CIP is the most common fatal adverse reaction to PD-1/PD-L1 inhibitors (7). Compared with patients with other cancers, patients with lung cancer have a greater incidence and severity of CIP (8). Compared with monotherapy, combination immunotherapy results in more frequent CIP (9, 10). Clinical studies have reported a 3%-5% overall incidence of CIP with 1% grade≥3 cases (9, 11, 12), whereas real-world data indicate higher rates of 13%-19% (13–15).

Moey et al. (16) conducted an analysis of the WHO drug safety database and reported that the mortality rate associated with CIP was significantly higher than that reported in clinical trials, with 20.4% in lung cancer patients, potentially due to diagnostic challenges distinguishing CIP from tumor progression or infections (17). CIP may occur anytime post-ICI therapy, with a median onset of 2.8 months (range: 9 days–19.2 months) (18), rarely extending to 36 months (19). Onset occurs earlier in non-small cell lung cancer (NSCLC) patients (1.1 vs 3.1 months). Severe CIP is associated with a 61.5% mortality rate, with a median survival of 104 days (20). Consequently, for patients with a history of ICI treatment, CIP should be prioritized as a differential diagnosis if they exhibit respiratory-related symptoms or abnormal chest imaging findings.

Identifying the risk factors associated with CIP is crucial for facilitating early diagnosis and effective management of high-risk patients. Previous retrospective studies have reported that a high incidence of immune checkpoint inhibitor-related pneumonia (CIP) is associated with various antitumor treatments, tumor pathological types, chronic underlying lung diseases, age, smoking history, radiotherapy history, levels of certain inflammatory factors, and genetic susceptibility ( Figure 1 ).

PD-1/PD-L1 inhibitors can augment the antitumor efficacy of T cells (47, 48). Accumulating evidence indicates that the upregulation of T cells may play a role in the pathogenesis of CIP (4, 49). The increased activity of these targeted T cells can lead to the attack of cross-reactive antigens shared between tumors and normal lung tissue, resulting in off-target toxicity.

Hiroyuki and colleagues reported the presence of T-cell-predominant lymphocytosis in the BALF of patients with CIP (50). Notably, the proportions of CD8⁺ T cells expressing immune checkpoint proteins, a hallmark of tumor-infiltrating T lymphocytes (TILs), are greater in patients with CIP than in those with other immune-related adverse events (irAEs) (51–53). Naidoo et al. proposed increased infiltration of highly proliferative CD8⁺ T cells in the lung biopsy tissue of NSCLC patients who developed chronic bronchiolitis obliterans organizing pneumonia subsequent to nivolumab therapy (54). Subudhi et al. demonstrated a significant correlation between the abundance of CD8⁺ T-cell clones in the peripheral blood and the incidence of irAEs (P=0.01), particularly grade 2–3 irAEs (P<0.0001), among patients treated with ipilimumab, a CTLA-4 inhibitor (55, 56). Suresh and colleagues reported a significant increase in CD4⁺ T cells in the BAL samples of CIP patients, primarily those with NSCLC, following treatment with anti-PD-1/PD-L1 inhibitors (57). These findings support a lymphocyte-driven hyperimmune response.

In addition, the expression levels of CTLA-4 and PD-1 on Treg cells in CIP patients were significantly decreased, suggesting an attenuated Treg suppressive phenotype (57, 58). When regulated by ICIs, CTLA-4, a key molecule in Treg cell function, can alter the suppressive capabilities of Treg cells within the tumor immune microenvironment. This alteration may lead to the development of CIP by relieving the immunosuppressive effects of Treg cells (2, 59). Moreover, deficiency of CTLA-4, in addition to PD-1, may further compromise the ability of Treg cells to control the proinflammatory responses of conventional T cells and macrophages, thereby exacerbating the unchecked immune dysregulation observed in CIP (57).

T-cell activation and infiltration within the lung tissue of CIP patients signify enhanced antitumor immunity. However, this heightened immune activity poses a risk of normal tissue damage due to overly aggressive immune reactions (60, 61), which support the hypothesis that ICI-related pneumonitis may be induced when ICI-activated T cells recognize self-peptides or epitopes that are shared between the tumor and the host (62). This hyperresponsive immune state, triggered by T-cell activation and infiltration, may lead to the misidentification of self-antigens in lung tissue, thereby triggering an autoimmune reaction (63).

An increasing number of studies indicate that the occurrence of CIP may be linked to elevated levels of both preexisting and newly emerging autoantibodies within the human immune system (58, 64). These autoantibodies, which may be present at low concentrations before the initiation of ICI therapy or generated de novo during treatment, are believed to contribute to the pathogenesis of certain nonpulmonary irAEs (58).

PD-1/PD-L1-targeted therapy has been shown to induce Treg dysfunction and foster the generation of pathological autoantibodies, as evidenced in both PD-1-knockout mice and patients undergoing such treatment (65, 66). Adverse effects such as pneumonitis, arthralgia, vitiligo, and hypothyroidism are frequently observed in patients receiving PD-1/PD-L1 inhibitor treatment (67, 68). Additionally, Toi et al. demonstrated that NSCLC patients have preexisting autoantibodies, such as rheumatoid factor (69).

Klocke et al. investigated the role of CTLA-4 by ablating CTLA-4 expression in adult mice and compared the resulting autoimmunity with that observed in congenital CTLA-4 deficiency. They identified autoimmune disease phenotypes, including pneumonitis, accompanied by organ-specific autoantibodies (70). In a retrospective study of 22 patients with CTLA-4 insufficiency and COVID-19, Ochoa et al. examined autoantibodies against type 1 IFNs at baseline. They reported that most patients with CTLA-4 insufficiency and COVID-19 had nonsevere disease and lacked autoantibodies against type 1 Interferons (71). These findings suggest a potential explanation for the risk of developing autoimmune complications in cancer patients during treatment with the CTLA4-blocking checkpoint inhibitor ipilimumab (72).

Emerging evidence implicates multiple distinct autoantibodies in the development of CIP. Notably, CD74, an autoantibody-active protein, functions as an intracellular chaperone for major histocompatibility complex class II (MHC-II) and can stimulate the release of inflammatory mediators. Although primarily intracellular, CD74 is also expressed on the cell membrane of immune cells, including macrophages (73). It serves as a high-affinity receptor for macrophage inhibitory factors, thereby inducing inflammatory mediator release and cell proliferation (74, 75). In normal human lung tissue, CD74 is expressed at modest levels; however, its expression is dramatically increased in lung tissues affected by ICI-induced pneumonitis (76). A large-scale screening study of ICI-treated patients revealed that serum levels of anti-CD74 antibodies were elevated in patients with CIP compared with their pretreatment levels (77). Salahaldin A. Tahir et al. conducted a high-throughput serological analysis of recombinant cDNA expression to investigate autoantibodies in patients receiving ICIs. They reported a significant median 1.34-fold increase in anti-CD74 antibody levels post-ICI treatment in patients with CIP, whereas no significant changes were detected in a comparison group of 20 patients without pneumonitis, suggesting a pathogenic role for CD74 autoantibodies in the development of pneumonitis (78). These findings indicate the pathogenic involvement of CD74 and its autoantibodies in the development of CIP.

The administration of ICIs can activate T cells, resulting in excessive release of cytokines and potent proinflammatory responses, which in turn promote the development of various adverse events, including CIP. Elevated levels of various cytokines have been observed in patients experiencing irAEs following ICI treatment, highlighting significant changes in cytokine profiles (79). Moreover, certain cytokines have shown promise as predictive biomarkers for irAEs, offering promising avenues for early detection and management strategies.

Tumor necrosis factor-α (TNF-α), interleukin-6 (IL-6) and interferon-γ (IFN-γ) collectively contribute to tissue damage and exacerbate the immune-mediated lung injury characteristic of CIP (63, 80, 81). Dysregulated cytokine production disrupts immune homeostasis by initiating inflammatory cascades, which recruit neutrophils and macrophages to injury sites and suppress anti-inflammatory counterregulation, ultimately worsening CIP severity (82–84). Mechanistically, Lin et al. demonstrated that the infiltration of Th2 cells in the BALF of CIP patients drives the overproduction of interleukin clusters (IL-4, IL-5, IL-6, IL-9, IL-10, and IL-13), thereby creating a self-sustaining inflammatory loop (85).

Similarly, other studies have also identified specific cytokines associated with the development of irAEs, including pneumonitis. Lim’s team, who studied melanoma patients receiving ICIs, identified 11 plasma cytokines (including granulocyte–colony-stimulating factor [G-CSF], granulocyte–macrophage colony–stimulating factor [GM–CSF], and IL–13) that dynamically increase before or during the onset of high-grade irAEs (86). Notably, NSCLC studies have revealed both similarities and divergences in cytokine patterns: while elevations in IL-1β and IL-8 align with findings in melanoma, decreased G-CSF has emerged as a unique predictor in lung cancer cohorts (87). This discrepancy underscores microenvironment-driven variations in the cytokine network. Furthermore, a study of 204 NSCLC patients with or without PD-L1 inhibitor monotherapy (43 of whom developed irAEs) revealed a proinflammatory increase in IL-1β and elevated levels of IL-5, IL-8, IL-10, IL-12p70, and granzyme A, along with decreased G-CSF, as predictors of irAEs, including pneumonitis (86). Receptors for IL-8 are present on neutrophils, Tregs, monocytes, and NK cells, indicating their potential involvement in the complex biology of CIP (49, 88). However, the role of IL-5 and IL-10 in lung injury associated with pneumonitis remains uncertain.

IL-6 has garnered significant attention in CIP research because of its potential role in its pathogenesis. One study reported elevated IL-6 levels in CIP patients compared with those at baseline, whereas another examination of BALF cytokines in 12 CIP patients revealed significantly higher IL-6 levels than in controls (89, 90). However, it is important to note that IL-6 elevation is not universal in the BALF of CIP patients (57). Despite this variability, tocilizumab, an IL-6 inhibitor, has shown efficacy in treating steroid-refractory CIP in a single-center study (91). C-reactive protein (CRP), an acute-phase protein secreted by liver cells in response to inflammatory cytokines such as IL-6 and TNF-α, is also relevant. Given the observed elevated IL-6 levels in some CIP patients, it is unsurprising that CRP levels are increased in NSCLC patients who develop CIP postatezolizumab treatment compared with baseline (92, 93). These findings suggest that CIP development may be attributed to excessive immune system activation induced by both CRP and IL-6, along with their potent proinflammatory properties (81).

IL-17, a cytokine with diverse functions, plays crucial roles in autoimmune diseases and inflammation. Its abnormal expression has been implicated in the pathology of various lung diseases, including asthma, pneumonia, and pulmonary fibrosis (94). Lou et al. demonstrated a significant increase in serum IL-17 levels in NSCLC patients who developed CIP following ICI treatment (95). This phenomenon may be attributed to the disruption of immune tolerance and enhanced T-cell activation resulting from PD-1 and PD-L1 blockade (96, 97). A separate analysis of serum and BALF from 13 NSCLC patients with CIP post-PD-1/PD-L1 therapy revealed elevated levels of both IL-17A and IL-35 in these compartments. Furthermore, serum IL-17A levels were found to be positively correlated with the Th17 cell subtype (98). Given that IL-17A is implicated in other autoimmune disorders, acute lung injury, and lung fibrosis, it is plausible that it also contributes to the pathogenesis of CIP (49, 99, 100).

C-X-C chemokines (CXCLs) are pivotal in regulating the differentiation of primary T cells into Th1 cells and facilitating the migration of immune cells to tumor sites through interactions with molecules such as CXCL9, CXCL10, CXCL11, and CXCL13 (101, 102). By assessing 40 cytokines in plasma, Shaheen Khan and colleagues reported significant upregulation of various cytokines, particularly CXCL9, 10, 11, and 13, following ICI treatment, which was closely linked to the onset of irAEs (103).

Despite the variability in underlying tumor histology, host factors, and disease profiles among patients, emerging evidence suggests that cytokine dysregulation is involved in the pathogenesis of CIP. This dysregulation leads to the release of proinflammatory cytokines and the infiltration of immune cells into the lung parenchyma, thereby exacerbating tissue damage and contributing to the development of pneumonitis (81).

Given the widespread belief that the development of irAEs is linked to autoimmunity, genetic variations have emerged as a potential contributing factor to this phenomenon. Notably, human leukocyte antigen (HLA) variations, which are crucial at the immune cell interface, have garnered significant attention in recent research. In a cohort of 256 patients undergoing PD-1/PD-L1 treatment, including 29 CIP patients, HLA typing revealed a strong correlation between the incidence of CIP and the germline expression of HLA-B allele 35 and HLA-DRB1 allele 11 (39). These alleles have also been implicated in other autoimmune disorders, further underscoring the potential significance of genetic factors in the pathogenesis of CIP (104). Furthermore, specific genetic polymorphisms have been associated with susceptibility to CIP. These include polymorphisms in genes related to immune regulation and inflammation, such as those encoding gamma-aminobutyric acid type A receptor subunit pi (GABRP), desmocollin, and bromodomain adjacent to zinc finger domain 2B (BRD2B). However, further research is necessary to elucidate their precise roles in the development of CIP (82, 105). Collectively, these findings suggest that genetic variations are likely to contribute to the dysregulation underlying CIP.

Hakozaki et al. highlighted disparities in the gut microbiome between advanced NSCLC patients experiencing low- and high-grade irAEs (106). This has prompted a growing focus on the ‘gut–lung axis’, a concept describing the interaction between the gut and respiratory microbiomes and its potential impact on immune responses (107). Although research on respiratory microbiomes in the context of cancer immunotherapy is relatively less extensive than that on gut microbiomes, studies have indicated that anti-PD-1 treatment can modulate the diversity and abundance of specific microbiome populations (108, 109). Notably, Zhang et al. discovered that certain respiratory microbes, particularly enriched Streptococcus, might potentiate antitumor immune responses by enhancing antigen presentation and effector T-cell function, thereby potentially increasing the incidence of irAEs, including CIP (110).

Hsu et al. demonstrated that PD-1-expressing NK cells can mediate immunosuppression in tumor microenvironments by interacting with PD-L1-positive tumor cells. PD-1/PD-L1 blockade can release the cytotoxic potential of NK cells against tumors by increasing PD-1-induced immune inhibition. However, when ICI treatment triggers inflammation in nonmalignant lung tissue via diverse non-NK cell-mediated pathways, activated NK cells can directly eliminate infected cells and secrete proinflammatory cytokines. Consequently, NK cell activation mediated by PD-1/PD-L1 blockade might exacerbate inflammation and contribute to damage in normal lung tissues (111).

The 2021 Fleischner guidelines categorized the CT imaging patterns associated with drug-induced pneumonia into five distinct types (115), four of which can be observed in patients receiving immunotherapy.

In addition to the classic manifestations of CIP, radiation recall pneumonitis (RRP) has attracted increasing interest from oncologists and radiation therapists in recent years. The incidence of RRP has increased since immunotherapy, which is used as a consolidation treatment following chemoradiotherapy for locally advanced NSCLC, was promoted clinically. Some studies report that the incidence of RRP can reach 18.8%, with a median onset time of 450 days post-radiotherapy (116). RRP is characterized by CT-detected pneumonia outside the typical radiation pneumonitis window (4–12 weeks post-RT) and is localized to the irradiated area ( Figures 4A, B ). Imaging features include patchy ground-glass opacity/consolidation (120). Pathologically, it is characterized by mucosal congestion, leukocyte infiltration, alveolitis, type II alveolar cell hyperplasia, and fibrosis (121). ICI-induced RRP results in radiation field-aligned consolidation/ground-glass opacity with clear lesion-normal tissue borders. ICI-induced RRP mechanisms may involve overactivation of T lymphocytes post-ICI binding to irradiated lung tissue, triggering hypersensitivity; radiation-induced endothelial damage, increasing vascular permeability and localized ICI accumulation in irradiated areas; and synergistic damage to alveolar epithelial cells caused by both therapies (122, 123).

Bronchiolitis has been reported exclusively in case studies. Histologically, bronchiolitis is obliterative and characterized by centripetal fibrosis of the submucosa and peribronchial tissues of the terminal and respiratory bronchioles (124). CT findings include thickened bronchial walls and diffuse tree-in-bud opacities in the peripheral lung zones, often accompanied by bronchiectasis or sharply demarcated mosaic attenuation patterns of the lung parenchyma, which are more pronounced during expiration (125). SLR represents a drug-induced multisystem granulomatous reaction that is histologically characterized by noncaseating epithelioid granulomas with multinucleated giant cells at the center and surrounding scattered lymphocytes (126). It occurs in approximately 5% to 7% of patients receiving nivolumab therapy (127). Typical HRCT features include miliary nodules predominantly distributed in the upper and middle lung zones along the lymphatic vessels, which are associated with mediastinal and bilateral hilar lymphadenopathy (128). Most SLR cases require no treatment owing to minimal symptoms. ICI-induced AEP is rare and presents as bilateral ground–glass opacities, consolidation, septal thickening, and pleural effusion on CT. Diagnosis requires elevated eosinophils in the blood and bronchoalveolar fluid. Monitoring blood eosinophils during ICI therapy aids early detection of CIP (129, 130).

Imaging exams are crucial for diagnosing CIP, with manifestations linked to disease progression, hormone therapy response, and prognosis. Studies have shown that organizing pneumonia (OP) is the most common CT pattern in CIP patients (114). Clinically, the severity of CIP follows the order of DAD/AIP > NSIP/HP > OP (131). Capaccione et al. (132) demonstrated through multimodal imaging analysis that the imaging characteristics of CIP patients are significantly associated with the type and duration of immunotherapy. Specifically, CIP linked to anti-PD-1/PD-L1 therapy predominantly exhibits an OP pattern, whereas CIP associated with anti-CTLA-4 therapy typically manifests as an NSIP pattern. Research indicates that individuals in CIP stages G1–2 have higher survival rates than those in G3–4. Hormone therapy benefits HP/OP-like patients, whereas DAD-like patterns show rapid progression and poor treatment response (33, 115).

Clinically, patients with CIP often exhibit no specific symptoms, and approximately one-third of them may remain asymptomatic. Therefore, differential diagnosis is essential to exclude other potential conditions, such as pulmonary infection, tumor pseudoprogression, radiation-induced lung injury, and pulmonary edema. Among them, opportunistic pulmonary infections, including tuberculosis (TB), aspergillosis, cytomegalovirus pneumonia (CMVP), and Pneumocystis jirovecii pneumonia (PJP), present significant diagnostic challenges and constitute important differential diagnoses (133–135). A meta-analysis revealed that patients treated with PD-1/PD-L1 inhibitors are not only at risk of developing various types of immune-related pneumonitis but also exhibit an increased incidence of infectious pneumonia compared with patients in the chemotherapy or placebo groups (136). The presence of symptoms such as fever, expectoration, and elevated blood counts may indicate infection in patients. Obstructive pneumonia represents a common type of pulmonary infection among lung cancer patients, with bacterial pathogens being the most prevalent cause. P. jirovecii infection may result in bilateral ground–glass opacities and hypoxemia, whereas viral infections can also cause diffuse pulmonary involvement. Furthermore, several case reports have documented fungal infections, fungal airway disease, and active pulmonary tuberculosis in patients receiving ICI therapy (137–139). Pulmonary infection and CIP may present with overlapping imaging features, making differential diagnosis challenging. Therefore, it is essential to integrate findings from sputum microbiology and serum biomarkers, including Legionella antibodies, Mycoplasma antibodies, (1, 3)-β-D-glucans, Aspergillus galactomannan, Cryptococcus capsular polysaccharide antigen, and interferon-γ release assays, to facilitate accurate differentiation (140). In patients who develop immune checkpoint inhibitor-associated pneumonitis (CIP), particularly those with grade 2 or higher CIP, bronchoscopy and/or bronchoalveolar lavage (BAL) are recommended as initial diagnostic procedures to exclude infectious etiologies. When concurrent infection is suspected, prompt initiation of empirical anti-infective therapy is also warranted (120). In cases where CIP and infectious pneumonia cannot be definitively differentiated, occur concurrently, or are followed by secondary infection, empirical antibiotic therapy should be initiated following appropriate diagnostic evaluation, while simultaneous efforts are made to identify the etiological agent. Moreover, throughout the management of CIP, continuous monitoring is needed for opportunistic infections that may arise as a consequence of immunosuppressive therapy (141).

In 2019, Colen et al. (146) pioneered a radiomic model for predicting CIP in 32 patients. Using 3D Slicer for CT image segmentation, they extracted 1,860 texture features on the basis of histograms and the GLCM. The mRMR method selected predictive features, achieving 100% accuracy (p = 0.0033). However, the small sample size limits the generalizability of the results. Building on Cole’s work, Zhang et al. (147) increased the robustness of the NSCLC prediction model by integrating greyscale-dependent matrix radiomic features. Features were selected via combined MRMR and RFE methods, with added model evaluation techniques. The ICC consistency assessment ensured reliability. Ultimately, the model constructed using the RS, clinical, and SF features demonstrated excellent performance in terms of ROC curves, calibration curves, and DCA. This advancement enables physicians to predict the likelihood of CIP in NSCLC patients with greater speed and accuracy.

Tan et al. (148) developed a multimodal deep learning model using a 3D ResNet18 architecture to predict chemotherapy-induced pneumonia in lung cancer patients. The model integrated nine clinical and radiomic features through 18 convolutional layers and fully connected layers. Compared with two-stage transfer learning with contrastive learning, it achieves an AUC of 0.918 via fivefold cross-validation. Similarly, Cheng et al. (149) developed a multimodal nomogram integrating clinical and deep imaging radiomics to predict CIP risk in 141 lung cancer patients. They optimized ResNet-50-V2 with a feature pyramid network (FPN), showing that the multimodal model outperformed the radiomic-only and clinical-only models, achieving AUCs of 0.910 vs 0.871 vs 0.778 (training set) and 0.900 vs 0.856 vs 0.869 (test set), respectively.

Chen et al. (150) constructed a radiomic prediction vector for lung cancer immunotherapy (LCI-RPV) on the basis of radiomic tumor features from enhanced chest CT scans, CD274 counts in NSCLC patients treated with anti-PD-1/PD-L1 therapy, and RNA expression levels of the PD-L1 protein-coding gene as the response variable. Subgroup analysis revealed that the LCI-RPV could predict CIP in PD-L1 inhibitor patients, with an AUC of 0.74 (95% CI: 0.53–0.95). Thomas et al. pioneered functional lung radiomics in locally advanced NSCLC, assessing its role in pneumonia risk stratification for combined radiation and ICI therapy. Baseline COPD significantly increased pneumonia risk (HR 4.59). COPD was the primary predictor, with a c-index of 0.69 (0.59–0.80) (151).

The incidence of RP and CIP is greater than that of other types of treatment-related pneumonitis in NSCLC patients. RP is dose dependent with radiation field confinement, whereas CIP presents bilateral multilobe involvement without dose correlation (120, 152, 153). Concurrent RT-ICI-induced pneumonia combines the features of CIP and RP (11, 154) rather than simple superposition. In the PEMBRO-RT trial and PACIFIC study, the incidence of pneumonia in NSCLC patients receiving radiotherapy (RT) and ICI was reported to be as high as 26% to 33.9% (95, 155). Differentiating CIP from RP is crucial for clinical management, guiding steroid dosage and ICI rechallenge or discontinuation (156).

Chen et al. (157) analyzed CT data from 82 NSCLC patients with pneumonia (immunotherapy/radiotherapy/combined therapy). They built a radiomic model using LASSO-selected features and grid search optimization, achieving a training set AUC of 0.76 for distinguishing ICI- and RT-induced pneumonia. Cheng et al. (158) analyzed CT images and clinical data from 73 patients with pneumonia related to ICIs, radiotherapy or combined therapy. Three CT texture features (intensity histogram, GLCM-based, and Bow) were extracted, with the Bow showing the best cross-validation performance (AUC 0.937). Logistic regression, random forest, and linear SVM models underwent 10-fold validation in ICI/radiotherapy-only patients. Testing in combined therapy patients yielded AUCs of 0.765, 0.848, and 0.937, respectively. The optimal model achieved an AUC of 0.896 in the combined treatment cohort, outperforming prior studies. QIU et al. (159) developed a CT radiomic model that integrates radiomic, clinical, and radiological factors to differentiate CIP from RP, with excellent performance (AUC 0.953 and 0.947). Bilateral CT changes were more likely in patients with ICI-induced pneumonia than in those with RP (p<0.001). CIP patients had less well-defined boundaries (p=0.001), which aligns with other studies (157). The radiomic nomogram outperformed the Rad-score in distinguishing CIP from RP.

Wang et al. (160) developed a CT-based radiomic model for RP and CIP patients. They built two models (random forest and linear discriminant classifiers), achieving test cohort AUCs of 0.851 and 0.842, respectively, with 83% accuracy. Additionally, this study identified four highly discriminative features, including one first-order feature, one gray-level run-length matrix (GLRLM)-based feature, one gray-level co-occurrence matrix (GLCM)-based feature, and one neighborhood gray-tone difference matrix (NGTDM)-based feature. Previous studies (161, 162) have shown that multifrequency CT decomposition effectively decodes the underlying phenotypic differences between CIP and RP in patients undergoing RT and ICI therapy.

The De Ruysscher team published a study in 2021 at the European Society for Medical Oncology (ESMO), which distinguished CIP from other causes of pneumonia (163). They extracted 837 radiomic features and used recursive feature elimination (RFE) to select the 42 most correlated features, and the areas under the ROC curves of the clinical, radiomic, and combined models for predicting ICI-R pneumonitis were 0.99, 0.65, and 0.99, respectively.

Early identification and diagnosis of CIP are crucial but complicated by its broad onset window and clinical and radiological heterogeneity. The imaging features of CIP frequently overlap with those of infectious pneumonia. Since the emergence of the COVID-19 pandemic, this diagnostic challenge has become even more significant. Sumeet Hindocha et al (164) developed a rigorously validated machine learning tool capable of differentiating CIP from radiation pneumonitis (RP), COVID-19, and other infectious pneumonia. Compared with radiologists, the models for distinguishing RP from COVID-19, CIP from COVID-19, and CIP from non-COVID-19 interstitial pneumonia (IP) demonstrated superior performance, as evidenced by test set AUCs of 0.92 versus 0.8 and 0.8; 0.68 versus 0.43 and 0.4; and 0.71 versus 0.55 and 0.63, respectively.

Recent research has expanded the application scope of radiomics in the treatment of CIP and improved its value in immunotherapy monitoring. The application of radiomics in the treatment of CIP has gradually expanded from initial risk stratification to automatic identification, differential diagnosis, and prognosis prediction.

For Grade 1 CIP, ICI discontinuation is advised with self-monitoring of symptoms and weekly physical examination and oxygen saturation monitoring during the first 3 weeks. If symptoms occur, chest CT should be implemented in advance. Without clinical progress, a follow-up chest CT after 3–4 weeks should be performed. If imaging shows improvement, resumption of ICI therapy with close follow-up is recommended. In cases of no radiographic improvement, treatment should be escalated to Grade 2, and ICI therapy should be paused.

The recommended approaches for Grade 2 CIP include baseline assessment, such as blood tests (complete blood count, liver and kidney function, electrolytes), and pulmonary function analysis. ICI therapy should be paused until the condition improves to Grade 1 or lower. Intravenous methylprednisolone at 1–2 mg/(kg·d) is advised for 48–72 hours. If symptoms improve, the steroid dose should be tapered over 4–6 weeks and reduced by 5–10 mg weekly. If there is no improvement, treatment should be escalated to Grade 3–4 protocols. If infection cannot be ruled out, empirical antibiotics should be considered. Bronchoscopy with bronchoalveolar lavage is recommended for atypical lesions.

Generally, for Grade 3–4 CIP, ICI therapy should be permanently discontinued. Intravenous methylprednisolone at 2 mg/(kg·d) is recommended. If clinical improvement is observed within 48 hours, continue steroid therapy until symptoms improve to Grade 1 or lower and then taper over 4–6 weeks. Bronchoscopy with bronchoalveolar lavage is advised for atypical lesions, and biopsy may be considered. Empirical antibiotic therapy should be initiated if infection cannot be ruled out. If there is no significant improvement with steroids, alternative therapies such as tocilizumab (8 mg/kg, repeatable after 14 days), infliximab (5 mg/kg, repeatable after 14 days), mycophenolate mofetil (1–1.5 g twice daily), or intravenous immunoglobulin (IVIG) should be considered.

Antimicrobial agents constitute a critical component of the pharmacological management of CIP. Patients who have received antitumor immunotherapy, chemotherapy, radiotherapy, as well as steroid hormones and immunosuppressive therapies for CIP are frequently immunocompromised, rendering them highly susceptible to a range of infectious diseases, including opportunistic infections. Respiratory infections can potentially trigger or worsen CIP, and a subset of patients may concurrently present with both CIP and pulmonary infectious diseases, thereby complicating the clinical management of CIP. In cases of coexisting infection, an individualized antibiotic regimen should be formulated on the basis of the clinical and radiological characteristics of CIP, etiological findings, and the patient’s baseline therapeutic strategy (140). Some experts propose that during the treatment of CIP with steroid hormones and/or immunosuppressants, prophylactic sulfonamide therapy can be administered. In the absence of contraindications, a prophylactic dose of sulfonamide should be routinely administered orally. Common high-risk factors are poorly controlled diabetes mellitus, long-term use of hormones (prednisone equivalent ≥ 20 mg/d) and/or immunosuppressants, bone marrow suppression following oncological treatment, a history of Pneumocystis jirovecii infection, and a peripheral blood CD4⁺ T-cell count of less than 400 cells/μl (167–170).

Corticosteroids are generally recommended as the initial treatment for symptomatic checkpoint inhibitor pneumonia (CIP) patients. Retrospective data indicate that over 90% of Grade 1–2 CPI events can improve or resolve with either the cessation of the drug or the use of corticosteroids, with or without additional interventions. Furthermore, 60%-86% of patients who experience Grade 3 or more severe CIP can achieve remission or cure through corticosteroids or supplementary immunosuppressive therapy (11, 153). Steroid-refractory CIP is defined as the absence of clinical improvement following high-dose corticosteroids administered for 48 hours, which necessitates the introduction of additional immunosuppressive treatments (131).

Some researchers propose dividing the clinical phenotypes of CIP into acute, subacute, and chronic phases and dividing the 2 pathological processes of CIP into inflammatory, profibrotic, and fibrotic stages (179, 180). Early-onset CIP often occurs within 6 weeks of patients receiving ICI treatment, and the symptoms are generally severe with a poor prognosis. Late-onset CIP typically appears after 6 weeks of ICI treatment, with fewer symptoms and a better prognosis. Chronic CIP is defined as CIP that persists or worsens after the reduction of steroid treatment, and such patients require more than 12 weeks of immunosuppressive therapy after the discontinuation of ICIs (181). During the chronic CIP phase, fibrotic interstitial lung diseases (such as complete damage to normal lung structure, thickening and obstruction of blood vessels) are histologically dominant, which may impair the efficiency of blood–gas exchange in the lungs (182, 183). Typical sequelae may include persistent pulmonary interstitial fibrosis and reduced pulmonary function caused by severe CIP (54). These concepts provide new directions for the treatment of CIP beyond immunosuppressive therapy, namely, antifibrotic treatment. Anti-fibrosis agents such as pirfenidone have demonstrated efficacy (184). In a retrospective study (185), compared with the glucocorticoid-only group, the glucocorticoid-pirfenidone group demonstrated significant improvements in forced vital capacity (FVC), diffusing capacity of the lung for carbon monoxide (DLCO), 6-minute walk distance (6MWD), and oxygen saturation without supplemental oxygen (P < 0.05). Among patients with grade 2 immune checkpoint inhibitor-related pneumonia (CIP), those receiving combination therapy exhibited a significantly shorter duration of symptom improvement than those treated with glucocorticoids alone did. Furthermore, following the resumption of immunotherapy, the incidence of CIP recurrence was lower in the glucocorticoid-pirfenidone group than in the glucocorticoid-only group. Nintedanib is a small molecule tyrosine kinase inhibitor that has been approved for the antifibrotic treatment of patients with idiopathic pulmonary fibrosis (IPF) and chronic interstitial lung disease (ILD) (186). Additionally, nintedanib has demonstrated efficacy against solid tumors in multiple clinical trials. For patients with advanced squamous carcinoma, the LUME-Lung 3 study reported that with first-line combination treatment of nintedanib plus cisplatin plus gemcitabine, the disease control rate (DCR) was 81.3%, the median progression-free survival (mPFS) was 4.2 months, and the median overall survival (mOS) was 6.7 months (187). In lung adenocarcinoma, the combination treatment of nintedanib and docetaxel has been approved by the European Union as a first-line chemotherapy regimen (https://www.ema.europa.eu/en/medicines/human/EPAR/vargatef). Several case reports have documented the use of nintedanib for salvage treatment in refractory and chronic CIP patients, with clinical efficacy (186, 188).

When grade 1–2 CIP events resolve or revert to grade 1, patients may continue their previous ICI regimen. The recurrence of CIP should be closely monitored, particularly for early-onset irAEs (189). Delaunay et al. reported that among 10 patients with non-small cell lung cancer (NSCLC) who were rechallenged with ICIs after initially managing grade 1–2 CIP with ICI withdrawal or corticosteroids, 3 patients (30%) experienced recurrent interstitial lung disease (ILD) (153). A larger retrospective cohort study revealed that 9 patients (20.0%) developed recurrent pneumonitis, and 11 patients (24.4%) experienced new irAEs among 45 CIP patients who underwent ICI rechallenge (190). A more recent analysis (191) revealed that while ICI rechallenge beyond disease progression did not introduce new safety concerns, it did not improve clinical outcomes in patients with locally advanced or metastatic NSCLC. However, patients who demonstrate a favorable response to initial ICI treatment still benefit from subsequent ICI rechallenge therapy. Similarly, systematic analysis revealed that the incidence rates of all-grade and high-grade (grade 3 or 4) irAEs were not significantly different between initial and readministered ICIs, and no significant difference in overall survival (OS) was observed between the ICI rechallenge and discontinuation cohorts (192). For patients who have recovered from grade 2 pneumonitis, cautious reintroduction of ICIs is advised, particularly for those with a positive response to initial ICIs and without feasible alternative therapies. This resumption may be accompanied by concomitant prednisone treatment, typically at a dosage of 20–30 mg/day, and should be conducted under close monitoring with chest CT (189). Although reintroduction of ICIs has been proven safe in rare and carefully selected cases (193), it is not recommended for patients who have recovered from grade 3 CIP events. For patients with grade 4 CIP, the use of ICIs should be permanently avoided.

Overall, most cases of CIP can be effectively managed through the withholding of ICIs and the administration of corticosteroids. Although severe cases are relatively uncommon, they are associated with poor prognosis and high mortality, underscoring the need for prompt recognition and intervention. Currently, there are no prospective trials evaluating the efficacy of treatments for severe cases; thus, no universal guidelines for optimal therapy beyond corticosteroids exist. For patients who have recovered from CIPs, immunotherapy rechallenge may be feasible, particularly for those who experienced low-grade CIPs and benefit from initial ICI treatment; however, the recurrence of CIP or the emergence of new irAEs should be closely monitored.