Disproportionality analysis and “perpetrator” or “accomplice” review of Baloxavir marboxil in public FAERS database

Abstract

Background

Baloxavir marboxil (Xofluza™) was initially approved for marketing in Japan and the United States in 2018, which is a first-in-class cap-dependent endonuclease inhibitor of the influenza virus polymerase PA subunit. Despite indicating that Baloxavir marboxil is effective and well-tolerated through a series of clinical trials, studies on its safety in real-world settings are scarce.

Methods

We conducted a mining analysis of the FDA Adverse Event Reporting System (FAERS) database from Q4 2018 to Q2 2024 to extract adverse events associated with Baloxavir marboxil. Descriptive statistics were first performed to characterize the included reports, covering demographic variables, clinical outcomes, and concomitant medications. We subsequently employed four disproportionality analysis algorithms—ROR, PRR, BCPNN, and MGPS—for signal detection. To verify the robustness of the identified signals, we performed comprehensive subgroup and sensitivity analyses.

Results

A total of 1,765 reports comprising 3,624 adverse events with baloxavir marboxil as the primary suspected drug were identified. Most reports originated from the United States (57.98%), followed by Japan (39.58%) and China (1.82%). Over 65% of reports were submitted by non-consumers. Fatal outcomes were reported in 10.89% of cases. In both the overall study population and the deceased subgroup, hypotensive drugs represented the most frequently reported concomitant medications alongside influenza therapeutics. Eight robust safety signals were ultimately detected: altered state of consciousness, anaphylactic reaction, anaphylactic shock, colitis ischaemic, drug eruption, intentional product use issue, melaena, and no adverse event. Among these, “altered state of consciousness” is not listed in the product label. Several labelled adverse reactions with unestablished causality—rash, urticaria, erythema multiforme, vomiting, delirium, abnormal behavior, and hallucinations—disappeared during subgroup and sensitivity analyses. Additionally, the Ω shrinkage measure for the baloxavir marboxil-warfarin interaction yielded Ω₀₂₅ = 1.42.

Conclusion

The eight ultimately detected signals represent robust positive findings. Reactions including rash, urticaria, erythema multiforme, vomiting, delirium, abnormal behavior, and hallucinations were determined to be false positives. The most severe outcome, death, primarily reflected the presence of pre-existing cardiovascular conditions. A potential interaction between baloxavir marboxil and warfarin was identified, warranting further investigation.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12879-026-12724-w.

Introduction

Influenza viruses often threaten public health on a seasonal epidemic basis, and occasional global pandemic. Although symptoms are typically mild, there remains a substantial risk of morbidity and mortality among vulnerable populations, including the elderly, neonates, pregnant individuals, and those with compromised immune systems or pre-existing medical conditions [1]. It has been estimated that globally, up to approximately 650,000 respiratory-related fatalities each year can be attributed to influenza infections [2, 3]. Even today, we must never forget that the influenza epidemic during the First World War was supposed to have killed more soldiers than the war itself [4]. The development of anti-influenza virus drugs is extremely important.

In February 2018, Baloxavir marboxil (Xofluza™; BXM) was approved in Japan and in October of the same year it was approved in the United States. BXM (formerly S-033188) is a prodrug which can be metabolized to active form baloxavir acid (BXA, formerly S-033447) in vivo, and the latter is a first-in-class cap-dependent endonuclease inhibitor of the influenza virus polymerase PA subunit [5, 6]. This makes it the first new drug approved global for the treatment of recently circulating influenza viruses since the marketing of the neuraminidase inhibitors (NAIs) oseltamivir and zanamivir in 1999 (USA), laninamivir in 2010 (Japan), and peramivir in 2014 (USA) [4]. This novel mechanism partially alleviates the current issue of influenza viruses developing resistance to neuraminidase inhibitors. A series of previous clinical trials and several years of post-marketing use have confirmed generally its efficacy and good tolerability [7–9]. BXM is now in post-marketing pharmacovigilance phase [4, 10], to address safety issues that remained undetected in clinical trials because of the rigorous exclusion criteria employed and limited clinical practice.

The FDA Adverse Event Reporting System (FAERS) is currently one of the world’s largest public pharmacovigilance databases, contains tens of millions of adverse drug event and medication errors reported by physicians, pharmacists, manufacturers, and others [11, 12]. Increasing evidence indicates that FAERS plays a significant role in detecting rare adverse drug reactions and in early identification of drug-related safety signals from a large population in real world [13–18].

The aim of this study is to delineate the safety signal profile of BXM and thoroughly investigate the underlying information of the detected signals through FAERS data mining, which can provide some reference for its clinical surveillance and risk identification.

Materials and methods

Data sources

The study is designed as an observational, retrospective vigilance analysis, involving data from the fourth quarter (Q4) of 2018 (FDA marketing approval of BXM) to the second quarter (Q2) of 2024 (the most recent update of the FAERS database at the time this study was performed) were downloaded from the FAERS database, that is free and publicly accessible (https://fis.fda.gov/extensions/FPD-QDE-FAERS/FPD-QDE-FAERS.html). Five types of datasets were used, including patient demographic and administrative information (DEMO), drug/biologic information (DRUG), adverse event encodings (REAC), patient outcomes (OUTC), and indication/diagnosis for use (INDI). All documents were downloaded from the FAERS database in ASCII format.

Data cleaning and processing

As FAERS is a spontaneous reporting system, duplicate records are inevitable and need to be deleted. To ensure the uniqueness of the reports, firstly, we deleted seriatim the cases based on the caseid provided in the additional deleted file for each quarterly document. Then, we performed the deduplication process according to the FDA recommendations by selecting the higher primaryid when the caseid are the same and selecting the most recent FDA_DT when the primaryid are the same [19]. Finally, by checking whether the number of dataset entries matches the count of unique primaryid values, we can verify that all duplicate entries have been completely excluded.

Cases of BXM as the primary suspected (PS) drug were identified using the generic name (prod_ai column as Baloxavir Marboxil). Once we have obtained cases of BXM as the PS drug, we can utilize the common identifier “primaryid” across datasets to extract data from other datasets. In the same report, if BXM was identified as the PS drug, other drugs in the role_cod field as “secondary suspect”, “concomitant”, or “interacting” were considered concomitant drugs. The adverse events were coded using the preferred terms (PTs), which were mapped to the corresponding primary system organ class (SOC) level based on the standardized Medical Dictionary for Regulatory Activities (MedDRA) version 27.0.

Pharmacovigilance signal detection

Disproportionality analysis is widely utilized in pharmacovigilance studies to identify safety signals and explore potential causal relationships between drugs and adverse events [20, 21]. Its fundamental principle involves considering reports of the target population or target medications in the pharmacovigilance database as cases, while regarding reports of non-target populations or non-target medications as non-cases (or background). This framework enables the calculation of whether the frequency of a specific adverse event (observed frequency) in the case population is disproportionate to its frequency in the non-case population (anticipated frequency). When the observed frequency of an adverse event significantly exceeds the anticipated frequency, it becomes a suspicious safety signal requiring further investigation.

In case-non-case studies, there are four primary algorithms, including the reporting odds ratio (ROR), the proportional reporting ratio (PRR), the Bayesian confidence propagation neural network (BCPNN), and the multi-item gamma Poisson shrinker (MGPS). The first two are the most used frequency counting methods, they have been offered as a way of addressing the non-selective underreporting of certain medications or adverse events. However, one disadvantage is that ROR and PRR may be undefined if there are no instances of a condition in non-case population (b = 0). It may also produce abnormally high results for rare events (when the values for a, b, and c are too low). Both the BCPNN and MGPS methods employ Bayesian estimation to adjust observed-to-anticipated ratios, thereby improving the stability of the results. The BCPNN is particularly effective for detecting signals associated with rare adverse events or sparse reports, even with missing data. In contrast, the MGPS is designed to identify drug-adverse event combinations that have “interestingly large counts in a large frequency table containing millions of cells, most of which have an observed frequency of 0 or 1.” [22].

In the present study, all four algorithms were employed simultaneously to minimize biases caused by the limitations of any single method. A PT was considered a valid signal of disproportionate reporting (SDR) only when it concurrently met the criteria of all four algorithms. The fourfold table for disproportionality analysis, along with the formulas and criteria for the four algorithms used in BXM signal detection, are provided in Supplementary Tables 1 and 2.

Subgroup and sensitivity analysis

Recognizing that spontaneous reporting databases are susceptible to various confounders and biases [23], we further conducted a series of subgroup and sensitivity analyses, consistently applying all four disproportionality algorithms across these assessments. The case and non-case groups were rigorously matched to maintain homogeneity. For instance, if cases were defined as males associated with BXM, non-cases comprised male reports associated with other drugs.

The subgroup analyses included: males, females, age < 18 years, age ≥ 18 to < 65 years, age ≥ 65 years, reports excluding consumer sources, and reports from 2021 onward. To address confounding by indication, a sensitivity analysis restricted to influenza patients was performed. Influenza cases were identified in the INDI sub-dataset using the ‘indi_pt’ field with the following terms: ‘Influenza’, ‘Influenza immunisation’, ‘Influenza like illness’, ‘H1N1 influenza’, ‘Influenza A virus test positive’, ‘H3N2 influenza’, ‘Avian influenza’, ‘Influenza B virus test positive’, ‘Influenza virus test positive’, ‘Influenza A virus test’, ‘Influenza B virus test’, and ‘Influenza virus test’.

Additionally, to minimize interference from concomitant medications, a case-non-case analysis was conducted excluding users of hypotensive drugs. The list of 37 hypotensive medications was derived from those co-reported in cases where BXM was the primary suspected drug, including: Amlodipine, Losartan, Telmisartan, Bisoprolol, Valsartan, Candesartan, Irbesartan, Carvedilol, Azilsartan, Enalapril, Metoprolol, Cilnidipine, Lisinopril, Olmesartan, Azelnidipine, Eplerenone, Guanfacine, Imidapril, Nifedipine, Tolvaptan, Verapamil, Aliskiren, Arotinolol, Benidipine, Bevantolol, Diltiazem, Doxazosin, Esaxerenone, Flecainide, Manidipine, Nebivolol, Niacin, Nicorandil, Nisoldipine, Ranolazine, Trapidil, and Urapidil. Different salt forms were disregarded during screening.

Signals that disappeared in these subgroup or sensitivity analyses were considered potential false positives attributable to various biases and confounding factors. Signals that persisted in at least one age subgroup, one gender subgroup, and across all sensitivity analyses were considered robust.

Statistical analysis

Descriptive analysis was employed to summarize the clinical features of cases occurred BXM-related adverse events, including sex, age, weight, occurred country, reporters, reporting year, outcomes and concomitant drugs. Fishers exact test was used for hypothesis testing of categorical variables, while t-test is for the continuous variables. A p value less than 0.05 was considered statistically significant. All data processing and statistical analyses were performed using Jupyter Notebook 6.5.4, providing a Python 3 (ipykernel) computational environment.

Results

Descriptive analysis

Demographic characteristics

The comprehensive data collection and processing workflow of this study was illustrated in Fig. 1.After data cleaning, 1765 BXM-associated case reports and 3624 adverse events with BXM as the PS drug were filtered out. The clinical characteristics of these 1765 case reports are shown in Table 1.

Fig. 1.

Flow diagram of data collection and processing of BXM-associated adverse events. AEs: adverse events; ROR: reporting odds ratio; PRR: proportional reporting ratio; BCPNN: Bayesian confidence propagation neural network; MGPS: multi-item gamma Poisson shrinker; PS: primary suspected drug; SOC: system organ class

Table 1.

Characteristics BXM-associated AE reports (n = 1,765)

Number of events	Available number, n	Case number, n	Case proportion, %
Sex	1388		78.64
Female		739	53.24
Male		649	46.76
NA	377		21.36
Age (years)	1149		65.10
< 5		23	2.00
≥ 5		328	28.55
≥ 12		106	9.23
≥ 18		451	39.25
≥ 65		142	12.36
≥ 80		99	8.62
NA	616		34.90
Median (IQR)	32 (11, 60)
Weight (kg)	253		14.33
< 20		7	2.77
20 ~ 80		178	70.36
> 80		68	26.88
NA	1512		85.67
Median (IQR)	65.83(53, 81)
Reported countries	1761		99.77
US		1021	57.98
JP		697	39.58
CN		32	1.82
Others		11	0.62
NA	4
Reporters	1761		99.77
MD		670	38.05
PH		232	13.17
HP		176	9.99
CN		597	33.90
OT		86	4.88
NA	4		0.23
Reporting year	1765		100.00
2024Q1 & Q2		83	4.70
2023		132	7.48
2022		77	4.36
2021		39	2.21
2020		647	36.66
2019		777	44.02
2018Q4		10	0.57

MD: physician, PH: pharmacist, HP: health-professional, CN: Consumer, OT: Other

Sex data were available for 1388 cases. Among these, females accounted for 53.24% (n = 739) and males accounted for 46.76% (n = 649). Age data was available for 1149 cases, with the median age of 32 years (IQR: 11–60). Dividing the patients into six different age groups using 5, 12, 18, 65 and 80 years as cutoffs, we observed that the cases aged 18–65 years accounted for a higher proportion compared with other age groups (n = 451, 39.25%). It is a remarkable fact that 23 cases under 5 years old contributed 2%, but FDA has not yet approved BXM used for these children. Weight data was available for only 14.33% of cases, with a median weight of 65.83 kg (IQR: 53–81). Regarding the reporting countries, the highest number of cases (n = 1021, 57.98%) was reported from the USA, followed by Japan (n = 697, 39.58%) and China (n = 32, 1.82%). Excluding 4 reports with unknown reporters, physicians (MD) reported the most adverse event reports at 38.05% (n = 670), consumers and pharmacists reported the adverse events at 33.90% (n = 597) and 13.17% (n = 232), respectively. Approximately 80% of adverse event reports associated with BXM were concentrated in the years 2019 and 2020, with reports plummeting sharply thereafter, suggesting the potential presence of the Weber effect [23].

Concomitant drugs profile

We consider that concomitant medication profiles can serve as a proxy for the distribution of underlying conditions in the study population. A total of 1,698 concomitant medication records were documented, involving 470 cases in this study. Due to the presence of multi-component medications, we processed the data on concomitant medications. In simple terms, we split most of the multi-component medications, for example, “amlodipine besylate\azilsartan” and “amlodipine besylate\telmisartan”, each with one record, would be split into “amlodipine besylate” with 2 records, “azilsartan” with one record, and “telmisartan” with one record. However, there are some fixed multi-component medications that we did not split, such as “carbidopa\levodopa”, “sulfamethoxazole\trimethoprim”, “piperacillin sodium\tazobactam sodium”, etc. It is worth noting that all multi-component medications containing “acetaminophen” were split. Finally, we gathered 418 types of concomitant medications through case-by-case examination, disregarding the distinction between specific salts of the same drug. As these drugs exhibited significant overlap in their therapeutic categories, we proceed to classify them into 66 therapeutic categories with reference to the ATC codes (https://atcddd.fhi.no/atc_ddd_index/) and the MCDEX Clinical Drug Reference (MCDEX, version 2008), as well as their indications in INDI, proprietary drug names, dose, and route in DRUG (where available on a case-by-case basis). Thus, we can conduct a more precise analysis. Figure 2A presents the top 20 concomitant drugs, along with the case counts and their respective percentages within the complete set of cases. Similarly, Fig. 2B illustrates the top 20 therapeutic categories, detailing the case counts and their proportion within the complete set of cases. Acetaminophen, carbocysteine, oseltamivir, dextromethorphan, amlodipine, were the most frequently concomitant drugs, accounting for 11.73% (n = 207), 3.80% (n = 67), 3.0% (n = 53), 2.15% (n = 38) and 1.98% (n = 35), respectively. However, the top five categories are indeed Nonsteroidal Anti-Inflammatory Drugs (NSAIDs), antitussives & expectorants & antiasthmatics (ATEA), hypotensives, antiallergics, as well as antibiotics, accounting for 15.35% (n = 271), 15.01% (n = 265), 7.99% (n = 141), 6.35% (n = 112) and 4.65% (n = 82), respectively.

Fig. 2.

Concomitant drugs and Outcomes analysis. A presents the top 20 concomitant drugs. B illustrates the top 20 categories of concomitant drugs. For A and B, color green represents case counts, and color pink represents corresponding proportions within the complete set of cases. C displays the incidence of various severe outcomes, including death (DE), disabilities (DS), Hospitalization-initial or prolonged (HO), life-threatening (LT), and other serious outcomes (OT). The number of cases and the percentage of various serious outcomes are shown on the right side and in the middle of the bars, respectively. D illustrates the age distribution and gender differences among 80 cases of death outcomes. ATEA: Antitussives & expectorants & antiasthmatics; NSAIDs: Nonsteroidal anti-inflammatory drugs

Outcome distribution

We evaluated the potential harm of BXM-associated adverse events by investigating the reporting rate of severe outcomes, including death (DE), disabilities (DS), hospitalization-initial or prolonged (HO), life-threatening (LT), and other serious outcomes (OT). The number and percentage of serious outcomes were shown in Fig. 2C. Removing two cases of congenital anomaly, in 907 cases with reported outcomes, the death, disabilities, hospitalization and life-threatening rates were 10.89%, 1.32%, 34.21% and 6.93%, respectively.

Because of the gravity of mortality, we conducted a distinct analysis on the demographic traits and clinical medications for fatal cases. In the 80 cases of death outcomes that mentioned sex, the proportion of death outcomes in males was significantly higher than that in females (male, n = 48/649; female, n = 32/739; p = 0.0154). The age distribution of these cases exhibited a severe skewness. Among the 79 cases that reported age, the median age was 73 years (IQR: 55.5–86.5). Furthermore, there was no significant statistical difference in the age distribution between males and females. Figure 2D effectively illustrates these details.

We further analyzed the concomitant medications and their corresponding indications in the cases with a fatal outcome, aiming to investigate the patterns of polypharmacy and comorbidity in these cases. For the DRUG file and the INDI file, the entries can be linked using the “prod_ai” and “drug_seq” fields. Ultimately, among the 99 cases with a fatal outcome, 40 individuals reported concomitant medications, ranging from one to dozens of different medications, totaling 195 records of concomitant medication involving 119 types. However, only 108 records (33 types) of indications were found, and the remaining entries without indications were filled with “unknown”. As mentioned in our Concomitant drugs profile, we manually classified each concomitant medication case-by-case into a therapeutic category by referencing the ATC codes (https://atcddd.fhi.no/atc_ddd_index/) and the MCDEX Clinical Drug Reference (MCDEX, version 2008), as well as their indications in INDI, proprietary drug names, dose, and route in DRUG (where available on a case-by-case basis). Finally, these 119 medications (multi-component medications were not split because their proportion is small) fell into 40 therapeutic categories. Therefore, we used a Sankey diagram to display the data flow from these concomitant medications to their indications and therapeutic categories (Fig. 3). The diagram clearly shows the comorbidity of the patients, such as the top three indications being influenza, hypertension and fever, and the top five therapeutic categories of concomitant medications are indeed NSAIDs, antitussives & expectorants & antiasthmatics (ATEA), antibiotics, gastric antisecretory drugs, as well as hypotensives. In the right pink bar of this plot, we have specifically highlighted therapeutic categories with high flow in other colors, such as hypotensives and diuretic & hypotensives, which are shown in red.

Fig. 3.

Sankey plot illustrating the concomitant medications and their corresponding indications in the cases with a fatal outcome. The green bar in the middle represents 119 kinds of concomitant medications (multi-component medications not separated), the sky-blue bar on the left represents 33 indications, and the pink bar on the right represents 40 therapeutic categories. The size of the bars represents the number of reports they have, and the gray arcs illustrate the flow of the drug between indications and therapeutic categories. The total flow size is 195 entries involving combination medications. In the right pink bar, we have specifically highlighted therapeutic categories with high flow in other colors, such as hypotensives and diuretic & hypotensives, which are shown in red. ISDN: Isosorbide dinitrate, AMC\CLVP: Amoxicillin\clavulanate potassium, CHLPM\CDN\MEPH: Chlorpheniramine maleate\dihydrocodeine phosphate\methylephedrine hydrochloride, APAP: acetaminophen, APAP\CAF\PMZ\SALAMIDE: Acetaminophen\caffeine\promethazine\salicylamide, ATEA: Antitussives & expectorants & antiasthmatics

Signal detection

Among all 3,624 BXM-associated adverse events (representing 562 PTs), 50 PT signals across 18 SOC categories concurrently met the criteria of all four disproportionality algorithms used in this study. The case numbers and signal strengths for these detected PT-level signals are presented in Table 2. These 50 PTs constitute the crude signals, representing the broadest set of hypothesized safety signals generated. Notably, the no adverse event (ROR = 57.03, n = 47) signal was detected. This is attributable to the packaging of the BXM suspension and its unique dosing regimen (40 mg for patients > 5 years and > 20 kg; 80 mg for those > 40 kg), which resulted in the signal being accompanied by reports such as off-label use (ROR = 6.50, n = 38), intentional product use issue (ROR = 31.67, n = 27), product administered to patient of inappropriate age (ROR = 40.17, n = 19), product packaging quantity issue (ROR = 26.04, n = 18), and medication error (ROR = 7.23, n = 13).

Table 2.

The case number and signal strength of BXM-associated adverse events at the PT level

pt	a	b	c	d	ROR		PRR		BCPNN		MGPS		SOC
					ROR	95%CI-L	PRR	X2	IC	IC025	EBGM	EBGM05
Abnormal behaviour	26	3598	14,199	36,938,626	18.80	12.78	18.67	434.22	4.22	2.88	18.64	12.67	Psychiatric disoders
Altered state of consciousness	18	3606	11,487	36,941,338	16.05	10.10	15.98	252.42	4.00	2.42	15.95	10.04	Nervous system disorders
Anaphylactic reaction	35	3589	28,974	36,923,851	12.43	8.91	12.32	363.80	3.62	2.70	12.30	8.82	Immune system disorders
Anaphylactic shock	19	3605	12,623	36,940,202	15.42	9.82	15.35	254.55	3.94	2.44	15.33	9.76	Immune system disorders
Bronchitis	22	3602	48,259	36,904,566	4.67	3.07	4.65	63.05	2.22	1.34	4.65	3.06	Infections and infestations
Cardio-respiratory arrest	7	3617	15,540	36,937,285	4.60	2.19	4.59	19.68	2.20	0.45	4.59	2.19	Cardiac disorders
Colitis ischaemic	29	3595	3054	36,949,771	97.60	67.61	96.83	2724.59	6.58	3.95	95.92	66.45	Gastrointestinal disorders
Cyanosis	5	3619	6611	36,946,214	7.72	3.21	7.71	29.19	2.95	0.42	7.71	3.20	Vascular disorders
Cystitis haemorrhagic	4	3620	2097	36,950,728	19.47	7.30	19.45	69.88	4.28	0.44	19.41	7.28	Renal and urinary disorders
Delirium	16	3608	17,899	36,934,926	9.15	5.60	9.11	115.55	3.19	1.84	9.11	5.57	Psychiatric disoders
Delirium febrile	7	3617	45	36,952,780	1589.22	716.22	1586.15	9596.33	10.42	1.81	1372.77	618.67	Psychiatric disoders
Depressed level of consciousness	11	3613	19,446	36,933,379	5.78	3.20	5.77	43.35	2.53	1.08	5.77	3.19	Nervous system disorders
Diarrhoea	106	3518	400,547	36,552,278	2.75	2.27	2.70	114.53	1.43	1.11	2.70	2.22	Gastrointestinal disorders
Disseminated intravascular coagulation	7	3617	5856	36,946,969	12.21	5.81	12.19	71.82	3.61	1.13	12.18	5.80	Blood and lymphatic system disorders
Drug eruption	12	3612	9332	36,943,493	13.15	7.46	13.11	134.12	3.71	1.85	13.10	7.43	Skin and subcutaneous tissue disorders
Encephalopathy	8	3616	12,347	36,940,478	6.62	3.31	6.61	38.05	2.72	0.89	6.60	3.30	Nervous system disorders
Enterocolitis	6	3618	3846	36,948,979	15.93	7.15	15.91	83.70	3.99	1.03	15.88	7.13	Gastrointestinal disorders
Enterocolitis haemorrhagic	4	3620	654	36,952,171	62.43	23.35	62.37	240.06	5.95	0.62	61.99	23.18	Gastrointestinal disorders
Erythema multiforme	14	3610	4258	36,948,567	33.65	19.89	33.53	440.38	5.06	2.56	33.42	19.76	Skin and subcutaneous tissue disorders
Face oedema	9	3615	7555	36,945,270	12.17	6.33	12.15	91.97	3.60	1.46	12.13	6.31	General disorders and administration site conditions
Facial paralysis	4	3620	6952	36,945,873	5.87	2.20	5.87	16.14	2.55	-0.04	5.86	2.20	Nervous system disorders
Febrile convulsion	5	3619	482	36,952,343	105.92	43.86	105.77	513.60	6.71	1.08	104.70	43.35	Nervous system disorders
Haematochezia	15	3609	40,435	36,912,390	3.79	2.28	3.78	30.73	1.92	0.87	3.78	2.28	Gastrointestinal disorders
Hallucination	16	3608	43,427	36,909,398	3.77	2.31	3.76	32.39	1.91	0.90	3.76	2.30	Psychiatric disoders
Hepatic function abnormal	16	3608	21,205	36,931,620	7.72	4.73	7.69	93.16	2.94	1.67	7.69	4.70	Hepatobilliary disorders
Hyperpyrexia	4	3620	2101	36,950,724	19.43	7.28	19.41	69.73	4.28	0.44	19.38	7.26	General disorders and administration site conditions
Ileus paralytic	3	3621	1893	36,950,932	16.17	5.21	16.16	42.60	4.01	-0.10	16.14	5.20	Gastrointestinal disorders
Influenza	36	3588	82,391	36,870,434	4.49	3.23	4.46	96.65	2.16	1.51	4.45	3.21	Infections and infestations
Intentional product use issue	278	3346	96,682	36,856,143	31.67	28.02	29.32	7602.48	4.87	4.54	29.24	25.87	Injury, poisoning and procedural complications
International normalised ratio increased	7	3617	7479	36,945,346	9.56	4.55	9.54	53.50	3.25	0.99	9.54	4.54	Investigations
Lip swelling	8	3616	15,605	36,937,220	5.24	2.62	5.23	27.35	2.39	0.70	5.23	2.61	Gastrointestinal disorders
Loss of consciousness	36	3588	61,523	36,891,302	6.02	4.33	5.97	148.99	2.58	1.88	5.96	4.29	Nervous system disorders
Lower respiratory tract congestion	4	3620	1950	36,950,875	20.94	7.85	20.92	75.71	4.38	0.46	20.88	7.82	Respiratory, thoracic and mediastinal disorders
Medication error	13	3611	18,380	36,934,445	7.23	4.20	7.21	69.54	2.85	1.44	7.21	4.18	Injury, poisoning and procedural complications
Melaena	23	3601	11,054	36,941,771	21.35	14.16	21.22	442.27	4.40	2.87	21.17	14.05	Gastrointestinal disorders
Mouth haemorrhage	3	3621	3943	36,948,882	7.76	2.50	7.76	17.65	2.95	-0.33	7.75	2.50	Gastrointestinal disorders
No adverse event	473	3151	97,014	36,855,811	57.03	51.76	49.71	22528.52	5.63	5.34	49.48	44.91	General disorders and administration site conditions
Off label use	380	3244	654,420	36,298,405	6.50	5.84	5.92	1581.23	2.57	2.38	5.92	5.32	Injury, poisoning and procedural complications
Pneumonia	92	3532	195,082	36,757,743	4.91	3.99	4.81	278.88	2.27	1.89	4.81	3.91	Infections and infestations
Pneumonia bacterial	16	3608	6367	36,946,458	25.73	15.74	25.62	377.72	4.68	2.60	25.56	15.63	Infections and infestations
Pneumonia influenzal	10	3614	809	36,952,016	126.39	67.69	126.04	1225.37	6.96	2.34	124.51	66.68	Infections and infestations
Prescribed underdose	14	3610	17,496	36,935,329	8.19	4.84	8.16	87.92	3.03	1.62	8.15	4.82	Injury, poisoning and procedural complications
Product administered to patient of inappropriate age age	19	3605	4848	36,947,977	40.17	25.57	39.96	719.04	5.32	3.04	39.81	25.34	Injury, poisoning and procedural complications
Product packaging quantity issue	18	3606	7082	36,945,743	26.04	16.38	25.92	430.18	4.69	2.74	25.85	16.26	Product issues
Rhabdomyolysis	25	3599	17,774	36,935,051	14.43	9.74	14.34	310.01	3.84	2.62	14.32	9.66	Musculoskeletal and connective tissue disorders
Seizure	25	3599	81,142	36,871,683	3.16	2.13	3.14	36.57	1.65	0.91	3.14	2.12	Nervous system disorders
Shock	6	3618	10,526	36,942,299	5.82	2.61	5.81	23.90	2.54	0.47	5.81	2.61	Vascular disorders
Swelling of eyelid	4	3620	5132	36,947,693	7.96	2.98	7.95	24.28	2.99	0.12	7.94	2.98	Eye disorders
Urticaria	37	3587	87,965	36,864,860	4.32	3.13	4.29	93.50	2.10	1.47	4.29	3.10	Skin and subcutaneous tissue disorders
Vomiting	88	3536	243,663	36,709,162	3.75	3.03	3.68	173.04	1.88	1.51	3.68	2.98	Gastrointestinal disorders

Subgroup and sensitivity analysis

In the sex-stratified subgroup analyses, 32 signals were detected in the male group and 38 in the female group. Compared to the crude signals, 8 signals were not detected in either the male or female subgroup, suggesting these 8 signals are false positives. The signal detection parameters and case count for the male and female subgroups are presented in Supplementary Tables 3 and 4.

In the age-stratified analyses, 14 signals were detected in the < 18 years group, 30 in the ≥ 18 to < 65 years group, and 17 in the ≥ 65 years group. A total of 15 crude signals were not detected in any age stratum. The signal detection parameters and case count for each age group are shown in Supplementary Tables 5–7.

Given that a substantial proportion of reports (597 cases, 33.90%) originated from consumers, a subgroup analysis was conducted using only non-consumer reports to mitigate potential reporting source bias. This analysis yielded 45 signals (Supplementary Table 8), with 10 crude signals not being detected.

Due to the potential influence of the Weber effect within the reporting timeframe, reports filed more than two years post-BXM approval were analyzed separately (Supplementary Table 9). However, the case count in this time-restricted subgroup was substantially reduced (n = 331). Consequently, we conclude that this subgroup requires a longer follow-up period with more accumulated cases to provide reliable results.

Several crude signals were related to influenza and its complications (e.g., Influenza, Influenza pneumonia, Bronchitis, Pyrexia), suggesting substantial confounding by indication. To investigate whether these signals were drug-specific or indication-specific, a sensitivity analysis restricted the case and non-case populations to influenza patients. This analysis detected only 12 signals, and all influenza-related signals disappeared. These results are provided in Supplementary Table 10.

Analysis of concomitant medications, including those in fatal cases, indicated that antihypertensive drugs were prominent. This suggested that antihypertensive drug use, or the underlying hypertension itself, could influence signal detection. Therefore, a sensitivity analysis was performed, restricting the population to non-users of the specified hypotensive drugs. This analysis detected 45 signals, with 6 crude signals disappearing (Supplementary Table 11).

The presence or absence of the 50 crude signals across these subgroup and sensitivity analyses is summarized in Table 3. This summary allows for a clearer identification of signals affected by biases and confounders, as well as signals specific to particular subpopulations. After accounting for various confounding factors and biases, we ultimately identified eight robust positive safety signals (bolded in Table 3). These signals persisted in sensitivity analyses and were detected in at least one sex subgroup and one age subgroup. It should be emphasized that all signals ultimately identified represent disproportional reporting signals and do not indicate confirmed adverse drug reactions, even after multiple subgroup and sensitivity analyses.

Table 3.

The 50 crude signals across these subgroup and sensitivity analyses

Crude signals	Sex - subgroups		Age - subgroups			Sensitivity analyses
	Male	Female	< 18 years	≥ 18 to < 65 years	≥ 65 years	Non-consumer-reports	Influenza patients	Hypotensives’s non-users
Abnormal behaviour	+	+	+	-	-	+	-	+
Altered state of consciousness	+	+	-	+	+	+	+	+
Anaphylactic reaction	+	+	-	+	-	+	+	+
Anaphylactic shock	+	+	+	+	-	+	+	+
Bronchitis	+	+	-	+	-	+	-	+
Cardio-respiratory arrest	-	-	-	-	-	-	-	-
Colitis ischaemic	+	+	+	+	-	+	+	+
Cyanosis	-	-	-	-	-	+	-	+
Cystitis haemorrhagic	-	+	-	-	-	+	-	+
Delirium	+	+	+	-	-	+	-	+
Delirium febrile	+	-	+	-	-	+	-	+
Depressed level of consciousness	-	+	-	-	+	-	-	+
Diarrhoea	-	+	-	+	-	+	-	+
Disseminated intravascular coagulation	-	+		+	-	+	-	+
Drug eruption	+	+	-	+	-	+	+	+
Encephalopathy	-	+	-	-	-	+	-	+
Enterocolitis	-	+	-	+	-	+	-	+
Enterocolitis haemorrhagic	-	+	-	+	-	+	-	+
Erythema multiforme	+	+	+	+	-	+	-	+
Face oedema	-	+	-	+	-	+	-	+
Facial paralysis	-	-	-	-	-	-	-	-
Febrile convulsion	-	+	+	-	-	+	-	+
Haematochezia	-	+	-	+	-	+	-	+
Hallucination	-	+	-	-	-	+	-	+
Hepatic function abnormal	+	+	-	+	-	+	-	+
Hyperpyrexia	+	-	-	-	-	-	-	+
Ileus paralytic	-	+	-	-	-	-	-	-
Influenza	+	-	-	-	+	+	-	+
Intentional product use issue	+	+	+	-	+	+	+	+
International normalised ratio increased	+	-	-	-	-	+	-	-
Lip swelling	+	-	-	-	-	+	-	+
Loss of consciousness	+	+	-	+	+	+	-	+
Lower respiratory tract congestion	-	-	-	-	-	-	-	+
Medication error	+	+	+	+	-	-	-	+
Melaena	+	+	-	+	-	+	+	+
Mouth haemorrhage	-	-	-	-	-	-	-	-
No adverse event	+	+	+	+	+	+	+	+
Off label use	+	+	+	-	-	-	+	+
Pneumonia	+	+	-	+	+	+	-	+
Pneumonia bacterial	-	-	-	-	-	+	+	+
Pneumonia influenzal	+	+	+	+	-	+	-	+
Prescribed underdose	+	+	-	+	-	+	-	+
Product administered to patient of inappropriate age	+	+	+	-	-	-	+	+
Product packaging quantity issue	+	+	+	-	-	+	-	+
Rhabdomyolysis	+	+	-	+	+	+	-	+
Seizure	+	-	-	-	-	+	-	+
Shock	-	+	-	+	-	+	-	-
Swelling of eyelid	-	-	-	-	-	+	-	+
Urticaria	+	+	-	+	-	+	-	+
Vomiting	-	-	-	+	-	+	-	+

Discussion

This study first conducted a detailed characterization of the included reports, identifying key biases and confounders—including reporting source bias, the Weber effect, confounding by indication, and the influence of hypertension pathology and related medications—to be considered in the disproportionality analysis of BXM-associated adverse events. A total of eight robust positive safety signals were identified through a series of stratified designs to mitigate these factors. These signals, which reflect disproportionate reporting rather than a confirmed causal association, were: altered state of consciousness (ROR = 16.05, n = 18), anaphylactic reaction (ROR = 12.43, n = 35), anaphylactic shock (ROR = 15.42, n = 19), colitis ischaemic (ROR = 97.60, n = 29), drug eruption (ROR = 13.15, n = 12), intentional product use issue (ROR = 31.67, n = 27), melaena (ROR = 21.35, n = 23), and no adverse event (ROR = 57.03, n = 47). Among these, “altered state of consciousness” is not listed in the product label. Furthermore, this study provides negating evidence for labelled adverse reactions with unestablished causality—such as rash, urticaria, erythema multiforme, vomiting, delirium, abnormal behavior, and hallucinations.

Notably, the detection of the “no adverse event” signal indicates disproportionate reporting of medication-related events unrelated to BXM’s pharmacological action. It reflects reporting behavior rather than a biological or clinical phenomenon, does not constitute a safety outcome, and should not be interpreted as evidence of a protective or clinical effect of BXM. It may suggest that for BXM, a single-dose, newly approved medication, patient education on drug use is particularly important.

This work identified eight crude signals that disappeared in both sex- and age-stratified analyses, including cardiorespiratory arrest, cyanosis, facial paralysis, hallucination, lower respiratory tract congestion, mouth haemorrhage, pneumonia bacterial, and swelling of eyelid. These findings indicate these signals represent definitive false positives.

Five signals—depressed level of consciousness, hyperpyrexia, medication error, off-label use, and product administered to patient of inappropriate age—were no longer detected after excluding consumer reports, suggesting they are attributable to reporter bias. The shock signal disappeared after excluding users of antihypertensive drugs, indicating its association with underlying patient pathology.

Restricting the analysis to influenza patients resulted in the disappearance of most crude signals, confirming substantial confounding by indication in BXM safety reports. For instance, while Yunsong Li and colleagues reported rhabdomyolysis as an adverse reaction not listed in the BXM prescribing information in their drug safety comparison of oseltamivir and BXM [24], this signal disappeared in our influenza-restricted analysis. This suggests rhabdomyolysis represents a signal specific to the influenza population, consistent with known influenza-related complications [25–27]. The signals for rash, urticaria, erythema multiforme, vomiting, delirium, and abnormal behavior—which are listed with unestablished causality in the BXM prescribing information—are also attributable to influenza itself.

Prior to conducting subgroup and sensitivity analyses, we observed an unlisted signal: “International normalised ratio increased,” involving 7 patients. Given that INR is typically monitored in patients taking anticoagulants like warfarin, we investigated the concomitant medications of these cases. As anticipated, 6 of them reported concurrent warfarin use.

Since anticoagulants represented a substantial proportion of concomitant medications, with warfarin ranking just below the top 20, we matched DEMO and REAC data for all warfarin users. A comprehensive case-by-case review identified 12 patients with documented warfarin use (excluding one case reporting INR increase without warfarin documentation). Among these, 6 exhibited increased INR, while the remaining 6 reported distinct bleeding events: internal haemorrhage, small intestinal haemorrhage, cystitis haemorrhagic, renal haemorrhage, acute kidney injury, and pulmonary alveolar haemorrhage. Notably, 4 of these bleeding cases required hospitalization, and 2 were life-threatening. All 12 reports originated from healthcare professionals (6 physicians and 6 pharmacists).

Although the INR increase signal was subsequently identified as a false positive in formal analyses, these findings suggest a potential drug-drug interaction between BXM and warfarin. Supporting this, Tohoku University Hospital reported a case of a patient with an implantable ventricular assist device whose INR significantly increased beyond the target range following BXM administration, despite previously stable self-monitored INR and precise warfarin dosing. The authors also suggested a potential warfarin-BXM interaction [28].

The Ω shrinkage measure, utilized by WHO-UMC for drug-drug interaction signal detection [29], demonstrates conservative performance among frequentist-based algorithms [30, 31]. A signal is considered positive when the lower limit of the 95% confidence interval for Ω (Ω₀₂₅) exceeds zero. Application of this algorithm to REAC data for BXM and warfarin yielded an Ω (Ω₀₂₅) value of 3.48 (1.42).

While the Ω shrinkage signal and case-by-case review suggest a potential interaction, they do not establish the risk magnitude or frequency of a BXM-warfarin interaction, which should be confirmed through clinical or experimental studies—an endeavor beyond the scope of this study.

In the analysis of concomitant medications in all cases, hypotensive agents hold a 3rd position, even outperformed both antiallergics and antibiotics against flu symptoms. It is important to emphasize that the hypotensive agents referred to in this study include alfa or beta blocking agents, agents acting on the renin-angiotensin system, calcium channel blockers, vasodilators and agents acting principally on the central nervous system. Therefore, the hypotensives mentioned here are not only for hypertension but also for all cardiovascular diseases that require these treatments. In the Sankey diagram analysis of deceased patients, hypotensive drugs ranked fifth among concomitant medications, and the indications connected to the “diuretic & hypotensives” (such as hydrochlorothiazide), excluding that were not mentioned, were all hypertensions. This means that if the “diuretics & hypotensives” drugs were combined with the “hypotensives” drugs, the hypotensives agents would be the third most common concomitant medication in terms of flow, even surpassing the antibiotics and following only ATEA related to influenza (see Fig. 3, red color). These analyses appear to indicate that these outcome act more as proxies for pre-existing cardiovascular conditions, which make individuals susceptible to severe influenza complications. Therefore, similarly to research based on vaccines [32–34], annual timely vaccination and other preventive measures, are the best choice for these individuals, even if the first-in-class anti-influenza drugs have been launched.

This study has several potential limitations. First, although we endeavored to account for possible confounders and biases, residual confounding factors persist that could not be fully addressed through subgroup or sensitivity analyses. These include the inability to stratify by indication severity, unmeasured confounding from polypharmacy (beyond hypotensives and warfarin), and a lack of adjustment for reporting competition and notoriety bias. Second, since BXM has a fixed dosing regimen with relatively uniform doses for different body weight categories (e.g., 40 mg for 20–80 kg; 80 mg for > 80 kg), data on dosage and body weight are substantially limited, rendering the THER dataset largely uninformative. This precluded the assessment of the influence of dosage and body weight on adverse events. Third, the study may also be subject to false negatives, as sporadic, subgroup-specific signals emerged in subgroup and sensitivity analyses that were not present among the crude signals. Despite these limitations, this study provides supporting evidence to refute numerous BXM safety signals with unconfirmed causality.

Conclusion

Eight robust positive safety signals were ultimately identified. Among these, “altered state of consciousness” is not listed in the product label. Furthermore, several labelled adverse reactions with unestablished causality—rash, urticaria, erythema multiforme, vomiting, delirium, abnormal behavior, and hallucinations—were determined to be false positives. The analysis integrating concomitant medications with indications successfully characterized the clinical profile of the target population. In deceased patients, this approach suggested that their outcomes primarily served as proxies for pre-existing cardiovascular conditions, which increase susceptibility to severe influenza complications. Additionally, this study presents multiple lines of evidence suggesting a potential warfarin-BXM interaction.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

BXM

Baloxavir marboxil

FAERS

FDA Adverse Event Reporting System

DEMO

Demographic and administrative information

DRUG

Drug/biologic information

REAC

Adverse event encodings

OUTC

Patient outcomes

INDI

Indication/diagnosis for use

PS

Primary suspected drug

PTs

Preferred terms

SOC

System organ class

ROR

Reporting odds ratio

PRR

Proportional reporting ratio

BCPNN

Bayesian confidence propagation neural network

MGPS

Multi-item gamma Poisson shrinker

SDR

Signal of disproportionate reporting

INR

International normalised ratio

NSAIDs

Nonsteroidal Anti-Inflammatory Drugs

ATEA

Antitussives & expectorants & antiasthmatics

DE

Death

DS

Disabilities

HO

Hospitalization-initial or prolonged

LT

Life-threatening

OT

Other serious outcomes

ISDN

Isosorbide dinitrate

AMC\CLVP

Amoxicillin\clavulanate potassium

CHLPM\CDN\MEPH

Chlorpheniramine maleate\dihydrocodeine phosphate\methylephedrine hydrochloride

APAP

Acetaminophen

APAP\CAF\PMZ\SALAMIDE

Acetaminophen\caffeine\promethazine\salicylamide

Author contributions

FS, JX: Conceptualization. FS, H-LG, Y-HH: Data collection, collation, and analysis. FS: Writing the manuscript. YZ: Reviewing the data and methods. FC, H-LG: Revised the manuscript. FC: Project administration, Funding acquisition. All authors contributed to the article and approved the final submitted manuscript.

Funding

This study was supported by the Specially Appointed Medical Expert Project of the Jiangsu Commission of Health (2019).

Data availability

The datasets analyzed during the current study are available in the US FAERS database (https://fis.fda.gov/extensions/FPD-QDE-FAERS/FPD-QDE-FAERS.html). All data generated by the present study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study utilized publicly available and anonymized datasets, which do not require ethical approval under the Declaration of Helsinki guidelines. All data were accessed and analyzed in compliance with the original data providers ‘terms of use’.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.