Eosinophilia-related adverse events reporting associated with medications: a disproportionality and regression analysis of the FDA Adverse Event Reporting System

Abstract

Introduction

Eosinophilia encompasses a wide spectrum of hematologic and non-hematologic disorders, and may lead to clinically significant organ damage. Although numerous medications have been implicated in eosinophilia and related hypersensitivity reactions, their overall drug-associated risk remains unclear. Real-world pharmacovigilance data can provide important insights into rare but serious eosinophilia-related adverse events (AEs). This study investigated the demographic and drug-related risk factors for eosinophilia-associated AEs using a large dataset from the US Food and Drug Administration Adverse Event Reporting System (FAERS).

Aim

To comprehensively evaluate the associations between a broad range of medications and eosinophilia-related AEs.

Method

FAERS reports from 2004 Q1 to 2025 Q1 were analyzed. Four disproportionality algorithms—reporting odds ratio, proportional reporting ratio, Multi-item Gamma Poisson Shrinker, and Bayesian Confidence Propagation Neural Network—were applied to identify drugs with significant safety signals. LASSO regression was used for variable selection, followed by a multivariate logistic regression analysis. The Bonferroni correction was applied for multiple comparisons. The time to onset of eosinophilia-related AEs was evaluated.

Results

In total, 55,456 eosinophilia-related AEs were identified. Using all four disproportionality methods, 173 drugs were significantly associated with eosinophilia. These included anti-infective medications (82/173), cardiovascular agents (20/173), nervous system drugs (16/173), antineoplastic and immunomodulatory therapies (14/173), digestive system drugs (9/173), respiratory system drugs (9/173) and other medications (23/173). LASSO followed by multivariate logistic regression identified 60 drugs that were independently associated with higher reporting odds within the FAERS database along with several demographic characteristics. Male sex and age 56–69 years were also significantly associated with higher reporting odds (adjusted p < 0.01). Among reports with available timelines, the median time to onset of eosinophilia-related AEs was 20 days (interquartile range 7–44 days).

Conclusion

This large pharmacovigilance study identified the key demographic and drug-related risk factors for eosinophilia-associated AEs. These findings highlight the need for increased clinical vigilance, especially during the early weeks of therapy and among patients with identified risk factors. Although causality cannot be established from the FAERS data, the results offer valuable real-world evidence to support the early detection, risk mitigation, and post-marketing safety monitoring of medications associated with eosinophilia.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11096-026-02109-z.

Impact statements

• This study identified several medications associated with eosinophilia-related adverse events, including agents not previously labeled for this risk, highlighting the need for strengthened post-marketing safety monitoring.

• The inclusion of demographic predictors enables clinicians to prioritize monitoring of higher-risk patients and to recognize eosinophilia-related reactions earlier in therapy.

• The analytical framework provides a practical model for detecting rare adverse drug events using real-world data, supporting safer prescribing and informing regulatory decision-making.

Introduction

Eosinophilia, defined by the World Health Organization as an absolute eosinophil count > 500/μL and hypereosinophilic syndrome (> 1500/μL on two occasions at least two weeks apart), encompasses a wide spectrum of conditions, ranging from asymptomatic laboratory abnormalities to severe, life-threatening organ damage [1]. In some cases, eosinophilic infiltration may be confined to specific tissues, causing organ injury even when peripheral eosinophil counts are normal [2]. Drug exposure is the major cause of secondary eosinophilia. Among drug-induced reactions, Drug Rash with Eosinophilia and Systemic Symptoms (DRESS) is particularly important, with an estimated incidence of 1/1000 to 1/10,000 depending on the culprit medication [3]. Although uncommon, DRESS accounts for nearly one-quarter of hospitalized cutaneous adverse drug reactions and has a mortality rate of 1–10% [4–6]. Other drug-associated eosinophilic conditions—including eosinophilic pneumonia, eosinophilic gastroenteritis, and eosinophilic granulomatosis with polyangiitis—may also result in persistent morbidity or prolonged hospitalization [7–9]. Therefore, early identification of eosinophilia-related adverse reactions is essential for preventing irreversible harm.

Post-marketing surveillance systems, such as the US Food and Drug Administration Adverse Event Reporting System (FAERS), play a key role in detecting rare, serious, and unexpected adverse events (AEs) [10, 11]. Because eosinophilia-related reactions are infrequent and often unsuitable for randomized controlled trials, FAERS provides valuable real-world data to support early safety signal detection and guide medication risk assessment in clinical practice.

Previous pharmacovigilance studies have typically focused on a single phenotype (e.g., DRESS) or a narrow drug class (e.g., antibiotics and antiseizure medications), limiting their generalizability [12–14]. To date, no study has systematically evaluated eosinophilia-related AEs across all medications or integrated demographic characteristics into a unified risk-factor framework.

Aim

This study aimed to comprehensively evaluate the associations between a broad range of medications and drug-induced eosinophilia-related AEs by analyzing the FAERS database using disproportionality methods and multivariate regression modeling.

Method

Data source and case identification

FAERS data from the first quarter of 2004 to the first quarter of 2025 were downloaded from the US FDA website. Duplicate reports were removed according to FDA recommendations. When multiple records shared the same CASEID, the report with the most recent FDA_DT was retained. When both CASEID and FDA_DT were identical, the record with a larger PRIMARYID was preserved.

Eosinophilia-related AEs were identified based on predefined MedDRA-preferred terms (Supplemental Table S1) (MedDRA version 27.1). Only reports that designated the medication as the primary suspected drug were included in the analysis.

Disproportionality analysis

Associations between individual drugs and eosinophilia-related AEs were assessed using four established signal detection algorithms: reporting odds ratio (ROR), proportional reporting ratio (PRR), Multi-item Gamma Poisson shrinker (MGPS, expressed as EBGM), and Bayesian Confidence Propagation Neural Network (BCPNN, expressed as an information component [IC]). All calculations were based on 2 × 2 contingency tables, and the corresponding formulas and significance thresholds are provided in Supplemental Table S2. Statistical significance was evaluated using Bonferroni-adjusted p values derived from chi-square or Fisher’s exact tests. To minimize false-positive signals, only associations meeting all predefined criteria across the ROR, PRR, EBGM, BCPNN, and Fisher’s exact test were considered meaningful. Volcano plots were generated to visualize the signal intensity.

Data preparation

Reports containing complete information on age, sex, and weight were extracted for regression analyses. Patients with implausible values, defined as age greater than 120 years or weight greater than 400 kg, were also excluded. The resulting dataset was used in a multistage modeling framework designed to evaluate the associations between a broad range of medications and eosinophilia-related AEs.

Variable selection and regression modeling

The first stage involved identifying candidate drugs using disproportionality analysis. Medications were retained only if they met all of the following criteria: ROR > 1 with a lower 95% confidence interval (CI) above 1 and p-adjust < 0.01; PRR > 2 with χ² > 4 and a lower 95% CI > 1; EBGM05 > 2; IC025 > 0, and at least 100 associated reports.

In the second stage, Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression was applied to the candidate drugs to identify the most predictive variables. Because LASSO selects binary predictors, demographic variables were not included at this stage. A tenfold cross-validation procedure was used to determine the optimal penalty parameter (λ). LASSO modeling was performed using the glmnet package (version 4.1-10) in R 4.4.2, with α = 1, standardization enabled, a convergence threshold of 1 × 10⁻⁶, and a maximum iteration limit of 10,000.

In the final stage, drugs selected by LASSO were entered into a multivariable logistic regression model along with demographic covariates (age, sex, and weight). Stepwise regression using the Akaike Information Criterion (AIC) was performed to identify independent predictors. Model performance was assessed using the area under the receiver operating characteristic curve (AUC).

Sensitivity analysis

Owing to the heterogeneity among the included PT terms, a sensitivity analysis was performed on the primary outcomes. After excluding 'Drug Reaction with Eosinophilia and Systemic Symptoms' (DRESS) from the PT terms, the signal detection analysis was re-conducted. Medications that tested positive across all four detection methods were compared with the primary results to verify the stability of the findings.

Statistical analysis

Descriptive statistics were used to summarize patient characteristics. Bonferroni correction was applied for all multiple comparisons. Disproportionality analyses, LASSO regression, logistic regression, and visualization were performed using R, version 4.4.2.

Ethics approval

This study used publicly available, de-identified FAERS data. As no identifiable patient information was accessed, institutional ethics approval was not required.

Results

Characteristics of eosinophilia-related reports

The workflow is presented in Supplemental Fig. S1, a total of 55,456 FAERS reports involving eosinophilia-related AEs were identified between 2004 Q1 and 2025 Q1. The MedDRA-preferred terms (PTs) covered 30 eosinophilia-related conditions, including eosinophilia, eosinophilic pneumonia, eosinophilic fasciitis, eosinophilic hepatitis, eosinophilic myocarditis, and DRESS (Supplemental Table S1).

The patient demographics and characteristics are presented in Table 1. The median age was 56 years (inter-quartile range [IQR] 37–69 years), with 44.7% female and 44.1% male. Weight information was available for 21.7% of the patients (median, 70 kg). Nearly half (49.3%) of the patients required hospitalization and 5.3% died. Reports primarily originated from the United States (23.5%), France (22.6%), Japan (9.5%), and Canada (7.6%). The reported occupations were physicians (41.7%), other healthcare professionals (15.6%), pharmacists (6.5%), and consumers (10.3%).

Table 1.

Basic characteristics

Characteristics	Overall (n = 55,456)
Age, years
Median (Q1, Q3)	56.0 (37.0, 69.0)
Unknown	10,140 (18.3%)
Gender
Female	24,772 (44.7%)
Male	24,473 (44.1%)
Unknown	6211 (11.2%)
Weight, kg
Median (Q1, Q3)	70.0 (58.0,85.0)
Unknown	43,408 (78.3%)
Outcome
Hospitalization (Initial or Prolonged)	27,362 (49.3%)
Disability	292 (0.5%)
Life-Threatening	4435 (8.0%)
Death	2942 (5.3%)
Required Intervention	98 (0.2%)
Congenital Anomaly	27 (0.0%)
Other Serious (Important Medical Event)	17,381 (31.3%)
Unknown	2919 (5.3%)
Occupation of the reporter
Physician	23,099 (41.7%)
Pharmacist	3584 (6.5%)
Other health-professional	8628(15.6%)
Lawyer	120 (0.2%)
Consumer	5696 (10.3%)
Unknown	14,449 (26.1%)
Country of the reporter
United States	13,013 (23.5%)
France	12,551 (22.6%)
Japan	5249 (9.5%)
Canada	4237 (7.6%)
United Kingdom	3005 (5.4%)
Spain	2003 (3.6%)
Italy	1361 (2.5%)
Germany	1209 (2.2%)
Portugal	896 (1.6%)
Australia	795 (1.4%)
Korea	761 (1.4%)
Netherlands	746 (1.3%)
Switzerland	724 (1.3%)
India	596 (1.1%)
Other countries	8310 (15.0%)

Temporal, demographic, diagnostic, and clinical outcome distributions are shown in Supplemental Fig. S2. Annual reporting has increased steadily over time, with higher reporting in adult and older adult age groups and a substantial proportion involving hospitalization, life-threatening events, or death. With respect to the reported indications, the five most common were asthma, epilepsy, schizophrenia, seizures, and hyperuricemia.

Disproportionality analysis

A total of 173 drugs demonstrated significant signals in the disproportionality analysis (Supplemental Table S3). Drug classifications are summarized in Table 2, showing that anti-infective medications comprised the largest proportion (82 drugs; 47.4%), followed by cardiovascular (11.6%), nervous system (9.3%), antineoplastic/immunomodulatory (8.1%), digestive (5.2%), and respiratory (5.2%) system agents. Sixty-four of them had more than 100 associated reports.

Table 2.

Classification of drugs associated with eosinophilia-related adverse events

Drug classification	Quantity (percentage)
Anti-infective medication	82 (47.40%)
Cardiovascular system medication	20 (11.56%)
Nervous system medication	16 (9.25%)
Antineoplastic and immunomodulatory medication	14 (8.09%)
Digestive system medication	9 (5.20%)
Respiratory system medication	9 (5.20%)
Other medication	23 (13.29%)

Drugs were grouped into therapeutic categories based on their primary clinical use for descriptive purposes

Figure 1 shows a distinct cluster of high-signal medications in the upper-right quadrant of the volcano plot, including clozapine, lamotrigine, vancomycin, daptomycin, allopurinol, carbamazepine, montelukast, and dupilumab. These drugs exhibited large disproportionality estimates and strong statistical significance, indicating a robust association with eosinophilia-related AEs. The darker dot coloration further reflects a higher reporting frequency. These patterns confirm the strong eosinophilic risk profiles of several antibiotics, antiepileptic drugs, antipsychotics, leukotriene receptor antagonists, and type-2 inflammation biologics.

Fig. 1.

Volcano plot of drug-associated eosinophilia-related adverse events. Each point represents the drug. The x-axis shows the log-transformed reporting odds ratio (ROR), and the y-axis shows the –log₁₀(p-adjusted) value. A positive value indicates a significant difference. The color of each dot reflects the logarithm of the number of case reports, with darker colors indicating a higher number of reports. Accordingly, the drugs located in the upper-right region of the plot demonstrate both strong signal intensity and marked differences. ROR, reporting odds ratio

Variable selection using LASSO regression

To refine the list of candidate drugs for multivariable modeling, all 64 signal-positive medications were entered into a tenfold cross-validated LASSO logistic regression model. At the optimal penalty parameter (λ), the model retained 60 drugs with the strongest predictive contribution to eosinophilia-related AEs. These drugs were classified as antibiotics (23/60), antiviral agents (3/60), nervous system drugs (9/60), antineoplastic and immunomodulatory therapies (7/60), digestive system drugs (6/60), respiratory system drugs (2/60), cardiovascular system drugs (6/60), or other medications (4/60). Hydroxychloroquine, doxycycline, budesonide, and azathioprine were excluded from the model. These LASSO-selected drugs corresponded to the major therapeutic categories identified in the disproportionality analysis, indicating the structural consistency between the two analytical approaches. The coefficient paths and optimal λ selections are shown in Fig. 2.

Fig. 2.

Results of the LASSO regression analysis. LASSO, least absolute shrinkage and selection operator

Multivariable logistic regression

A multivariable logistic regression model was constructed, including 60 LASSO-selected drugs along with demographic covariates (age, sex, and weight). Male sex, age 56–69 years, and lower reported body weight (< 58 kg) were associated with higher reporting odds of eosinophilia-related AEs in FAERS (p < 0.01).

All the 60 drugs selected by LASSO retained statistical significance in the final multivariate logistic regression model. Several medications demonstrated particularly strong associations with eosinophilia-related AEs, including antibiotics such as vancomycin, daptomycin, cefotaxime, piperacillin–tazobactam, sulfamethoxazole–trimethoprim, rifampicin, meropenem, and linezolid; antiepileptic agents such as carbamazepine and lamotrigine; proton pump inhibitors (PPIs) such as pantoprazole, the urate-lowering agent allopurinol; and biologics targeting type-2 inflammation, including dupilumab, benralizumab, and mepolizumab. A forest plot of the adjusted odds ratios and 95% CIs is shown in Fig. 3.

Fig. 3.

Forest plot of the multivariable logistic regression analysis identifying drugs with higher reporting odds for eosinophilia-related adverse events. CI, confidence interval; OR, odds ratio; P-adjusted, p value (P-adjust < 0.01 indicates statistical significance)

Sensitivity analysis

After excluding DRESS from the included PTs, the drugs that tested positive across all four disproportionality analyses are presented in Supplemental Table S4. Compared with the 60 drugs identified in the primary analysis, seven drugs (ibuprofen, esomeprazole, telaprevir, lansoprazole, metronidazole, moxifloxacin, and oxcarbazepine) were no longer significant in the sensitivity analysis. However, the signals for these drugs were not entirely negative; their RORs remained positive, whereas one or more of the other three detection methods yielded negative results, suggesting a potentially weaker signal strength. This also indicates that these drugs are associated with a higher proportion of DRESS cases than other PTs, a factor that should be considered when interpreting the primary results.

Model performance

The final multivariable logistic regression model demonstrated good discriminatory capability. The ROC curve (Supplemental Fig. S3) yielded an AUC of 0.764, indicating a good model discrimination for reporting.

Collinearity assessment

Variance inflation factor (VIF) values are presented in Supplemental Table S5. All VIFs were below five, indicating that multicollinearity in the multivariable logistic regression model was acceptable.

Time to onset of eosinophilia-related adverse events

Among the reports with available onset dates (those with missing information or implausible values were excluded from the analysis), the median time to onset was 20 days (IQR 7–44). Nearly 75% of eosinophilia-related AEs occur within 45 days of drug initiation. A small subset had an onset several months to years later, suggesting possible delayed hypersensitivity reactions or a prolonged therapy course.

The distribution of the onset times is shown in a violin plot (Fig. 4A), whereas the cumulative incidence curve (Fig. 4B) illustrates a rapid early rise in cases following exposure.

Fig. 4.

Time interval between drug exposure and onset of eosinophilia-related adverse events. A Violin plot showing the distribution of time to onset across all reports. B Cumulative incidence curve of eosinophilia-related adverse events following drug initiation

Discussion

Principal findings

This study presents the first comprehensive pharmacovigilance assessment of drug-induced eosinophilia-related AEs using FAERS data collected over 20 years. Sixty drugs across anti-infective, antiepileptic, antineoplastic/immunomodulatory, cardiovascular, digestive, and respiratory classes, as well as male sex and age 56–69 years, were independently associated with higher reporting odds of eosinophilia-related AEs within the FAERS database. Owing to the substantial amount of missing data and the inherent limitations of data collection in the FAERS database, the interpretation of results related to body weight requires increased caution. A median onset time of 20 days highlights the importance of close monitoring after initiating high-risk medications.

Although FAERS cannot confirm causality, consistent signals across methods and clinically plausible associations offer valuable guidance for real-world risk mitigation and post-marketing safety surveillance.

Trends in eosinophilia-related adverse events

Eosinophilia-related AE reports increased steadily over time, with a sharp rise in 2017–2018 following the introduction of dupilumab, a type-2 inflammation biologic known to cause eosinophilia and occasional organ involvement [15]. The decline in 2020 is likely due to reduced antibiotic use during the COVID-19 pandemic [16]. As biologics and targeted therapies continue to expand, vigilance for eosinophilia-related AEs is necessary.

Hospitalization was the most common outcome (49.3%), and 5.3% of cases resulted in death. Many patients require drug discontinuation and corticosteroid therapy [6], although eosinophilia associated with agents such as dupilumab may be asymptomatic in most cases, with only a minority progressing to organ involvement [17]. While the overall mortality rate of eosinophilia-related AEs remains undefined, the 1–10% mortality rate reported for DRESS [6] highlights the importance of early detection and appropriate management.

Demographic risk factors

Male sex had higher reporting odds, despite similar reporting numbers for men and women. This aligns with previous studies on hypereosinophilic syndromes, which showed a higher risk in males [18, 19], although the underlying mechanisms are unclear. Age was also strongly associated with risk, with individuals aged 56–69 and ≥ 70 years showing significantly higher reporting odds, consistent with DRESS studies, in which most patients were between 50 and 75 years of age [20], likely reflecting immune senescence and age-related metabolic changes [21].

Lower body weight (< 58 kg) was associated with higher reporting odds; however, this finding should be interpreted cautiously, as it may reflect frailty, comorbidity burden, or reporting artifacts rather than a direct biological effect. Prior studies have suggested a higher DRESS risk in immunocompromised or renally impaired patients [22] and demonstrated that both very low and very high BMI adversely affect immune function [23]. Therefore, although low body weight (< 58 kg) was associated with higher reporting odds, it was not categorized as a high-risk factor. In clinical practice, risk assessment should be individualized and include additional patient-specific factors, such as BMI and renal function.

Drug classes associated with higher risk

Antibiotics accounted for the largest proportion of the associated drugs. Strong and consistent signals were observed for vancomycin, daptomycin, piperacillin–tazobactam, sulfamethoxazole–trimethoprim, minocycline, and amoxicillin, which are known triggers for DRESS and eosinophilic pneumonia. A previous study on DRESS identified vancomycin, amoxicillin, sulfamethoxazole–trimethoprim, and piperacillin–tazobactam as high-risk agents [12]. Our findings also highlight daptomycin, which is more frequently associated with eosinophilic pneumonia than DRESS and is considered one of the drugs most likely to induce eosinophilic pneumonia [7]. One proposed mechanism is that daptomycin binds irreversibly to pulmonary surfactants, activates macrophages, and triggers an inflammatory cascade that leads to eosinophil recruitment [13]. Male sex, high doses, prolonged treatment duration, and hemodialysis have also been identified as risk factors for daptomycin-induced eosinophilic pneumonia [24]. These data support close monitoring and early intervention (e.g., drug withdrawal and systemic corticosteroids) when eosinophilia develops during antibiotic therapy.

Among nervous system drugs, the antiepileptic agents carbamazepine, levetiracetam, zonisamide, and lamotrigine showed strong positive signals. Previous studies have demonstrated a strong association between several antiepileptic drugs and DRESS [14], including phenobarbital, which showed a positive signal in our study, but was not included in the final regression model because of its small sample size (< 100). Numerous human leukocyte antigen (HLA) alleles have been implicated in the pathogenesis of antiepileptic-induced DRESS: HLA-B15:02 and HLA-A31:01 increase the risk associated with carbamazepine and oxcarbazepine [25–28], HLA-B51:01 and HLA-B56:04 are associated with PHT-induced DRESS [29, 30], and HLA-A01:01 and HLA-B13:01 have been linked to phenobarbital reactions [31]. Pretreatment HLA screening, which is already recommended for certain populations, may reduce the risk of severe eosinophilic hypersensitivity reactions.

Dupilumab, mepolizumab, and omalizumab, all of which target the type-2 inflammation pathway, are strongly associated with eosinophilia. Although these drugs are designed to reduce eosinophilic inflammation, paradoxical eosinophilia has been reported in patients receiving these treatments. One proposed mechanism for dupilumab involves reduced expression of vascular cell adhesion molecule-1 (VCAM-1) after interleukin (IL)-4/IL-13 blockade, which decreases eosinophil migration into tissues and thereby increases circulating eosinophil counts [32]. Dupilumab-associated eosinophilic pneumonia has also been reported, and lung biopsy findings have consistently shown eosinophilic infiltration [33]. Eosinophilia has similarly been noted with mepolizumab and omalizumab, although they target different pathways and the exact mechanisms remain unclear [34]. Expert guidance recommends that patients receiving dupilumab with sustained eosinophilia (> 3000 cells/µL for ≥ 3 months or a single value > 5000 cells/µL) undergo further evaluation, including echocardiography and chest imaging, to assess organ involvement [35]. With the increasing use of type-2 inflammation–targeted biologics, clinicians should be particularly attentive to eosinophil trends in treated patients.

Among the other medications, allopurinol deserves special attention. It showed one of the strongest positive signals and the largest number of eosinophilia-related reports. Allopurinol-induced DRESS was first reported in 1977 [36], and subsequent studies have linked this reaction to HLA-B*58:01 hypersensitivity [37]. Patients with renal impairment are at a heightened risk of multi-organ involvement [38]. Therefore, the benefits and risks of allopurinol should be carefully evaluated in high-risk populations.

The predominant gastrointestinal drugs in our dataset were PPIs and sulfasalazine. Previous reports have described associations between pantoprazole, omeprazole, lansoprazole, and DRESS [39, 40], and STAT6 variants may increase susceptibility to eosinophilic esophagitis during long-term PPI use [41]. Sulfasalazine, used to treat rheumatoid arthritis and inflammatory bowel disease, has also been associated with eosinophilia-related AEs. In one series, female patients were slightly more common, with renal and pulmonary involvement occurring in approximately 30% and 25% of cases, respectively [42]. These findings demonstrate the need to monitor eosinophil counts as well as renal and pulmonary function in patients receiving sulfasalazine.

Respiratory drugs, particularly salbutamol and montelukast, have also been used. Salbutamol has been associated with increased eosinophil counts, possibly through the β₂-adrenergic receptor–mediated induction of interleukin-8 in airway epithelial cells and macrophages, which in turn promotes eosinophil release from the bone marrow and migration to airway tissues [43, 44]. Montelukast and other leukotriene receptor antagonists have been associated with eosinophilia and eosinophilic granulomatosis with polyangiitis (EGPA) [45]. One hypothesis is that blockade of cysteinyl leukotriene receptors 1 and 2 (CysLT₁_/₂) shifts leukotriene profiles toward leukotriene B₄ (LTB₄), a potent eosinophil chemoattractant that may contribute to systemic vasculitis [46]. However, other researchers have suggested that many patients may have had subclinical EGPA before drug exposure and that leukotriene receptor antagonists may unmask, rather than cause, the disease [46]. Regardless, these agents appeared to increase the risk of eosinophilia-related AEs, warranting close monitoring.

Among cardiovascular drugs, amiodarone and enoxaparin exhibited the strongest signals. Amiodarone, class III antiarrhythmic and vasodilator, is limited by its potential for pulmonary toxicity, including eosinophilic pneumonia, characterized by increased eosinophils in bronchoalveolar lavage fluid [47]. The risk is related to the cumulative dose and treatment duration, and patients receiving high-dose or long-term therapy require particular attention [48]. Enoxaparin has been reported to induce hypereosinophilic syndrome or DRESS, although these events are rare. The mechanism remains unclear, and existing evidence is largely based on case reports; however, this association has been recognized in the package insert [49, 50].

Time to onset and clinical implications

The median onset time of eosinophilia-related AEs was 20 days (IQR 7–44 days), and a small proportion of cases occurred several years after treatment initiation. Previous DRESS studies have reported a similar median onset time of 20 days [20], and daptomycin-induced eosinophilic pneumonia had a median onset of approximately 19 days [51]. These findings suggest that for high-risk medications, close attention should be paid to patient symptoms during the first month of treatment, while continued vigilance is also warranted during long-term therapy. However, time-to-onset estimates derived from the FAERS database are subject to reporting bias inherent to spontaneous reporting systems, and may not accurately reflect the true latency of AEs. Consequently, more definitive conclusions require integration with data from clinical trials and electronic health records.

Clinical and pharmacovigilance implications

Although eosinophilia-related AEs are uncommon with most medications, their potential severity and associated mortality require heightened monitoring. In addition to well-established high-risk medications, such as allopurinol and various antibiotics, our analysis identified positive safety signals for dupilumab, mepolizumab, and omalizumab. Therefore, clinicians should maintain heightened vigilance for eosinophilia-related AEs when prescribing these agents. When symptoms such as fever, rash, dyspnea, or unexplained eosinophilia occur, careful review of the patient’s medication history, along with appropriate laboratory evaluation, is warranted. Early identification of the suspected causative drug, prompt discontinuation, and timely clinical intervention are essential for preventing disease progression.

The multivariate model developed in this study provides a structured framework for prioritizing pharmacovigilance signals and identifying patient and drug characteristics associated with higher reporting likelihood. By integrating demographic and drug-related risk factors across multiple eosinophilia-related AEs, this study supports proactive monitoring and early detection of adverse reactions in clinical and post-marketing safety settings.

Limitations

This study has several limitations. First, FAERS is a spontaneous reporting system prone to underreporting, duplicate submissions, reporting bias, stimulated reporting, and variable data completeness. Second, the absence of height data prevented BMI assessment, limiting the interpretation of the weight-related findings. Third, the MedDRA PT hierarchy does not include “EOSINOPHIL COUNT ABNORMAL” potentially omitting relevant events coded under more specific terms. Fourth, important confounders, including race, comorbidities, concomitant medications, dosage, and treatment duration, were inconsistently reported and could not be fully adjusted. Fifth, there is potential for channeling bias. For example, clinicians may be more likely to prescribe type 2 inflammation–targeted therapies, such as dupilumab, to patients with preexisting elevated eosinophil levels, which could artificially inflate the observed safety signals for these agents. Therefore, further investigation into the underlying pharmacological mechanisms is needed to clarify the causal relationships between these therapies and reported AEs. Sixth, substantial heterogeneity existed among the included MedDRA-PTs. Caution is therefore warranted when interpreting the results, as the identified drugs may not be strongly associated with all included terms. Certain medications may exhibit stronger associations with specific eosinophilia-related phenotypes while showing little or no association with others. Finally, FAERS reports are unverified, and both disproportionality analyses and logistic regression reflect associations within spontaneous adverse event reports and does not represent population-level risk estimation or causal inference.

Conclusion

This large-scale pharmacovigilance analysis identified key demographic and drug-related risk factors for eosinophilia-related adverse events in FAERS. Male sex, age of 56–69 years, and 60 medications were independently associated with higher reporting odds within the FAERS database. Most events occur within the first few weeks of treatment, highlighting the need for early monitoring when initiating high-risk drugs. These findings provide valuable real-world evidence to support clinical decision making and strengthen post-marketing safety surveillance for eosinophilia-related adverse reactions.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Haimo Liu contributed to the conceptualization, methodology, software development, formal analysis, data visualization, and writing of the original draft. Xingang Li was responsible for conceptualization, supervision, project administration, and critical review and editing of the manuscript. Sun and Song performed the data curation, investigation, validation, and formal analysis. Sheng Cheng, Yin Liao, Weina Wang, and Fang Yi contributed to writing, reviewing, and editing, and Fang Yi and Xingang Li provided the final approval of the manuscript.

Funding

This study was supported by grants from the Beijing Natural Science Foundation (Z230021).

Data availability

The datasets generated and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval

This study used publicly available de-identified data from the FAERS database. Because no patient-identifiable information was accessed, institutional ethics approval was not required.