Abstract

Background:

Tarlatamab is a DLL3-targeted bispecific T-cell engager approved for previously treated extensive-stage small cell lung cancer (ES-SCLC). However, its post-marketing safety profile in routine practice remains incompletely characterized.

Methods:

We conducted a retrospective pharmacovigilance study using the FDA Adverse Event Reporting System from Q2 2024 to Q3 2025, with a descriptive cross-database comparison using WHO-VigiAccess. Disproportionality analyses were performed using reporting odds ratio, proportional reporting ratio, Bayesian confidence propagation neural network, and multi-item gamma Poisson shrinker algorithms. Clinical priority scoring, subgroup analysis, time-to-onset analysis, multivariable logistic regression, and interpretable machine learning with SHAP (SHapley Additive exPlanations) were further applied.

Results:

A total of 942 reports with tarlatamab as the primary suspect drug were identified, comprising 1,346 adverse events. At the preferred-term level, 30 signals met all four disproportionality criteria. The most frequent and strongest signals were cytokine release syndrome (CRS; n = 201; ROR 223.84, 95% CI 192.15–260.77) and immune effector cell-associated neurotoxicity syndrome (ICANS; n = 106; ROR 312.43, 95% CI 254.68–383.26). Common additional signals included pyrexia, dysgeusia, hypotension, and ageusia. Potentially under-recognized events, such as intestinal perforation, dyspnoea at rest, and incontinence, were also detected. Most adverse events occurred early after treatment initiation, with a median time to onset of 3 days; CRS and ICANS both showed a median onset of 1 day. In multivariable analysis, concomitant medication use was associated with higher reported odds of CRS (OR 2.551, 95% CI 1.353–4.811), whereas reports from Japan and the year 2025 were associated with lower reported odds. The CRS report-level classification model showed acceptable discrimination (AUC 0.733), whereas the best machine-learning model for adverse events of immune disorders classification within FAERS reports showed only modest performance (validation AUC 0.639). SHAP analysis indicated that country and therapy type contributed more to model output than age and sex, suggesting that the model primarily captured reporting context and treatment complexity rather than intrinsic biological susceptibility. Cross-database comparison with WHO-VigiAccess showed a broadly concordant reporting pattern, with CRS and ICANS remaining the most commonly reported toxicities.

Conclusion:

Post-marketing data indicate that tarlatamab has a distinct safety profile dominated by early-onset immune-mediated and neurologic toxicities, particularly CRS and ICANS. Early monitoring and prompt supportive management during initial treatment cycles are essential. These findings broaden current knowledge of tarlatamab safety in real-world practice and support prospective studies to pharmacovigilance-based signal stratification and monitoring strategies.

1 Introduction

Small cell lung cancer (SCLC) is a highly aggressive neuroendocrine malignancy characterized by rapid proliferation, early metastatic spread, and poor prognosis. It accounts for approximately 14% of all lung cancers, and 60%–70% of patients present with extensive-stage disease at initial diagnosis. The incidence of brain metastases during the disease course reaches up to 40%–50%, and the median overall survival for extensive-stage SCLC remains limited to 6–12 months (Bolte et al., 2025). Despite advances in surgery, radiotherapy, chemotherapy, and immunotherapy (Kemper et al., 2025), durable disease control remains uncommon, highlighting a critical unmet need for novel targeted therapies.

Tarlatamab (tarlatamab-dlle; IMDELLTRA™) is a novel half-life–extended bispecific T-cell engager (BiTE) antibody that simultaneously targets Delta-like ligand 3 (DLL3) on tumor cells and CD3 on cytotoxic T lymphocytes (CTLs). By physically bridging tumor cells and T cells, tarlatamab induces T-cell activation, promotes cytokine release, and facilitates targeted cytotoxicity against DLL3-expressing malignancies (Dhillon, 2024). Under physiological conditions, DLL3 is an intracellular protein involved in the negative regulation of the Notch signaling pathway; however, it is aberrantly expressed on the surface of SCLC cells. Notably, DLL3 expression has been reported in approximately 85%–94% of patients with SCLC, making it an attractive therapeutic target (Ahn et al., 2023).

In May 2024, tarlatamab received its first approval from the U.S. Food and Drug Administration (FDA) for adult patients with extensive-stage SCLC (ES-SCLC) who have progressed on or after platinum-based chemotherapy (Dhillon, 2024). Emerging evidence suggests that tarlatamab may also have therapeutic potential in other neuroendocrine malignancies, including neuroendocrine prostate cancer and head and neck neuroendocrine carcinoma (Aggarwal et al., 2025). Given its unique mechanism of action and promising efficacy, tarlatamab represents an important advancement in the treatment landscape of SCLC. However, the immune activation induced by BiTE therapies is inherently associated with immune-related toxicities. Clinical trials, including DeLLphi-301 and DeLLphi-303, have reported a high incidence of adverse events, particularly cytokine release syndrome (CRS), pyrexia, decreased appetite, dysgeusia, anemia, and immune effector cell-associated neurotoxicity syndrome (ICANS) (Sands et al., 2025; Lawrence, 2026). Current evidence regarding the safety profile of tarlatamab is largely derived from clinical trials, meta-analyses, and case reports. These studies are limited by strict inclusion criteria, relatively small sample sizes, and short follow-up durations, which may not fully reflect real-world safety outcomes. In particular, immune-related adverse events may lead to prolonged hospitalization, increased susceptibility to infection, or even death (Bajwa et al., 2025). Therefore, comprehensive post-marketing pharmacovigilance studies based on real-world data are essential to better characterize its safety profile.

Traditional pharmacovigilance approaches, such as disproportionality analysis, are effective for detecting population-level safety signals but provide limited insight into individual-level risk prediction. Recently, interpretable machine learning (ML) has emerged as a promising complementary strategy, enabling improved predictive performance while maintaining model transparency. This approach has been successfully applied in multiple biomedical domains, including drug discovery, proteomics, and genomics (Medina-Ortiz et al., 2025). These advances may offer new opportunities for exploring report-level patterns related to tarlatamab-associated immune adverse events and improving the interpretability of pharmacovigilance-based signal stratification (Liu et al., 2024). In this study, we conducted a comprehensive pharmacovigilance analysis based on the U.S. Food and Drug Administration (FDA) Adverse Event Reporting System (FAERS) database, complemented by descriptive cross-database comparison using the WHO VigiAccess platform. By integrating disproportionality analysis with interpretable ML, we aimed to systematically evaluate the safety profile of tarlatamab and provide evidence to support its rational clinical use.

2 Methods2.1 Data collection, FAERS database description, and MedDRA terminology

This study conducted a retrospective pharmacovigilance analysis based on the FAERS database. Post-marketing adverse event reports submitted between the second quarter of 2024 and the third quarter of 2025 were included. Given that tarlatamab received its first FDA approval in May 2024, this study period captures both early post-marketing and recent real-world safety data. The FAERS database comprises multiple data files, including DEMO, DRUG, REAC, OUTC, RPSR, THER, and INDI, which respectively contain patient demographic information, drug exposure data, adverse event reports, clinical outcomes, report sources, therapy dates, and indications. In this study, the PRIMARY ID was used as the unique identifier to link and integrate records across these datasets. Reports associated with the target drug were identified by searching the DRUG file using both generic names (“TARLATAMAB”, “TARLATAMAB DLLE”) and the brand name (“IMDELLTRA”). As the FAERS database relies on spontaneous reporting, duplicate or withdrawn/deleted reports are inevitable. 1n this study, reports were excluded if tarlatamab was not explicitly identified as the primary suspected (role_code: PS) drug, if adverse event descriptions were vague or critical information was missing (e.g., lacking specific event names or showing no temporal relationship with drug administration), or if cases shared identical events, event dates, age, sex, and country of origin (duplicate report) (Cirmi et al., 2020). Following FDA guidance on data refinement, duplicates were systematically removed using a hierarchical protocol: unique identifiers (PRIMARY ID), case identifiers (CASEID), and FDA receipt dates (FDA_DT) were extracted from the DEMO file and sorted by CASEID, FDA_DT, and PRIMARY ID; for duplicate CASEIDs, the report with the most recent FDA_DT was retained, and when CASEID and FDA_DT were identical, the entry with the highest PRIMARY ID was selected. An additional refinement step was applied by cross-referencing reports with the quarterly “deleted files” list, implemented since the first quarter of 2019, to remove withdrawn reports and further enhance dataset accuracy (Sui et al., 2025). In our analysis, missing categorical variables were not imputed. To avoid selection bias resulting from excluding cases with incomplete information, all 942 unique case reports were retained in the analysis cohort. During the generation of descriptive statistics, missing values were automatically treated as a separate “missing” category. This approach ensures transparency in data completeness and aligns with standard practices in pharmacovigilance research (Shen et al., 2025). Adverse events were coded using the Medical Dictionary for Regulatory Activities (MedDRA), version 28.1 (Brown, 2003). Preferred Terms (PTs) and their corresponding System Organ Classes (SOCs) were standardized and mapped accordingly. After data preprocessing, a total of 2,164,043 DEMO records, 9,545,268 DRUG records, and 5,814,066 REAC records were obtained. Among these, 942 reports identified tarlatamab as the primary suspect drug, encompassing 1,346 tarlatamab-related adverse events. The study workflow is illustrated in Figure 1.

Flowchart of the entire study.

2.2 Disproportionality analysis

Disproportionality analysis is a classical approach in pharmacovigilance studies for identifying potential associations between drugs and specific adverse drug events (ADEs) (Scosyrev et al., 2025). The fundamental principle involves comparing the reporting proportion of a specific ADE associated with the target drug to that observed in a reference group, in order to determine whether the ADE is reported more frequently than expected among individuals exposed to the target drug. A significantly higher reporting proportion suggests the presence of a potential disproportionality signal.

In this study, four widely used pharmacovigilance signal detection methods were employed, including the reporting odds ratio (ROR), proportional reporting ratio (PRR), Bayesian confidence propagation neural network (BCPNN), and multi-item gamma Poisson shrinker (MGPS) algorithms. ROR and PRR are traditional ratio-based methods, characterized by computational simplicity, intuitive interpretation, and strong transparency (van Puijenbroek et al., 2002; Trippe et al., 2017). BCPNN reduces the impact of data sparsity on estimate stability and is particularly suitable for detecting rare adverse event signals (Bate, 2007). MGPS, through Bayesian shrinkage, enhances control of the signal-to-noise ratio, thereby improving both the accuracy and sensitivity of signal detection (Ahlmann-Eltze and Huber, 2021). Given that each algorithm has distinct strengths in signal identification, all four methods were applied in combination to improve the comprehensiveness and robustness of the analysis (Chen et al., 2025). All signal calculations were based on 2 × 2 contingency tables (Table 1). The criteria for defining a positive signal were as follows: for ROR, the lower limit of the 95% confidence interval (CI) > 1 with a minimum of three co-occurrences (n ≥ 3); for PRR, PRR >2 with χ2 ≥ 4; for BCPNN, the 2.5th percentile of the information component (IC025) > 0; and for MGPS, the lower limit of the 95% CI of the empirical Bayesian geometric mean (EBGM05) > 2. These thresholds were established according to commonly accepted pharmacovigilance practices. In general, higher values of these metrics indicate a stronger potential association between the target drug and the corresponding adverse event, reflecting greater signal strength.

​Target adverse drug eventNon-target adverse drug eventSumsTarlatamababa + bNon-tarlatamabcdc + dTotala + cb + da + b + c + dMethodsFormulaThresholdROR​PRRχ2 = [(ad - bc)2](a + b + c + d)/[(a + b) (c + d) (a + c) (b + d)]PRR>2BCPNNEBGM

We summarize the methodologies, corresponding formulas, and signal detection thresholds applied for the reporting odds ratio (ROR), proportional reporting ratio (PRR), Bayesian confidence propagation neural network (BCPNN), and empirical Bayesian geometric mean (EBGM).

In this 2 × 2 contingency table, a denotes the number of reports containing both tarlatamab and the target adverse event; b represents the number of reports involving tarlatamab in association with other adverse events; c refers to the number of reports describing the target adverse event in relation to other drugs; and d indicates the number of reports involving other drugs combined with non-target adverse events.

95% CI, 95% confidence interval; χ², chi-squared statistic; IC, information component; IC025, 2.5th percentile of the information component; E(IC), expected value of IC; V(IC), variance of IC; and EBGM05, 5th percentile of the empirical Bayes geometric mean.

2.3 Clinical priority assessment

In this study, a semi-quantitative evaluation framework was developed to prioritize clinically relevant ADEs, based on the guidance of the European Medicines Agency (EMA) regarding Designated Medical Events (DMEs) and Important Medical Events (IMEs) (Cecco et al., 2024). This framework integrates both the clinical significance of ADEs and the robustness of the detected signals, enabling a stratified assessment of different adverse events while also facilitating the identification of potential inconsistencies in reporting patterns. According to the composite scoring system, all ADEs were categorized into three priority levels: low priority (0–2 points), moderate priority (3–5 points), and high priority (6–10 points). Detailed scoring criteria and classification standards are presented in Table 2.

Criterium2 points1 point0 pointReporting rate (cases/non-cases)>10%1–10%0–1%Signal stability (consistency across disproportionality analyses)3/4 of 42 of 31 of 3Reported case fatality rate (proportion of reports with death as outcome)>50%25–50%<25%Clinical relevance (serious likely drug-attributable ADEs)DMEIMENone

Criteria and scoring framework applied to prioritize adverse drug events detected via disproportionality analysis.

ADEs, adverse drug events; DME designated medical event; IME important medical event.

2.4 Machine learning modeling and model interpretation

To explore whether report-level characteristics could help classify immune-related adverse event reports associated with tarlatamab, multiple MLs algorithms were applied for model construction and performance evaluation. The primary ML task is the binary classification of tarlatamab-associated “immune disorders” in SOCs, specifically distinguishing between reported (positive class) and unreported immune disorders (negative class) associated with tarlatamab. Prior to analysis, categorical variables (such as sex) were processed using one-hot encoding. For continuous features (such as age), Z-score standardization was performed to ensure that each feature had a mean of 0 and a standard deviation of 1, facilitating comparison across features with different units of measurement. This study utilized 13 commonly used supervised learning algorithms, including logistic regression, Lasso regression, discriminant analysis, support vector machine (SVM), random forest, gradient boosting, extreme gradient boosting (XGBoost), neural networks, Bayesian methods, and k-nearest neighbor algorithms. The dataset was randomly split into a training set and a validation set in a 1:1 ratio. To mitigate overfitting and improve generalizability, 5-fold cross-validation with 10 repetitions was employed on the training set, and grid search was used to optimize the hyperparameters of all models (Chen et al., 2024; Wu et al., 2025). The final models were trained using the best-performing hyperparameters, and evaluation was conducted on the validation set. The optimal hyperparameters after the 5-fold cross-validation for each ML model are provided in Supplementary Table S1. Model performance was rigorously quantified using metrics such as area under the receiver operating characteristic (ROC) curve (AUC), sensitivity, specificity, accuracy, and F1 score. The discriminatory ability of the models was assessed by the AUC and its 95% confidence interval (CI), with results visualized using forest plots. Additionally, confusion matrices were constructed to evaluate classification performance in both the training and validation sets, and overall accuracy was calculated. To further assess the consistency between predicted probabilities and observed outcomes, calibration curves were generated to evaluate model calibration performance. The selection of the best model was based on its overall performance on the independent validation set. A higher AUC under the ROC curve was the primary criterion for selecting the optimal model, with F1 score, sensitivity, and specificity serving as secondary metrics for evaluating model performance (Yang et al., 2026). After identifying the best-performing model, the SHAP (SHapley Additive exPlanations) method was integrated to enhance model interpretability (Li Y. et al., 2025). By calculating the SHAP values for each feature, we precisely quantified the contribution of each feature to the model’s predictions, providing a practical and interpretable example of ML-based feature identification. First, feature importance was assessed by calculating the mean absolute SHAP value for each variable, followed by ranking the features. SHAP beeswarm plots were then used to illustrate the direction and magnitude of each feature’s contribution to the model’s predictions across different value ranges. Additionally, dependence plots were generated to explore the relationships between key variables (e.g., age, sex, country, and therapy type) and predicted risk. At the individual level, waterfall plots and force plots were employed to visualize the contribution of each feature to the prediction for a single sample, thereby offering interpretable insights into the model’s decision-making process.

2.5 Logistic regression analysis

To investigate the risk of CRS associated with tarlatamab and to identify potential risk factors for mortality following CRS onset, univariate and multivariate logistic regression models were constructed. Independent variables included age, sex, reporter type, treatment regimen (with or without concomitant medications), reporting country, and reporting year. To ensure the reliability of the analysis, reports with missing values in any of these variables were excluded. For the purpose of statistical modeling, the original count data based on System Organ Class (SOC) were transformed into binary outcome variables. Specifically, for the analysis of CRS occurrence, “CRS following tarlatamab use” was coded as 1, and “no CRS occurrence” as 0. For the mortality risk analysis, “death following CRS” was coded as 1, whereas “survival after CRS” was coded as 0.

2.6 Time-to-onset (TTO) analysis

In this study, TTO was defined as the interval between the date of adverse event occurrence (recorded in the DEMO file) and the initiation date of drug therapy (recorded in the THER file). Reports with invalid date records (e.g., erroneous entries or incomplete date information lacking year, month, or day) or negative calculated intervals were excluded to ensure analytical accuracy. At the overall level, TTO was summarized using the median and interquartile range (IQR) to describe its central tendency and dispersion. In addition, the temporal distribution of TTO was characterized using the shape parameter (β) of the Weibull distribution model. In Weibull analysis, a β value <1 with a 95% confidence interval (CI) entirely below 1 indicates a decreasing hazard over time, corresponding to an “early failure” type; a β value approximately equal to 1 with a 95% CI including 1 suggests a constant hazard over time, consistent with a “random failure” type; and a β value >1 with a 95% CI excluding 1 indicates an increasing hazard over time, corresponding to a “wear-out failure” type (Shu et al., 2022). At the PT level, differences in TTO among various PTs were compared using the Kruskal–Wallis rank-sum test. Furthermore, Kaplan–Meier survival analysis was applied to estimate the cumulative incidence of adverse events across different subgroups, with comparisons performed between groups.

2.7 Cross-database descriptive comparison using the WHO VigiAccess

The data used in this study were obtained from VigiAccess (https://www.vigiaccess.org/), a publicly accessible online platform developed by the World Health Organization (WHO). This platform is based on VigiBase, the WHO global database of individual case safety reports (ICSRs), and provides aggregated information on ADEs related to medicines and vaccines. On 31 January 2026, we accessed the platform and systematically retrieved adverse event reports associated with tarlatamab. The extracted data included key demographic characteristics of the reporting population, such as age groups, sex distribution, reporting year, and geographic distribution by continent. All adverse events were coded using the Medical Dictionary for Regulatory Activities (MedDRA) and were categorized and analyzed according to SOCs and PTs.

2.8 Statistical analysis and software

All statistical analyses and data visualizations were performed using R software (version 4.2.3). All statistical tests were two-sided, with a P value <0.05 considered statistically significant. The study design and reporting were conducted in accordance with the READUS-PV guidelines, ensuring transparency and reproducibility of the research methodology and findings (Fusaroli et al., 2024).

3 Results3.1 Description of baseline information from adverse reaction reports relating to tarlatamab in the FAERS database

A detailed analysis of the baseline characteristics of 942 ADE reports associated with tarlatamab in the FAERS database was performed. As of the third quarter of 2025, the number of reports in 2025 (635 reports) had more than doubled compared to 2024 (307 reports) (Figure 2A). Among reports with specific age information, the majority of reporters were aged 65–85 years (n = 194, 20.59%), followed by the middle-aged group (18–65 years) (n = 118, 12.53%) (Figure 2B). In reports containing sex information, male reporters accounted for a higher proportion (n = 344, 36.52%) compared to female reporters (n = 265, 28.13%) (Figure 2C). However, over 90% of the reports lacked detailed weight information (n = 904, 95.97%) (Figure 2D). Regarding the reporting countries, the top four countries by report count were the United States (n = 672), Japan (n = 181), Canada (n = 24), and South Korea (n = 22) (Figure 2E). In terms of route of administration, intravenous injection was the most common (Figure 2F), consistent with the drug’s label. Among reports that included outcome information, the top three outcomes were hospitalization (n = 172, 18.26%), death (n = 152, 16.14%), and life-threatening (n = 48, 5.10%) (Figure 2G). Regarding the type of reporter, the majority of reports were submitted by physicians (n = 415, 44.06%) and healthcare professionals (n = 264, 28.09%), with reports from consumers accounting for 10.30% (Figure 2H). The proportions of death and severe outcomes were 16.14% (Figure 2I) and 81.95% (Figure 2J), respectively. Most reports did not include information on concomitant medications (n = 601, 63.80%) (Figure 2K). The most common recorded indication was SCLC (n = 592), which is consistent with the approved indication for tarlatamab (Figure 2L). Because missingness varied substantially across variables, different analytic subsets were used for different downstream analyses. Descriptive summaries were based on all available reports for each variable, whereas subgroup, regression, and machine-learning analyses were restricted to reports with complete data for the variables required by each analysis. Body weight was not used in subgrouping or predictive modeling because 95.97% of reports lacked weight information.

Clinical characteristics of ADE reports associated with tarlatamab. (A) Year of report submission. (B) Age of reporters. (C) Sex distribution. (D) Body weight. (E) Country of origin of the report. (F) Administration route. (G) Outcome of the adverse event. (H) Reporter type. (I) Fatal outcome of the event. (J) Serious outcome of the event. (K) Therapy type. (L) Medical indications for tarlatamab use. Kg, kilogram; US, United States; JP, Japan; CA, Canada; KR, Korean.

3.2 Signal value detection based on the disproportionate analysis3.2.1 Signal detection at the SOC level

At the SOC level, the most frequently reported SOCs were General disorders and administration site conditions (n = 232), Nervous system disorders (n = 231), and Immune system disorders (n = 204). Using four disproportionality analysis methods, three SOCs met the predefined positive signal criteria across all algorithms (Supplementary Table S2). These were Nervous system disorders (ROR = 2.70, 95% CI: 2.35–3.11; PRR = 2.41; EBGM05 = 2.09; IC025 = 1.05), Immune system disorders (ROR = 14.28, 95% CI: 12.30–16.58; PRR = 12.27; EBGM05 = 10.54; IC025 = 3.32), and Neoplasms benign, malignant and unspecified (including cysts and polyps) (ROR = 4.02, 95% CI: 3.24–4.99; PRR = 3.82; EBGM05 = 3.08; IC025 = 1.57) (Figure 3A). In addition, Metabolism and nutrition disorders met the positive signal thresholds for the ROR and BCPNN methods (ROR = 1.73, 95% CI: 1.29–2.31; IC025 = 0.32), but not for PRR or MGPS.

Disproportionality analysis results at the SOC and PT levels. (A) The bar chart on the left illustrates the distribution of adverse event reports associated with tarlatamab across System Organ Class (SOC) categories, while the heatmap on the right displays signal strength identified by four disproportionality analysis methods. SOCs meeting the predefined algorithm-specific thresholds are highlighted with dark borders. (B) The bar chart presents the top 30 preferred terms (PTs) ranked by the number of reported cases. Percentages indicate the proportion of each specific adverse event relative to all reported adverse events, with different colors representing their corresponding SOC classifications. (C) Signal detection at the PT level. The circular heatmap depicts the 30 PTs that met the positive signal thresholds across all four disproportionality algorithms, illustrating case counts, signal strength, clinical priority, and whether they were classified as emerging signals. The innermost ring indicates the SOC classification of each PT.

3.2.2 Signal detection at the PT level

At the PT level, the most frequently reported adverse events were CRS (n = 201, 14.93%), ICANS (n = 106, 7.88%), pyrexia (n = 49, 3.64%), fatigue (n = 20, 1.49%), dysgeusia (n = 19, 1.41%), headache (n = 14, 1.04%), hypotension (n = 14, 1.04%), decreased appetite (n = 13, 0.97%), ageusia (n = 12, 0.89%), and weight decreased (n = 12, 0.89%) (Figure 3B). A total of 30 PTs met the positive signal criteria across all four disproportionality analysis methods (Table 3). Frequently reported PTs with positive signals included CRS (n = 201; ROR = 223.84, 95% CI: 192.15–260.77; PRR = 190.56; EBGM05 = 156.70; IC025 = 6.36), ICANS (n = 106; ROR = 312.43, 95% CI: 254.68–383.26; PRR = 287.90; EBGM05 = 220.07; IC025 = 5.96), pyrexia (n = 49; ROR = 7.10, 95% CI: 5.34–9.44; PRR = 6.88; EBGM05 = 5.16; IC025 = 2.20), dysgeusia (n = 19; ROR = 20.45, 95% CI: 12.99–32.20; PRR = 20.18; EBGM05 = 12.76; IC025 = 2.71), hypotension (n = 14; ROR = 3.90, 95% CI: 2.31–6.61; PRR = 3.87; EBGM05 = 2.29; IC025 = 0.95), and ageusia (n = 12; ROR = 26.04, 95% CI: 14.73–46.05; PRR = 25.82; EBGM05 = 14.52; IC025 = 2.34). Several PTs not listed in the drug label also met the positive signal criteria, including incontinence (n = 4; ROR = 21.60, 95% CI: 8.08–57.78; PRR = 21.54; EBGM05 = 8.01; IC025 = 0.78), dyspnoea at rest (n = 3; ROR = 41.09, 95% CI: 13.16–128.24; PRR = 41.00; EBGM05 = 13.01; IC025 = 0.44), intestinal perforation (n = 3; ROR = 13.61, 95% CI: 4.38–42.32; PRR = 13.58; EBGM05 = 4.35; IC025 = 0.26), and unresponsive to stimuli (n = 3; ROR = 6.83, 95% CI: 2.20–21.22; PRR = 6.82; EBGM05 = 2.19; IC025 = 0.03). Several PTs with small numbers of reports showed high disproportionality estimates, including tumour lysis syndrome (TLS) (n = 9; ROR = 43.37, 95% CI: 22.44–83.81; PRR = 43.09; EBGM05 = 22.08; IC025 = 2.12), neurotoxicity (n = 7; ROR = 16.49, 95% CI: 7.84–34.71; PRR = 16.41; EBGM05 = 7.77; IC025 = 1.46), mental status changes (n = 5; ROR = 12.32, 95% CI: 5.11–29.70; PRR = 12.28; EBGM05 = 5.08; IC025 = 0.91), and central nervous system lesion (n = 3; ROR = 18.90, 95% CI: 6.07–58.81; PRR = 18.86; EBGM05 = 6.03; IC025 = 0.34) (Figure 3C).

SOC namePT nameCase numberROR (95% CI)PRR (χ²)EBGM (EBGM05)IC(IC025)DeadFatality rateCytokine release syndromeImmune system disorders201223.84 (192.15–260.77)190.56 (36330.39)182.55 (156.7)7.51 (6.36)280.139303483Immune effector cell-associated neurotoxicity syndromeNervous system disorders106312.43 (254.68–383.26)287.9 (28420.04)269.97 (220.07)8.08 (5.96)120.113207547PyrexiaGeneral disorders and administration site conditions497.1 (5.34–9.44)6.88 (247.03)6.87 (5.16)2.78 (2.2)20.040816327DysgeusiaNervous system disorders1920.45 (12.99–32.2)20.18 (344.99)20.09 (12.76)4.33 (2.71)20.105263158HypotensionVascular disorders143.9 (2.31–6.61)3.87 (29.91)3.87 (2.29)1.95 (0.95)10.071428571AgeusiaNervous system disorders1226.04 (14.73–46.05)25.82 (284.71)25.67 (14.52)4.68 (2.34)00Tumour lysis syndromeMetabolism and nutrition disorders943.37 (22.44–83.81)43.09 (366.42)42.67 (22.08)5.42 (2.12)40.444444444HypoxiaRespiratory, thoracic and mediastinal disorders810.5 (5.23–21.05)10.44 (68.17)10.42 (5.19)3.38 (1.38)10.125Taste disorderNervous system disorders89.18 (4.58–18.41)9.13 (57.85)9.12 (4.55)3.19 (1.29)10.125NeurotoxicityNervous system disorders716.49 (7.84–34.71)16.41 (100.95)16.35 (7.77)4.03 (1.46)00DeliriumPsychiatric disorders67.65 (3.43–17.07)7.62 (34.47)7.61 (3.41)2.93 (0.87)00Aspartate aminotransferase increasedInvestigations66.9 (3.09–15.39)6.87 (30.07)6.86 (3.08)2.78 (0.81)00HyponatraemiaMetabolism and nutrition disorders55.35 (2.22–12.89)5.34 (17.6)5.33 (2.21)2.41 (0.45)00Mental status changesPsychiatric disorders512.32 (5.11–29.7)12.28 (51.69)12.25 (5.08)3.61 (0.91)20.4Alanine aminotransferase increasedInvestigations54.93 (2.05–11.86)4.91 (15.58)4.91 (2.04)2.3 (0.39)00IncontinenceaRenal and urinary disorders421.6 (8.08–57.78)21.54 (77.96)21.44 (8.01)4.42 (0.78)00DisorientationPsychiatric disorders47.47 (2.8–19.96)7.46 (22.33)7.44 (2.79)2.9 (0.41)00Acute respiratory failureRespiratory, thoracic and mediastinal disorders49.1 (3.41–24.31)9.08 (28.7)9.06 (3.39)3.18 (0.5)