Among the 5,326 patients receiving ICPis, 186 patients (3.49%) were diagnosed with irH, and 51 (0.96%) were diagnosed with severe irH. Table 1 delineates the baseline characteristics of the severe irH group and non-irH group, both pre- and post-propensity score matching. Subsequent conditional logistic regression analysis of other unmatched factors (Supplementary Table S1) revealed that patients undergoing PD-L1 inhibitor therapy exhibited a reduced risk of developing severe irH compared to those receiving other ICPis (p = 0.0343; odds ratio: 0.230, 95% confidence interval 0.05899–0.8969).

The detailed baseline characteristics of the severe irH group are documented in Table 2. Twenty-one (41.2%) patients experienced grade 3 irH, and 30 (58.8%) experienced grade 4 irH. The median age was 65 years (IQR, 57–68), and 37 patients (72.5%) were male. Twelve patients (23.5%) had HBV infection and 11 (21.6%) had a history of alcohol consumption. The vast majority of the primary tumors were lung cancer (20, 39.2%) and hepatobiliary carcinoma (8, 15.7%). Most of the patients were at stage IIIb–IV (46, 90.2%). Thirteen (25.5%) patients presented with liver lesions, of which nine had liver metastasis. The main ICPi administered were PD-1 inhibitors (46, 90.2%), although three patients (5.9%) received PD-L1 inhibitors and two (3.9%) received a combined anti-CTLA plus anti-PD-1 regimen. Fourteen (27.5%) patients received targeted therapy, and 26 (51%) received chemotherapy. The details about the primary tumor type and its therapeutic regimen are listed in Supplementary Table S2. There was no significant difference between patients with grade 3 and 4 irH in terms of general characteristics.

According to the RUCAM assessment method, a causal relationship between immunotherapy and irH was highly probable in 34 (66.7%) patients and probable in 17(33.3%). irH occurred after a median duration of 36 (TTO; IQR, 21–85) days and a median of two (IQR, 1– 4) doses. The most common symptoms were jaundice (24, 47.1%), anorexia (18, 35.3%), fever (15, 29.4%), and vomiting (10, 19.7%). Most patients with grade 3 irH were diagnosed in routine examinations before each immunotherapy cycle, with only seven individuals (33.3%) experiencing symptoms compatible with irH. In contrast, in patients with grade 4 irH, most of them (27, 90%) experienced obvious systemic symptoms. Based on the biochemical classification of liver injury, the proportions of hepatocellular, cholestatic, and mixed patterns in grade 3 irH were 47.6%, 23.8%, and 28.6%, respectively. In grade 4 irH, these proportions were 50%, 10%, and 40%. Most of the patients showed a negative result of autoantibodies tests. However, 11 patients presented with positive ANA at low titers (1:80, n = 10; 1:100, n = 1). EBV-DNA and CMV-DNA tests were negative in all patients.

Immunotherapy was discontinued, and GCS was subsequently administered to all patients. The maximum GCS dose administered and patient outcomes are shown in Fig. 2A and B, respectively. All patients with grade 3 irH were steroid-sensitive and were completely cured afterward. The median maximum dose of GCS (calculated as methylprednisolone equivalent) was 0.7 (IQR, 0.5–1.3) mg/kg/day, and TTR was 1.37 (IQR, 1.00–1.57) months. A higher dose of glucocorticoids (GCS) was administered to grade 4 irH than grade 3 irH (2.6 [IQR, 1.3–2.7] mg/kg/day vs. 0.7 [IQR, 0.5–1.3] mg/kg/day, p = 0.001). In patients with grade 4 irH, 21 (70%) patients were steroid-sensitive with a median maximum corticosteroid dose of 2.4 (IQR, 1.3–2.7) mg/kg/day and TTR of 1.40 (IQR, 1.05–1.98) months. Even with a higher GCS dose (median maximum dose: 2.8 [IQR, 1.8–3.2] mg/kg/day), nine (30%) patients still presented resistance to GCS and required second immunosuppression with a TTR of 2.05 (IQR, 1.61–2.32) months. For patients responsive to GCS, those with grade 4 severity required a larger dose of steroids compared to those with grade 3 (p = 0.007), but there was no significant difference in the time taken to achieve remission. For patients resistant to GCS, they needed additional immunosuppressants on top of a larger dose of steroids for treatment, and ultimately, they required a longer time to achieve remission compared to patients of the same severity level (p = 0.036). The curative effect of GCS therapy on different clinical subtypes of irH showed a great difference. The effective rate of GCS in the hepatocellular pattern and the mixed pattern was 92% (23/25) and 100% (8/8), while that in cholestatic patterns was 44.4% (8/18). The detailed general characteristics and treatment regimens of the steroid-refractory patients are listed in Supplementary Table S3 and Supplementary Table S4.

Treatment and outcome of the patients with high-grade immune-related hepatotoxicity (irH). A The violin plot shows the time to resolution (TTR) and the maximum glucocorticoid dose applied to the patients in the steroid-sensitive group and steroid-refractory group with grade 3 irH or grade 4 irH, respectively. B The Sankey diagram shows the outcomes of the patients with grade 3 and grade 4 irH with different biochemical classifications of liver injury

As salvage therapy for these patients, all patients (seven patients) with cholestatic patterns were administered intravenous immunoglobulin (IVIG), and one patient was also administered tocilizumab. The two patients with a hepatocellular pattern were treated with MMF and tacrolimus, respectively. Five patients (55.6%) improved within 1 week after the administration of the second immunosuppressive agent, and six patients (66.7%) were cured within 3 months (Supplementary Table S4). However, three patients died: Patient 4, with severe systemic symptoms, eventually died of tumor progression and infection 16 weeks after the diagnosis of irH. Patient 6 died of pseudomembranous invasive tracheobronchial aspergillosis during the treatment of irH. Patient 9 did not receive timely GCS after the discovery of grade 4 irH as GCS was administered after 6 weeks and eventually died because of liver failure (histopathology and imaging details are shown in Supplementary Fig. S1), which indicated that delayed management would result in poor prognosis.

Six of the nine steroid-refractory patients underwent liver biopsies at the time of diagnosis as steroid-refractory by the authoritative oncologist. The histological features and immunocytochemistry are listed in Table 3.

The histopathology of Patient 1 and Patient 5 showed a hepatitic pattern. There was no obvious bile duct injury in the portal area, and the main pathological finding was lobular hepatitis, exhibiting spotty necrosis and focal necrosis (Fig. 3A).

Liver biopsies of steroid-refractory patients. A Patient 1: lobular inflammation associated with spotty and focal necrosis with moderate steatosis (hematoxylin–eosin–saffron [HES] ×100). B–C Patient 2: B severe balloon degeneration and cholestasis (HES ×100); C bile duct reaction (immunohistochemistry CK19 ×400). D Patient 4: the bile duct loss can be observed. Only the vessels appeared in the portal area (CK AE1/AE3&CD34 double-staining ×100). E–F The immunohistochemistry of the patient 2. E CD3+ T cells mainly concentrated around the portal (immunohistochemistry CD3 ×40), F CD8 + T cells mainly concentrated around the portal (immunohistochemistry CD3 ×40)

The remaining four patients showed a cholangiopathic pattern. Mild inflammatory cells were observed in the hepatic sinusoid and portal area, and the manifestation of lobule injury was mild, with only one patient having severe balloon degeneration (Fig. 3B). The main pathological changes were concentrated in the bile ducts (including bile duct reaction and loss). The bile duct reaction was observed in Patient 2 and Patient 3 (Fig. 3C). In Patient 4, a remarkable bile duct loss was reported (Fig. 3D). There was significant fibrous hyperplasia, small blood vessel hyperplasia, and bile duct reaction, while mild inflammatory cell infiltration in the portal area.

Immunohistochemical results show that in steroid-refractory irH, CD8 + T cells constitute the predominant inflammatory cell type.

ICPi was rechallenged in 12 (23.5%) patients (7 patients in the grade 3 irH group and 5 patients in the grade 4 irH group) after recovery from irH. Their characteristics are listed in Supplementary Table S5. Eleven patients were readministered the same ICPi, whereas one patient was changed from PD-1 to PD-L1 inhibitors, and three received concomitant chemotherapy. At the time of the last follow-up, the median rechallenge ICPi dose was 6 (IQR, 3–12), and the median time between severe irH and ICPi retreatment was 26 (IQR, 10–45) weeks. Overall, with a median follow-up time of 9.3 (IQR, 7.5–16.3) months, no irH reoccurred, two patients presented with grade two hypothyroidism, and one presented with grade one eosinophilia and grade two hypothyroidism during the rechallenge. It is worth emphasizing that, compared with patients with mixed or cholestatic patterns, patients with hepatocellular patterns were more likely to resume ICPi treatment after irH (p = 0.035). It may suggest that the type of liver injury may correlate with the probability of retreatment requirement.