An updated pharmacovigilance evidence from U.S. tobacco problem reports following the FDA safety communication on e-cigarettes and seizures

Abstract

Objective

Following the 2019 FDA safety communication on e-cigarette–related seizures, we conducted an updated pharmacovigilance analysis of FDA Tobacco Problem Reports using multimethod disproportionality approaches.

Design

Pharmacovigilance study of spontaneously reported adverse events.

Setting

FDA Tobacco Problem Reports database, publicly available online.

Methods

Reports in which the affected individual was not the product user were excluded. Cases were classified as e-cigarette exposure if the product description mentioned “electronic cigarette,” while all others were grouped as other tobacco products. Reported health problems were harmonized into standardized AE categories. Disproportionality analyses were conducted using four methods: Reporting Odds Ratio (ROR), Proportional Reporting Ratio (PRR), Bayesian Confidence Propagation Neural Network (BCPNN), and shrinkage-based Observed-to-Expected ratio (O/E). A signal was defined when the lower bound of the 95% confidence or credibility interval exceeded the null.

Results

A total of 1,082 reports were analyzed, of which 891 (82.3%) involved electronic cigarettes and 191 (17.7%) involved other tobacco products. Combustible tobacco products accounted for 99 cases (9.15%) of other tobacco products, while other or unknown tobacco products represented 60 cases (5.55%). Smokeless or other nicotine products were least frequently reported, accounting for 32 cases (2.96%).

Reports involving electronic cigarettes were significantly more likely to have been submitted before 2020 compared with reports involving other tobacco products (80.8% vs 50.8%; OR = 4.07, 95% CI 2.93–5.67; p < 0.001). Seizures emerged as the strongest and only consistent safety signal across all disproportionality methods. Respiratory outcomes, including shortness of breath, cough, pneumonia, and respiratory failure, were more frequently reported in e-cigarette cases and showed signals in ROR and PRR analyses, but did not remain significant when assessed with Bayesian approaches.

Conclusions

Seizures represent a pharmacovigilance signal associated with e-cigarette use. Respiratory events were reported more often among e-cigarette users but did not meet conservative Bayesian thresholds. These findings highlight the need for continued surveillance, studies, and regulatory vigilance regarding the safety of e-cigarettes.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-026-27386-1.

Introduction

Electronic cigarettes (e-cigarettes) have become one of the most disruptive innovations in the nicotine market over the past decade. Initially promoted as harm-reduction tools and smoking cessation aids, they are now widely used recreationally, especially among adolescents and young adults. Global prevalence has risen steadily, with estimates suggesting that millions of individuals regularly use e-cigarettes in North America, Europe, and increasingly in low- and middle-income countries [1, 2]. Their rapid diffusion has outpaced regulatory frameworks, creating ongoing debate about their benefits and risks.

Although e-cigarettes eliminate combustion, they deliver nicotine and other chemicals through aerosolization of e-liquids. Experimental studies have demonstrated that e-cigarette aerosols contain potentially harmful substances, including carbonyls, heavy metals, and volatile organic compounds [3, 4]. Clinical and epidemiological studies have raised concerns regarding respiratory effects, such as airway inflammation and impaired pulmonary function [5, 6], as well as cardiovascular effects, including increased heart rate, blood pressure, and endothelial dysfunction [7, 8]. Neurological and psychiatric outcomes have also been reported, with seizures being highlighted in several case reports and regulatory alerts [9, 10].

In 2019, the U.S. Food and Drug Administration (FDA) issued a safety communication reporting cases of seizures temporally associated with e-cigarette use, particularly among youth and young adults, and called for continued surveillance to better characterize this potential safety signal [11].

Moreover, the 2019 outbreak of e-cigarette, or vaping, product use-associated lung injury (EVALI) in the United States, which caused over 2,800 hospitalizations and 68 deaths, highlighted the acute risks linked to certain e-cigarette products, reinforced the urgent need for ongoing surveillance of e-cigarette safety [12, 13].

Post-marketing pharmacovigilance systems are useful instruments for monitoring emerging risks of tobacco and nicotine products. The FDA Tobacco Problem Reports database collects voluntary submissions from consumers, clinicians, and manufacturers, capturing a broad spectrum of product-related concerns. While spontaneous reporting systems are subject to under-reporting, duplication, and reporting biases, they remain useful for detecting unexpected or rare adverse events at the population level [14, 15]. Signal detection in such databases typically relies on disproportionality analysis, which compares the observed frequency of a given adverse event with its expected frequency under independence [14].

Several complementary approaches to disproportionality have been developed. Frequentist methods, such as the Reporting Odds Ratio (ROR) and Proportional Reporting Ratio (PRR), are intuitive and widely applied but can be unstable in the presence of sparse data [16, 17]. Bayesian methods, including the Bayesian Confidence Propagation Neural Network (BCPNN) and shrinkage-based observed-to-expected (O/E) ratios, mitigate this limitation by incorporating prior distributions and stabilizing estimates [18, 19]. Combining these approaches allows cross-validation of potential safety signals and improves the robustness of interpretation: outcomes consistently flagged across both frameworks are more likely to represent a signal of associations, whereas discordant results may reflect statistical artifacts or reporting bias. Thus, integrating multiple disproportionality methods improves the robustness and reliability of interpretation, guiding researchers and regulators toward signals most worthy of further investigation [16, 17].

Despite growing concerns, systematic pharmacovigilance analyses of e-cigarette safety remain scarce, and a comprehensive updated surveillance assessment that harmonizes reported health problems into standardized categories and evaluates their disproportionality across multiple methodological frameworks is lacking.

Previous analyses of the FDA Tobacco Problem Reports and related surveillance systems have provided descriptive summaries of reported adverse events or have focused on specific outcomes, such as seizures or pulmonary complications associated with vaping [20]. While these studies contributed important early signals, they often relied on limited event categorization or single-method signal detection approaches [21, 22]. The present study instead expands this evidence by providing a updated analysis covering the most recent reporting period, applying a structured harmonization of reported health problems into standardized adverse event categories. Following the 2019 FDA safety communication on e-cigarette–related seizures, we conducted an updated pharmacovigilance analysis of the FDA Tobacco Problem Reports database for the most recent reporting period, applying multiple complementary signal detection approaches [23]. Applying both frequentist and Bayesian disproportionality methods, we sought to identify potential pharmacovigilance signals, assess their consistency across approaches, and provide a timely updated overview of the safety profile of e-cigarettes in real-world use.

Methods

Data source

We conducted a secondary analysis of spontaneous reports collected in the FDA Tobacco Problem Reports database, which contains individual case safety reports (ICSRs) submitted voluntarily by consumers, healthcare professionals, and manufacturers. This database is publicly accessible online (https://open.fda.gov/data/downloads/) and includes structured information on the product(s) involved, as well as unstructured narrative fields describing health problems, product malfunctions, and circumstances of use. Each entry may contain multiple adverse events and product-related issues.

For this study, we used the complete dataset of 1314 available records downloaded on 31 December 2025 in JSON format from https://open.fda.gov/data/downloads/. Reports in which the field indicated that the person experiencing the adverse event was not the actual product user (for example, a family member or bystander) were excluded. The remaining 1082 records were treated as unique case reports (Fig. 1).

Fig. 1.

Data selection process for the FDA (Food and Drug Administration) Tobacco Problem Reports dataset

The exposure definition was based on the product description field. Reports mentioning “electronic cigarette” were classified as e-cigarette cases, irrespective of the presence of other tobacco products. All remaining cases were grouped under other tobacco products. In mixed-product reports, e-cigarettes took precedence, meaning any mention of e-cigarette use placed the case in the exposed category.

Calendar years were extracted from the submission date, and reports were categorized as “Before 2020” (≤ 2020) or “After 2020” (> 2020). This temporal classification was used to account for potential changes in reporting intensity associated with the 2019 outbreak of e-cigarette or vaping product use–associated lung injury (EVALI) and subsequent regulatory communications. Because reporting activity remained elevated during the period immediately following the outbreak, the chosen cutpoint allows comparison between the reporting period encompassing the EVALI event and the subsequent phase [12, 13].

Adverse event harmonization

Health problems are often inconsistently described in spontaneous reports, with variations in spelling, terminology, and granularity. To achieve consistent categorization, we developed a synonym dictionary based on regular expression rules. This dictionary grouped reported terms into clinically meaningful adverse event (AE) categories (Appendix Table S1). Categories were defined a priori and included respiratory (e.g., shortness of breath, cough, pneumonia), neurologic (e.g., seizure, headache, tremor), psychiatric (e.g., anxiety, psychosis, depression), cardiovascular (e.g., palpitations, tachycardia, chest pain), general/systemic (e.g., fatigue, fever), gastrointestinal, dermatologic/burns, infections, oncologic diagnoses, substance/nicotine-related problems, and a residual “other adverse event” group for uncategorized or ambiguous terms.

Each AE category was coded as a binary variable indicating whether at least one synonymous term was reported. This allowed both frequency tabulation (Table S2) and quantitative signal detection analyses.

Descriptive analyses

Baseline characteristics of the case reports were summarized overall and by exposure group (e-cigarette vs. other tobacco). Variables with fewer than four unique categories (e.g., reporting year, number of health problems) were analyzed as categorical factors. Comparisons were made using chi-square or Fisher’s exact tests, and results are presented in Table 1. Frequencies of adverse events by category are detailed in Table S2.

Table 1.

Characteristics of reported cases by product type (E-cigarettes vs. other tobacco products). Results are shown as counts and percentages, with odds ratios (OR) and 95% confidence intervals for comparisons

	Overall	Others tobacco products	E-cigarettes
	N = 1082	N = 191	N = 891	OR	P-Value
Year of reporting:					< 0.001
After 2020	265 (24.5%)	94 (49.2%)	171 (19.2%)	Ref.
Before 2020	817 (75.5%)	97 (50.8%)	720 (80.8%)	4.07 [2.93;5.67]
Number of tobacco products:					0.671
<=1	1045 (96.6%)	183 (95.8%)	862 (96.7%)	Ref.
> 1	37 (3.42%)	8 (4.19%)	29 (3.25%)	0.76 [0.36;1.82]
Number of health problems:					0.235
<=1	670 (61.9%)	126 (66.0%)	544 (61.1%)	Ref.
> 1	412 (38.1%)	65 (34.0%)	347 (38.9%)	1.24 [0.89;1.72]
Number product problems:					< 0.001
<=1	894 (82.6%)	120 (62.8%)	774 (86.9%)	Ref.
> 1	188 (17.4%)	71 (37.2%)	117 (13.1%)	0.26 [0.18;0.36]

Disproportionality analyses

For each AE category, we constructed 2 × 2 contingency tables of exposure (e-cigarette vs. other products) and outcome (presence vs. absence of the AE). To mitigate instability from sparse data, zero cell counts were replaced with a value of one only when necessary.

We applied four widely used signal detection methods.

Reporting odds ratio (ROR)

The ROR compares the odds of a specific adverse event being reported for a product of interest (here, e-cigarettes) with the odds of the same event being reported for a comparator group (other tobacco products). Conceptually, it is equivalent to the odds ratio in a case–non-case design, where “cases” are reports containing the adverse event of interest and “non-cases” are reports of all other events [16].

Mathematically, the ROR is defined as:

Where:

• ​: number of reports with the event of interest and the product of interest (e-cigarettes),

• ​: reports without the event but with the product of interest,

• ​: reports with the event but with comparator products,

• ​: reports without the event and with comparator products.

The ROR is presented with Wald 95% confidence intervals (CIs). A signal is considered present when the lower bound of the CI exceeds 1 [15], indicating that the event is disproportionately reported for the product of interest compared with the comparator.

Proportional reporting ratio (PRR)

Another frequentist measure, which represents the ratio of event reporting proportions across exposure groups, is the PRR. This Indicator compares the proportion of a specific adverse event among all reports for the product of interest with the corresponding proportion among comparator products [24].

Like the ROR, it is calculated from a 2 × 2 table and expressed with 95% CIs. A signal is considered when the lower CI bound exceeds 1 [15].

Bayesian confidence propagation neural network (BCPNN)

The BCPNN is a Bayesian approach developed within the WHO Uppsala Monitoring Centre. It calculates the Information Component (IC), a log-transformed measure of the observed-to-expected ratio of reporting [18]. The IC is

Where .

A signal was defined when the lower 95% credibility limit exceeded 0 [15].

Observed-to-expected ratio with shrinkage (O/E)

The O/E is another disproportionality method designed to stabilize estimates for rare events. It compares the number of observed reports of an event for a product with the number expected under independence, applying a Bayesian shrinkage adjustment to account for random variation. The expected reports under independence are:

The final ratio is:

A Bayesian shrinkage model that corrects for random variation, especially in low-frequency cells [19]. A signal was defined when the lower bound of the 95% credibility interval was above 0 [15].

Forest plots of ROR and PRR estimates are displayed in Figs. 2 and 3, while BCPNN and O/E results appear in Figs. 4 and 5. In all plots, red markers denote outcomes meeting the predefined signal thresholds.

Fig. 2.

Forest plot of disproportionality results expressed as Reporting Odds Ratios (ROR) with 95% confidence intervals for adverse-event preferred terms. Points are ROR estimates and horizontal lines are 95% CIs; the dashed vertical line marks no disproportionality (ROR = 1). Terms are ordered by point estimate. Red markers denote signals with CIs entirely above 1. Right-hand columns report the 2 × 2 table counts for each term: ​ (event+exposed), ​ (event+unexposed), ​ (other events+exposed), ​ (other events+unexposed)

Fig. 3.

Forest plot of disproportionality results expressed as Proportional Reporting Ratio (PRR) with 95% confidence intervals for adverse-event preferred terms. Points are ROR estimates and horizontal lines are 95% CIs; the dashed vertical line marks no disproportionality (PRR = 1). Terms are ordered by point estimate. Red markers denote signals with CIs entirely above 1. Right-hand columns report the 2 × 2 table counts for each term: ​ (event+exposed), ​ (event+unexposed), ​ (other events+exposed), ​ (other events+unexposed)

Fig. 4.

Disproportionality analysis of adverse-event preferred terms using the Bayesian Confidence Propagation Neural Network (BCPNN) approach with Normal approximation. Effect estimates are shown as Information Components (IC) with 95% credibility intervals on a linear scale. The vertical dashed line indicates the null (IC = 0). Red markers denote significant signals where the lower bound of the credibility interval exceeds 0. Right-hand columns report the 2 × 2 table counts for each term: ​ (event+exposed), ​ (event+unexposed), ​ (other events+exposed), ​ (other events+unexposed)

Fig. 5.

Disproportionality analysis of adverse-event preferred terms using the Observed-to-Expected (O/E) ratio with simple shrinkage, following the method of Norén (2013). Estimates are displayed with 95% credibility intervals on a linear scale. The vertical dashed line corresponds to no disproportionality (O/E = 0 after shrinkage adjustment). Red markers indicate signals where the lower bound of the credibility interval is above 0. Red markers denote significant signals where the lower bound of the credibility interval (IC025_{025}025​) exceeds 0. Right-hand columns report the 2 × 2 table counts for each term: ​ (event+exposed), ​ (event+unexposed), ​ (other events+exposed), ​ (other events+unexposed)

To reduce the likelihood of spurious findings arising from sparse data or multiple comparisons, signal interpretation did not rely on a single disproportionality metric. Instead, results should be evaluated across multiple complementary methods, including both frequentist (ROR, PRR) and Bayesian (BCPNN and shrinkage-based O/E) approaches, which provide higher stability in the presence of low reporting counts. Signals consistently identified across these frameworks were considered more robust, whereas isolated findings were interpreted cautiously as exploratory signals.

All analyses were performed using R 3.4.2 [25] with the pharmsignal [26] package.

Sensitivity analysis

To assess the robustness of the exposure classification and address potential bias related to mixed-product reports, several sensitivity analyses were performed. First, analyses were repeated restricting the dataset to reports involving a single tobacco product, thereby excluding cases in which multiple products were reported simultaneously. Second, the comparator group was restricted to combustible tobacco products (e.g., cigarettes, cigars, roll-your-own tobacco, and pipe tobacco), excluding other non-combustible or nicotine products to reduce heterogeneity in the reference category. Third, reports categorized as “other” or with unclear product classification were excluded. Disproportionality analyses using the same methodological framework (ROR, PRR, BCPNN, and shrinkage-based observed-to-expected ratio) were repeated under these alternative definitions to evaluate whether the main safety signals remained consistent across exposure specifications. Results of these analyses are presented in Supplementary Figures S1-S4.

Ethics

The study analyzed de-identified, spontaneously reported cases from a publicly available regulatory database. No patient contact occurred, and institutional review board approval was not required.

Results

Data description

The data selection process is presented in Fig. 1, resulting in 1,082 eligible reports (Table S1); 82.3% involved electronic cigarettes, while 17.7% involved other tobacco products. Among the latter, combustible tobacco accounted for 9.15% of reports, followed by other or unknown products (5.55%) and smokeless or other nicotine products (2.96%). Reports involving electronic cigarettes were significantly more likely to have been submitted before 2020 compared with reports involving other tobacco products (80.8% vs. 50.8%; OR = 4.07, 95% CI 2.93–5.67; p < 0.001). In addition, reports with more than one product-related problem were significantly less likely to involve electronic cigarettes than other tobacco products (13.1% vs. 37.2%; OR = 0.26, 95% CI 0.18–0.36; p < 0.001) (Table 1).

Disproportionality analysis

The disproportionality analysis ROR (Fig. 2) revealed several signals. The most striking was for seizures, with a substantially elevated ROR and a confidence interval entirely above unity, consistent with the descriptive data in Table S2, where seizures were reported in nearly one-third of e-cigarette cases versus only 3.3% of other tobacco cases (p < 0.001). Respiratory outcomes also showed excess reporting. Respiratory failure was observed exclusively in e-cigarette reports, and pneumonia was more frequently reported among e-cigarette users. Similarly, both shortness of breath and cough displayed RORs above one, with confidence intervals suggesting potential disproportionality. By contrast, other adverse events, including psychiatric and cardiovascular outcomes, showed ROR estimates closer to unity with wide intervals, indicating no disproportionality.

Results were consistent when evaluating PRR (Fig. 3), which again highlighted seizures as the dominant signal and supported disproportionate reporting of several respiratory outcomes as in ROR analysis.

More conservative methodologies restricted the list of positive signals. The Bayesian Confidence Propagation Neural Network (BCPNN) analysis (Fig. 4) identified seizures as the only outcome with a positive information component ( > 0). Similarly, the shrinkage-based Observed-to-Expected ratio (Fig. 5) confirmed seizures as the sole signal after adjustment. Thus, while crude disproportionality analyses suggested possible associations with respiratory events, only seizures emerged as a stable and statistically significant signal across all approaches.

Sensitivity analyses restricting the dataset to single-product reports and limiting the comparator group to combustible tobacco products yielded results broadly consistent with the primary analysis. Seizures remained the most prominent signal in the frequentist disproportionality analyses (ROR and PRR) and were also supported by the BCPNN approach; under the more conservative shrinkage-based O/E method, the credibility interval only marginally crossed the null threshold, suggesting attenuation rather than disappearance of the signal. In contrast, shortness of breath remained significant only under the less conservative frequentist approaches. These findings indicate that the primary signal for seizures is largely robust to alternative exposure specifications (Figures S1-S4).

Discussion

In the FDA Tobacco Problem Reports database, updated adverse event reporting patterns identified for seizures were consistent with the safety signal previously highlighted by the FDA for e-cigarettes [11]. Across more than 1,000 case reports, seizures emerged as the only consistent, statistically significant signal across all disproportionality approaches, including both frequentist and Bayesian methods. Respiratory events such as shortness of breath, cough, pneumonia, and respiratory failure were more frequently reported in e-cigarette cases and showed signals in crude ROR and PRR analyses, but these associations did not persist when assessed with Bayesian shrinkage methods.

Our findings strengthen previous regulatory alerts and case series suggesting a possible disproportionality signal indication between e-cigarette use and seizures. In 2019, the U.S. Food and Drug Administration (FDA) reported clusters of seizure cases temporally associated with e-cigarette use, primarily in adolescents and young adults [9]. While the biological mechanism remains uncertain, hypotheses include acute nicotine toxicity, particularly with high-concentration formulations, rapid absorption, or product manipulation that alters delivery [10]. The signal observed in our study, persisting across multiple methodological approaches, supports the plausibility of a pharmacovigilance signal indication that warrants further mechanistic and epidemiological investigation. A possible toxicological explanation relates to the pharmacokinetic properties of nicotine formulations used in many modern e-cigarette liquids. In particular, nicotine salts, formed by combining nicotine with organic acids such as benzoic acid, may allow higher nicotine concentrations while maintaining a lower aerosol pH, which reduces throat irritation and facilitates inhalation [27]. These formulations can enhance systemic nicotine delivery and may promote more rapid absorption into the bloodstream compared with traditional free-base nicotine solutions [28]. Nicotine readily crosses the blood–brain barrier and binds to neuronal nicotinic acetylcholine receptors (nAChRs), modulating synaptic transmission and neuronal excitability. At high systemic levels, excessive activation of these receptors may increase neuronal firing and glutamatergic signaling, potentially lowering the seizure threshold. The rapid central nervous system exposure associated with nicotine salts may therefore contribute to acute neurologic toxicity, particularly among adolescents and young adults, who represent a substantial proportion of e-cigarette users [29]. Moreover, user-driven product manipulation may also influence nicotine exposure and toxicity. Many e-cigarette devices allow modification of voltage or wattage, and practices such as “dripping,” in which e-liquid is applied directly to the heating coil, can increase aerosol temperature and nicotine delivery. These behaviors may lead to higher systemic nicotine concentrations and potentially enhance neurotoxic effects [30].

Respiratory events were also disproportionately reported among e-cigarette users in descriptive analyses. Shortness of breath, cough, and pneumonia showed elevated frequencies and positive signals in traditional disproportionality methods. These findings are consistent with prior literature linking e-cigarette use to respiratory symptoms in adolescents [6], increased risk of bronchitic symptoms [6], and clinical case reports of pneumonia or organizing pneumonia associated with vaping.

We stratified reports by reporting period to account for potential changes in reporting intensity related to the EVALI outbreak. However, the attenuation of these signals under Bayesian methods may indicate that some of these associations may be driven by stimulated reporting during the 2019 EVALI outbreak or other forms of reporting bias [12, 13]. However, the limited number of reports did not allow robust temporal disproportionality analyses for individual respiratory outcomes. Therefore, the temporal stratification was primarily aimed at contextualizing reporting patterns rather than at formally assessing changes in signal strength over time.

In this context, the EVALI outbreak may have increased awareness among clinicians, regulators, and the public, resulting in a temporary rise in the reporting of respiratory symptoms associated with vaping products. Such reporting inflation could amplify the apparent prominence of respiratory outcomes in spontaneous reporting systems, even when causal relationships remain uncertain.

Nonetheless, the exclusivity of respiratory failure reports in e-cigarette cases highlights the need for vigilance and continued monitoring, even if conservative statistical thresholds were not surpassed.

Our study illustrates the value of combining frequentist and Bayesian disproportionality approaches. The ROR and PRR are sensitive to emerging patterns but may produce spurious signals in sparse data. At the same time, Bayesian methods such as the BCPNN and shrinkage O/E ratios provide more conservative, stable estimates [14, 16, 19]. The convergence of evidence across methods strengthens confidence in seizures as a signal, while the discordance for respiratory outcomes highlights the importance of contextual interpretation. These results underscore that pharmacovigilance signal detection should not be based on a single metric, but rather on triangulation of complementary approaches, clinical plausibility, and consistency with external data.

Several limitations must be acknowledged. First, spontaneous reporting systems are subject to under-reporting, reporting biases, and missing or incomplete information. The database does not provide denominators of exposed users, precluding incidence rate estimation or risk quantification [31]. Second, the exposure definition relied on product descriptions that may vary in accuracy, and misclassification cannot be excluded. Third, confounding by co-exposures (such as other tobacco products or illicit substances) could not be fully addressed. In particular, the database [11] does not capture detailed information on concomitant exposures, such as THC-containing vaping products or other substances, which have been implicated in vaping-related health events. As a result, some reported neurologic or respiratory outcomes may reflect combined exposures rather than e-cigarette use alone.

In addition, reporting patterns may have been influenced by stimulated reporting, whereby adverse event reporting increases following regulatory alerts or heightened media attention, such as during the 2019 EVALI outbreak. These dynamics should be considered when interpreting pharmacovigilance signals derived from spontaneous reporting systems. Finally, disproportionality analyses are hypothesis-generating tools and cannot establish causality.

Despite these limitations, this study provides novel insights by systematically harmonizing adverse event categories and applying multiple disproportionality methods. To our knowledge, this is the first comprehensive pharmacovigilance evaluation of e-cigarettes in the FDA Tobacco Problem Reports database [23].

Implications for research and policy

The identification of seizures as a signal highlights a need for studies and population-based research to clarify causality, dose–response patterns, and susceptible populations. Clinicians should be aware of the potential neurologic effects of e-cigarettes, particularly in adolescents and young adults, and inquire about vaping in cases of new-onset seizures.

Moreover detailed information on device settings or user practices is not systematically captured in the FDA Tobacco Problem Reports database, these factors should be considered when interpreting adverse event reports related to e-cigarette use. From a clinical and public health perspective, clinicians evaluating new-onset seizures, particularly among adolescents and young adults, should consider screening for e-cigarette use and inquire about device type and nicotine concentration, while regulatory surveillance should continue to monitor high-nicotine products and youth exposure.

The excess reporting of respiratory outcomes, while not statistically robust under Bayesian methods, should not be dismissed, given consistency with experimental and clinical evidence. These findings support ongoing regulatory monitoring of e-cigarettes, strengthening post-marketing surveillance.

Conclusion

In this pharmacovigilance analysis of the FDA Tobacco Problem Reports database, seizures emerged as the most consistent safety signal linked to e-cigarette use, remaining evident across both traditional and Bayesian disproportionality methods. Respiratory issues like shortness of breath, cough, pneumonia, and respiratory failure were reported more often in e-cigarette cases. They showed signals in standard analyses, but these associations lost significance when assessed with more conservative Bayesian techniques. These results highlight the importance of using multiple, complementary disproportionality methods to confirm potential safety signals in spontaneous reporting systems. While seizures require further investigation into their mechanisms and epidemiology, the significance of respiratory events underscores the need for ongoing monitoring, given their clinical plausibility and consistency with existing evidence.

Authors’ contributions

DA conceived and designed the study, performed the statistical analyses, and drafted the manuscript. PB and MBM contributed to data interpretation and manuscript revision. DG provided supervision and critical revision. All authors approved the final manuscript.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Data availability

The data used in this study are publicly available from the U.S. Food and Drug Administration Tobacco Problem Reports database (https://open.fda.gov/data/downloads/).

Declarations

Ethics approval and consent to participate

This study analyzed de-identified data from the FDA Tobacco Problem Reports database, which is publicly available. No patient contact occurred, and no identifiable personal information was accessed. According to applicable regulations, ethical approval and informed consent were not required.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.