Sweet’s syndrome (SS), also known as acute febrile neutrophilic dermatosis, was first described by Dr. Robert Douglas Sweet in 1964 (1). It is a rare inflammatory dermatosis characterized clinically by the abrupt onset of fever and painful erythematous plaques or nodules, typically affecting the upper limbs, trunk, and head or neck, and histopathologically by a dense neutrophilic infiltrate in the upper dermis (2). The pathogenesis of SS is considered multifactorial, involving dysregulated cytokine signaling, immune dysregulation, genetic susceptibility, and various external triggers, including infections, malignancies (particularly hematologic), autoimmune disorders, and pharmacologic agents (3). Based on etiologic factors, SS is classified into three clinical subtypes: classical (idiopathic), malignancy-associated, and drug-induced Sweet’s syndrome (DISS). Among these, DISS represents an especially important yet underrecognized subtype that presents unique diagnostic challenges and distinct therapeutic implications. A definitive diagnosis of DISS requires fulfillment of all of the following criteria: (A) abrupt onset of painful erythematous skin lesions; (B) histopathologic evidence of dense neutrophilic infiltration without leukocytoclastic vasculitis; (C) fever >38°C; (D) a clear temporal relationship between drug exposure and symptom onset, or recurrence upon rechallenge; and (E) resolution of lesions following drug withdrawal or treatment with systemic corticosteroids (4). To date, an increasing number of drugs have been implicated in the development of DISS, with granulocyte colony-stimulating factors (G-CSFs), azathioprine, all-trans retinoic acid (ATRA), and Fms-like tyrosine kinase 3 (FLT3) inhibitors among the most frequently reported (3, 5). Notably, clinical improvement and symptom resolution can often be achieved by discontinuation of the suspected agent alone. However, the current understanding of drug-related risk in DISS remains limited, as most data are derived from individual case reports or small case series (3, 5–7). The absence of large-scale, systematic investigations limits accurate assessment of the incidence and causality of DISS across different drug classes. Therefore, real-world pharmacovigilance data are urgently needed to expand our knowledge of drug-induced triggers and to improve early recognition and clinical management.

The U.S. Food and Drug Administration Adverse Event Reporting System (FAERS) is one of the largest publicly available pharmacovigilance databases, designed to collect reports on adverse drug events (ADEs), medication errors, and product quality issues related to both pharmaceutical drugs and biologic therapies. FAERS plays a critical role in post-marketing surveillance, supporting signal detection, risk evaluation, and regulatory decision-making (8). It enables identification of potential drug safety concerns that may not be evident during pre-approval clinical trials.

To our knowledge, this is the first FAERS-based study to systematically evaluate drugs associated with DISS while accounting for polypharmacy and key confounding factors. Disproportionality analysis, least absolute shrinkage and selection operator (LASSO) regression, and multivariable logistic regression were applied, supplemented by sensitivity analyses, external assessment using VigiBase, targeted literature review, and product label examination, to generate robust real-world evidence. These findings provide valuable evidence to support early identification and prevention of DISS and offer critical insights to inform clinical risk stratification and pharmacovigilance strategies.

DISS is an uncommon but clinically significant neutrophilic dermatosis, typically characterized by acute febrile onset and potential systemic involvement (5). However, early recognition remains challenging due to its heterogeneous clinical manifestations and the broad spectrum of implicated drugs. In the present study, a comprehensive pharmacovigilance analysis was conducted using FAERS data from 2004 to 2024 to clarify drug-associated risks. A total of 2,018 eligible reports were identified, involving 342 suspected drugs. Disproportionality analysis was employed as the primary signal detection method, resulting in 90 drugs disproportionately reported in association with DISS. To reduce confounding from SS-related comorbidities, the main analysis was restricted to reports without malignancy- or immune-related indications. Sensitivity analyses were subsequently performed by reintroducing these conditions. Across all models, consistent results were obtained through LASSO regression and multivariable logistic analysis, with substantial overlap in identified signals, thereby reinforcing the robustness and specificity of the associations. External validation was further conducted using the VigiBase database, supplemented by a targeted literature review and product label examination. This multi-step approach not only confirmed previously recognized associations but also revealed underreported or novel drug-related risks, underscoring the utility of pharmacovigilance data in characterizing the safety profile of DISS.

Regarding the baseline characteristics of DISS, our analysis revealed that the majority of cases occurred in middle-aged and older adults, with a marked female predominance—an epidemiological pattern consistent with previous reports (28, 29). Due to a high proportion of missing data on body weight in the FAERS database, further evaluation of constitutional risk factors was limited. Nearly half of the reports were submitted by medical doctors, and a disproportionately number originated from the United States. This geographic bias likely reflects reporting bias inherent to the FAERS database, as well as differences in pharmacovigilance systems, drug accessibility, and healthcare practices across countries. The most common underlying indications were inflammatory bowel disease (IBD; including CD and UC), hematologic malignancies (such as AML, plasma cell myeloma, and MDS), and RA. These conditions have been well-documented: SS is recognized as a cutaneous extraintestinal manifestation (EIM) of IBD (30), AML and MDS are established predisposing factors (31), and SS manifestations have been observed in RA (32). Given these associations, the potential confounding effects of underlying diseases and their treatments should be carefully considered when interpreting positive drug signals in DISS. In line with the nature of SS as an idiopathic, chronic systemic inflammatory response syndrome, most DISS cases in our cohort were associated with prolonged hospitalization or other non-life-threatening complications, while fatal or disabling outcomes were rare. The majority of cases occurred within 0–30 days after drug exposure, and clinical improvement was commonly observed after discontinuation of the suspected agent (33). Systemic corticosteroids are generally considered the first-line therapy, whereas potassium iodide or colchicine may serve as alternative options for patients with contraindications to corticosteroids (3).

DISS-associated drugs span a broad spectrum of pharmacological categories. Antineoplastic and immunomodulating agents accounted for the highest number of cases and the greatest diversity of implicated drugs. These agents may contribute to the pathogenesis of SS through disruption of immune homeostasis, cytokine dysregulation, and enhanced neutrophil maturation and proliferation (6). Anti-infective agents, including those for systemic and gastrointestinal use, also constituted a substantial proportion of DISS cases. The proposed mechanism involves cytokine-mediated hypersensitivity reactions (26) triggered either by infectious pathogens, such as Yersinia enterocolitica (34), or by the pharmacological activity of certain antibiotics (33), including sulfamethoxazole/trimethoprim (4) and ciprofloxacin (35). However, distinguishing infection-related from drug-induced remains a challenge with spontaneous report data. Several other drug classes were also frequently reported, including the antihypertensive hydralazine (36), the diuretic furosemide (37), the gastrointestinal agent mesalazine (20), the antiepileptic drug gabapentin (38) and the analgesic acetaminophen (17). Despite unclear mechanisms, growing case-based evidence suggests their potential role in DISS and highlights the need for further investigation.

To address potential confounding factors, a structured analytical approach was employed, comprising a main analysis and two complementary sensitivity analyses. Following initial signal detection through disproportionality analysis, drugs with positive associations—along with age and sex—were entered into LASSO regression, followed by multivariable logistic regression. In the main analysis, which excluded reports involving malignancy- or immune-related indications, 24 drugs were found to be significantly associated with DISS. The two sensitivity analyses, which reintroduced each exclusion criterion independently, identified 26 and 37 significant signals, respectively. Notably, 22 drugs were consistently detected across all three models, indicating robust and stable associations unlikely to be driven by indication-related bias.

To further support our findings, a cross-database consistency assessment was performed in VigiBase, where 10 of the 24 drugs identified in the main model were observed as positive signals. This partial overlap may reflect complementary signal detection across pharmacovigilance systems, adding contextual value to signal interpretation. However, this consistency should be interpreted with caution, as previous comparative studies have reported considerable redundancy among international pharmacovigilance databases such as EudraVigilance Data Analysis System (EVDAS), FAERS, and VigiBase, with signal overlap rates ranging from 85% to 97% (39). In addition, 15 drugs had prior documentation in published case reports or reviews, and 9 drugs explicitly listed SS in the adverse reaction section of their regulatory product labels. Taken together, these external sources provide supportive—but not confirmatory—evidence, which enhances the plausibility of the detected associations and may help guide future mechanistic research and clinical risk evaluation.

In the main analysis, we identified 24 drugs associated with DISS. Among anti-infective agents for systemic use, sulfamethoxazole/trimethoprim, levofloxacin, and amoxicillin (16) have been consistently reported in the literature as associated with SS, while ofloxacin (23) has been sporadically implicated. Notably, valacyclovir, linezolid, and azithromycin were also identified as positive signals, suggesting previously unrecognized role in DISS. More than half of the identified drugs were classified as antineoplastic and immunomodulating agents. Among hypomethylating agents commonly used to treat hematologic malignancies such as MDS, azacitidine has frequently been associated with DISS (10), while decitabine has recently emerged as a potential trigger (21), with its underlying mechanisms still unclear. G-CSF agents, such as filgrastim and pegfilgrastim, are the most commonly reported drugs associated with DISS. They are thought to excessively amplify the inflammatory response by promoting the proliferation, maturation, and activation of neutrophils, thereby substantially increasing the risk of SS following exogenous administration (6). Ruxolitinib, a Janus kinase (JAK) 1/2 inhibitor, has also been recently implicated in the onset of SS, particularly in patients with myelofibrosis (19). Azathioprine has been widely recognized as a significant precipitating factor, especially in patients with steroid-dependent or refractory IBD (13). In these cases, SS is often misdiagnosed as a flare of the underlying disease (13); however, oral rechallenge tests have substantiated a causal relationship between azathioprine and DISS (40), although the exact immunopathological mechanisms remain unclear. Additionally, mesalamine, a first-line therapeutic agent for IBD, has been documented to induce SS cutaneous manifestations within just 3 days of treatment initiation, with subsequent rapid progression to myopericarditis, resulting in multi-system involvement (20). Proteasome inhibitors, such as bortezomib (25), have also been associated with SS, possibly via inhibition of the β5 and β5i proteasomal subunits, leading to dysregulation of IL-6 and IFN-γ signaling pathways (41). Other newly identified drugs with currently unclear mechanisms included hydroxycarbamide (an antimetabolite), venetoclax (a B-cell lymphoma-2 inhibitor). Within the category of antiparasitic products, insecticides, and repellents, hydroxychloroquine (22) has been sporadically reported, although the precise mechanism remains to be clarified. Among NSAIDs, diclofenac was identified in the present study. Although reports remain limited (18), association between other NSAIDs, such as celecoxib (42), and SS has been documented, suggesting a potential class effect. Paracetamol was also identified but only in isolated cases, and current evidence remains limited. Oral contraceptives containing ethinylestradiol and levonorgestrel have been previously reported to increase the risk of SS (27). The proposed mechanism involves reduced neutrophil apoptosis during hormonal treatment or pregnancy, resulting in elevated counts of neutrophils, monocytes, and lymphocytes (27).

This study identified several monoclonal antibodies associated with DISS, including adalimumab and infliximab (both targeting TNF-α) as well as vedolizumab (targeting integrins). Although multiple case reports have implicated adalimumab in the induction of SS (11, 43), other studies have suggested its potential therapeutic efficacy in refractory or immune-mediated cases—particularly those associated with IBD (44, 45). Similarly, Infliximab has also been associated with both the induction (12, 46) and treatment (47, 48) of SS in clinical settings. Moreover, vedolizumab has been reported to both induce (24) and treat (49) SS in patients with UC. This paradoxical “inductive/therapeutic dual role” appears to be particularly prominent among anti-TNF-α agents (5, 50). TNF-α is a key pro-inflammatory cytokine implicated in the pathogenesis of several autoimmune and inflammatory diseases. Elevated TNF-α levels contribute to chronic immune activation and tissue damage in IBD and RA, justifying the use of anti-TNF-α therapies (51). Furthermore, TNF-α has been shown to promote neutrophil recruitment and activation in SS, suggesting an overlap in inflammatory pathways (52, 53). Anti-TNF-α agents may influence SS development through mechanisms involving neutrophil dysfunction or imbalances in cell-mediated immune regulation (54). As such, the role of such agents in SS should be understood as complex and highly individualized clinical phenomenon. These findings highlight the challenges of translating pharmacovigilance data into clinical practice and underscore the importance of multifactorial risk assessments that consider disease background, treatment indications, and individual susceptibility when evaluating DISS signals.

Additionally, several drugs previously implicated in case reports or small series, including minocycline, celecoxib, and thalidomide (33), were not retained after LASSO selection and multivariable analysis. Potential explanations include low reporting frequencies, variations in patient characteristics, and the application of stricter model selection criteria.

To enhance clinical interpretability, we conducted a supplementary disproportionality analysis based on reports submitted by medical doctors, which are generally considered to possess higher diagnostic specificity and may reflect confirmed clinical assessments. Among 215 evaluated drugs, 57 met the disproportionality threshold, and the distribution of positive signals demonstrated strong concordance with the full dataset. Although several drugs with high reporting frequencies—such as lenalidomide, etanercept, ruxolitinib, dexamethasone, paracetamol, methotrexate, tocilizumab, lamotrigine, dupilumab, and amlodipine—did not meet the threshold for a positive signal in this subset, this may, in part, be attributable to the limited number of physician-submitted reports, which could reduce the statistical power to detect disproportionality. Notably, two of these drugs (lenalidomide and ruxolitinib) were confirmed as positive signals in the main analysis. Furthermore, six of the ten most frequently reported but non-signaling drugs in this subset had previously been documented in the literature, lending additional support to their clinical relevance. This observation suggests that drugs frequently reported by clinicians, even if not statistically flagged in restricted analyses, may still represent clinically suspected signals and warrant further attention. These findings underscore the potential value of medical doctor–reported data in guiding signal prioritization for pharmacovigilance and follow-up, particularly in the context of rare adverse events such as DISS.

This study has several notable strengths. It utilized a large, publicly available pharmacovigilance database (FAERS), capturing a broad and heterogeneous population over a 20-year period. The substantial sample size enabled the detection of rare adverse events such as DISS and supported robust signal estimation based on real-world reporting patterns. A key methodological strength lies in the implementation of a comprehensive, multi-step analytical framework. Disproportionality analysis served as an initial screening tool to identify candidate drug signals, which were then refined through LASSO regression and confirmed via multivariable logistic analysis. Specifically, LASSO regression facilitated efficient variable selection by reducing the number of candidate signals and minimizing the risk of overfitting. The subsequent multivariable analysis enabled adjustment for confounding factors and identification of independently associated drug signals, thereby strengthening the internal validity of the results. Moreover, to further reduce bias from underlying conditions commonly associated with SS, the main analysis excluded reports involving malignancy- or immune-related indications. Sensitivity analyses that reintroduced these exclusions yielded consistent findings, reinforcing the robustness and specificity of the associations. The validity of the results was further supported by external evidence. Cross-database validation using VigiBase enhanced the reliability of signal detection across independent pharmacovigilance systems. Literature review provided additional clinical support for many of the observed associations, while product label examination highlighted the potential of post-marketing surveillance data to uncover underrecognized or unlabeled adverse events. In addition, co-medication patterns were explored using network visualization to reflect real-world polypharmacy, and time-to-onset analysis offered clinically relevant insights into the temporal profile of DISS across different drug classes. Collectively, this study presents a rigorous, data-driven, and externally validated pharmacovigilance assessment, offering a scalable framework for investigating rare but clinically important drug-related safety signals.

Despite the strengths of this study, several limitations should be acknowledged. First, as with all analyses based on spontaneous reporting systems, the FAERS database is subject to well-established biases, including underreporting, duplicate and selective reporting, lack of denominator data, and the absence of standardized event adjudication. These limitations hinder accurate estimation of incidence rates and constrain the ability to assess the absolute risk of adverse events. Second, the dataset lacks detailed patient-level clinical information, such as comorbidities, concomitant medications, disease severity, and time-to-onset data. The absence or incompleteness of these variables reduced interpretability in certain domains and limited the ability to perform more granular subgroup analyses or risk stratification. In particular, cases not submitted by medical doctors may involve diagnostic uncertainty, as they are less likely to reflect confirmed clinical assessments. Third, the core analytical method—disproportionality analysis—detects statistical associations rather than establishing biological or causal relationships. Signal detection can be influenced by external factors such as reporting behavior, masking effects, and patterns of polypharmacy. This study is exploratory in nature and reflects the inherent limitations of pharmacovigilance research; the analyses were intended to generate hypotheses rather than to confirm causality between drug exposure and adverse outcomes. Although LASSO regression was used prior to multivariable modeling to reduce dimensionality and address multicollinearity, the selection of covariates remained constrained by the structure of the FAERS dataset. Fourth, while a subset analysis of medical doctor–submitted reports was performed to enhance clinical interpretability, this approach remains subject to the inherent limitations of spontaneous reporting data. Frequently reported drugs that did not meet the disproportionality threshold should be regarded as clinically suspected signals rather than confirmed associations, and their interpretation requires caution in the absence of corroborating evidence. Fifth, despite efforts to mitigate confounding through indication-based exclusions and sensitivity analyses, residual confounding—particularly confounding by indication—cannot be fully excluded. Finally, although external sources such as VigiBase were incorporated to support the main findings, potential overlap with FAERS case reports may limit the independence. Thus, cross-database consistency signals should be interpreted with caution. Looking ahead, future research should prioritize well-designed cohort studies and rigorously documented case series, which are essential for confirming causality, elucidating underlying mechanisms, and guiding clinical risk assessment.

This study systematically identified and characterized drugs associated with DISS by leveraging the large pharmacovigilance database FAERS, integrating disproportionality analysis, LASSO regression, and multivariable modeling. Several previously recognized drugs, such as sulfamethoxazole/trimethoprim; azacitidine; azathioprine; filgrastim; pegfilgrastim; levofloxacin; amoxicillin; paracetamol; diclofenac; mesalazine; decitabine; hydroxychloroquine; ofloxacin; bortezomib; ethinylestradiol and levonorgestrel were confirmed, while novel signals, including azithromycin; hydroxycarbamide; linezolid and venetoclax were newly identified. Despite the inherent limitations of spontaneous reporting systems and retrospective observational designs, these findings provide a valuable foundation for future prospective studies and mechanistic investigations. This study represents a comprehensive effort to delineate the drug risk landscape of DISS using real-world data, providing new insights into its pharmacological triggers and laying the groundwork for improving pharmacovigilance, early recognition, and management of DISS in clinical practice.