Recently, immune checkpoint inhibitors (ICIs) have been developed as novel therapeutic agents for malignancies, achieving significant anti-tumor responses and extending survival in patients with certain tumor groups [1]. ICIs encompass monoclonal antibodies that target programmed cell death-1 (PD-1), programmed cell death ligand 1 (PD-L1), and cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4). Their anti-cancer effect is mediated by selectively blocking these key immune regulatory pathways, thereby releasing T cell to recognize and destroy tumor antigens [1]. However, this enhancement of anti-tumor immunity can paradoxically lead to nonspecific immune system activation, resulting in a group of toxicities collectively termed immune-related adverse events (irAEs) [2].

Cutaneous irAEs (cirAEs) are the most common irAEs, with a reported incidence approaching 30% in patients treated with ICIs [3]. While the most common cirAEs include nonspecific rash or pruritus, diseases such as eczema, psoriasis and vitiligo are also observed [4]. The mechanism of cirAEs may include epitope spreading and altered T cell subsets [5–7]. Although emerging evidence suggests that cirAEs are associated with enhanced anti-tumor response and improved patient survival outcomes in patients receiving ICIs [8]. CirAEs frequently compromise patients’ quality of life and potentially necessitate discontinuation of ICIs therapy. Bullous pemphigoid (BP) is a subepidermal autoimmune blistering disease and it may also occur from ICIs therapy (ICI-related BP, irBP). ICIs targeting the PD-L1/PD-1 axis can elicit BP in about 0.3%–0.6% patients [9]. In a cohort study of 5636 patients treated with ICIs, 35 (0.6%) developed BP [10]. Notably, irBP patients exhibits distinct clinical features compared to classical BP, such as a prolonged pruritic prodromal phases and extended corticosteroids treatment requirements [3]. Current understanding of irBP remains limited due to small sample sizes in existing studies, and the low prevalence of this condition continues to pose significant challenges in comprehensive clinical characterization.

The US Food and Drug Administration Adverse Event Reporting System (FAERS) is a publicly available database that aggregates voluntary reports of drug-associated AEs from health-care professionals and patients globally. Existing studies on irBP demonstrates notable limitations: (1) The work by Aggarwal et al. [11] while establishing FAERS as a viable data source, was constrained to PD-1 inhibitors (pembrolizumab and nivolumab), with modest case numbers (n = 118). (2) Tan et al. ’s comprehensive FAERS-based study (2011 Q1–2024 Q1), despite employing reporting odds ratio (ROR) methodology across 13-year data, exhibited three key constraints: (a) exclusive reliance on a single disproportionality analysis without complementary method, (b) lack of intra-class agent differentiation, (3) absence of temporal risk quantification.

This study provides a comprehensive pharmacovigilance analysis of irBP by leveraging the FAERS database over an extended period (Q1 2004–Q1 2025). We employed a multi-methodological approach for both signal detection and temporal risk assessment, which included the reporting odds ratio (ROR), proportional reporting ratio (PRR), Bayesian confidence propagation neural network (BCPNN), and multi-item gamma Poisson shrinker (MGPS). By integrating four complementary disproportionality algorithms, we enhanced the robustness and reliability of signal identification. Moreover, we integrated Kaplan–Meier analysis with Weibull shape parameter (WSP) modeling to quantitatively delineate temporal risk patterns. A key advancement in our study was the extension of evaluation beyond the ICI class level to encompass individual agent-level analyses, allowing direct comparisons of clinical characteristics and signal strengths among agents within the same class. Notably, disproportionality analyses consistently showed that PD-1 inhibitors exhibited a higher ROR for irBP compared to CTLA-4 inhibitors, which in turn showed higher ROR values than PD-L1 inhibitors. Collectively, these methodological refinements significantly enhance the depth and breadth of data analysis, providing a solid evidence base for more precise identification and understanding of irBP risk. This, in turn, facilitates the optimization of clinical monitoring and preventive strategies.

The increasing application of ICIs has significantly improved oncological outcomes, but various irAEs have also been reported. In particular, BP represents a rare but potentially serious cirAE, with this study identifying 51% mortality and 24% life-threatening outcomes among affected patients. Importantly, considering that ICIs are indicated for high mortality diseases, the primary cause of death and other detrimental outcomes may be attributed to disease progression rather than direct treatment toxicity.

In the current study, we provided a comprehensive pharmacovigilance analysis of irBP encompassing 850 documented cases. Consistent with previous findings [12], irBP occurred more commonly in males (67.8%) than females (22.0%). However, the global incidence rates of classical BP reveal a slightly higher rate in females (0.0202 per 1,000 person-years) compared to males (0.0181 per 1,000 person-years) [13]. This discrepancy may be attributed to the male predominance of certain types of cancer, such as melanoma, lung cancer, and renal cell carcinoma, which are major indications for ICIs [14–16]. The utilization patterns of ICIs in Korea also showed that the proportion of males (76.3%) was higher than that of females [17]. The most common age group was over 60 years (70.1%), which is consistent with the global incidence for different age groups [13]. Geographically, Japan accounted for the largest share of reports (29.4%), followed by the United States (26.7%) and France (16.2%). Notably, genetic polymorphism increases the risk of irBP [18], while ethnic differences play a role in genetic susceptibility to BP [19], which may also be the case in irBP.

Among irBP cases treated with ICIs, the majority occurred in patients treated for skin/melanoma (30.6%, mainly malignant melanoma), lung (22.6%, mainly non-small cell lung cancer), and kidney/renal malignancies (14.4%, mainly metastatic renal cell carcinoma), with smaller proportions in gastrointestinal, head and neck, bladder/urinary, and liver cancers. This distribution is consistent with prior epidemiological reports [3, 10, 12], confirming melanoma as the most prevalent underlying malignancy. Melanoma was associated with significantly increased odds of developing irBP after ICI treatment (adjusted OR = 3.21; 95% CI, 1.51–6.58) [10], potentially attributable to tumor-specific express of BP180 autoantigen triggering the production of anti-BP180 autoantibodies upon ICI-induced loss of immune tolerance [20]. While this mechanistically explains melanoma’s predisposition, the pathophysiological links between lung/renal cancer and BP remain unestablished, warranting further studies investigation.

Our disproportionality analysis detected significant BP signals across all four pharmacovigilance metrics (RORs, PRRs, BCPNN, and MGPS). PD-1 inhibitors consistently demonstrated the strongest class-level association with BP, exceeding signals from CTLA-4 and PD-L1 inhibitors across all methodologies. This result is concordant with previous pharmacovigilance studies about cirAEs [21]. At the agent level, cemiplimab (PD-1 inhibitor) monotherapy (ROR 37.96, PRR 37.81, EBGM₀₅ 25.21, IC₀₂₅ 3.4) and the combination of atezolizumab (PD-L1 inhibitor) with ipilimumab (CTLA-4 inhibitor, ROR 24.14, PRR 24.08, EBGM₀₅ 24.07, IC₀₂₅ 4.59) constitutes the most significant risks for irBP.

These findings corroborate previous FAERS-based analyses indicating a significant association between ICIs and BP, with PD-1 inhibitors generally showing elevated signal intensities (our ROR = 22.66; Tan et al. ROR = 24.45), supporting PD-1 blockade’s distinct role in BP pathogenesis. Methodologically, our study’s concurrent reporting of PRR, BCPNN, and MGPS allows for robust cross-algorithm validation of the ROR signals, reducing bias from reliance on a single method. Through agent-level stratification, cemiplimab and nivolumab are recognized as high-risk agents—an advancement beyond the class-level analysis by Tan et al.

Our analysis further identified a distinct hierarchy: PD-1 > CTLA-4 > PD-L1 inhibitors (ROR: 22.66 > 8.79 > 6.55). This contrasts with Tan et al.'s reported ranking (PD-1 > PD-L1 > CTLA-4) [12]. These discrepancies highlight the importance of methodological transparency in pharmacovigilance studies. Likewise, at the agent level, the extremely high-risk magnitudes demonstrated by cemiplimab (ROR = 37.96) and nivolumab (ROR = 29.99) demand the highest level of clinical vigilance.

TTO analysis indicated that the median onset time of irBP was 165.2 days (IQR 56–410), with 17.7% (44/249) of cases occurring within the first month and 50.6% (126/249) emerging after 6 months. This profile is generally consistent with the 204-day median (IQR 57–426) reported by Tan et al., [12] with minor differences possibly reflecting variations in observation periods (our inclusion of earlier cases from 2004 onward) and varying proportion of cases with valid TTO records.

To better understand the temporal dynamics of BP risk, we employed WSP modeling. All β values were <1, suggesting a declining hazard pattern—a characteristic of early-onset events. Among different ICI classes, CTLA-4 inhibitors exhibited the most marked early risk concentration (β = 0.48), whereas PD-L1 (β = 0.69) and PD-1 (β = 0.83) inhibitors exhibited a more extended risk period. Our quantitative confirmation of early failure patterns (β < 1) across all ICI classes complements Tan et al.’s clinical recommendation for long-term monitoring while emphasizing an early high-risk window, particularly for CTLA-4 blockade. Clinically, these findings highlight the importance of surveillance strategies stratified by risk magnitude. For instance, the rapid (median 35.7 days) and highly concentrated early-onset risk window for CTLA-4 inhibitors (β = 0.48), necessitates high-frequency dermatologic evaluation within the first month of initiating blockade. In contrast, PD-1 inhibitors not only carry the highest risk magnitude but also exhibit a much broader risk period (β = 0.83, median TTO 190.5 days), with 50.6% of cases emerging after 6 months. This risk magnitude profile compels the need for long-term, continued vigilance for patients on PD-1/PD-L1 therapies, extending well beyond the initial 6 months. Recognizing and leveraging the distinct “risk magnitude” and “temporal magnitude” across ICI classes and individual agents to design stratified surveillance strategies directly improves the timely detection and effective management of irBP, which is critical for optimizing clinical outcomes.

The limitations of this study inherent to pharmacovigilance databases. First, FAERS database has a voluntary nature with non-peer-reviewed AE data, potentially introducing unmeasured confounding. Second, a causal relationship cannot be established between ICIs and the onset of BP because of a disproportionality analysis. Third, absence of prescription denominator data precludes incidence calculation. Given these limitations, prospective studies are required to confirm these findings.