Skip to main content
NCBI home page
NPJ Breast Cancer logoLink to NPJ Breast Cancer
. 2025 Dec 3;11:144. doi: 10.1038/s41523-025-00849-1

Multi-institutional analysis of incidence and risks for late-onset immune toxicity in breast cancer

Saya Jacob 1,2,✉, Samantha Fisch 2, Carolyne Face 2, Kelly Blum 2, Madhuri Chengappa 3, Saliha Chaudhry 3, Alexis LeVee 4, Nikita V Baclig 5, Andrew Soliman 5, Laura A Huppert 2, Zoe Quandt 6, Laura Quintal 2, Dame Idossa 3, Michelle Melisko 2, A Jo Chien 2, Mi-Ok Kim 2, Joanne Mortimer 4, Kelly McCann 5, Anne Blaes 3, Hope S Rugo 2,4,✉
PMCID: PMC12714743  PMID: 41339346

Abstract

Immune checkpoint inhibitors (ICI) improve survival in triple-negative breast cancer (TNBC) but cause late-onset toxicity, with unknown incidence in breast cancer. This retrospective study included 700 patients (61%, n = 424 early-stage; 39%, n = 276 metastatic; 77% TNBC) from four NCI-designated centers treated with ICI between 2014–2021. Chart review identified immune toxicities, defined as ICI-related, as noted by the oncology provider or steroid-treated. 61% (n = 430) had toxicity: 37% (n = 257) early (≤90 days after ICI start), 34% (n = 240) delayed (>90 days), 10% (n = 67) both. Of the delayed, 144 (60%) were on-treatment, 56 (23%) off-treatment, 40 (17%) both. Twenty-two (9.2%) had off-treatment toxicity >1 year post-ICI. Median onset: 138 days (range 90–1380) on-treatment; 76 days (21–1144) off-treatment. Risk increased with more ICI cycles, especially >4 (early OR 1.104; metastatic OR 1.06; p < 0.0001) and higher baseline eosinophils (OR 3.46, p = 0.0484). Metastatic disease (OR 0.18, p < 0.0001) and early toxicity (OR 0.53, p = 0.0008) reduced risk. Findings support the need for high clinical suspicion for late toxicity.

Subject terms: Breast cancer, Cancer immunotherapy

Introduction

Immune checkpoint inhibitors (ICIs) have revolutionized the field of solid tumor oncology and have led to improvement in survival for many different cancer types1–3. Within breast oncology, the Keynote 522 and Keynote 355 trials demonstrated improved overall survival using pembrolizumab in combination with chemotherapy for patients with early-stage and PD-L1 positive metastatic triple negative breast cancer (TNBC), respectively4–7. The duration of pembrolizumab in early-stage disease is typically one year; the duration in metastatic disease is indefinite until disease progression or intolerable side effects. Although highly effective, ICI can cause immune-related adverse events (irAEs) with reported rates in patients with breast cancer ranging from 27–64%5–9. The most common irAEs in patients with breast cancer include dermatitis, affecting up to 49% of patients, followed by colitis in up to 20%, hypothyroidism in up to 18%, and adrenal insufficiency/hypophysitis (AI) in up to 10%10,11. Endocrine toxicities are of particular concern due to the need for lifelong replacement therapy with an impact on quality of life, although rarer toxicities affecting vital organs can be equally impactful. Thus, the benefit of these agents must be carefully considered alongside the risk of irAE.

Real-world use of ICI has led to the recognition of late-onset immune toxicities. These toxicities have been noted to occur months after the start of treatment and even after treatment discontinuation, and thus are not well captured by traditional clinical trial follow-up periods. For example, in the Keynote 522 trial, adverse events were monitored for a maximum of 90 days after treatment discontinuation4, but clinical practice has led to recognition of irAEs with onset well beyond this time point. Thus, the incidence of irAEs is likely higher than reported in pivotal phase III trials.

Prior analyses of late-onset immune toxicities have focused primarily on non-breast tumor types. Retrospective studies of ICI in lung cancer and melanoma found that 6.9% of patients experienced irAEs at least one year after the ICI start date12. Additional studies of patients with melanoma have shown that up to 4-6% of patients can experience irAEs more than 100 days after drug discontinuation and that 5.3% of patients experienced irAE one year after therapy initiation, of which 39% were grade 313,14. Interestingly, a meta-analysis evaluating 8,436 patients with several cancer types, not including breast cancer, found that in patients treated with PD-1 agents, there was a longer median time to onset for severe (grade 3) compared to any grade irAEs (27.5 weeks versus 8.4 weeks, respectively)15. Late onset toxicities may be serious, such as pneumonitis with a median onset of 38 weeks in a population with mixed tumor types16.

The incidence of late-onset irAEs in patients with breast cancer is not well studied. This represents an area of unmet need, given that ICIs are now part of the standard of care for patients with stage II and III TNBC in the neoadjuvant/adjuvant setting and as first-line treatment for patients with PDL1+ metastatic TNBC. Patients with early-stage TNBC are treated with curative intent and often survive many years after therapy, with a resulting large potential for impact from late-onset toxicities. The frequency of clinic visits decreases over time, so late toxicities may be missed until they are quite severe. Thus, a high rate of suspicion and patient education is needed to diagnose late-onset irAEs. To address this need, we conducted a multicenter retrospective cohort study of patients treated with ICI to further explore the timing and prevalence of delayed onset irAEs in patients with breast cancer.

Results

We identified 700 patients with breast cancer undergoing ICI at four different NCI-designated cancer centers between 2014–2021. Demographics of the overall cohort as well as patients with delayed toxicity are outlined in Table 1. In the overall cohort, the median number of ICI cycles was 10, the median age was 52, 448 patients (64%) were white, 276 (39%) had metastatic disease, and 423 (60%) had early-stage disease. Disease subtypes of the entire cohort are outlined in Table 1. All patients received a PD1 or PDL1 inhibitor, and one patient received combination therapy with PD1/CTLA4 (ipilimumab and nivolumab). This includes 613 patients (88%) who received pembrolizumab and 32 patients (5%) who received atezolizumab. Other checkpoint inhibitors received included dostarlimab (n = 15), cemiplimab (n = 14), durvalumab (n = 14), avelumab (n = 4), and nivolumab (n = 7).

Table 1.

Patient characteristics

OVERALL COHORT n = 700 DELAYED TOXICITY N = 240
Metastatic 276 (39% of overall) 62 (26% of delayed)
Early stage 423 (60%) 177 (74%)
 Stage 1 42 (6%) 23 (10%)
 Stage 2 233 (33%) 89 (37%)
 Stage 3 146 (21%) 64 (27%)
 Unknown 2 (<1%) 1 (<1%)
Subtype
 Triple negative 539 (77%) 200 (83%)
 Hormone positive 137 (20%) 30 (13%)
 HER2 positive, Any HR 24 (3%) 10 (5%)
Median number of ICI cycles 10 10
Median age at time of ICI treatment 52 52
Gender
 Male 2 (<1%) 0
 Female 698 (99%) 240 (100%)
Menopausal status
 Premenopausal/peri-menopausal 358 (51%) 117 (49%)
 Postmenopausal 332 (47%) 120 (50%)
 Unknown/Not applicable 10 (1%) 3 (1%)
Race
 White 448 (64%) 170 (71%)
 Asian 66 (9%) 26 (11%)
 Black or African American 52 (7%) 18 (8%)
 Other 19 (3%) 13 (5%)
 Unknown 17 (2%) 7 (3%)
 Hispanic or Latino 10 (1%) 1 (0.4%)
 American Indian or Alaska Native 6 (1%) 3 (1%)
 Native Hawaiian or Other Pacific Islander 3 (0%) 1 (0.4%)
Median length of follow-up after first ICI dose 18 months 21 months
DELAYED TOXICITY INCIDENCE (N = 240 patients)
>1 Delayed toxicity 62 (26%)
On treatment delayed toxicity ONLY 144 (60%)
Off treatment delayed toxicity ONLY 56 (23%)
Both on and off treatment delayed toxicity 40 (17%)
Off treatment delayed toxicity occurring >1 year after ICI start 22 (9.2%)
Open in a new tab

Of those with delayed toxicity (Fig. 1), the median number of ICI cycles was 10 (9 for early stage, 10 for metastatic), the median age at the time of ICI treatment was 52, and most patients were white (n = 170, 71% of those with delayed toxicity). Most patients had early-stage disease (177, 74%), and 62 (26%) had metastatic disease, with 1 of unknown stage. The majority had TNBC (n = 200, 83%), 30 (13%) had hormone-receptor positive (HR+) disease, and 10 (4%) had HER2+ disease. The median length of follow-up after the first ICI dose was 21 months (range 1-81).

Fig. 1. Definitions of toxicity groups.

Fig. 1

Open in a new tab

Delayed toxicity was defined as any toxicity with onset occurring at least 90 days after the start of immune checkpoint inhibitors. Early toxicity was any toxicity with onset occurring prior to 90 days after the start of the immune checkpoint inhibitor. Delayed toxicity was divided into two subgroups: on-treatment toxicity and off-treatment toxicity, which was defined as a toxicity occurring 21 days after the last dose of ICI.

Out of all patients, 430 (61%) developed irAE at any time point, 257 (37%) developed early immune toxicity, and 240 (34%) developed a delayed toxicity (Fig. 2). Sixty-seven patients (10%) had both an early and delayed toxicity. Of those with delayed toxicity, 144 patients (60% of the delayed toxicity group) had on-treatment toxicity, 56 (23%) had off-treatment toxicity and 40 (17%) had both an on and off-treatment delayed toxicity. Of those with off-treatment delayed toxicity, 22 had an off-treatment toxicity occurring >1 year after ICI initiation. Of patients with early-stage disease, 177 developed delayed toxicity, which consisted of 42% of all the early-stage disease patients in the cohort (n = 423 patients in the entire cohort). Sixty-two patients with metastatic disease had delayed toxicity, representing 26% of all patients with metastatic disease (N = 276 patients in the entire cohort).

Fig. 2. Incidence of toxicity.

Fig. 2

Open in a new tab

A total of 700 patients with breast cancer undergoing treatment with immune checkpoint inhibitors were identified across 4 NCI-designated cancer centers with stage and subtype information outlined above. Patients within the box represent those who had an immune-related toxicity at any time point and are further categorized as those having early toxicity depicted within the blue circle with blue text, delayed toxicity depicted in the orange circle with orange text, or both. Stage and subtype information are outlined for both early and delayed toxicity. Patients within the subgroups of on and off treatment delayed toxicity are outlined in the orange box. TNBC triple negative breast cancer, HR+ hormone receptor positive, HER2+ human epidermal growth factor receptor 2, irAE immune related adverse event, pts patients.

Median time to onset

Median time to onset of delayed toxicities for both on and off treatment toxicities is outlined in Fig. 3. For on-treatment delayed toxicities, the median time to onset was 138 days (range 90–1380) after the start of ICI. The median time to onset for on-treatment dermatitis was 123.5 days (range 90–1154), colitis was 127.5 days (90–584), AI was 158 days (90–354), and thyroiditis was 129 days (range 90–1380). Delayed toxicities occurring off treatment occurred at a median of 76 days after the end of treatment (range 21–1144). Off-treatment dermatitis occurred 45 days (21–327) after treatment end, colitis occurred 89.5 days (range 21–1042), AI occurred 78 days (21–785), and thyroiditis occurred 81 days (21–1144) after treatment end.

Fig. 3. Median time to onset of delayed toxicities.

Fig. 3

Open in a new tab

The median time to onset of delayed toxicities, both on and off treatment, is depicted above for both groups overall, as well as the most common types of toxicity. Text within the bars represents the numerical value of the median time to onset. The error bars extending to the right depict the range in days, with the upper limit of the range written to the right of the error bars. Orange bars depict median time to onset since the start of immune checkpoint inhibitors for those with on-treatment delayed toxicity. The blue bar depicts time to onset from the end of immune checkpoint inhibitor treatment, specifically for those toxicities occurring off treatment. AI Adrenal insufficiency.

Types of delayed toxicity

There was a total of 394 delayed toxicity events occurring in 240 patients. The majority of these were grade 1–2 (n = 298, 76%) with some grade 3 or higher toxicities noted (96, 24%). The most common types of delayed toxicity of any grade were thyroiditis (n = 109, 28% of all delayed toxicity events), dermatitis (n = 68, 17%), colitis (n = 50, 13%) and AI (n = 38, 10%), (Table 2, Figure 4). Of the 109 patients who developed thyroiditis, 72 (77%) had early-stage disease. Of the 38 patients who developed AI, 33 (92%) had early-stage disease. Toxicities included in the “Other” category are outlined in Table 2.

Table 2.

Types of delayed toxicity

Delayed toxicity overall
N = 394 events		 On treatment delayed toxicity
N = 281 events		 Off treatment delayed toxicity
N = 113 events
Types of toxicity Any Grade Grade ≥3 Any Grade Grade ≥3 Any Grade Grade ≥3
Thyroiditis 109 (28%)1 3 (1%)1 77 (27%)2 2 (1%)2 32 (28%)3 1 (1%)3
Dermatitis 68 (17%) 7 (2%) 56 (12%) 5 (2%) 12 (11%) 2 (2%)
Colitis 50 (13%) 10 (3%) 34 (10%) 4 (1%) 16 (14%) 6 (5%)
Adrenal insufficiency/hypophysitis 38 (10%) 18 (5%) 19 (5%) 8 (3%) 19 (17%) 10 (9%)
Hepatitis 31 (8%) 8 (2%) 27 (7%) 6 (2%) 4 (4%) 2 (2%)
Arthritis 22 (6%) 18(5%) 15 (3%) 3 (1%) 7 (6%) 1 (1%)
Pneumonitis 16 (4%) 5 (1%) 9 (2%) 3 (1%) 7 (6%) 2 (2%)
Nephritis 8 (2%) 4 (1%) 6 (2%) 3 (1%) 2 (2%) 1 (1%)
Diabetes Mellitus 9 (2%) 9 (2%) 5 (2%) 5 (2%) 4 (4%) 4 (4%)
Myalgia 6 (2%) 1( < 1%) 6 (2%) 1 ( < 1%) 0 (0%) 0 (0%)
Mucositis 5 (1%) 1( < 1%) 5 (2%) 1 ( < 1%) 0 (0%) 0 (0%)
Fatigue 6 (1%) 1( < 1%) 6 (2%) 1 ( < 1%) 0 (0%) 0 (0%)
Encephalitis 3 (1%) 2( < 1%) 1 ( < 1%) 0 (0%) 2 (2%) 2 (2%)
Other4 23 (6%) 9 (2%) 15 (6%) 4 (1%) 8 (7%) 5 (4%)
Open in a new tab

1 Percent of total delayed toxicity events 2 Percent of total on treatment delayed toxicity events 3 Percent of total off treatment delayed toxicity events 4 Other toxicities include: i) for total delayed toxicities: thrombocytopenia (n = 2), Brachial plexitis (n = 1), Neutropenia (n = 2), cardiomyopathy (n = 2), neuropathy (n = 2), Bell’s Palsy (n = 1), Sjogren’s syndrome (n = 1), duodenitis/gastritis (n = 2), fever (n = 2), neuromuscular weakness (n = 1), sarcoidosis (n = 1), vitiligo (n = 1), eosinophilia (n = 1), shock (n = 1), autoimmune hemolytic anemia (n = 1), pancreatitis (n = 1), bronchitis (n = 1) ii) on treatment: Thrombocytopenia (n = 2), Brachial plexitis (n = 1), Neutropenia (n = 1), cardiomyopathy (n = 1), neuropathy (n = 1), Bell’s Palsy (n = 1), Sjogren’s syndrome (n = 1), duodenitis (n = 1), fever (n = 2), neuromuscular weakness (n = 1), sarcoidosis (n = 1), vitiligo (n = 1), eosinophilia (n = 1). iii) off treatment: Shock (n = 1), autoimmune hemolytic anemia (n = 1), pancreatitis (n = 1), myocarditis (n = 1), neutropenia (n = 1), gastritis (n = 1), bronchitis (n = 1), neuropathy (n = 1).

Fig. 4. Types of delayed toxicity.

Fig. 4

Open in a new tab

Bar graph depicting specific subtypes of delayed toxicity as outlined by the x-axis. The Y axis depicts the number of patients. Toxicities of any grade are depicted in the blue bars. Toxicities of grade 3 or greater are depicted in the orange bars. Numbers above each bar represent the absolute value of patients experiencing that toxicity. The percentage of total delayed toxicities is listed to the right of each absolute value. Those toxicities representing less than 5% of all delayed toxicities are depicted with an asterisk. Specific “other” toxicities are outlined in Table 2. AI Adrenal insufficiency.

There was a total of 281 on-treatment delayed toxicities occurring in 184 patients. The most common on-treatment delayed toxicities were thyroiditis (77, 27% of all on-treatment toxicity), dermatitis (56, 12%), colitis (34, 10%), hepatitis (27, 7%), and AI (19, 5%). There was a total of 113 off-treatment delayed toxicity events in 96 patients. The most common off-treatment delayed toxicity events were thyroiditis (32, 28% of all off-treatment delayed toxicity events), AI (19, 17%), colitis (16, 14%), and dermatitis (12, 11%). Eighteen off-treatment toxicities occurred >1 year after ICI start and included thyroiditis (n = 4), colitis/diarrhea (n = 3), dermatitis (n = 3), Arthralgia (n = 2), encephalitis (n = 1), neuropathy (n = 1), elevated transaminases (n = 1), pneumonitis (n = 1), bronchitis (n = 1), autoimmune hemolytic anemia (n = 1).

Risk factors for delayed toxicity

A multivariable logistic regression analysis was performed to determine risk factors for developing delayed toxicities, with results outlined in Fig. 5. These analyses were controlled for geographic site and stage of disease. The treatment institution was controlled to account for heterogeneity across the sites in order to remove site-level confounding in the evaluation of risk factors. A statistically significant site-level confounding existed (p = 0.03) with a significant difference in the number of menopausal women noted in one site (p = 0.0045). Patients with metastatic disease were less likely to develop delayed toxicity compared to those with early-stage disease (odds ratio (OR) 0.18, 95% confidence interval (CI) 0.10–0.35, p < 0.0001), despite similar median follow-up in both early and metastatic groups (24 months and 21 months, respectively). In addition, patients who developed early toxicity were less likely to develop delayed toxicity compared to those without early toxicity (OR 0.46, CI 0.31–0.68, p < 0.01). This remained true even after adjusting for those patients who stopped treatment prior to 90 days due to early toxicity (OR 0.53, p = 0.0008). An increasing number of cycles was associated with increased risk of developing delayed toxicity in both metastatic and early-stage disease (OR 1.06, CI 1.05–1.14, p < 0.01; OR 1.04, CI 1.00–1.08, p = 0.0269, respectively). Elevated baseline eosinophil count was also predictive of developing delayed toxicity (OR 3.46, CI 1.01-11.89, p = 0.0484). Age at treatment, hormone receptor-positive disease, baseline BMI, and baseline absolute neutrophil count were not associated with risk of delayed toxicity (see Fig. 5).

Fig. 5. Risk factors for developing delayed toxicity.

Fig. 5

Open in a new tab

Multi-variable analysis controlled for institution of treatment and metastatic vs early-stage disease was performed with results outlined above. The y-axis outlines specific clinical risk factors evaluated, and the x-axis depicts odds ratios additionally adjusted for the rest of the candidate risk factors included in the multivariable analysis. Each plot depicts point estimates of specific odds ratios with error bars depicting confidence intervals. Absolute values of odds ratios, confidence intervals and p-values are shown to the right of each plot. Plots in red indicate those that show a statistically significant result, while those in black are not significant. CI confidence intervals, No number, HR+ hormone receptor positive, BMI body mass index.

Analysis was also performed to determine a “cut-off” number of ICI cycles at which the risk of development of delayed toxicity is significantly increased. After testing of multiple cut-offs, this was determined to be 4 cycles in both the early-stage (OR 0.29, CI 0.155–0.538, p < 0.0001) and metastatic setting (OR 0.04, CI 0.010–0.188, p < 0.0001).

Management of delayed immune-related toxicities

Ninety-nine (41%) of the patients with delayed toxicity required steroid treatment, of which 16 (7%) patients required intravenous (IV) steroid treatment. Three patients received biologic therapy, including infliximab and dupilumab. One patient received IVIG. Excluding patients who developed endocrinopathies, 23 (10%) patients experienced ongoing symptoms at the time of data collection after initial treatment, including symptoms of dermatitis, pneumonitis, mucositis, colitis, arthritis, neuromuscular weakness, fatigue, transaminase elevation, vitiligo, thrombocytopenia, myocarditis, and neuropathy.

Early toxicity

Of the 257 patients with early immune toxicity, 67 had subsequent delayed toxicity (Fig. 2). Characteristics of patients developing early toxicity are outlined in Supplementary Table 1. Of the patients developing early toxicity, 91 (35%) discontinued further ICI therapy prior to 90 days. Of these patients, eight (3%) went on to develop a delayed toxicity during an off-treatment time point. Of the patients who continued treatment after developing early toxicity, 59 (23%) developed a delayed toxicity. The most common early toxicities of any grade were dermatitis (n = 96), colitis (n = 62), thyroiditis (n = 59), and hepatitis (n = 43). The most common grade ≥3 toxicity was thyroiditis (n = 43).

Discussion

Immune checkpoint inhibitors in combination with chemotherapy have improved outcomes for patients with TNBC and are an active area of investigation in other subsets of breast cancer. However, immune toxicities remain a significant clinical challenge that often impacts the delivery of potentially curative treatment5,6. While clinical trials have characterized these toxicities in the immediate time-points after treatment start, real-world use has led to recognition of late-onset toxicities occurring long after treatment start or in some cases after ICI treatment cessation17,18. This retrospective study is a rigorous evaluation of a large cohort of patients with breast cancer involving primary data collection and review, and incorporating four distinct cancer centers.

Of the 700 patients with breast cancer who underwent ICI, we report an overall irAE rate of 61% (n = 430), which is consistent with other real-world analyses in patients with breast cancer showing rates of 48–64%8,19. Most patients experiencing immune toxicities in our cohort had TNBC, reflecting the current approval for chemo-immunotherapy specifically in TNBC. A sizable portion experienced delayed toxicities (n = 240, 34% of total cohort), indicating that delayed toxicities are common in this patient population. This may also account for the discrepancy of high rates of toxicity seen in real-world (48-64%) settings compared to the lower rates reported in clinical trials (~35%) where later toxicities may not have been well-captured. In fact, the frequency of observed late toxicities in the real world has led to discussion about whether dose-limiting toxicity observation periods should be lengthened in immunotherapy dose-finding clinical trials, with evidence to support longer observation periods20.

Delayed toxicities are pharmacologically possible due to the long-lasting effects of ICI. Pharmacokinetic studies of pembrolizumab, nivolumab, and other PD-1 inhibitors have shown that the half-life of these antibodies ranges from 12–25 days with estimated drug clearance by around day 9021–23. However, these same studies also showed that over 70% of PD-1 molecules occupied circulating T cells for over two months from the last drug administration, with ~40% remaining more than eight months after the last dose of therapy21. Therefore, it is plausible that new irAEs might manifest a year following the date of drug discontinuation.

Sixty-seven patients experienced both an early and delayed toxicity, indicating that, at least in some patients, there may be multiple time windows during which autoimmunogenicity may be elicited. Prior analysis of patients with melanoma and non-small lung cancer reported that a subset of patients developed more than one immune toxicity, with the onset of the second toxicity occurring a median of 20 weeks later24. Interestingly, in this cohort of melanoma and non-small cell lung cancer patients, the presence of multiple immune toxicities per patient was associated with a higher rate of disease response, pointing to multiple time-points of immune activation both against the tumor and the host. Finally, we note that 96 patients had an off-treatment delayed toxicity, indicating that ongoing exposure to ICI may not be necessary to elicit long-term immune activation. This is underscored by the fact that 22 patients experienced off-treatment delayed toxicities over 1 year after starting therapy.

For those with on-treatment delayed toxicity, the median time to onset was 138 days, with a range out to 1380 days, and a median of 158 days and 129 days for AI and thyroiditis specifically. We also observed AI and thyroiditis off-treatment delayed toxicities, occurring a median of 78 and 81 days after ICI stop, respectively. While it is possible that some delays are due to delayed diagnosis rather than true delayed onset, these findings are similar to prior non-breast cancer studies showing probability of irAE onset at 6, 12, and 24 months to be 42.8%, 51.0% and 57.3% respectively12,25. This poses significant clinical challenges as many patients with early-stage disease undergo frequent clinical monitoring during neoadjuvant therapy encompassing the first 5-6 months of treatment (152-182 days), with a subsequent drop-off in the frequency of visits in the post-operative period, when many patients are still receiving ICI. Thus, identification of these delayed immune toxicities can be challenging with fewer opportunities for patients to communicate with providers about perplexing and often difficult-to-diagnose new symptoms. This is particularly true in the case of endocrinopathies, where symptoms are often subtle and insidious but can lead to severe complications without prompt treatment26. Our analysis highlights the need for heightened clinician and patient awareness about the incidence of delayed toxicity to improve communication, monitoring, and timely identification and treatment of irAEs.

The most frequent delayed toxicities were thyroiditis, dermatitis, colitis, and AI, both overall and in the on-treatment and off-treatment subgroups. Excluding AI, this overlaps with the most common types of early toxicity, which were dermatitis, colitis, and thyroiditis. One reason that AI may have been seen more commonly in the delayed toxicity group is that diagnosis can be difficult, prone to late identification in real-world clinical practice; however, late onset of AI may be partly explained by the use of PD(L)1 inhibition in breast cancer and lack of CTLA4 inhibition, which is associated with earlier onset of AI27. Of note, the delayed time frame for AI is consistent with findings in non-breast tumor types, where the median time to onset for hypophysitis is ~27 weeks (190 days)28. Of note, 91.7% of patients with adrenal insufficiency had early-stage disease, meaning that a large portion of the patients who develop AI require life-long adrenal hormone replacement. It should also be noted that 9 patients developed ICI-induced diabetes mellitus (all grade 3 or higher), another disorder typically requiring life-long therapy. These serious endocrinopathies add a burden to survivorship as well as the risk of possible life-threatening complications.

Patients with metastatic disease were less likely to develop delayed toxicity compared to those with early-stage disease. This is consistent with prior analyses showing a lower rate of immune toxicity in general in those with metastatic disease compared to early-stage disease in breast cancer5,6. This may be due to shorter exposure to therapy, but another explanation is the downregulation of the immune landscape in metastatic disease. Metastatic tumors are inherently more immunologically inert, as shown by prior RNA analysis of both metastatic and primary breast tissue29,30, leading to reduced immune activation towards tumor antigens and less potential for cross-reactivity to normal tissue antigens. In addition, patients with metastatic disease are relatively immunologically suppressed due to both disease and treatment. For example, in patients with localized melanoma, baseline autoantibody levels were correlated with a higher risk of irAE development31, whereas in stage 4 metastatic melanoma, this was not true32, potentially pointing to an increased ability in early-stage disease for the expansion of pre-existing immune responses.

In our analysis, we found that an increasing number of ICI cycles was predictive of developing delayed toxicity in patients with both early and late-stage disease, and in particular, the risk significantly increased after four cycles. Prior studies in non-breast tumor types have also noted a median cumulative dose of four cycles in patients developing irAE25, indicating that four cycles may be an important clinical cut-off with regard to irAE risk. Currently, the minimum number of ICI cycles to achieve a therapeutic response is not well understood, though the current standard of care is to provide one year of therapy in the early-stage setting based on prior phase III trials5. However, analysis of patients with melanoma has shown immune activation even within one week of the first cycle of ICI33. Of note, the ISPY 2 trial in patients with high-risk early-stage breast cancer treated patients with four cycles of neoadjuvant pembrolizumab combined with chemotherapy (and no adjuvant pembrolizumab) and reported a pathologic complete response rate of 60% in those with TNBC, compared to a rate of 64.8% in the Keynote 522 trial in which eight cycles were given neoadjuvantly, followed by adjuvant therapy to complete one year34. Similarly, in GeparNuevo, in which five cycles of durvalumab (anti-PD1) was administered neoadjuvantly (and not adjuvantly), pCR rates were 53.4% (95% CI 42.5% to 61.4%)35. Thus, there may be a subset of patients where a shorter duration of ICI therapy is adequate to improve outcome. Taken together, this data provides further support for de-escalation trials of immunotherapy to see if fewer cycles may be as efficacious with less toxicity risk in excellent responders. One such trial, OptimICE-PCR (NCT05812807), is evaluating the omission of pembrolizumab in the adjuvant setting in patients who achieved pathologic complete response to neoadjuvant chemo-immunotherapy.

A higher absolute eosinophil count was predictive of developing delayed toxicity. This finding is supported by another retrospective study in non-breast tumor types where eosinophilia was predictive of irAE development, notably endocrine irAEs (p = 0.0287)36. Prior work has shown a link between eosinophils and the development of multiple types of autoimmune disease37–39, with proposed mechanisms being antibody-mediated cellular toxicity by killing host cells bound by autoantibodies40,41, direct antigen presentation to T cells42, and release of pro-inflammatory cytokines43,44. One prior real-world analysis demonstrated an association of developing irAE, of which eosinophilia was one of the irAE seen, and achievement of pathologic complete response in patients with breast cancer undergoing ICI8. These data indicate that eosinophils may play an important role in both anti-tumoral response as well as immune toxicity after immune checkpoint inhibitor treatment.

Development of early toxicity was negatively correlated with delayed toxicity, even after adjusting for those who stopped treatment early. This may be explained by confounding due to delays in diagnosis, causing toxicities to be categorized as delayed when they were, in fact, early in onset. Another possibility is differential mechanisms of toxicity in early vs delayed toxicity such as the existence of underlying auto-antibodies in those developing early toxicity, which has previously been implicated in development of severe toxicities and endocrine toxicities in particular31,45,46 Of note, eight patients who developed early toxicity and stopped treatment prior to 90 days still went on to develop off-treatment delayed toxicity, indicating that for a minority of patients, early toxicity is not protective for developing delayed toxicity. Future prospective studies with analysis of immune markers are needed to better understand the differential mechanisms between early and delayed toxicity.

Limitations of this study include that it is a retrospective analysis, and therefore, identification and grading of irAE depended on the treating physicians’ diagnosis and recognition of toxicity events related to ICI treatment. This could have led to both over- and under-diagnosis of true events and may have been prone to provider-dependent biases with regard to diagnosis. Furthermore, there may have been differences between treatment institutions with regard to treatment patterns, toxicity recognition, and documentation, as suggested by the fact that there were significant differences between treatment sites for which our analyses were adjusted (though this could also be related to differences in the number of menopausal women between sites). The real-world nature of the analysis also means that there may have been delays in toxicity recognition, impacting the median time of toxicity onset, given that follow-up intervals are not standardized outside of the trial setting. In cases where grade was not specifically documented, we determined grade by clinical description, which introduces additional variability. In addition, the reasons why ICI was discontinued or if there were treatment holds or delays were not always clear, thus this variable was not collected in our analysis. There is more heterogeneity in a real-world analysis, and patients in this cohort received a variety of different treatments, including different ICI agents and different ICI therapy partners, which were not always comprehensively reported, making it difficult to tease out the effects of specific treatments. Furthermore, some patients in the metastatic setting received subsequent treatment, which may have impacted delayed ICI toxicity. Finally, this analysis focuses on clinical factors and does not incorporate biomarker analysis.

This analysis found a high rate of irAEs, both early and delayed, in a large cohort of patients with breast cancer and identified risk factors for delayed toxicity, including early-stage disease, longer ICI exposure and higher baseline absolute eosinophil count. These results highlight the need for better awareness and understanding of delayed toxicities, particularly for patients with early-stage disease in the post-operative period when the frequency of visits typically decreases. They also emphasize the importance of ICI de-escalation trials, particularly for those patients with robust, early responses to therapy. Further prospective studies incorporating correlative analyses are needed to better understand timing, incidence, and predictive clinical risk factors/biomarkers, and biologic underpinnings of immune toxicity after immune checkpoint inhibitor.

Methods

This was a retrospective cohort study of patients across four different NCI-designated comprehensive cancer centers. This study was performed in accordance with the Declaration of Helsinki and approved by the institutional review board (IRB) at each site (University of California, San Francisco IRB #17-22987, University of Minnesota IRB #STUDY00004918, City of Hope IRB #22702, University of Californi,a Los Angeles IRB# 24-0388). Given that this study was a retrospective chart review with no anticipated risks to participants, the need for informed consent was waived by each institutional IRB. Patients were identified through pharmacy administration records and institutional database queries. Eligible patients had histologically proven breast cancer and underwent treatment with ICI between 2014 and 2021. All breast cancer subtypes and stages were included. Recipients of any type of ICI were allowed, including anti-PD1, anti-PDL1, and anti-CTLA4 agents. A clearly documented start and stop date of ICI treatment was required. Patients could undergo any type of combination treatment with ICI, including any chemotherapy or targeted therapy partners. Patients on clinical trials were included.

The charts of all patients meeting the inclusion criteria were manually reviewed with data stored in a secure REDCap database. Charts were reviewed by 1-3 individuals per institution who were provided instructions about collecting variables of interest. During the chart review process, the most recent oncology progress note was reviewed for documentation of immune toxicity to ensure maximal follow-up after ICI administration. If immune toxicity was not documented in this note, all oncology provider notes during the treatment period and for one year following the end of treatment were reviewed for documentation of irAEs. Finally, if no toxicity was found after manual review of progress notes, charts were also searched for each of the following terms: “immune related toxicity,” “pembrolizumab,” “rash,” “irAE,” “thyroid,” “adrenal insufficiency,” “diarrhea,” “colitis,” “dexamethasone,” “prednisone,” “hydrocortisone.”

Baseline clinical and demographic data were collected, including age at start of ICI, gender, race, ethnicity, menopausal status, body mass index, stage at ICI treatment start, breast cancer subtype, smoking history, baseline CBC with differential, baseline albumin and baseline TSH. Baseline co-morbidity data were collected including incidence of underlying autoimmune disease, type 2 diabetes mellitus, hypertension, hyperlipidemia, chronic kidney disease, coronary artery disease, and pre-existing lung disease. Follow-up data, including date of death or last encounter and length of follow-up since ICI start, as well as treatment, including ICI type, number of cycles, and combination agents were recorded.

Patient charts were reviewed for incidence of irAE, defined as any toxicity determined by a treating oncology provider as related to ICI treatment, or toxicity treated with high-dose steroids where the attribution was not clearly identified. Toxicities that could represent immune toxicities but were not specifically documented as being related to ICI or were not treated with steroids were not counted as irAE. Date of onset and grade of irAE were recorded; grade was either determined from documentation by the treating provider or, in cases where a grade was not documented, was based on clinical reports using CTCAE v547. Treatment of irAE was recorded, including the use of topical/oral steroids as well as immune-modulating agents.

All variables were collected at each institution and then de-identified prior to data sharing between institutions. IRB approval was obtained at each institution prior to data extraction.

Classification of toxicity by time to onset

Toxicities were categorized based on time from ICI initiation. Early toxicities were defined as occurring within 90 days of ICI start, and delayed toxicity as occurring at least 90 days after ICI start (Fig. 1). The cut-off of 90 days was selected based on prior definitions of delayed toxicities in patients with melanoma25. Delayed toxicities were also categorized based on whether they occurred on or off treatment. Off-treatment delayed toxicities were defined as toxicities occurring >21 days after cessation of ICI.

Outcome analysis

The primary outcome of this analysis was the incidence of delayed toxicities. Secondary outcomes included the incidence of irAE at any time point, incidence of early toxicity, incidence of on-treatment delayed toxicity, off-treatment delayed toxicity, median time to onset of delayed toxicities, and types of irAEs. Risk factors for delayed toxicity, including baseline labs, comorbidities, and demographic factors were assessed.

Statistical analysis

Descriptive analysis was used to determine the incidence of irAEs, types of irAEs, and time to onset of irAEs. To determine risk factors of delayed toxicity, a multivariable logistic regression was performed to determine the odds of developing delayed toxicities among the survivors. This logistic regression excluded 16 patients who were deceased prior to 90 days follow-up. All analyses were adjusted for early vs metastatic disease and treatment site. Variables included in the model were chosen based on characteristics associated with risk of irAE in non-breast tumor types, including age at start of ICI treatment48, baseline BMI48, breast cancer subtype, baseline eosinophil count49, baseline neutrophil count50, and number of cycles of immunotherapy51 in both the early-stage and metastatic setting. With regard to determining a cut-off number of cycles at which there was increased risk of developing delayed toxicities, we split samples to identify optimal cutoff points. This was done separately for patients with early-stage and metastatic disease. Optimal cutoff points were defined by the Akaike information criterion (AICs)52. Data were split differently 100 times to use one of each of the split samples to find the optimal cut-off for increased risks separately for early-stage and metastatic disease. The latter of the split samples was used to refit the model with the dichotomized high versus low cycles using the cutoff points found at high frequencies. Rubin’s formula53, which is used to summarize multiple imputation analysis results, was used to summarize the results from this refitting analysis repeated over 100 split samples in order to properly account for randomness both within and between multiple split samples. All associations were tested for significance using an alpha level of 0.05. Statistical analysis was performed using R54.

Supplementary information

Supplementary information (168.4KB, pdf)

Acknowledgements

S.J., S.F., C.F., K.B., M.C., S.C., A.L., N.B., A.S., L.Q., L.A.H., Z.Q., D.I., M.K., J.M., K.M., and A.B. have no relevant disclosures. M.M. reports institutional research funding from Pfizer, K.C.R.N. Research, Puma, and OBI Pharma. M.M. also reports COI for spouse who has stock ownership in Merrimack and speaker bureau/honoraria for Gilead, AstraZeneca/Daichi-Sankyo. A.J.C. reports institutional research funding from Merck, Pfizer, Puma, Seagen, Amgen, Olema, and advisory board participation for Genentech, Ellipses. HSR reports institutional research support (to UCSF) from AstraZeneca, Daiichi Sankyo, Inc., F. Hoffmann-La Roche AG/Genentech, Inc., Gilead Sciences, Inc., Lilly, Merck & Co., Inc., Novartis Pharmaceuticals Corporation, Pfizer, Stemline Therapeutics, OBI Pharma, Ambryx, as well as consultancy/advisory to Chugai, Sanofi, Napo.

Author contributions

S.J. conceived of the study design, performed primary data collection, performed data analysis, created all figures, and wrote the main manuscript text. S.M., C.F., M.C., S.C., A.L., N.B., A.S., L.H., L.Q. contributed to primary data extraction. KB and MK performed data analysis and assisted in figure creation. HSR was responsible for overall project supervision, study design, figure review, and manuscript review. All authors, including Z.Q., D.I., M.M., A.J.C., J.M., K.M., and A.B. reviewed the manuscript.

Data availability

Data will be available upon reasonable request of the corresponding authors. Data will not be available in a publicly accessible repository, given the clinical nature.

Competing interests

The authors declare no competing interests.

Footnotes

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Contributor Information

Saya Jacob, Email: saya.jacob@pennmedicine.upenn.edu.

Hope S. Rugo, Email: hrugo@coh.org, Email: hope.rugo@ucsf.edu

Supplementary information

The online version contains supplementary material available at 10.1038/s41523-025-00849-1.

References

  • 1.Larkin, J. et al. Five-year survival with combined Nivolumab and Ipilimumab in advanced melanoma. N. Engl. J. Med.381, 1535–1546 (2019). [DOI] [PubMed] [Google Scholar]
  • 2.Gandhi, L. et al. Pembrolizumab plus chemotherapy in metastatic non–small-cell lung cancer. N. Engl. J. Med.378, 2078–2092 (2018). [DOI] [PubMed] [Google Scholar]
  • 3.Motzer, R. et al. Lenvatinib plus Pembrolizumab or Everolimus for advanced renal cell carcinoma. N. Engl. J. Med.384, 1289–1300 (2021). [DOI] [PubMed] [Google Scholar]
  • 4.Schmid, P. et al. Overall survival with Pembrolizumab in Early-Stage Triple-Negative Breast Cancer. N. Engl. J. Med.391, 1981–1991 (2024). [DOI] [PubMed] [Google Scholar]
  • 5.Schmid, P. et al. Pembrolizumab for early triple-negative breast cancer. N. Engl. J. Med.382, 810–821 (2020). [DOI] [PubMed] [Google Scholar]
  • 6.Cortes, J. et al. Pembrolizumab plus chemotherapy in advanced triple-negative breast cancer. N. Engl. J. Med387, 217–226 (2022). [DOI] [PubMed] [Google Scholar]
  • 7.Schmid, P. et al. Event-free survival with Pembrolizumab in early triple-negative breast cancer. N. Engl. J. Med.386, 556–567 (2022). [DOI] [PubMed] [Google Scholar]
  • 8.Marhold, M. et al. Emergence of immune-related adverse events correlates with pathological complete response in patients receiving pembrolizumab for early triple-negative breast cancer. Oncoimmunology12, 2275846 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Zou, Y. et al. Efficacy and predictive factors of immune checkpoint inhibitors in metastatic breast cancer: a systematic review and meta-analysis. Ther. Adv. Med Oncol.12, 1758835920940928 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Gumusay, O., Callan, J. & Rugo, H. S. Immunotherapy toxicity: identification and management. Breast Cancer Res. Treat.192, 1–17 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Castillero, F. et al. Cancer immunotherapy-associated hypophysitis. Future Oncol.15, 3159–3169 (2019). [DOI] [PubMed] [Google Scholar]
  • 12.Ghisoni, E. et al. Late-onset and long-lasting immune-related adverse events from immune checkpoint-inhibitors: An overlooked aspect in immunotherapy. Eur. J. Cancer149, 153–164 (2021). [DOI] [PubMed] [Google Scholar]
  • 13.Ascierto, P. A. et al. Adjuvant nivolumab versus ipilimumab in resected stage IIIB-C and stage IV melanoma (CheckMate 238): 4-year results from a multicentre, double-blind, randomised, controlled, phase 3 trial. Lancet Oncol.21, 1465–1477 (2020). [DOI] [PubMed] [Google Scholar]
  • 14.Owen, C. N. et al. Delayed immune-related adverse events with anti-PD-1-based immunotherapy in melanoma. Ann. Oncol.32, 917–925 (2021). [DOI] [PubMed] [Google Scholar]
  • 15.Tang, S. Q. et al. The pattern of time to onset and resolution of immune-related adverse events caused by immune checkpoint inhibitors in cancer: a pooled analysis of 23 clinical trials and 8,436 patients. Cancer Res Treat.53, 339–354 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Schneider, B. J. et al. Management of immune-related adverse events in patients treated with immune checkpoint inhibitor therapy: ASCO Guideline Update. J. Clin. Oncol.39, 4073–4126 (2021). [DOI] [PubMed] [Google Scholar]
  • 17.Gohel, S. et al. The immune system’s echo: the phenomenon of delayed Neurological Immune-Related Adverse Events (NIRAEs) in adjuvant immunotherapy. Cureus15, e44529 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Wu, Y. et al. Late‑onset immune checkpoint inhibitor‑related pneumonitis after cessation of sintilimab: A case report and literature review. Exp. Ther. Med.25, 83 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Krishnan, J. et al. Real-world efficacy and adverse events of neoadjuvant immunotherapy in patients with early-stage triple-negative breast cancer: A multi-center experience. J. Clin. Oncol.42, e23293–e23293 (2024). [Google Scholar]
  • 20.Kanjanapan, Y. et al. Delayed immune-related adverse events in assessment for dose-limiting toxicity in early phase immunotherapy trials. Eur. J. Cancer107, 1–7 (2019). [DOI] [PubMed] [Google Scholar]
  • 21.Brahmer, J. R. et al. Phase I study of single-agent anti-programmed death-1 (MDX-1106) in refractory solid tumors: safety, clinical activity, pharmacodynamics, and immunologic correlates. J. Clin. Oncol.28, 3167–75 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Hurkmans, D. P. et al. A prospective cohort study on the pharmacokinetics of nivolumab in metastatic non-small cell lung cancer, melanoma, and renal cell cancer patients. J. Immunother. Cancer7, 192 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Dang, T. O. et al. Pembrolizumab for the treatment of PD-L1 positive advanced or metastatic non-small cell lung cancer. Expert Rev. Anticancer Ther.16, 13–20 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Laparra, A. et al. Multiple immune-related toxicities in cancer patients treated with anti-programmed cell death protein 1 immunotherapies: a new surrogate marker for clinical trials?. Ann. Oncol.32, 936–937 (2021). [DOI] [PubMed] [Google Scholar]
  • 25.Couey, M. A. et al. Delayed immune-related events (DIRE) after discontinuation of immunotherapy: diagnostic hazard of autoimmunity at a distance. J. Immunother. Cancer7, 165 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Sznol, M. et al. Endocrine-related adverse events associated with immune checkpoint blockade and expert insights on their management. Cancer Treat. Rev.58, 70–76 (2017). [DOI] [PubMed] [Google Scholar]
  • 27.Gubbi, S. et al. Hypophysitis: An update on the novel forms, diagnosis and management of disorders of pituitary inflammation. Best. Pr. Res Clin. Endocrinol. Metab.33, 101371 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Di Dalmazi, G. et al. Hypophysitis induced by immune checkpoint inhibitors: a 10-year assessment. Expert Rev. Endocrinol. Metab.14, 381–398 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Szekely, B. et al. Immunological differences between primary and metastatic breast cancer. Ann. Oncol.29, 2232–2239 (2018). [DOI] [PubMed] [Google Scholar]
  • 30.Garcia-Recio, S. et al. Multiomics in primary and metastatic breast tumors from the AURORA US network finds microenvironment and epigenetic drivers of metastasis. Nat. Cancer4, 128–147 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Johannet, P. et al. Baseline serum autoantibody signatures predict recurrence and toxicity in melanoma patients receiving adjuvant immune checkpoint blockade. Clin. Cancer Res28, 4121–4130 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Ghosh, N. et al. Lower baseline autoantibody levels are associated with immune-related adverse events from immune checkpoint inhibition. J. Immunother. Cancer10 (2022). [DOI] [PMC free article] [PubMed]
  • 33.Anderson, T. M. et al. FDG PET/CT Imaging 1 week after a single dose of Pembrolizumab predicts treatment response in patients with advanced melanoma. Clin. Cancer Res30, 1758–1767 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Nanda, R. et al. Effect of Pembrolizumab Plus Neoadjuvant chemotherapy on pathologic complete response in women with early-stage breast cancer: an analysis of the ongoing Phase 2 adaptively randomized I-SPY2 Trial. JAMA Oncol.6, 676–684 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Loibl, S. et al. A randomised phase II study investigating durvalumab in addition to an anthracycline taxane-based neoadjuvant therapy in early triple-negative breast cancer: clinical results and biomarker analysis of GeparNuevo study. Ann. Oncol.30, 1279–1288 (2019). [DOI] [PubMed] [Google Scholar]
  • 36.Kizawa, R. et al. Eosinophilia during treatment of immune checkpoint inhibitors (ICIs) to predict succeeding onset of immune-related adverse events (irAEs). J. Clin. Oncol.37, e14110–e14110 (2019). [Google Scholar]
  • 37.Truelove, S. C. & Richards, W. C. Biopsy studies in ulcerative colitis. Br. Med J.1, 1315–8 (1956). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Tsukadaira, A. et al. Eosinophil active cytokines and surface analysis of eosinophils in Churg-Strauss syndrome. Allergy Asthma Proc.20, 39–44 (1999). [DOI] [PubMed] [Google Scholar]
  • 39.Corradi, D. et al. Eosinophilic myocarditis in a patient with idiopathic hypereosinophilic syndrome: insights into mechanisms of myocardial cell death. Hum. Pathol.35, 1160–3 (2004). [DOI] [PubMed] [Google Scholar]
  • 40.Diny, N. L., Rose, N. R. & Čiháková, D. Eosinophils in autoimmune diseases. Front Immunol.8, 484 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Porrillo, J. E. & Fauci, A. S. Human Eosinophils. purification and cytotoxic capability of eosinophils from patients with the Hypereosinophilic Syndrome. Blood51, 457–473 (1978). [PubMed] [Google Scholar]
  • 42.Pozo, V. et al. Eosinophil as antigen-presenting cell: activation of T cell clones and T cell hybridoma by eosinophils after antigen processing. Eur. J. Immunol.22, 1919–25 (1992). [DOI] [PubMed] [Google Scholar]
  • 43.Rosenberg, H. F., Dyer, K. D. & Foster, P. S. Eosinophils: changing perspectives in health and disease. Nat. Rev. Immunol.13, 9–22 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Jacobsen, E. A. et al. Differential activation of airway eosinophils induces IL-13-mediated allergic Th2 pulmonary responses in mice. Allergy70, 1148–59 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.Daban, A. et al. Preexisting autoantibodies as predictor of immune related adverse events (irAEs) for advanced solid tumors treated with immune checkpoint inhibitors (ICIs). OncoImmunology12, 2204754 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Labadzhyan, A. et al. Endocrine autoantibodies determine immune checkpoint inhibitor-induced endocrinopathy: a prospective study. J. Clin. Endocrinol. Metab.107, 1976–1982 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Services, U.S.D.o.H.a.H. Common Terminology Criteria for Adverse Events (CTCAE) Version 5.0. 2017; Available from: https://ctep.cancer.gov/protocolDevelopment/electronic_applications/docs/CTCAE_v5_Quick_Reference_5x7.pdf.
  • 48.Samani, A. et al. Impact of age on the toxicity of immune checkpoint inhibition. J. Immunother. Cancer8 (2020). [DOI] [PMC free article] [PubMed]
  • 49.Chu, X. et al. Association of baseline peripheral-blood eosinophil count with immune checkpoint inhibitor-related pneumonitis and clinical outcomes in patients with non-small cell lung cancer receiving immune checkpoint inhibitors. Lung Cancer150, 76–82 (2020). [DOI] [PubMed] [Google Scholar]
  • 50.Fujimoto, A. et al. Risk factors indicating immune-related adverse events with combination chemotherapy with immune checkpoint inhibitors and platinum agents in patients with non-small cell lung cancer: a multicenter retrospective study. Cancer Immunol. Immunother.72, 2169–2178 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51.Bazarbachi, A.-H. et al. Evidence for a reduction in number of cycles of immune checkpoint inhibitors. Lancet Oncol.26, 9–11 (2025). [DOI] [PubMed] [Google Scholar]
  • 52.Portet, S. A primer on model selection using the Akaike Information Criterion. Infect. Dis. Model5, 111–128 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53.Roderick Little, D. R. Statistical Analysis with Missing Data, Third Edition. 1987, New York: Wiley and Sons, Inc.
  • 54.Foundation, T. R. The R Project for Statistical Computing. [cited 2025 March 1]; Available from: https://www.r-project.org/.

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary information (168.4KB, pdf)

Data Availability Statement

Data will be available upon reasonable request of the corresponding authors. Data will not be available in a publicly accessible repository, given the clinical nature.

Articles from NPJ Breast Cancer are provided here courtesy of Nature Publishing Group

RESOURCES