayed or wear-out pattern, whereas donanemab showed an approximately random pattern because the 95% CI of β included 1. To further characterize onset patterns, PT-level TTO analyses were performed for the five most frequently reported PTs of each drug (). Overall, variation in onset timing was observed among PTs within the same drug. For anti-Aβ monoclonal antibodies, infusion-related reactions and constitutional symptoms generally occurred earlier than ARIA-related events. In contrast, conventional symptomatic drugs tended to show earlier onset of gastrointestinal and neuropsychiatric events, whereas cardiovascular and fall-related events occurred relatively later. PT-level Weibull analyses were broadly consistent with the drug-level findings, with most PTs exhibiting early-failure patterns (β < 1). Following sex stratification, differences in TTO between female and male reports were not significant for most drugs. Statistically significant differences were only observed for memantine (both 17 days in female and male reports, P = 0.016) and rivastigmine (70 days in female reports vs. 83 days in male reports, P = 0.043). For memantine, the median TTO has no differences in female and male, suggesting limited clinical relevance despite statistical significance. For rivastigmine, TTO was approximately 13 days shorter in female reports than in male reports, suggesting a potential difference in the timing of treatment intolerance; however, this finding should be interpreted cautiously because TTO data were available only for a subset of reports. Discussion This study used data from the FAERS database to perform signal detection for seven anti-AD drugs: aducanumab, donanemab, lecanemab, donepezil, galantamine, rivastigmine, and memantine. A total of 39,301 valid adverse event reports were included in the final analysis; our goal was to provide real-world evidence for the safety evaluation of pharmacotherapy for AD and to inform the rational selection of drugs, risk identification, and safety management for anti-dementia treatment (; ; ; ). Overall, women accounted for 52.3% of all reports, and patients aged 65 years or older constituted the predominant age population. Reports for anti-Aβ monoclonal antibodies were more concentrated among patients aged 65–85 years, whereas conventional symptomatic drugs were noted in a relatively higher proportion of reports among patients aged > 85 years. This distribution is broadly consistent with current real-world application in that monoclonal antibodies are used more often in earlier stages of disease and in more strictly selected populations, whereas AChEIs and memantine are used in broader and older dementia populations (; ; ; ). At both the SOC and PT levels, the adverse event profiles of the seven anti-AD drugs showed clear class-specific differences. Anti-Aβ monoclonal antibodies were characterized by concentrated signals involving the nervous system, whereas conventional symptomatic drugs involved multiple systems, including cardiac, neuropsychiatric, gastrointestinal, and administration-site disorders. These differences were closely related to drug targets and the treated populations. Anti-Aβ monoclonal antibodies act directly on cerebral Aβ pathology, and adverse events associated with these drugs are therefore more likely to involve the central nervous system and cerebral microvascular changes. In contrast, AChEIs and memantine predominantly improve symptoms by modulating neurotransmitter systems, which may lead to peripheral cholinergic reactions, cardiovascular conduction abnormalities, and fluctuations in neuropsychiatric status (; ). ARIA-related events were the most important safety signals for the three anti-Aβ monoclonal antibodies. ARIA-E, ARIA-H, cerebral microhemorrhage, and superficial siderosis were highly prominent for aducanumab, donanemab, and lecanemab, indicating that cerebral microvascular injury-related events were the predominant reporting signals for this class of drug. This finding is consistent with a mechanism in which the removal of perivascular amyloid deposition may induce changes in vascular permeability, cerebral edema, and microhemorrhage. Donanemab and lecanemab were also associated with strong infusion-related reaction signals, indicating that the clinical risk of anti-Aβ monoclonal antibodies is not limited to imaging abnormalities but also includes systemic immune reactions during early administration (; ; ). Therefore, safety management for anti-Aβ monoclonal antibodies should emphasize baseline Magnetic resonance imaging (MRI) assessment, serial MRI follow-up during early treatment cycles, patient education regarding ARIA warning symptoms (such as headache, confusion, visual disturbance, gait change, or seizures), and close observation for infusion-related reactions, particularly after initial administrations. The exceptionally strong and concordant ARIA-related signals, together with their biological plausibility, reinforce ARIA as the defining safety concern of anti-Aβ monoclonal antibodies. The signals observed for donepezil and galantamine reflected the systemic effects of enhanced cholinergic activity induced by AChEIs. Bradycardia, syncope, and falls may not be isolated events; instead, they may reflect a clinically relevant cluster of reporting signals in which reduced heart rate or the suppression of conduction causes syncope and subsequently fall-related injury (; ; ). These risks are clinically relevant for patients with AD because these patients are often older, have cardiovascular comorbidities or orthostatic hypotension, and receive multiple types of medication. Falls may lead to fractures, hospitalization, and functional decline. Therefore, monitoring during AChEI therapy should not be limited to gastrointestinal reactions but should also include baseline heart rate, electrocardiographic findings, conduction function, the history of syncope and falls, and a structured review of concomitant rate-limiting drugs (such as beta-blockers, non-dihydropyridine calcium-channel blockers, and digoxin). These measures may be particularly important in older adults, patients with low body weight, and those with pre-existing cardiovascular disease. Rivastigmine and memantine represented two additional monitoring priorities. The most characteristic signal for rivastigmine was reactions at the site of administration, suggesting that although transdermal patches can improve gastrointestinal tolerability in some patients, local skin irritation remains an important issue in clinical application (; ; ). Furthermore, higher reporting burdens for pneumonia, falls, and death should be interpreted cautiously in the context of longer disease duration, a poorer functional status, dysphagia, and a higher risk of infection, rather than being attributed directly to the drug. Memantine-related adverse events were more often reflected as fluctuations in consciousness and behavior, including confusional state, agitation, dizziness, somnolence, and falls. This pattern is consistent with the potential effects of NMDA receptor modulation on the central excitatory-inhibitory balance. Clinically, for rivastigmine patch therapy, routine rotation of application sites, early recognition of erythema or pruritus, and assessment of frailty-related risks such as aspiration, infection, and falls are advisable. For memantine, changes in consciousness, dizziness, somnolence, neuropsychiatric symptoms, and fall tendency should be monitored particularly during treatment initiation and dose escalation, and dose adjustment should be considered in patients with renal impairment (). The asterisk-marked PTs in provide an additional label-comparison perspective for interpreting the PT-level signal spectrum. These signals were positive in disproportionality analyses but were not listed as adverse reactions in the reviewed prescribing information for the corresponding drugs. Therefore, they may represent potential unlabeled reporting signals that deserve further pharmacovigilance attention. However, such findings should not be interpreted as newly confirmed adverse reactions, because FAERS data cannot determine incidence, causality, or the influence of disease progression, comorbidities, and concomitant medications. Similarly, low-frequency signals should be interpreted cautiously because disproportionality estimates may be unstable and should be regarded as signals requiring further validation rather than confirmed high-risk adverse events. Clinically, these signals should be regarded as hypothesis-generating findings and may help guide targeted monitoring in future observational studies or clinical practice. Subgroup analyses further indicated the existence of sex-related differences although this did not constitute a major basis for risk stratification overall (; ; ). Similarly, age subgroup analysis indicated that the principal PT spectrum of each drug remained largely consistent across age groups. Although modest rank shifts were observed, particularly for aducanumab and donanemab, no age-specific dominant reporting signals emerged. These findings suggest that age may influence the relative prominence of established adverse events rather than fundamentally altering the overall safety profile. In terms of onset time, only memantine and rivastigmine showed statistically significant differences between sex. The median TTO for memantine was both 17 days in female male reports, thus indicating limited clinical relevance. The median TTO for rivastigmine was shorter in female reports than in male reports (70 vs. 83 days, P = 0.043), thus suggesting a potentially earlier reporting onset in women, but this should be interpreted cautiously because of the inherent limitations of spontaneous reporting data. PT heatmaps also revealed that the core signal spectra were generally consistent when compared between female and male reports. ARIA-related events dominated for anti-Aβ monoclonal antibodies in both sexes, thus suggesting that risk is primarily driven by drug mechanism and cerebral microvascular pathology rather than by sex itself. Some sex-specific differences observed for conventional symptomatic drugs, such as stronger cholinergic syndrome signals in male reports and more prominent donepezil-related arrhythmia signals in female reports, may be associated with differences in body weight, frailty, baseline cardiovascular status, renal function, electrolyte status, and concomitant medications. However, these findings should be viewed as reporting signals rather than incidence-based risk estimates (). For clinical practice, closer monitoring of pulse rate and conduction abnormalities may be warranted in patients receiving AChEIs, especially when additional rate-limiting drugs are co-prescribed; women receiving rivastigmine may benefit from earlier follow-up after initiation. For memantine, careful observation of consciousness, behavioral status, dizziness, and fall risk during the early treatment period remains advisable. PT-level TTO analyses further indicated that onset timing varied across different adverse events within the same drug. Early-occurring infusion-related or gastrointestinal events may require closer monitoring shortly after treatment initiation, whereas delayed ARIA-related and cardiovascular events support continued safety surveillance during follow-up. Collectively, these findings provide not only statistical evidence for signal detection but also clinically relevant information regarding drug-specific risk profiles, subgroup consistency, and temporal characteristics that may facilitate individualized pharmacovigilance and safety monitoring. This study has certain limitations that should be considered, particularly because this study was based on a spontaneous reporting pharmacovigilance system. First, the FAERS database can allow for the identification of statistical associations between drugs and adverse events but cannot establish causality. Second, under-reporting, duplicate reports, stimulated reporting, notoriety effects after the market entry of new drugs, comorbidities, and polypharmacy can all influence the interpretation of signal strength. Although indication-related confounding and false-positive signals related to product packaging were removed as far as possible and four algorithms (ROR, PRR, BCPNN, and MGPS) were used for cross-validation to improve robustness, outcomes such as falls, confusional state, and pneumonia should still be interpreted cautiously (; ; ). Similarly, asterisk-marked signals in should be interpreted as signals requiring further validation rather than evidence of newly established adverse reactions. In addition, PT-level multiplicity was addressed using the Benjamini–Hochberg FDR procedure, but residual confounding and reporting bias cannot be eliminated in spontaneous reporting systems. Conclusion Using the FAERS database, we systematically compared adverse event signals for seven anti-AD drugs by applying four disproportionality analysis methods. Our findings indicated heterogeneity in safety profiles between drugs. Risks associated with anti-Aβ monoclonal antibodies were predominantly concentrated in the nervous system, with ARIA-related events and infusion-related reactions as core signals. Conventional symptomatic drugs were associated with broader multi-system safety signals. In addition, identified several potential unlabeled reporting signals. Collectively, our findings provide real-world reporting evidence to support individualized and risk-differentiated safety assessment and clinical use of anti-AD drugs, while acknowledging the exploratory nature and inherent limitations of spontaneous reporting data. Funding Statement The author(s) declared that financial support was received for this work and/or its publication. This work was supported by the program of State Key Laboratory of Neurology and Oncology Drug Development (No. SKLSIM-F-2025113). Footnotes Generative AI statement The author(s) declared that generative AI was not used in the creation of this manuscript. Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us. Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.