Incidence, Clinical Characteristics, Management, and Outcomes of Hematologic Immune‐Related Adverse Events Associated With Immune Checkpoint Inhibitors: A Single‐Center Japanese Retrospective Cohort Study

ABSTRACT

Introduction

Immune‐related adverse events (irAEs) are complications associated with immune checkpoint inhibitors (ICIs). Hematologic irAEs (heme‐irAEs) are rare but potentially serious and may interfere with cancer treatment. However, their clinical characteristics remain insufficiently described.

Methods

We conducted a single‐center retrospective study using the National Cancer Center Hospital East medical database from April 1, 2014, to March 31, 2024. Potential heme‐irAE cases were identified using diagnostic name searches, bone marrow examination records, and positive DAT results and were followed by chart review to exclude alternative etiologies, such as chemotherapy‐related myelosuppression, marrow involvement, infection, and hematologic malignancy.

Results

Of the 3371 patients treated with ICI‐containing regimens, 40 patients developed heme‐irAEs, indicating an observed incidence of 1.2%. The most common heme‐irAE reported was immune thrombocytopenia, and 85.0% of these events were Grade 3 or higher. The median time from ICI initiation to heme‐irAE onset was 12 weeks. For all affected patients included in the regimen‐specific onset‐time comparison, the median time to heme‐irAE onset appeared shorter with dual ICI therapy without chemotherapy than with ICI monotherapy, although this descriptive comparison was based on a small number of cases. ICI‐containing regimens were interrupted or discontinued in most patients, and systemic corticosteroids were administered to 25 patients, 18 of whom achieved hematologic improvement. Cancer treatment was resumed or continued for 20 patients, with no recurrent heme‐irAEs observed during the follow‐up period. In an exploratory landmark analysis, hematologic response was associated with longer survival after heme‐irAE onset.

Conclusion

Heme‐irAEs were rare but frequently severe in this Japanese single‐center cohort. Their occurrence was associated with prolonged interruption of cancer treatment, underscoring the importance of early recognition, careful diagnostic evaluation, and multidisciplinary management of heme‐irAEs. Our findings regarding dual ICI therapy, survival, and subsequent ICI therapy after heme‐irAE onset should be interpreted as exploratory.

This single‐center retrospective study evaluated hematologic immune‐related adverse events (irAEs) among 3371 patients treated with immune checkpoint inhibitors. Forty patients developed hematologic irAEs (1.2%); 85.0% of events were Grade 3 or higher, and the median time to onset was 12 weeks. The survival curves compare patients according to their response to treatment for hematologic irAEs.

1. Introduction

Immune checkpoint inhibitors (ICIs), targeting PD‐1/PD‐L1 or CTLA‐4 pathways, have become an important treatment modality in oncology.

Since the approval of the first ICI in Japan in 2014, multiple antibodies targeting immune checkpoint pathways, including PD‐1, PD‐L1, and CTLA‐4, have been incorporated into routine cancer treatment. Although ICIs are important cancer treatment drugs, they are known to cause immune‐related adverse events (irAEs) that have not been observed with previous cytotoxic anticancer drugs. It has been reported that 54.0%–76.0% of patients treated with ICIs experience some form of irAEs [1]. It is known that irAEs occur relatively frequently in the skin, gastrointestinal tract, liver, lungs, and endocrine organs. irAEs in these organs are widely recognized, thus various guidelines have been developed and treatment strategies have been established [2, 3, 4, 5, 6, 7].

However, with the increased use of ICIs, rare irAEs have been identified, and in such cases, it can be difficult to decide on a treatment plan. Hematologic irAEs (heme‐irAEs) are a typical example, with a previously reported incidence of 0.6%–3.7% in clinically reported cohorts [8, 9, 10]. However, reported frequencies can vary depending on case definitions, diagnostic methods, and whether studies evaluate clinically adjudicated heme‐irAEs or broader hematologic laboratory toxicities [11, 12]. Heme‐irAEs are more likely to become severe, with 76.0%–85.0% of all cases being Common Terminology Criteria for Adverse Events (CTCAE) Grade 3–4, 2.0%–9.0% being Grade 5, and the majority of cases requiring inpatient care, which can have a major impact on the patient's treatment plan [9, 10, 13, 14]. In contrast, some studies have reported heme‐irAEs to be generally less severe and more common than previously recognized. These conflicting findings highlight the need for further research to clarify the epidemiology and clinical spectrum of heme‐irAEs [12].

Owing to the rarity of heme‐irAEs, there are few reports of large‐scale comprehensive analyses of heme‐irAE. Moreover, diagnosis and treatment selection for heme‐irAEs are made based on the autoimmune blood disorders without background or underlying disease, but it is unclear whether this is the optimal management for heme‐irAEs [15]. To address these clinical questions, we conducted a single‐center retrospective study of real‐world data on heme‐irAEs using the medical database of the National Cancer Center Hospital East to obtain more detailed clinical data.

2. Materials and Methods

Heme‐irAE patients treated at the National Cancer Center Hospital East, Kashiwa, Japan from April 1, 2014, to March 31, 2024, were selected for a retrospective study. Heme‐irAEs are defined as new‐onset hematologic abnormalities occurring during or after ICI therapy that were judged to be immune‐mediated after eliminating all alternative etiologies, including cytotoxic chemotherapy‐related myelosuppression, radiotherapy‐related marrow suppression, bone marrow involvement by malignancy, hematologic malignancy, active infection, disseminated intravascular coagulation, bleeding, nutritional deficiency, and concomitant medications. In patients undergoing concomitant cytotoxic chemotherapy, chemotherapy‐related myelosuppression was considered less likely when cytopenia occurred outside the expected nadir period, was disproportionate to the chemotherapy regimen, showed isolated lineage involvement inconsistent with conventional myelosuppression, was accompanied by immune‐mediated features such as hemolysis or a positive direct antiglobulin test (DAT), persisted despite treatment delay or dose modification, or improved after immunosuppressive therapy. Conversely, patients whose cytopenia was temporally consistent with the expected chemotherapy nadir and recovered without immunosuppressive treatment were not classified under having heme‐irAEs unless additional immune‐mediated features were observed. Case ascertainment was performed using three complementary approaches: diagnostic name searches, identification of patients who underwent bone marrow examination, and identification of patients with a positive direct antiglobulin test. In addition to the predefined diagnostic terms of autoimmune hemolytic anemia (AIHA), aplastic anemia (AA), immune thrombocytopenia (ITP), hemophagocytic lymphohistiocytosis (HLH), and pure red cell aplasia (PRCA), cases of neutropenia, pancytopenia, Evans syndrome, and DRESS‐associated cytopenia were identified by hematologists reviewing bone marrow examination records, laboratory abnormalities, and clinical documentation. Severity was described according to the CTCAE (version 5.0).

Treatment response for heme‐irAEs was classified into the following three groups: complete response (CR), partial response (PR), and no response. CR was defined as fulfilling all of the following criteria: neutrophil count of > 1.5 × 10⁹/L, platelet count of > 100 × 10⁹/L, and Hb of > 11 g/dL. PR was defined as fulfilling all of the following criteria: neutrophil count of > 0.5 × 10⁹/L, platelet count of > 20 × 10⁹/L, and Hb of > 8 g/dL. Criteria for assessing the therapeutic response to heme‐irAEs were defined based on a previously reported response criteria for AA [16] and institutional laboratory reference ranges. Because the cohort included heterogeneous heme‐irAE entities, we used predefined global hematologic recovery criteria based on neutrophil count, platelet count, and hemoglobin level to conduct pooled analyses across all heme‐irAEs. These pooled criteria were intended to provide a uniform descriptive measure of hematologic recovery and not considered validated response criteria for each individual heme‐irAE subtype. Time to onset was descriptively summarized for patients who developed heme‐irAEs. Regarding heme‐irAE onset time, heme‐irAEs onset after 12 months from the start of ICI administration was defined as delayed onset. In terms of regimen‐specific onset‐time comparisons, one patient who received dual ICI therapy combined with chemotherapy was excluded as this regimen did not fit into the mutually exclusive categories of ICI monotherapy, ICI plus chemotherapy, and dual ICI therapy without chemotherapy. However, this patient was retained in the overall cohort analyses. Because patients without heme‐irAEs were not included in this onset‐time comparison, the analysis was not intended to estimate cumulative incidence in the full ICI‐treated cohort. To calculate the observed incidence of heme‐irAEs, the total number of patients treated with ICI‐containing regimens during the study period was obtained from the institutional medical database. Because case ascertainment relied on diagnostic coding and selected diagnostic procedures, the calculated incidence should be interpreted as a minimum estimate. Regarding ICIs, we searched for pembrolizumab and nivolumab as anti‐PD‐1 antibodies, atezolizumab and durvalumab as anti‐PD‐L1 antibodies, and ipilimumab and tremelimumab as anti‐CTLA‐4 antibodies. Other ICI agents approved during or after the study period were not included if they were not used in the study cohort.

Using the medical database, we assessed patient demographics (gender, age, tumor type, performance status, etc.), treatment for the primary disease, details of irAEs (onset time, diagnosis name, blood test findings, etc.), treatment management of irAEs (use of systemic corticosteroids and other immunomodulatory drugs), and the impact of heme‐irAE onset on cancer treatment (suspension of cancer treatment due to heme‐irAE, selection of cancer treatment drugs after heme‐irAE onset, etc.). We also considered the need for bone marrow examination in diagnosing heme‐irAE patients and the feasibility of ICI re‐administration in treatment. Exact 95% confidence intervals for proportions were calculated using the Clopper–Pearson method. Differences in the proportions of the treatment groups were evaluated using Fisher's exact test. Median time to heme‐irAE onset and corresponding 95% confidence intervals were estimated descriptively for patients who developed heme‐irAEs using the Kaplan–Meier method, with all included patients treated as event cases. Survival time was defined as the interval from heme‐irAE onset to death from any cause. Survival curves were estimated using the Kaplan–Meier method and compared between groups using the log‐rank test.

EZR was used for all statistical analyses. EZR is a statistical software that extends the functions of R and R Commander and is distributed free of charge on the website of Jichi Medical University Saitama Medical Center [17]. Statistical significance was defined as a p value of < 0.05. This study was approved by the ethics committee of the National Cancer Center Hospital East (Research Project Number: 2018‐416). This retrospective observational study employed an opt‐out consent process. Details of the study, including the possibility of publishing anonymized data, were publicly disclosed, allowing patients the opportunity to refuse participation.

3. Results

Patient characteristics are summarized in Table 1 and Table S1. Of the 3371 patients who underwent treatment with ICI‐containing regimens at our institution, 40 were identified as having heme‐irAEs, resulting in an observed incidence rate of 1.2% (95% CI, 0.8%–1.6%). Further, the observed incidence was 1.0% (34 of 3301 patients; 95% CI, 0.7%–1.4%) among patients treated with single ICI with or without chemotherapy and 8.6% (6 of 70 patients; 95% CI, 3.2%–17.7%) among those treated with dual ICI therapy, defined as combined PD‐1/PD‐L1 and CTLA‐4 blockade. An exploratory comparison revealed that the observed incidence of heme‐irAEs was higher in patients receiving dual ICI therapy than in those receiving single ICI with or without chemotherapy (Fisher's exact test, p < 0.001). Because only six events occurred in the dual ICI group, this comparison should be interpreted as exploratory. Among the 40 patients with heme‐irAEs, 25 (62.5%) were male, and the median age was 67.5 years (range: 37–81 years). No patient had a history of autoimmune disease before the administration of ICIs. The primary tumor types of patients who underwent ICI treatment and developed heme‐irAEs were lung cancer (13 patients), renal cell carcinoma (five patients), hepatobiliary and pancreatic cancer (five patients), gastric cancer (four patients), cervical cancer, malignant mesothelioma and ovarian cancer (two patients each), cancer of unknown primary origin, esophageal cancer, esophagogastric junction cancer, nasal cancer, oropharyngeal cancer, parotid gland cancer, and prostate cancer (one patient each). The best tumor response before heme‐irAE onset was complete response in one patient, partial response in eight patients, stable disease in five patients, progressive disease in 20 patients, and not evaluable in six patients. The objective response rate, defined as complete or partial response, was 22.5%. Tumor response before heme‐irAE onset was summarized to contextualize subsequent survival and the feasibility of resuming cancer treatment.

TABLE 1.

Patient characteristics.

	Patients (n = 40), n (%)
Age in years, median (range)	67.5 (37–81)
Sex, n (%)
Male	25 (62.5)
Female	15 (37.5)
Tumor type, n (%)
Thoracic	15 (37.5)
Gastrointestinal	6 (15.0)
Urologic	6 (15.0)
Hepatobiliary and pancreatic	5 (12.5)
Gynecological	4 (10.0)
Head and neck	3 (7.5)
Cancer of unknown primary origin	1 (2.5)
ICI regimen, n (%)
ICI monotherapy	12 (30.0)
ICI plus chemotherapy	22 (55.0)
Dual ICI therapy	6 (15.0)
Number of chemotherapy regimens before ICI, n (%)
0	22 (55.0)
1	11 (27.5)
2	2 (5.0)
3	3 (7.5)
4	1 (2.5)
10	1 (2.5)
Radiation before ICI, n (%)
Yes	9 (22.5)
No	31 (77.5)

Abbreviation: ICI, immune checkpoint inhibitor.

ICIs used were anti‐PD‐1 antibodies in 65.0% (26 of 40 patients), anti‐PD‐L1 antibodies in 20.0% (8 of 40 patients), and anti‐PD‐1 antibodies + anti‐CTLA‐4 antibodies in 15.0% (6 of 40 patients). 30.0% (12 of 40 patients) received ICI monotherapy, 15.0% (6 of 40 patients) received dual ICI treatment, and 55.0% (22 of 40 patients) received ICI plus chemotherapy. 45.0% (18 of 40 patients) had a history of chemotherapy prior to ICI administration. The median number of chemotherapy regimens before ICI administration was 0 (range: 0–10). 22.5% (9 of 40 patients) had a history of radiotherapy prior to ICI administration.

Of the patients, 72.5% (29 of 40) underwent bone marrow examination. DAT was performed for 20 of the 40 patients, including those with anemia or suspected hemolysis, and was positive in 8 patients. Figure 1 shows the breakdown of heme‐irAEs. ITP was the most common heme‐irAE at 30.0% (12 of 40 patients), followed by AA/pancytopenia at 17.5% (7 of 40), AIHA at 12.5% (5 of 40), PRCA at 10.0% (4 of 40), HLH at 10.0% (4 of 40), neutropenia at 10.0% (4 of 40), Evans syndrome at 5.0% (2 of 40), DRESS syndrome at 2.5% (1 of 40), and immunological anemia at 2.5% (1 of 40). Work‐up completeness and diagnostic evidence by heme‐irAE subtype are summarized in Table S2. Figure 2 shows the severity of heme‐irAEs by disease. Adverse events of Grade 3 or higher occurred in 85.0% (34 of 40 patients, 95% CI, 70.2%–94.3%) (Grade 1: 0 patient, Grade 2: 6 patients, Grade 3: 20 patients, Grade 4: 14 patients, Grade 5: 0 patient). The percentage of heme‐irAEs of Grade 3 or higher with ICI monotherapy, dual ICI, and ICI plus chemotherapy was 83.0%, 100.0%, and 82.0%, respectively. Figure 3 shows the percentage of ICI monotherapy, dual ICI, and ICI plus chemotherapy for each heme‐irAE. 35.0% (14 of 40 patients) had concomitant non‐hematologic irAEs. These included pneumonitis in 12.5% (5 of 40, 2 of 5 preceded heme‐irAEs), hepatitis/cholangitis in 7.5% (3 of 40, 2 of 3 preceded heme‐irAEs), nephritis in 7.5% (3 of 40, 2 of 3 preceded heme‐irAEs), and enteritis in 5.0% (2 of 40, 1 of 2 preceded heme‐irAEs). The cumulative distribution of time to heme‐irAE onset among affected patients is shown in Figure S1. Half of the cases occurred within 4 months after starting treatment. The median time to heme‐irAE onset after ICI initiation was 12 weeks (range: 1–148). Among patients who developed heme‐irAEs and were included in the regimen‐specific onset‐time comparison, the median time to onset was approximately 24 weeks in the ICI monotherapy group (95% CI, 6–61 weeks), approximately 4 weeks in the dual‐ICI therapy without chemotherapy group (95% CI, 1 week–not estimable), and approximately 7 weeks in the ICI plus chemotherapy group (95% CI, 4–31 weeks). These regimen‐specific comparisons are shown in Figure 4A,B. One patient who received dual ICI therapy combined with chemotherapy was excluded from this regimen‐specific comparison because this regimen did not fit into the mutually exclusive regimen categories. Because this descriptive analysis included only patients who developed heme‐irAEs and did not include patients without heme‐irAEs as censored observations, the apparent shorter time to onset in the dual‐ICI group should be interpreted cautiously. A nominal p value was calculated for descriptive purposes. There was no significant difference between ICI monotherapy and ICI plus chemotherapy.

FIGURE 1.

Frequency of hematologic immune‐related adverse events (heme‐irAEs) (n = 40). Breakdown of hematologic immune‐related adverse events (heme‐irAEs) observed in this study. Immune thrombocytopenia (ITP) was the most common at 30.0% (12 of 40), followed by aplastic anemia (AA)/pancytopenia at 17.5% (7 of 40), autoimmune hemolytic anemia (AIHA) at 12.5% (5 of 40), pure red cell aplasia (PRCA) at 10.0% (4 of 40), hemophagocytic lymphohistiocytosis (HLH) at 10.0% (4 of 40), and neutropenia at 10.0% (4 of 40). ITP is shown in blue, AA/pancytopenia in orange, AIHA in gray, PRCA in yellow, HLH in light blue, neutropenia in green, Evans syndrome in navy blue, DRESS syndrome in brown, and immunological anemia in black.

FIGURE 2.

CTCAE grade of hematologic immune‐related adverse events (heme‐irAEs). Breakdown of CTCAE Grades for each heme‐irAE observed in this study. Grade 4 is shown in red, Grade 3 in yellow, and Grade 2 in blue. Across all conditions, the majority (85.0%) of heme‐irAEs were Grade 3 or higher.

FIGURE 3.

Proportion of ICI monotherapy, ICI plus chemotherapy, and dual ICI therapy by heme‐irAE subtype. Breakdown of immune checkpoint inhibitor (ICI) regimens for each heme‐irAE subtype observed in this study. ICI monotherapy is shown in blue, ICI plus chemotherapy in yellow, and dual ICI therapy in orange.

FIGURE 4.

Time to onset of heme‐irAEs after ICI treatment. (A) ICI monotherapy vs. dual ICI therapy without chemotherapy. (B) ICI monotherapy vs. ICI with chemotherapy. Time to onset among patients who developed heme‐irAEs, stratified by ICI regimen. This analysis was descriptive and limited to event cases. One patient who received dual ICI therapy combined with chemotherapy was excluded from the regimen‐specific comparison because this regimen did not fit into the mutually exclusive categories of ICI monotherapy, ICI plus chemotherapy, and dual ICI therapy without chemotherapy. (A) comparison of ICI monotherapy versus dual ICI therapy without chemotherapy; (B) comparison of ICI monotherapy versus ICI plus chemotherapy. The median time to onset was 171.5 days in the ICI monotherapy group, 29 days in the dual ICI therapy without chemotherapy group, and 47 days in the ICI plus chemotherapy group. Because patients without heme‐irAEs were not included as censored observations, this analysis was not intended to estimate cumulative incidence or compare event risks in the full ICI‐treated cohort. A nominal p value is shown for descriptive purposes only.

Regarding the treatment of heme‐irAEs, ICI therapy was interrupted or discontinued in 95.0% (38 of 40 patients), and systemic corticosteroids were administered in 62.5% (25 of 40 patients). In addition, secondary immunosuppressive therapy was performed in 17.5% (7 of 40) of patients. The second‐line or later agents used included IVIG in four patients, TPO‐RA in two patients, CyA in one patient, and rituximab in one patient; one patient received both CyA and TPO‐RA. Figure S2 shows the efficacy of systemic corticosteroids for heme‐irAEs. Hematologic improvement, defined as CR or PR, was achieved in 18 of 25 patients treated with systemic corticosteroids (72.0%; 95% CI, 50.6%–87.9%) (ITP: 4 of 8 patients, AA/pancytopenia: 5 of 6 patients, AIHA: 1 of 1 patient, PRCA: 4 of 4 patients, HLH: 3 of 3 patients, DRESS syndrome: 1 of 1 patient, Evans syndrome: 0 of 2 patients) (CR: 10 patients, PR: 8 patients). The median time of response to systemic corticosteroids was 14 days (range: 2–72). Of the 18 patients in whom systemic corticosteroids were effective, they could be tapered off in 17 patients. Among these 17 patients, systemic corticosteroids were discontinued in seven patients (3 with PRCA, 2 with pancytopenia, 1 with DRESS syndrome, 1 with AIHA), tapered to and maintained at ≤ 10 mg/day in nine patients, and tapered to 40 mg/day in one patient before death due to progression of the primary malignancy. Figure S3 shows the efficacy of second‐line treatment for heme‐irAEs. Intravenous immunoglobulin (IVIG) was used in four patients. In three patients, IVIG was administered concomitantly with systemic corticosteroids, and CR of heme‐irAEs was achieved. The remaining patient had steroid‐refractory ITP, and IVIG was also ineffective. Cyclosporine (CyA) was used in one patient with steroid‐refractory AA/pancytopenia as second‐line treatment, but it was ineffective. Thrombopoietin receptor agonist (TPO‐RA) was used in two patients. In 1 of the 2 patients, it was used as second‐line treatment for steroid‐refractory ITP, and CR was achieved with TPO‐RA. Rituximab was selected as second‐line treatment in one patient with steroid‐resistant Evans syndrome, and CR was achieved.

Heme‐irAEs recurred in four patients (2 with ITP, 1 with AIHA, 1 with Evans syndrome).

Chemotherapy or radiotherapy for cancer could be resumed in 50.0% (20 of 40 patients, 95% CI, 33.8%–66.2%) after the onset of heme‐irAEs. The median period of treatment interruption for cancer after the onset of heme‐irAEs was 83 days (range: 34–271). Of the 20 patients who resumed or continued treatment for cancer, seven underwent subsequent ICI therapy after heme‐irAE onset. Subsequent ICI therapies after the onset of heme‐irAEs are summarized in Table S3. Six patients underwent ICI rechallenge after temporary interruption of ICI therapy, whereas one patient continued ICI therapy without interruption because the heme‐irAE was not clinically recognized at the time and eventually improved spontaneously. Among these seven patients, one had Grade 2, four had Grade 3, and two had Grade 4 heme‐irAEs. Three patients received the same ICI class (PD‐L1 in two patients and PD‐1 in one patient), and four were switched to another ICI class (PD‐1 + CTLA‐4 to PD‐L1, PD‐1 to PD‐L1, PD‐L1 to PD‐1, and PD‐1 to other PD‐1 in 1 patient each). The decision to rechallenge or continue ICI therapy was made after multidisciplinary discussion or subsequent retrospective review, considering the degree of hematologic recovery, absence of effective alternative cancer therapy, expected oncologic benefit, and feasibility of close hematologic monitoring. ICIs were administered without systemic corticosteroids in all seven patients. No recurrent heme‐irAEs were observed among the seven patients who received subsequent ICI therapy after heme‐irAE onset during the follow‐up period (0/7; 95% CI for recurrence, 0%–41.0%).

Moreover, 60.0% (24/40) of patients had died at the time of data collection. The cause of death was primary disease progression in 21 patients, irAE of pneumonitis in two patients, and pulmonary embolism in one patient. Figure 5 shows survival after heme‐irAE onset. The median survival after heme‐irAE onset was 197.5 days (range: 36–822). Regarding survival by response to treatment for heme‐irAEs, survival was 249 days (range: 117–604) for patients with CR, 174 days (range: 36–822) for patients with PR, and 111 days (range: 40–257) for patients with no response. Based on response to treatment for heme‐irAEs, patients were divided into those who responded to treatment (CR + PR) and those who did not respond to treatment (no response), and their survival after heme‐irAE onset was compared. Since heme‐irAEs improved in a median of 14 days, the landmark time point was set at 14 days, and survival was analyzed by the landmark method using EZR. In an exploratory landmark analysis conducted at 14 days, median survival after heme‐irAE onset was 279 days in patients with hematologic response (CR + PR) (95% CI, 218–822 days) and 111 days in those with no response (95% CI, 40 days–not estimable; log‐rank p < 0.001). Given the small number of non‐responders and heterogeneity of cancer types and disease status, this analysis should be interpreted as descriptive. In the no‐response group, five of the six patients were unable to resume subsequent cancer treatment, and four of them died of their primary disease due to cancer progression.

FIGURE 5.

Overall survival by hematologic response to heme‐irAE treatment. Survival after heme‐irAE onset was analyzed using the Kaplan–Meier method with a 14‐day landmark. Median survival after heme‐irAE onset was 279 days in patients with hematologic response (complete response [CR] or partial response [PR]; 95% CI, 218–822 days) and 111 days in those with no response (95% CI, 40 days–not estimable; log‐rank p < 0.001). This comparison was descriptive and should be interpreted cautiously because of the small number of non‐responders and heterogeneity of cancer types and disease status.

4. Discussion

In this single‐center Japanese retrospective cohort of 3371 patients treated with ICI‐containing regimens, 40 patients developed heme‐irAEs, corresponding to an observed incidence of 1.2% (95% CI, 0.8%–1.6%). Although rare, they were frequently severe, with 85.0% of cases classified under Grade 3 or higher. Immune thrombocytopenia was the most common, followed by aplastic anemia/pancytopenia, autoimmune hemolytic anemia, pure red cell aplasia, hemophagocytic lymphohistiocytosis, and neutropenia. Most patients required interruption or discontinuation of ICI therapy, and systemic corticosteroids were administered in a substantial proportion of cases. Cancer treatment was resumed or continued in only half of the patients after heme‐irAE onset, and seven patients underwent subsequent ICI therapy after heme‐irAE onset, including six patients who underwent ICI rechallenge after temporary interruption and one patient who continued ICI therapy without interruption. These findings highlight the clinical importance of heme‐irAEs as rare but often severe complications that are associated with interruption of subsequent cancer treatment.

The incidence of heme‐irAEs was 1.2%, generally similar to previous reports [9]. In previously published studies of all irAEs, including non‐hematologic irAEs, it was reported that the incidence of irAEs was higher with dual ICI compared to ICI monotherapy [18], but in our study focusing on heme‐irAEs, the incidence of heme‐irAEs was 1.0% with single ICI ± chemotherapy and 8.6% with dual ICI, indicating that the incidence of heme‐irAEs may also be higher with dual ICI as in the case of non‐hematologic irAEs.

Previous studies have reported that pembrolizumab‐associated neutropenia occurs earlier in patients treated with pembrolizumab in combination with chemotherapy than in those treated with pembrolizumab monotherapy [19]. Earlier onset of heme‐irAEs has been reported with ipilimumab‐containing regimens than with anti‐PD‐1 or anti‐PD‐L1 monotherapy [20]. In the incidence analysis, patients who received dual ICI therapy combined with chemotherapy were included in the dual ICI group because the analysis was based on exposure to combined PD‐1/PD‐L1 and CTLA‐4 blockade. In contrast, for the regimen‐specific onset‐time comparison, one patient who received dual ICI therapy combined with chemotherapy was excluded because this regimen represented an overlapping category and did not fit into the mutually exclusive categories of ICI monotherapy, ICI plus chemotherapy, and dual ICI therapy without chemotherapy. Thus, the observed higher incidence refers to dual ICI therapy overall, whereas the apparent earlier onset refers to dual ICI therapy without chemotherapy. Both findings were based on small numbers of events and should be interpreted as exploratory and hypothesis‐generating. Larger multicenter cohorts or pharmacovigilance datasets are required to determine whether dual checkpoint blockade independently increases the risk of heme‐irAEs.

Regarding the breakdown of heme‐irAEs, although there are differences among reports, previous overseas reports have shown that ITP and AIHA are common, whereas AA, PRCA, and HLH are rare [9, 13]. In our study, ITP was the most common, whereas AA, AIHA, PRCA, and HLH were observed with a similar frequency. Compared with overseas reports, ITP occurred at a similar frequency, but AIHA may have occurred at a slightly lower frequency, while AA, PRCA, and HLH occurred at a slightly higher frequency. For AIHA, previous reports have demonstrated that 25.0% of the patients have an underlying disease [21], but in our study, no patients were found to have a history of autoimmune disease or hematological malignancy.

The mechanisms underlying heme‐irAEs are not fully understood, but the proposed mechanisms include loss of peripheral immune tolerance, expansion of autoreactive T and B cells, autoantibody production, hypercytokinemia, and immune‐mediated hematopoietic stem or progenitor cell dysfunction [13, 15]. These mechanisms may differ across heme‐irAE entities, which may partly explain the heterogeneity of clinical presentation and treatment response [14, 15].

Regarding the severity of heme‐irAEs, 85.0% of the patients were CTCAE Grade 3–4, which was generally similar to previous reports [9, 10, 13]. However, because case ascertainment in this retrospective study relied on diagnostic coding, selected diagnostic procedures, and clinical recognition, severe cases were likely to have been preferentially captured. Therefore, the observed proportion of Grade ≥ 3 events may be overestimated, and the true proportion among all heme‐irAEs may be lower. Heme‐irAEs that developed in patients receiving dual ICI were severe, with all patients being CTCAE Grade 3 or higher. Previous studies have reported 12.0% of deaths due to heme‐irAEs [13], but in our study, no deaths were directly attributed to heme‐irAEs. Most deaths were attributed to progression of the primary malignancy. However, because heme‐irAEs often led to interruption of cancer treatment, their potential indirect impact on oncologic outcomes warrants further investigation. For hematologic adverse events in patients with a history of ICI use, consultation with a hematologist and early diagnosis and treatment of heme‐irAEs are considered to be important.

Although irAEs usually occur within weeks to months (up to 4 months) after initiation of treatment, they may occur at any time, including after discontinuation of ICIs [22, 23]. The time from the start of immunotherapy to heme‐irAE onset ranged from 1 week to 34 months in our study. If the onset of irAEs after 12 months from the start of ICI administration is defined as delayed onset, there were six delayed onset cases. Therefore, heme‐irAEs also often appear within weeks to months (within 4 months) after initiation of treatment, and delayed onset cases were observed, so it is necessary to monitor for the appearance of heme‐irAEs during ICI treatment, even long after the start of treatment.

Regarding the diagnosis of heme‐irAEs, detailed tests such as bone marrow examination and DAT were not performed in all patients in this study. Heme‐irAEs were clinically diagnosed in 15.0% (6 of 40) of patients. Our findings support the importance of systematic evaluation to exclude alternative causes of cytopenia, including chemotherapy‐related myelosuppression, marrow involvement, hematologic malignancy, infection, and disseminated intravascular coagulation. Bone marrow examination may be useful in selected patients, particularly those with multilineage cytopenia, abnormal peripheral blood findings, suspected marrow involvement, lymphadenopathy, splenomegaly, or inadequate response to initial management. However, because diagnostic work‐up was not uniform in this retrospective cohort, our data do not establish a standardized diagnostic algorithm for heme‐irAEs. This nonuniform diagnostic work‐up may have introduced a risk of entity misclassification.

Among patients treated with systemic corticosteroids, hematologic improvement was achieved in 72.0% (18 of 25), whereas 17.5% (7 of 40) required second‐line treatment for heme‐irAEs. In our cohort, systemic corticosteroids were frequently used and were associated with hematologic improvement in several subtypes, including PRCA. However, the optimal immunosuppressive strategy for heme‐irAEs remains uncertain and should be individualized according to the specific hematologic diagnosis, severity, and oncologic context. Because the timing and choice of immunosuppressive therapy were not standardized in this retrospective cohort, the observed treatment responses should be interpreted descriptively.

With regard to the impact of heme‐irAEs on cancer treatment, treatment of the primary disease could be resumed in 50.0% of patients (20 of 40), despite the large number of severe heme‐irAE cases. Although some irAEs (especially endocrinologic irAEs) do not necessarily require discontinuation of ICIs, in some heme‐irAE cases, permanent discontinuation should be considered even at Grade 2, as described in the ASCO and ESMO guidelines [3, 4]. Previous reports have shown that 30.3%–60.0% of patients experienced recurrent heme‐irAEs after ICI rechallenge [9, 24]. Nevertheless, other reports have suggested that ICI rechallenge is feasible in carefully selected patients with immune‐related autoimmune hemolytic anemia (AIHA) [25, 26]. However, comparative data to guide management after irAE resolution—specifically regarding ICI rechallenge—remain insufficient. Since current recommendations for managing severe or steroid‐refractory irAEs are largely based on expert opinion, there is an urgent need for more structured, comparative evidence to refine these clinical approaches [8]. In this cohort, seven patients received subsequent ICI therapy after heme‐irAE onset. Six patients underwent ICI rechallenge after temporary interruption of ICI therapy, whereas one patient continued ICI therapy without interruption because the heme‐irAE was not recognized during the clinical course and improved spontaneously. No recurrent heme‐irAEs were observed during the follow‐up period. However, these findings should be interpreted with caution because the patients were highly selected, the sample size was small, and one case represented uninterrupted ICI continuation rather than conventional rechallenge. Larger pharmacovigilance and cohort studies have reported clinically meaningful rates of recurrent irAEs after ICI rechallenge [27, 28]. Therefore, our findings do not establish the safety of ICI rechallenge. Decisions regarding subsequent ICI therapy after heme‐irAE onset should be personalized and made only after careful clinical assessment and close hematologic monitoring. If ICI rechallenge is considered, close hematologic monitoring and multidisciplinary collaboration are warranted so that recurrent heme‐irAEs can be promptly recognized and managed.

The limitations of this study include the following: there are limitations to the number of cases and data collection because this was a retrospective study conducted at a single institution; there is no specific diagnostic tool for heme‐irAEs, and some patients were diagnosed by the clinical judgment of attending physicians; the timing and content of therapeutic intervention for heme‐irAEs are not standardized as they are based on the judgment of each attending physician. In this study, there were patients in whom heme‐irAEs were not diagnosed in real time, but after spontaneous resolution of hematologic toxicity. It is possible that low‐grade hematologic toxicity may not have been addressed, and that there may be more heme‐irAE patients who were actually unrecognized. We did not systematically evaluate the baseline predictors or soluble biomarkers associated with the development of heme‐irAEs, such as baseline blood counts, prior autoimmune predisposition, inflammatory markers, or treatment‐related factors. Therefore, this study could not identify patients at increased risk of developing heme‐irAEs. Limitations in the analysis of survival after heme‐irAE onset include the fact that there is a wide variety of primary cancer types, and that there is a wide variety of treatment histories for the primary diseases. The longer survival observed among patients with hematologic response should not be interpreted as evidence that improvement of heme‐irAEs directly prolongs survival. Patients who responded to heme‐irAE treatment may have had a more favorable disease status at baseline, better performance status, or greater opportunity to resume cancer therapy. Therefore, this finding should be considered hypothesis‐generating. Considering the rarity of heme‐irAEs, prospective studies are likely to be challenging, and further accumulation of cases will continue to be necessary.

Another limitation is that pooled response criteria may not fully capture disease‐specific recovery patterns. For example, platelet recovery is most relevant in ITP, hemolysis resolution in AIHA, reticulocyte recovery in PRCA, and inflammatory marker improvement in HLH. Therefore, the pooled response rate should be interpreted as a descriptive measure of hematologic recovery rather than as a validated disease‐specific treatment response.

In conclusion, heme‐irAEs associated with ICIs were rare but frequently severe in this single‐center Japanese retrospective cohort. Heme‐irAEs often required ICI interruption, systemic corticosteroid therapy, and prolonged interruption of subsequent cancer treatment. Because the observed associations involving dual ICI therapy, hematologic response, survival, and ICI rechallenge or continuation were based on small numbers of patients, these findings should be interpreted as exploratory. Early recognition, careful exclusion of alternative causes, and multidisciplinary management involving hematologists are essential for patients with suspected heme‐irAEs.

Author Contributions

Ryo Yoshimaru: conceptualization, methodology, writing – original draft, writing – review and editing. Hirotaka Nakamura: conceptualization, methodology, writing – original draft, writing – review and editing. Yong‐Mei Guo: investigation, writing – review and editing. Kensuke Matsuda: investigation, writing – review and editing. Tetsuya Sakai: investigation, writing – review and editing. Yoshitaka Zenke: investigation, writing – review and editing. Kiyotaka Yoh: investigation, writing – review and editing. Kohei Shitara: investigation, writing – review and editing. Nobuaki Matsubara: investigation, writing – review and editing. Kenichi Harano: investigation, writing – review and editing. Hiromichi Nakajima: investigation, writing – review and editing. Nobuyuki Takahashi: investigation, writing – review and editing. Masafumi Ikeda: investigation, writing – review and editing. Takao Fujisawa: investigation, writing – review and editing. Saki Nakamura: investigation, data curation, writing – review and editing. Kenta Akie: investigation, data curation, writing – review and editing. Masayuki Ito: investigation, data curation, writing – review and editing. Masashi Wakabayashi: formal analysis, writing – review and editing. Junichiro Yuda: conceptualization, methodology, writing – original draft, supervision, writing – review and editing.

Funding

The authors have nothing to report.

Ethics Statement

All procedures were performed in accordance with the ethical standards of the institutional and national research committee and the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. This study was approved by the review board of National Cancer Center Hospital (Approval Number: 2018‐416). This retrospective observational study employed an opt‐out consent process. Details of the study, including the possibility of publishing anonymized data, were publicly disclosed, allowing patients the opportunity to refuse participation.

Consent

The authors have nothing to report.

Conflicts of Interest

Kensuke Matsuda has received honoraria from Chugai Pharmaceutical. Tetsuya Sakai has received grants from Amgen, GlaxoSmithKline K.K., Merck, Daiichi Sankyo, NEC, Janssen; honoraria from AstraZeneca, Novartis, Thermo Fisher Scientific, Takeda, OLYMPUS, Taiho, Janssen, Chugai, MSD, Merck, Daiichi Sankyo, RIKEN GENESIS, Amco. Yoshitaka Zenke has received grants or contracts from AstraZeneca, Daiichi Sankyo, MSD, Merck, Amgen, GSK, Roche, Chugai, and Novocure; consulting fees from AstraZeneca, Boehringer‐Ingelheim, and Pfizer; and payment or honoraria from AstraZeneca, Lilly, Chugai, Ono, Bristol‐Myers Squibb, Takeda, Boehringer‐Ingelheim, Taiho, MSD, Novartis, Pfizer, Nippon Kayaku, Kyowa Kirin, Amgen, Guardant Health, Johnson & Johnson Innovative Medicine, and Merck Biopharma. Kiyotaka Yoh has received grants or contracts from AbbVie, ArriVent Biopharma, AstraZeneca, Boehringer‐Ingelheim, Chugai Pharma, Daiichi Sankyo, Lilly, MSD, Taiho, Takeda; consulting fees Boehringer‐Ingelheim, AbbVie; and honoraria from Amgen, AstraZeneca, Bristol‐Myers Squibb, Chugai Pharma, Daiichi Sankyo, Kyowa Kirin, Lilly, MSD, Ono Pharmaceutical, Takeda, Janssen, Taiho, Boehringer‐Ingelheim. Kohei Shitara has received personal fees for consulting and advisory roles from Bristol‐Myers Squibb, Takeda, Ono Pharmaceutical, Novartis, Daiichi Sankyo, Amgen, Boehringer‐Ingelheim, Merck Pharmaceutical, Astellas, Guardant Health Japan, Janssen, AstraZeneca, Zymeworks Biopharmaceuticals, ALX Oncology Inc., Bayer, GlaxoSmithKline K.K., Healios K.K., Moderna Inc., and Arcus Biosciences Inc.; honoraria from Bristol‐Myers Squibb, Ono Pharmaceutical, Janssen, Eli Lilly, Astellas, and AstraZeneca; and research funding (all to institution) from Astellas, Ono Pharmaceutical, Daiichi Sankyo, Taiho Pharmaceutical, Chugai, Merck Pharmaceutical, Amgen, Eisai, PRA Health Sciences, Syneos Health, AstraZeneca, PPD‐SNBL K.K. and TORAY. Nobuaki Matsubara has received honoraria (personal) from Sanofi; research funding (institution) from Amgen, Astellas Pharma, AstraZeneca, Bayer, Bristol‐Myers Squibb, Boehringer‐Ingelheim, Chugai Pharma, Eisai, Janssen, Lilly, MSD, Novartis, Pfizer, PRA Health Sciences, Roche, Seagen, Taiho, Takeda and Telix; and travel, accommodation, and/or expenses (personal) from Pfizer. Kenichi Harano has received honoraria from AstraZeneca, Chugai, Eisai, MSD, Taiho, Takeda; advisory roles from AstraZeneca, Chugai, Taiho, Eisai, Takeda; and Research Fund (institution) from AstraZeneca, Chugai, Daiichi Sankyo, MSD, Takeda. Hiromichi Nakajima has received consulting fees from Terumo Corporation; honoraria from Chugai, Takeda, Pfizer, Eli Lilly, AstraZeneca, Ono Pharmaceutical, Sanofi and Regeneron, and MSD; and institutional research funding from Merck Biopharma (all to the institution). H.N. also reports that an immediate family member is employed by Otsuka Pharmaceutical Co. Ltd. Masafumi Ikeda has received honoraria from AstraZeneca, Bristol‐Myers Squibb, Chugai, Eisai, Guardant Health Japan, Incyte Biosciences Japan, MSD, Nihon Servier, Novartis, Ono, Taiho, Takeda, Teijin Pharma; consulting or advisory role fees from AbbVie, Astellas Pharma, AstraZeneca, Boehringer‐Ingelheim, Bristol‐Myers Squibb, Eisai, Guardant Health Japan, MSD, Nihon Servier, Novartis, Ono Pharmaceuticals, Delta‐Fly Pharma, Rakuten Medical; and research funding (institution) from AbbVie, Amgen, AstraZeneca, Bristol‐Myers Squibb, Chiome Bioscience, Chugai, Eisai, Eli Lilly Japan, Incyte Biosciences Japan, Merck Biopharma, MSD, Nihon Servier, Novartis, Rakuten Medical, Syneos Health, Taiho. Takao Fujisawa has received honoraria from Amelieff Co. Ltd. and AstraZeneca. Masashi Wakabayashi has received honoraria from Nihon Medi‐Physics Co. Ltd. and Kyowa Kirin Co. Ltd. Junichiro Yuda has received research grants from AbbVie, Daiichi Sankyo, Chugai, Genmab, Novartis, Amgen, Takeda, Bristol‐Myers Squibb Ltd., Incyte, Janssen, Sumitomo, MSD, and Mitsubishi Tanabe. All other authors declare no conflicts of interest.

Supporting information

Figure S1: Cumulative distribution of time to heme‐irAE onset among affected patients.

Figure S2: Proportion of treatment response to systemic corticosteroids by heme‐irAE subtype.

Figure S3: Proportion of treatment response to second‐line therapy by heme‐irAE subtype.

Table S1: List of patients with hematologic immune‐related adverse events.

Table S2: Diagnostic work‐up completeness and diagnostic evidence by heme‐irAE subtype.

Table S3: Subsequent ICI therapy after heme‐irAE onset.

Data Availability Statement

The datasets generated and/or analyzed during the current study are not publicly available because they contain potentially identifiable patient information and are subject to institutional ethical restrictions. De‐identified data may be available from the corresponding author upon reasonable request and with approval from the institutional review board, where applicable.