Our study evaluated the feasibility of using the FAERS database for cannabis‐related safety surveillance. We grouped diverse CDP terminologies into relevant categories, stratified by pharmaceutical and recreational motivations for utilization, and descriptively summarized reporting characteristics across groups. We conducted hypothesis‐free disproportionality analysis of Epidiolex and non‐pharmaceutical CBD as a relative comparison of signal detection profiles for the same cannabinoid under different regulatory and utilization contexts to assess feasibility. We observed substantial heterogeneity in CDP reporting terminology, patterns, and signal detection results, which was also observed in a post hoc subgroup analysis stratified by seizure indication.

We identified 1160 terms describing CDPs, with fragmented and inconsistent nomenclature. Terminology included diversity of products, mentioning legal statuses (e.g., ‘illegal marijuana’ and ‘legal cannabis’), highlighting the additional complexity of studying CDPs in FAERS beyond the usual concerns of linking commercial and chemical terms with small proportions of spelling errors [22]. While we provide a preliminary effort for categorizing CDP terms into meaningful groupings for future research, we demonstrated that manual intervention and contextual standardization are additional necessary steps for data cleaning in CDP‐related signal detection analysis. These processes are more complex and time‐consuming than for pharmaceutical drugs, but the risk of information bias and exposure misclassification may further compromise the reliability of safety signal detection [23] and we have provided our mappings to support implementation in future studies in Online Appendix 1.

Despite these challenges, our descriptive analysis revealed reporting patterns consistent with known CDP indications and user demographics. Epidiolex reports were predominantly flagged as primary suspect and frequently involved pediatric patients, reflecting its approved indication for refractory epilepsy [1]. In contrast, non‐pharmaceutical CBD reports were more heterogeneous, with higher representation of women and older adults, and concomitant reporting alongside other products. This pattern may reflect broader use of CBD formulations for symptom management in conditions such as chronic pain, anxiety, and sleep disturbances, consistent with previous surveys and observational studies [24, 25].

Reports involving THC and Cannabis were associated with males aged 18 to 54 years and were often recorded as secondary suspects. This reflects national cannabis surveys, where males report higher percentages of cannabis use than females [26, 27]. Our findings highlighted that, while Epidiolex and non‐pharmaceutical CBD both contain cannabidiol as a primary active ingredient, they likely differ in composition, purity, and concentration. Moreover, reporters vary in utilization and demographic characteristics, which may introduce heterogeneity in motivation to report, impacting the nature of reporting bias.

The disproportionality analysis highlighted these differences as two distinct signal detection profiles were observed. Epidiolex exhibited strong signals for seizure‐related adverse events, reflecting indication and heightened reporting in clinical settings. Such findings were anticipated, as pharmaceutical drugs are subject to legislative requirements for mandatory reporting [28]. In contrast, non‐pharmaceutical CBD reports showed a broader spectrum of signals, including neoplasm events and neurological disorders, likely reflecting the use of CBD to treat or manage these symptoms or health conditions. Seizure‐related events were less prominent compared to Epidiolex reports. The indication‐specific reporting profiles were noted further in our subgroup analyses (Online Appendix 3) where, notably, in the non‐seizure subgroup, the top signals detected for the non‐pharmaceutical version of CBD had no counterparts in the Epidiolex analysis with one exception (behavior disorder).

While some signals reflect ADRs consistent with the Epidiolex product monograph [1] (e.g., drooling, sedation, fluctuations in weight), observed differences with non‐pharmaceutical CBD imply factors beyond indication alone. Reporting context, product composition, dose, product quality, and patient characteristics may influence reporting behaviors [29]. Reporting bias and attribution effects may contribute to non‐pharmaceutical CBD signals, as individuals without medical supervision may be less likely to recognize or report AEs, or may misattribute them to underlying conditions [30]. In contrast, patients treated with Epidiolex are monitored in specialized care settings, where AEs are systematically documented, increasing the capture of seizure‐related events and other known risks. Selection bias may contribute, as individuals with chronic or severe illnesses may proactively report AEs, particularly when concerns are overlooked by clinicians [31]. Furthermore, non‐pharmaceutical CBD formulations exhibit variability in composition, purity, and labeling accuracy, potentially contributing to ADRs themselves as well as to reports linked to contaminants or inconsistent cannabinoid content [32]. Therefore, differences in regulatory status, prescribing practices, utilization patterns, and reporting behaviors may influence the signals detected for the same cannabinoid.

Our study has limitations. First, grouping CDP terms relied on manual, subjective decision‐making based on reported terms and associated cannabinoid pharmacology and regulatory status. Although this process was designed to maximize sensitivity and specificity, it may have introduced some degree of exposure misclassification, particularly for ambiguous or inconsistently reported terms. Second, our analysis was a broad, hypothesis‐free screening approach intended to compare high‐level signal detection profiles between pharmaceutical and non‐pharmaceutical CBD, rather than a study robustly designed to identify or validate individual ADRs. Hence, the observed signals may be influenced by systematic biases inherent to spontaneous reporting, such as confounding by association, competition bias, and under‐reporting or stimulated reporting [33, 34, 35], that were not addressed given the exploratory nature of this design. The choice of comparators in disproportionality analysis has been shown to influence both the magnitude and directionality of signal estimates [36, 37], and we used the full‐data reference set, as this approach was not intended to optimize signal detection for any single ADR.

Accordingly, signals identified should not be interpreted as definitive evidence of the CDP safety profile or to infer differences between pharmaceutical and non‐pharmaceutical formulations. They reflect CDP reporting patterns within a specific timeframe and context. These findings are exploratory, and individual drug‐event combinations require validation in rigorously designed studies. Finally, disproportionality estimates alone cannot establish causality and must be evaluated in conjunction with complementary sources of evidence, including biological plausibility, controlled clinical data, patient‐level characteristics, and alternative explanations for the reported events [38].