2026 年 49 巻 1 号 p. 113-121
Idiosyncratic drug-induced liver injury (iDILI) is an unpredictable and potentially severe adverse drug reaction, in which immune-mediated mechanisms are suspected to play a central role. Although eosinophilia is often considered a marker of hypersensitivity reactions, the role of eosinophils in iDILI, including drug-specific risks, remains poorly understood. We conducted a case–control study using electronic medical records to evaluate drug-specific risks associated with drug-induced liver injury with eosinophilia (DILI-Eos). Among 17129 Japanese adult patients who underwent serial liver function tests and eosinophil counts, we extracted 631 DILI-Eos cases and 16498 non-DILI-Eos controls. Multivariable logistic regression analysis was performed for 38 drugs that were newly prescribed in more than 50 DILI-Eos cases within 60 d prior to liver injury onset. Sulbactam/cefoperazone showed the strongest association with DILI-Eos (adjusted odds ratio (OR) 14.51; 95% confidence interval (CI) 10.09–20.85), followed by meropenem (OR 5.68; 95% CI 4.10–7.82) and tazobactam/piperacillin (OR 3.55; 95% CI 2.63–4.75). Several commonly used drugs, such as mosapride, lansoprazole, furosemide, and ambroxol, were also significantly associated with increased risk. These findings suggest that DILI-Eos can be triggered by a wide range of drugs across various therapeutic classes potentially via immune-mediated pathways. Notably, substantial variability in risk was observed even within the same drug classes, such as β-lactam antibiotics and nonsteroidal anti-inflammatory drugs (NSAIDs), underscoring the importance of drug-specific evaluation. Further studies are needed to clarify the causality and mechanisms underlying eosinophilic responses in DILI.
Drug-induced liver injury (DILI) is a serious adverse drug event caused by a variety of commonly used drugs.1) DILI is typically classified into intrinsic (predictable) and idiosyncratic (unpredictable) types.2) Among these, idiosyncratic drug-induced liver injury (iDILI) remains a significant concern not only for patients but also in drug development, due to its low incidence but potentially severe outcomes and lack of predictability.3) iDILI is classified into two mechanistic subtypes: immune-mediated and metabolic.4) The precise pathophysiology of iDILI is still poorly understood, but it is presumed to mainly involve immune-mediated mechanisms, particularly hypersensitivity reactions triggered by reactive drug metabolites.5,6) Although eosinophilia is not universally observed in patients with DILI, it is a clinical feature associated with immune-mediated iDILI and can serve as an indicator of drug allergy and hypersensitivity.7)
Recent evidence has highlighted that eosinophilia, beyond its classical roles in parasitic infections and allergic diseases,8,9) may contribute to both hepatic injury and repair depending on the context.10) In immune-allergic iDILI, eosinophilia is often linked to type I (immunoglobulin E (IgE)-mediated) or type IV (T cell-mediated) hypersensitivity reactions,11) characterized by symptoms such as fever, rash, and lymphadenopathy.12) In addition, several studies have suggested that the presence of peripheral blood eosinophilia or hepatic eosinophilic infiltration is associated with a favorable prognosis, potentially reflecting an active role of eosinophils in resolving inflammation and promoting tissue remodeling.13,14) Conversely, experimental models have demonstrated that eosinophils can also contribute to liver injury by mediating immune activation, recruiting cytotoxic mediators, and exacerbating hepatocyte damage.15,16) Although very few eosinophils are detectable in the liver under homeostatic conditions,17) their accumulation has been observed in several pathological contexts, including DILI.13) The expression of specific chemoattractants, such as eotaxin, appears to mediate selective eosinophil recruitment into hepatic tissue.18) However, it is important to recognize that the number of peripheral eosinophils does not necessarily correlate with the extent of hepatic eosinophilic infiltration.18) These findings suggest a complex role of eosinophils in immune-mediated iDILI pathogenesis, involving not only drug-specific factors but also host immune responses.
To date, to the best of our knowledge, no previous studies have reported the identification of causative drugs, specifically in DILI cases accompanied by eosinophilia, using clinical laboratory data. This research focus is particularly important given the limitations of spontaneous reporting systems (SRS), which lack objective clinical laboratory data. Although SRS are widely used to collect information on adverse drug reactions and several studies have used SRS to evaluate suspected drugs and the risk of DILI,19,20) distinguishing between “intrinsic” and “idiosyncratic” types of DILI remains challenging. Previous studies have demonstrated that using the electronic medical record database of Nihon University School of Medicine enabled objective evaluation of correlations between liver function test values and drug use, allowing the detection of potential signals for DILI. Some high-risk and potentially hepatotoxic drugs have already been reported using this approach.21) The aim of this study was to identify high-risk drugs associated with immune-allergic iDILI by focusing on DILI cases accompanied by eosinophilia using a clinical database. Additionally, we sought to characterize the temporal patterns of eosinophilia onset in relation to liver injury, providing new insights into the immune dynamics of iDILI.
This study was conducted using Nihon University School of Medicine’s Clinical Data Warehouse (NUSM’s CDW), which contains anonymized electronic medical records from three hospitals affiliated with Nihon University in Japan. The database is a centralized data repository that integrates detailed clinical information including patient demographics, diagnoses, laboratory data, and prescription data. The experimental protocol was approved by the Ethical Committee of NUSM (Approval No. 2025-09), and the study was conducted in compliance with the Ethical Guidelines for Medical and Health Research Involving Human Subjects of the Ministry of Education, Culture, Sports, Science and Technology and the Ministry of Health, Labour and Welfare, Japan.
Study SubjectsThis case–control study included 295458 adult patients aged 18 years or older, who underwent liver function tests, at least two times within a 90-d period between September 1, 2004 and February 28, 2024 (Fig. 1). A timeline diagram of the study design is shown in Fig. 2. To identify patients with drug-induced liver injury with eosinophilia (DILI-Eos), the following criteria were applied: (1) DILI definition: The event date was defined as the first date on which alanine aminotransferase (ALT) and aspartate aminotransferase (AST) were measured to be at least three times the upper limit of normal (×3 ULN), and patients were considered to have DILI if their ALT and AST values were within the normal range within 90 d of before the event date (Fig. 2A). Furthermore, patients who had undergone ALT or AST measurements during the 180-d period between 90 and 270 d prior to the event date were excluded. (2) Eosinophilia definition: Among these patients, those who exhibited peripheral blood eosinophilia, defined as an eosinophil count of ≥6%, within 60 d before or within 20 d after the event date, were considered to have DILI-Eos (Fig. 2A). This eosinophil cutoff was determined based on previous DILI studies that defined eosinophilia as 4–6% depending on institutional reference ranges,22) and is consistent with the Digestive Disease Week-Japan (DDW-J) 2004 diagnostic criteria used in Japanese clinical settings.23) The index date was set at the time of both DILI and eosinophilia onset (Fig. 2B). To minimize temporal bias in the selection of the control group, patients who had AST and ALT measurements within 1 d before or after the index date were considered potential controls, and this day was defined as the reference date (Fig. 2B). Patients were further required to have at least one AST and ALT measurement within 90 d prior to the reference date, with all values within the normal range. Among these patients, those who had an eosinophil count of <6% within 60 d before or 20 d after the reference date were classified as controls. The 180 d prior to the index or reference date represented the exposure assessment period (Fig. 2B).
Flow diagram illustrating the selection process for the study cohort. As indicated by the asterisk, the index date was defined as the onset of both DILI and eosinophilia. For controls, patients with AST and ALT measured within ±1 d of the case index date were selected, and this date was defined as the reference date. NUSM’s CDW, Nihon University School of Medicine’s Clinical Data Warehouse; DILI: drug-induced liver injury; ALT: alanine aminotransferase; AST: aspartate aminotransferase.
(A) Definition of cases. The event date was defined as the first date on which both ALT and AST levels were >3 times the upper limit of normal (ULN), provided that both values were within the normal range within the prior 90 d. Furthermore, patients who had undergone ALT or AST measurements during the 180-d period between 90 and 270 d prior to the event date were excluded. Patients who exhibited peripheral blood eosinophilia (eosinophil count ≥6%) within 60 d before or 20 d after the event date were classified as having drug-induced liver injury with eosinophilia (DILI-Eos). (B) Definitions of the index date and controls. The index date was defined as the date of onset of both DILI and eosinophilia. Controls were defined as patients with ALT and AST measured within ±1 d of the index date defined in the case group (referred to as the reference date), with at least one ALT and AST measurement within the 90 d preceding the reference date, all within the normal range. Among these patients, those with an eosinophil count of <6% within 60 d before or 20 d after the reference date were classified as controls. The 180-d period prior to the index date (for cases) and the reference date (for controls) was defined as the exposure assessment period. During this period, patients were excluded if they had pre-existing liver, gallbladder, biliary tract, and pancreatic disease without DILI or eosinophilia-related conditions and those receiving hepatoprotective or eosinophil-suppressive medications. ALT: alanine aminotransferase; AST: aspartate aminotransferase; DILI: drug-induced liver injury; ULN: upper limit of normal.
Patients were excluded from study if, during exposure assessment period, they had (1) pre-existing liver, gallbladder, biliary tract, and pancreatic disease without DILI, (2) pre-existing disease associated with eosinophilia, (3) taken any drug for hepatoprotection: glycyrrhizin, glutathione, lactulose, l-arginine, l-glutamate hydrate, tiopronin, protoporphyrin, lactitol, rifaximin, ursodeoxycholic acid, liver hydrolysate, or branched chain amino acid preparation, or (4) taken any drug to suppress eosinophilia: anti-histamines, anti-allergic agents, leukotriene receptor antagonists, anti-IgE antibodies, anti-interleukin (IL)-5 antibodies, or corticosteroids (including glucocorticoids and their derivatives). The International Classification of Diseases tenth revision (ICD-10) codes used to define each disease are listed in Supplementary Tables S1 and S2. After applying the inclusion and exclusion criteria, a total of 17129 patients were enrolled in the study, including 631 DILI-Eos cases and 16498 non-DILI controls.
Suspected Drugs Causing DILI with EosinophiliaAmong the 17129 patients included in the study, a total of 968 drugs were newly prescribed within 60 d prior to the index date. Of these, 38 drugs had been prescribed to more than 50 DILI-Eos patients. These 38 drugs were considered to be suspected drugs and were selected for further analysis to evaluate their association with DILI-Eos.
CovariatesThe covariates included age, sex, number of concomitant drugs, diabetes mellitus (ICD-10 codes E10-E14), dyslipidemia (ICD-10 code E78), hypertension (ICD-10 codes I10-I15), ischemic heart disease (ICD-10 codes I20-I25), cerebrovascular disease (ICD-10 codes I60-I69), thyroid disease (ICD-10 codes E00-E07), kidney disease (ICD-10 codes N00-N29), inflammatory bowel disease (ICD-10 codes K50, K51), and cancer (ICD-10 codes C00-C97, D00-D09). Medical history was defined as the presence of a relevant diagnosis recorded within 365 d prior to the index or reference date. The number of concomitant drugs was defined as the number of drugs dispensed within 90 d prior to the index date.
Statistical AnalysisTo assess baseline comparability between the DILI-Eos case and control groups, Mann–Whitney U test was used for continuous variables such as age, number of concomitant drugs, and liver function test values while chi-squared test was applied to categorical variables such as sex and medical history. Multivariable logistic regression analysis was performed to assess the association between suspected drugs and the risk of DILI-Eos, adjusting for the above covariates. Adjusted odds ratios (ORs) and 95% confidence intervals (CIs) were calculated. Type I errors for multiple testing were controlled by false discovery rate (FDR) correction.24) For all statistical analyses, the significance level was set at 0.05. All statistical analyses were conducted with R software (version 4.2.2; R Foundation for Statistical Computing, Vienna, Austria).
A total of 17129 patients were included in the study; 631 DILI-Eos cases and 16498 non-DILI-Eos controls. The baseline characteristics of each group are summarized in Table 1. The mean age of patients in the DILI-Eos group was significantly older than that of the non-DILI-Eos group (63.1 ± 17.2 vs. 57.5 ± 19.7 years, p < 0.001). The proportion of male patients was also significantly higher in the DILI-Eos group (67.0%) compared with the non-DILI-Eos group (43.3%). In addition, the mean number of concomitant drugs was significantly greater in the DILI-Eos group than in the non-DILI-Eos group (13.3 ± 11.2 vs. 6.9 ± 7.2, p < 0.001). Regarding medical history, there was no statistically significant difference in the prevalence of diabetes between the DILI-Eos and non-DILI-Eos groups (45.3 vs. 46.1%, p = 0.745). Similarly, the prevalence of thyroid disease (14.7 vs. 16.1%, p = 0.407) and inflammatory bowel disease (0.32 vs. 0.50%, p = 0.772) did not differ significantly between the two groups. By contrast, patients in the DILI-Eos group had significantly higher prevalence of dyslipidemia (18.4 vs. 15.3%, p = 0.038), hypertension (30.6 vs. 20.9%, p < 0.001), ischemic heart disease (27.1 vs. 16.8%, p < 0.001), cerebrovascular disease (19.3 vs. 12.9%, p < 0.001), kidney disease (27.1 vs. 22.5%, p = 0.008), and cancer (45.0 vs. 33.9%, p < 0.001).
| Characteristics | DILI-Eos group (n = 631) |
Non-DILI-Eos group (n = 16498) |
p-Value |
|---|---|---|---|
| Age (years), mean ± S.D. | 63.1 ± 17.2 | 57.5 ± 19.7 | <0.001 |
| Male, n (%) | 423 (67.0%) | 7142 (43.3%) | <0.001 |
| Number of concomitant drugs, mean ± S.D. | 13.3 ± 11.2 | 6.9 ± 7.2 | <0.001 |
| Laboratory parameters | |||
| ALT (units/L), median (IQR) | 182.0 (150.0–273.0) | 15.0 (11.0–21.0) | <0.001 |
| AST (units/L), median (IQR) | 166.0 (104.0–321.0) | 19.0 (16.0–24.0) | <0.001 |
| ALP (units/L), median (IQR) | 409.0 (265.0–665.0) | 204.0 (162.0–257.0) | <0.001 |
| TB (mg/dL), median (IQR) | 0.71 (0.45–1.30) | 0.54 (0.39–0.76) | <0.001 |
| Medical history, n (%) | |||
| Diabetes | 286 (45.3) | 7599 (46.1) | 0.745 |
| Dyslipidemia | 116 (18.4) | 2521 (15.3) | 0.038 |
| Hypertension | 193 (30.6) | 3454 (20.9) | <0.001 |
| Ischemic heart disease | 171 (27.1) | 2773 (16.8) | <0.001 |
| Cerebrovascular disease | 122 (19.3) | 2121 (12.9) | <0.001 |
| Thyroid disease | 93 (14.7) | 2650 (16.1) | 0.407 |
| Kidney disease | 171 (27.1) | 3714 (22.5) | 0.008 |
| Inflammatory bowel disease | 2 (0.32) | 82 (0.50) | 0.772 |
| Cancer | 284 (45.0) | 5600 (33.9) | <0.001 |
ALP: alkaline phosphatase; ALT: alanine aminotransferase; AST: aspartate aminotransferase; DILI-Eos: drug-induced liver injury with eosinophilia; IQR: interquartile range; S.D.: standard deviation; TB: total bilirubin.
Thirty-eight drugs, each prescribed to ≥50 patients with DILI-Eos, were selected from 968 newly prescribed drugs for analysis. Among the suspected drugs, several drugs were significantly associated with an increased risk of DILI-Eos (Table 2, Supplementary Table S3). Among antibiotics, sulbactam/cefoperazone was associated with the highest risk of DILI-Eos (adjusted odd ratio (OR): 14.51, 95% CI: 10.09–20.85, PFDR-adjusted <0.001), followed by meropenem (adjusted OR: 5.68, CI: 4.10–7.82, PFDR-adjusted <0.001), and tazobactam/piperacillin (adjusted OR: 3.55, CI: 2.63–4.75, PFDR-adjusted <0.001) also showed a high risk of DILI-Eos. Other β-lactam antibiotics such as piperacillin, seftriaxone, and cefmetazole were also significantly associated with increased risk, whereas cefazolin was associated with a significantly lower risk (adjusted OR: 0.49, CI: 0.38–0.63, PFDR-adjusted <0.001). Levofloxacin did not show significant association (adjusted OR: 1.06, CI: 0.82–1.36, PFDR-adjusted = 0.651). Among anti-inflammatory drugs, both acetaminophen (adjusted OR = 1.56, 95% CI: 1.30–1.86, PFDR-adjusted <0.001) and aspirin (adjusted OR = 1.42, 95% CI 1.13–1.78, PFDR-adjusted = 0.005) were weakly associated with increased risk of DILI-Eos, whereas loxoprofen (adjusted OR: 0.68, CI: 0.56–0.82, PFDR-adjusted <0.001) was associated with a significantly lower risk of DILI-Eos. Diclofenac showed no significant association (adjusted OR: 0.98, CI: 0.79–1.20, PFDR-adjusted = 0.842). Several gastrointestinal drugs such as mosapride (adjusted OR: 2.07, CI: 1.54–2.73, PFDR-adjusted <0.001), lansoprazole (adjusted OR: 2.04, CI: 1.67–2.49, PFDR-adjusted <0.001), omeprazole (adjusted OR: 1.68, CI: 1.33–2.11, PFDR-adjusted <0.001), and famotidine (adjusted OR: 1.59, CI: 1.37–1.89, PFDR-adjusted <0.001) were associated with a statistically significant increased risk of DILI-Eos, while rebamipide (adjusted OR: 0.69, CI: 0.55–0.86, PFDR-adjusted = 0.002) and metoclopramide (adjusted OR: 0.75, CI: 0.59–0.94, PFDR-adjusted = 0.024) were associated with a significantly reduced risk of DILI-Eos. Amlodipine, a vasodilator, was associated with a moderate increase in DILI-Eos risk (adjusted OR: 1.30, 95% CI: 1.04–1.62, PFDR-adjusted = 0.028). Furosemide, a diuretic, was associated with a statistically significant increased risk of DILI-Eos (adjusted OR: 1.81, 95% CI: 1.45–2.26, P FDR-adjusted <0.001). Among expectorants, both ambroxol (adjusted OR: 1.96, 95% CI: 1.46–2.58, PFDR-adjusted <0.001) and l-carbocisteine (adjusted OR: 1.51, 95% CI: 1.18–1.90, PFDR-adjusted = 0.002) were significantly associated with an increased risk of DILI-Eos. In addition, several other drugs showed statistically significant associations with DILI-Eos. Insulin (adjusted OR: 1.63, 95% CI: 1.26–2.10, PFDR-adjusted <0.001), picosulfate (adjusted OR: 1.30, 95% CI: 1.06–1.60, PFDR-adjusted = 0.021), flumazenil (adjusted OR: 1.98, 95% CI: 1.54–2.51, PFDR-adjusted <0.001), and midazolam (adjusted OR: 1.29, 95% CI: 1.03–1.60, PFDR-adjusted = 0.032) were all associated with a modest but statistically significant increase in the risk of DILI-Eos, with flumazenil showing the highest odds ratio among them. Conversely, haloperidol, an antipsychotic drug, showed a modest but significant reduction in the risk of DILI-Eos (adjusted OR: 0.75, 95% CI: 0.57–0.97, PFDR-adjusted = 0.048), and remifentanil, an opioid, was also associated with a significantly lower risk (adjusted OR: 0.75, 95% CI: 0.60–0.93, PFDR-adjusted = 0.019). Adjusted ORs for DILI-Eos for all 38 suspected drugs and differences in patient characteristics between users and non-users of these drugs are shown in Supplementary Tables S4 and S5, respectively.
| Class | Generic name | DILI-Eos (n) | Non-DILI-Eos (n) | Adjusted OR [95% CI] | p-Value | FDR_p |
|---|---|---|---|---|---|---|
| Antibiotics | Sulbactam/Cefoperazone | 62 | 76 | 14.51 [10.09–20.85] | <0.001 | <0.001 |
| Meropenem | 73 | 142 | 5.68 [4.10–7.82] | <0.001 | <0.001 | |
| Tazobactam/Piperacillin | 69 | 194 | 3.55 [2.63–4.75] | <0.001 | <0.001 | |
| Piperacillin | 83 | 651 | 1.80 [1.40–2.30] | <0.001 | <0.001 | |
| Seftriaxone | 54 | 363 | 1.66 [1.22–2.21] | 0.001 | 0.002 | |
| Cefmetazole | 71 | 2295 | 1.54 [1.19–1.96] | 0.001 | 0.002 | |
| Ampicillin/Sulbactam | 88 | 932 | 1.30 [1.02–1.63] | 0.030 | 0.044 | |
| Cefazolin | 76 | 2202 | 0.49 [0.38–0.63] | <0.001 | <0.001 | |
| Levofloxacin | 52 | 869 | 1.06 [0.82–1.36] | 0.634 | 0.651 | |
| Anti-inflammatory drugs | Acetaminophen | 221 | 3226 | 1.56 [1.30–1.86] | <0.001 | <0.001 |
| Asprin | 88 | 1441 | 1.42 [1.13–1.78] | 0.003 | 0.005 | |
| Diclofenac | 100 | 2042 | 0.98 [0.79–1.20] | 0.842 | 0.842 | |
| Loxoprofen | 121 | 3103 | 0.68 [0.56–0.82] | <0.001 | <0.001 | |
| Gastrointestinal drugs | Mosapride | 51 | 292 | 2.07 [1.54–2.73] | <0.001 | <0.001 |
| Lansoprazole | 131 | 1111 | 2.04 [1.67–2.49] | <0.001 | <0.001 | |
| Omeprazole | 97 | 782 | 1.68 [1.33–2.11] | <0.001 | <0.001 | |
| Famotidine | 182 | 2436 | 1.59 [1.37–1.89] | <0.001 | <0.001 | |
| Metoclopramide | 104 | 2030 | 0.75 [0.59–0.94] | 0.015 | 0.024 | |
| Rebamipide | 74 | 2096 | 0.69 [0.55–0.86] | 0.001 | 0.002 | |
| Vasodilators | Amlodipine | 79 | 1461 | 1.30 [1.04–1.62] | 0.018 | 0.028 |
| Diuretics | Furosemide | 108 | 1009 | 1.81 [1.45–2.26] | <0.001 | <0.001 |
| Expectorants | Ambroxol | 57 | 410 | 1.96 [1.46–2.58] | <0.001 | <0.001 |
| l-Carbocisteine | 81 | 765 | 1.51 [1.18–1.90] | 0.001 | 0.002 | |
| Antidiabetics | Insulin | 98 | 685 | 1.63 [1.26–2.10] | <0.001 | <0.001 |
| Laxatives | Picosulfate | 118 | 1219 | 1.30 [1.06–1.60] | 0.012 | 0.021 |
| Benzodiazepine antagonists | Flumazenil | 58 | 229 | 1.98 [1.54–2.51] | <0.001 | <0.001 |
| Sedative-hypnotic drugs | Midazolam | 105 | 1164 | 1.29 [1.03–1.60] | 0.021 | 0.032 |
| Antipsychotics | Haloperidol | 86 | 1102 | 0.75 [0.57–0.97] | 0.034 | 0.048 |
| Opioids | Remifentanil | 131 | 2453 | 0.75 [0.60–0.93] | 0.010 | 0.019 |
Data are presented as number of cases of drug-induced liver injury with eosinophilia (DILI-Eos) and non-DILI-Eos cases, along with the adjusted odds ratios (OR), 95% confidence intervals (CI), p-values, and false discovery rate-adjusted p-value (FDR-p) for 29 drugs with therapeutic classes. A list of all 38 suspected drugs is provided in Supplementary Table S3.
We examined the temporal patterns of eosinophilia onset in relation to liver injury. The timing of eosinophilia onset was evaluated for each drug relative to the onset of liver injury (defined as the event date), and classified into three categories: “Before” (prior to liver injury), “On” (on the same day as liver injury), and “After” (following liver injury). The proportion of eosinophilia cases in each category was calculated for 38 suspected drugs (Supplementary Table S6). Among the 38 investigated drugs, eosinophilia was more frequently observed before the onset of liver injury than after or on the same day. The “on the same day” category was generally less frequent across all drugs, typically accounting for less than 14% of cases. Drugs showing a high frequency (approximately 60%) of eosinophilia occurring before the onset of liver injury included the antibiotic tazobactam/piperacillin (59.4%), the sedative-hypnotic drug diazepam (61.0%), and the antipsychotic haloperidol (59.3%). By contrast, the expectorant ambroxol showed eosinophilia more frequently after the onset of liver injury (59.6%).
In this study, we investigated the clinical characteristics and potential drug-related risk factors associated with DILI-Eos using a retrospective cohort of 17129 patients. A total of 631 cases of DILI-Eos were identified. To our knowledge, there are no prior studies reporting the incidence of DILI-Eos specifically, and therefore direct comparisons of incidence rates with other cohorts are not possible. Nevertheless, peripheral eosinophilia has been reported in approximately 30% of patients with DILI.13,25) Our analysis revealed that patients in the DILI-Eos group were significantly older, more likely to be male, and were prescribed a greater number of concomitant drugs compared with the non-DILI-Eos group. Although female sex has often been suggested to be a risk factor for certain types of DILI,26) current evidence does not support a consistent sex-related difference in overall DILI incidence.27) For instance, a cohort from the Spanish DILI registry, which included 603 cases, showed a nearly equal sex distribution, with 49% of patients being female.26) However, it is important to note that sex-related differences may exist for specific drugs or DILI phenotypes. For example, women have been found to be more susceptible to DILI from drugs such as minocycline and nitrofurantoin.28)
Our findings revealed several important insights regarding drug-specific risks for DILI-Eos. Among antibiotics, several broad-spectrum β-lactam antibiotics, including sulbactam/cefoperazone, meropenem, and tazobactam/piperacillin, were significantly associated with elevated risk of DILI-Eos. These antibiotics have been implicated in rare but serious hypersensitivity reactions, such as drug reaction with eosinophilia and systemic symptoms (DRESS) syndrome.29) All β-lactam antibiotics have the potential to cause transient elevations in hepatic enzymes such as ALT and AST.30) There have been case reports of cefoperazone causing DILI accompanied by eosinophilia,31) and the results of this study support the finding that the combination of sulbactam and cefoperazone is a high-risk drug for this phenotype. Therefore, the clinical use of these antibiotics should involve regular monitoring of liver function and vigilant observation for signs of hypersensitivity, include eosinophilia. By contrast, cefazolin, another β-lactam antibiotic, was associated with a significantly lower risk of DILI-Eos. This contrast underscores the heterogeneity within β-lactam antibiotics and suggests that drug-specific factors, including immunogenic potential, play a crucial role in modulating DILI risk.
The occurrence of DILI-Eos was not confined to antibiotics but was also observed with other drug categories. Among anti-inflammatory drugs, both acetaminophen and aspirin showed a modest increase in DILI-Eos risk, and the clinical and mechanistic interpretation of these associations warrants careful consideration. Acetaminophen-induced liver injury is a well-recognized example of dose-dependent, intrinsic hepatotoxicity characterized by glutathione depletion and oxidative stress, and aspirin is also known to cause dose-dependent hepatotoxicity.32) Interestingly, recent studies have demonstrated a protective role of eosinophils in acetaminophen-induced liver injury. Eosinophils were shown to attenuate liver injury via the production of IL-4 and IL-13, which suppress interferon (IFN)-γ-mediated inflammatory pathways through the p38 mitogen-activated protein kinase (MAPK)/cyclooxygenase (COX)/nuclear factor (NF)-κB signaling cascade.33) These findings suggest that eosinophil activation during acetaminophen-induced liver injury may represent a protective immune response. Although aspirin-induced hepatotoxicity is typically mild and asymptomatic, mild eosinophilia accompanied by an elevation of liver enzymes has been reported,34) which may also reflect a protective immune response. By contrast, loxoprofen was inversely associated with DILI-Eos. In addition, gastrointestinal drugs such as mosapride, lansoprazole, omeprazole, and famotidine were significantly associated with increased risk. Proton pump inhibitors (PPIs) and histamine 2-receptor antagonists are not traditionally considered hepatotoxic, but they have been implicated in rare cases of immune-mediated liver injury.35,36) Our results suggest the need for further investigation into their role in eosinophilic DILI, especially considering their widespread use. Additional drug classes associated with elevated risk included furosemide, insulin, and expectorants such as ambroxol and l-carbocisteine. A benzodiazepine antagonist, flumazenil, is generally considered non-hepatotoxic (“No-DILI-concern” category); however, rare cases of hepatotoxicity have been reported, which should be noted.37,38) These findings suggest that eosinophilic DILI may be triggered across diverse drug classes, potentially via shared mechanisms involving immune sensitization, hapten formation, or eosinophil activation.
In addition to drug-specific effects, each drug’s influence on DILI should be interpreted with consideration of concomitant drug use. Patients in the DILI-Eos group were prescribed an average of more than 10 concomitant drugs, which was higher than in the control group. Previous studies have shown that the concurrent use of multiple hepatotoxic drugs substantially increases the risk of acute liver injury compared with single-drug exposure, suggesting possible pharmacological and toxicological interactions between drugs.39) Furthermore, studies investigating synergistic interactions between two drugs on the risk of ALT elevation have reported that combinations such as diclofenac and famotidine are associated with an increased risk of ALT abnormalities.40) Although acetaminophen was one of the most frequently used drugs among DILI-Eos patients, there have been case reports indicating that concomitant use of acetaminophen with antiepileptic drugs such as phenytoin41) or carbamazepine42) may enhance its hepatotoxicity. Analyses of spontaneous adverse event data have also shown that sympathetic stimulants are associated with an increased fatality rate in acetaminophen-associated liver injury, whereas certain concomitant medications, including statins, beta-blockers, angiotensin-converting enzyme (ACE) inhibitors, angiotensin II receptor blockers, and NSAIDs, may be linked to a reduced fatality rate.43) These observations highlight the potential for complex pharmacological and toxicological interactions among drugs. Taken together, these findings suggest that future studies should investigate how specific concomitant drugs modify the hepatotoxic effects of particular drugs, for example by performing stratified analyses based on major concomitant drug classes or using data-driven approaches such as least absolute shrinkage and selection operator (LASSO) regression.
Furthermore, we examined the temporal patterns of eosinophilia onset in relation to liver injury, revealing that eosinophilia occurred before liver injury in a large number of cases (49.7%), followed by before (42.6%) and on the index date (7.7%). This temporal distribution suggests that eosinophilia may not only be a trigger of liver injury, but could also reflect immunological events that precede or follow hepatic damage. Previous studies have reported that eosinophils are involved in both pro-inflammatory and tissue-repair mechanisms, depending on the context of injury.10,44) For instance, in a halothane-induced liver injury model, eosinophils infiltrated necrotic areas of the liver and contributed to hepatocellular damage through degranulation and the release of cytotoxic mediators such as major basic protein (MBP), suggesting a pathogenic and pro-inflammatory role.15) Conversely, in animal models of acetaminophen-induced liver injury, it has been reported that eosinophils exert hepatoprotective effects through IL-33-mediated activation, followed by the secretion of IL-4, which promotes the induction of anti-inflammatory macrophages and tissue repair.10) This may explain why eosinophilia is not uniformly associated with worsening liver function and, in some cases, may even reflect a compensatory or protective immune response.10,13) The observed variability in the timing patterns of eosinophilia for each drug highlights the complex, drug-specific interaction between liver injury and eosinophila. We also acknowledge that the timing of laboratory testing and the frequency of eosinophil measurements varied among patients, which may have influenced the apparent temporal relationships. Therefore, these results should not be interpreted as demonstrating definitive causal sequences but rather as reflecting general trends in the timing of eosinophilia relative to liver injury. Collectively, these findings support the concept that DILI-Eos is not a homogeneous entity but rather a spectrum involving diverse immune mechanisms. Understanding these dynamics may help elucidate drug-specific risk and variations in risk based on individual immune responses and inform decisions on monitoring and drug discontinuation. Understanding the drug-specific risks associated with eosinophilia may provide valuable clues for early detection and therapeutic intervention in immune-mediated iDILI.
However, some limitations must be acknowledged. First, the diagnosis of DILI was based solely on liver enzyme thresholds without detailed clinical adjudication or liver biopsy confirmation, which could lead to misclassification. In addition, although we required that patients had no ALT and AST measurements within the 180-d period between 90 and 270 d prior to the event date an inclusion criterion, we could not completely exclude the possibility that some patients had experienced DILI before the observation period or had a prior history of DILI diagnosed at other medical institutions. Therefore, we cannot definitively confirm that all cases represented incident, initial onset DILI. Second, based on previous reports indicating that iDILI typically occurs within 1–2 months after the initiation of the causative agent,45) a 60-d window was used in this study to define the temporal association between drug exposure and liver injury. However, causality could not be definitively established. Third, while we adjusted for major covariates, residual confounding by unmeasured variables (e.g., alcohol use, body mass index, or genetic predisposition) remains possible. Fourth, medical history was extracted solely based on ICD-10 codes recorded in NUSM’s CDW. While ICD-10 codes provide a standardized framework for disease classification, their use alone may not fully capture the clinical complexity or accuracy of diagnoses, potentially leading to misclassification or incomplete recording of comorbidities. To assess how accurately medical history was captured, we examined the clinical records for HbA1c levels in patients with and without a diabetes diagnosis. The results showed a mean HbA1c of 6.6% in the diabetes group and 5.6% in the non-diabetes group, supporting the reliability of the ICD-10-based definition of diabetes. However, this validation was limited to diabetes, and the accuracy of diagnosis of other comorbidities remains uncertain. Lastly, by focusing on drugs prescribed to ≥50 DILI-Eos patients, our analysis may have underrepresented rare but potentially highly hepatotoxic agents. Additionally, eosinophilia itself can be transient and may not always reflect true hypersensitivity-related pathology. In the DILI-Eos cases used in this study, no cases with a definitive diagnosis of iDILI could be identified in the medical records. This likely reflects the practical limitations of clinical diagnosis, in which DILI is primarily a diagnosis of exclusion and no established biomarkers are available, making it extremely difficult to distinguish iDILI from intrinsic DILI in real-world clinical settings.
In the present study, we performed a risk assessment of DILI-Eos using electronic medical records. We identified several drugs that were significantly associated with increased risk of DILI-Eos, most notably sulbactam/cefoperazone, meropenem, and tazobactam/piperacillin. These findings indicate the potential role of immune-mediated mechanisms in the pathogenesis of DILI-Eos and suggest that commonly used agents, including gastrointestinal drugs and expectorants, may contribute to eosinophilic liver injury in susceptible individuals. Notably, marked differences in risk were observed even within the same drug class, such as β-lactam antibiotics and NSAIDs, indicating the importance of drug-specific factors. Further prospective and mechanistic studies are warranted to clarify causality, characterize immunological pathways, and improve prediction and prevention of eosinophilic DILI.
This research was supported by the Ministry of Education, Culture, Sports, Sciences and Technology (MEXT) of the Japanese Government and the Japan Agency for Medical Research and Development (AMED) under Grant numbers JP18km0605001 and JP223fa627011. The authors express special thanks to 4DIN Ltd. (Tokyo, Japan) and Phenogen Medical Corporation (Tokyo, Japan) for financial support. Both companies had no role in the research design, analysis, data collection, interpretation of data, or review of the manuscript, and no honoraria or payments were made for authorship.
The authors declare no conflict of interest.
Supplementary MaterialsThis article contains supplementary materials.
