Clinical, Radiologic, and Highest-Grade Assessment of Immune Checkpoint Inhibitor Pneumonitis in Non–Small Cell Lung Cancer

INTRODUCTION

Immune checkpoint inhibitors (ICIs) are an integral component of systemic therapy for non–small cell lung cancer (NSCLC) in tumors without actionable oncogenic alterations and have improved overall survival.[1–4] However, ICIs can also cause potentially fatal toxicities, including ICI-associated pneumonitis.[5] Accurate grading of pneumonitis is critical for management because severity assessment informs immunosuppression intensity, hospitalization decisions, prognosis, and the safety of potential ICI rechallenge.

The Common Terminology Criteria for Adverse Events (CTCAE) version 5.0 is widely used to grade adverse events and primarily reflects symptom severity, oxygen requirement, and functional impairment.[6] In contrast, guideline-based management frameworks, including those from the National Comprehensive Cancer Network (NCCN), incorporate both clinical features and radiologic extent of lung involvement.[7] Therefore, discordance can arise between clinical severity and radiographic extent, creating uncertainty regarding which assessment should drive risk stratification and management.

Current NCCN and European Society for Medical Oncology (ESMO) guidance integrates clinical and radiographic features and generally recommends management according to the highest severity assessment.[7,8] Clinical–radiologic discrepancies during immunotherapy-associated pneumonitis have also been described, highlighting the potential for discordance between anatomic extent and clinical severity.[9] However, this approach has not been well evaluated as a prognostic strategy in patients with NSCLC who develop ICI pneumonitis. We evaluated the concordance between clinical CTCAE grade and radiologic extent and compared clinical grade, radiologic grade, and a combined highest-grade rule for predicting 1-year all-cause mortality after ICI pneumonitis in NSCLC. Exploratory analyses assessed long-term pulmonary outcomes.

METHODS

Study Design and Population

This retrospective cohort study included adults with NSCLC diagnosed with ICI-associated pneumonitis at MD Anderson Cancer Center (2016–2022) under institutional review board (IRB)-approved protocols (PA15-0917, PA13-0589). Pneumonitis was diagnosed by multidisciplinary assessment incorporating chest computed tomography (CT) imaging, symptoms, oxygen requirements, and exclusion of alternative etiologies (infection, progression, pulmonary edema, embolism, radiation pneumonitis, and radiation recall). For patients with multiple episodes, only the first was retained. Asymptomatic (CTCAE grade 1) events were excluded because radiographic-only events were not uniformly captured.

Grading Definitions

Clinical severity was graded per CTCAE v5.0 (grades 2–4). Radiologic extent was categorized by percentage of lung involvement on chest CT (<25%, 25–50%, >50%) and mapped to radiologic grades 2, 3, and 4, respectively. The highest-grade rule was defined as the maximum of the clinical and mapped radiologic grades. Concordance denoted agreement between the two grades; discordance was classified as radiologic-higher or clinical-higher.

Radiologic Assessment

Chest CTs obtained at or nearest to pneumonitis diagnosis were independently reviewed by a thoracic radiologist, a pulmonologist, and a thoracic oncologist, all blinded to clinical grade and outcomes. Extent was estimated by visual semiquantitative assessment; disagreements were resolved by consensus. Radiographic pattern (e.g., organizing pneumonia-like, NSIP-like, hypersensitivity-like, and DAD-like) and overlap with prior radiation fields were recorded when available.

Outcomes

The primary outcome was 1-year all-cause mortality. All-cause rather than pneumonitis-specific mortality was used because cause-of-death adjudication was not uniformly available. Exploratory long-term pulmonary outcomes included residual radiographic abnormality or fibrosis (persistent parenchymal abnormality vs qualitative return to baseline), persistent hypoxemia (SpO₂ ≤ 88% on room air or new oxygen requirement), and persistent dyspnea (clinician-documented dyspnea not attributable to another acute process).

Statistical Analysis

Categorical variables were summarized as counts and percentages; continuous variables as means (SD) or medians (IQR). Correlation between clinical and radiologic grades was assessed with Kendall tau-b. Discriminative performance for 1-year mortality was compared across the three grading strategies using area under the ROC (AUROC) with 95% CIs; pairwise comparisons used DeLong testing. Bootstrap resampling provided internal precision estimates. Exploratory ROC analyses for long-term pulmonary outcomes were performed as complete-case analyses among patients with complete grading data and evaluable end point status. Given limited events and small high-severity strata, multivariable adjustment was not performed; results were interpreted as descriptive and hypothesis-generating. A two-sided alpha of < 0.05 was used for statistical significance. All analyses were performed in R version 4.3.2 (R Foundation for Statistical Computing, Vienna, Austria) using the pROC package for ROC analysis and DeLong testing. Specific ICI agents and median follow-up duration were not available in the analytic dataset.

RESULTS

Cohort Characteristics

After retaining only the first pneumonitis episode per patient, the analytic cohort included 74 unique patients, of whom 68 (91.9%) had complete clinical and radiologic grading data. Baseline characteristics are summarized in Table 1. Notably, prior thoracic radiation was common (54 of 70 patients, 77.1%), baseline ILAs were present in 25 of 71 patients (35.2%), and 1-year all-cause mortality occurred in 30 of 74 patients overall (40.5%) and 26 of 68 complete grading cases (38.2%).

Table 1.

Descriptive characteristics of the first-episode patient-level cohort

Variable	Value
Unique patients in first-episode cohort	74
Patients with complete clinical and radiologic grading	68/74 (91.9)
Age, mean (SD), y	65.9 (9.0)
Female sex	31/70 (44.3)
Tobacco history
Never smoker	15/70 (21.4)
Former smoker	50/70 (71.4)
Current smoker	5/70 (7.1)
Prior thoracic radiation before ICI	54/70 (77.1)
Adenocarcinoma histology	44/70 (62.9)
Metastatic disease at baseline	37/71 (52.1)
Prior systemic therapy line(s) before ICI	25/71 (35.2)
Combination CTLA-4 therapy	10/71 (14.1)
COPD	32/71 (45.1)
ILD	4/71 (5.6)
Baseline ILAs on pre-ICI imaging	25/71 (35.2)
Baseline hypoxemia	2/71 (2.8)
Baseline shortness of breath	19/71 (26.8)
Steroids at pneumonitis onset	58/71 (81.7)
Nonsteroidal second-line immunosuppression	1/71 (1.4)
ICU admission	8/71 (11.3)
Clinical CTCAE grade 2	41/68 (60.3)
Clinical CTCAE grade 3	23/68 (33.8)
Clinical CTCAE grade 4	4/68 (5.9)
Radiologic involvement < 25%	35/68 (51.5)
Radiologic involvement 25–50%	20/68 (29.4)
Radiologic involvement > 50%	13/68 (19.1)
Highest-grade rule grade 2	37/68 (54.4)
Highest-grade rule grade 3	27/68 (39.7)
Highest-grade rule grade 4	4/68 (5.9)
1-year all-cause mortality, overall cohort	30/74 (40.5)
1-year all-cause mortality, complete grading cases	26/68 (38.2)
Residual radiographic abnormality/fibrosis, evaluable patients	14/48 (29.2)
Persistent hypoxemia, evaluable patients	15/47 (31.9)
Persistent dyspnea, evaluable patients	33/47 (70.2)

Values are n (%) unless otherwise stated.

Baseline demographic, treatment, and pulmonary comorbidity variables are reported among patients with available data. Long-term pulmonary outcomes are reported among evaluable patients in the full first-episode cohort. Grading distributions and ROC analyses use the complete clinical and radiologic grading cohort. ECOG performance status, exact ICI agent, and exact time from ICI initiation to pneumonitis diagnosis were not available and were not included.

AUROC: area under the receiver operating characteristic curve; COPD: chronic obstructive pulmonary disease; CTCAE: Common Terminology Criteria for Adverse Events; CTLA-4: cytotoxic T-lymphocyte associated protein 4; ICI: immune checkpoint inhibitor; ICU: intensive care unit; ILA: interstitial lung abnormality; ILD: interstitial lung disease; NSCLC: non–small cell lung cancer; SD: standard deviation.

Concordance Between Clinical and Radiologic Grades

Among the 68 complete grading cases, clinical and radiologic grades showed modest-to-moderate correlation (Kendall tau-b 0.37, p = 0.001). Concordance was observed in 40 cases (58.8%), radiologic grade exceeded clinical grade in 20 cases (29.4%), and clinical grade exceeded radiologic grade in 8 cases (11.8%) (Table 2, Fig. 1).

Table 2.

Corrected clinical grade vs mapped radiologic grade among complete grading cases

Mapped Radiologic Grade	Clinical CTCAE Grade 2	Clinical CTCAE Grade 3	Clinical CTCAE Grade 4	Total
Radiologic grade 2 (<25%)	27	7	1	35
Radiologic grade 3 (25–50%)	10	10	0	20
Radiologic grade 4 (>50%)	4	6	3	13
Total	41	23	4	68

Radiologic extent was mapped to radiologic grade 2 for < 25% involvement, grade 3 for 25–50% involvement, and grade 4 for > 50% involvement. Grades were concordant in 40 of 68 cases (58.8%). Radiologic grade exceeded clinical grade in 20 of 68 cases (29.4%), and clinical grade exceeded radiologic grade in 8 of 68 cases (11.8%).

Figure 1.

Distribution and discordance between clinical CTCAE grade and mapped radiologic grade in the first-episode complete grading cohort. Radiologic extent was mapped to radiologic grade 2 for < 25% involvement, grade 3 for 25–50% involvement, and grade 4 for > 50% involvement.

Discrimination for 1-Year All-Cause Mortality

The highest-grade rule had the highest AUROC (0.804; 95% CI, 0.702–0.906), followed by clinical CTCAE grade (0.760; 95% CI, 0.650–0.870) and mapped radiologic grade (0.631; 95% CI, 0.502–0.760) (Table 3, Fig. 2A). On pairwise DeLong testing, the highest-grade rule significantly outperformed radiologic grade (AUROC difference 0.173, p = 0.005) but not clinical grade (AUROC difference 0.044, p = 0.174). Clinical grade did not significantly outperform radiologic grade (AUROC difference 0.129, p = 0.094).

Table 3.

Discrimination for 1-year all-cause mortality among complete grading cases

Grading Strategy	AUROC	95% CI
Clinical CTCAE grade	0.760	0.650–0.870
Mapped radiologic grade	0.631	0.502–0.760
Highest-grade rule	0.804	0.702–0.906

AUROCs are from complete-case analysis among 68 patients with complete clinical and radiologic grading data. Pairwise DeLong comparisons: highest-grade rule vs mapped radiologic grade, AUROC difference 0.173, p = 0.005; highest-grade rule vs clinical CTCAE grade, AUROC difference 0.044, p = 0.174; clinical CTCAE grade vs mapped radiologic grade, AUROC difference 0.129, p = 0.094.

Figure 2.

Receiver operating characteristic curves for clinical CTCAE grade, mapped radiologic grade, and the highest-grade rule. Panel A shows discrimination for the primary end point of 1-year all-cause mortality; DeLong p-values for pairwise AUROC comparisons are shown in the panel A legend. Panels B–D show exploratory complete-case ROC analyses for residual radiographic abnormality or fibrosis, persistent hypoxemia, and persistent dyspnea, respectively.

Exploratory Long-Term Pulmonary Outcomes

Among evaluable patients in the full cohort, residual radiographic abnormality or fibrosis was present in 14 of 48 (29.2%), persistent hypoxemia in 15 of 47 (31.9%), and persistent dyspnea in 33 of 47 patients (70.2%). In complete-case ROC analyses, mapped radiologic grade showed the strongest discrimination for residual radiographic abnormality or fibrosis (AUROC 0.672), whereas clinical grade and the highest-grade rule showed modest discrimination for persistent hypoxemia. All three strategies showed limited discrimination for persistent dyspnea (Table 4, Figs. 2B–D).

Table 4.

Exploratory discrimination for long-term pulmonary outcomes among evaluable complete grading cases

End Point	Evaluable Cases/Events	Clinical CTCAE Grade AUROC (95% CI)	Mapped Radiologic Grade AUROC (95% CI)	Highest-Grade Rule AUROC (95% CI)
Residual radiographic abnormality/fibrosis	45 / 14	0.474 (0.320–0.627)	0.672 (0.513–0.830)	0.491 (0.333–0.649)
Persistent hypoxemia	44 / 14	0.671 (0.510–0.832)	0.590 (0.412–0.769)	0.657 (0.494–0.821)
Persistent dyspnea	44 / 32	0.531 (0.372–0.690)	0.475 (0.305–0.646)	0.492 (0.327–0.657)

AUROCs are from complete-case analyses among patients with complete clinical and radiologic grading data and evaluable end point status. Residual radiographic abnormality/fibrosis was coded as absence of full radiographic resolution.

DISCUSSION

Clinical Versus Radiologic Grading

In this retrospective cohort of NSCLC patients with ICI-associated pneumonitis, clinical CTCAE grade and the highest-grade rule demonstrated stronger discrimination for 1-year all-cause mortality than radiologic extent alone, although the highest-grade rule was not statistically superior to clinical grading. These findings suggest that clinical severity (capturing symptoms, oxygen requirement, and functional impairment) may better reflect near-term mortality risk than radiographic extent in isolation. This is consistent with the broader principle that physiologic compromise, rather than anatomic disease burden alone, drives short-term outcomes in patients with limited pulmonary reserve.

Discordance and Its Implications

Clinical-radiologic discordance was observed in 41.2% of cases, with radiologic grade exceeding clinical grade in the majority of discordant pairs. This pattern may reflect differences in baseline pulmonary reserve, comorbidity burden, prior thoracic radiation, concurrent tumor-related changes, or compensatory mechanisms that buffer symptomatic expression despite extensive radiographic involvement. Conversely, clinical-higher discordance may indicate physiologic vulnerability disproportionate to visible parenchymal disease. These findings underscore that single-domain assessment (whether clinical or radiologic) incompletely characterizes pneumonitis severity.

Role of Radiologic Assessment

Although radiologic extent was a weaker mortality discriminator, it retains clinical value for characterizing disease burden, evaluating alternative diagnoses, identifying radiographic patterns, and detecting residual parenchymal sequelae. Notably, mapped radiologic grade showed the strongest discrimination for residual radiographic abnormality or fibrosis among the exploratory long-term outcomes, suggesting that radiologic extent may be more relevant to chronic pulmonary morbidity than to short-term mortality. However, visual estimation of lung involvement is inherently subjective, particularly in patients with prior radiation changes, tumor-related abnormalities, or baseline interstitial lung disease.

Limitations

Several limitations should be considered when interpreting these findings. The retrospective single-institution design introduces risk of selection bias, missing data, and nonstandardized follow-up. The cohort was small (n = 74; 68 with complete grading), with limited deaths (n = 30) and very small grade 4 and high radiologic involvement strata, resulting in imprecise AUROC estimates and precluding comprehensive multivariable adjustment for metastatic disease, treatment line, tumor burden, comorbidities, pulmonary reserve, prior thoracic radiation, baseline interstitial lung disease, or steroid response. ECOG performance status, exact ICI agent, time from ICI initiation to pneumonitis diagnosis, and median follow-up duration were not available; therefore, the independent contribution of each grading strategy cannot be determined.

The primary end point was all-cause mortality rather than pneumonitis-specific mortality, and cause-of-death adjudication was not uniformly available. In NSCLC, mortality may be driven by cancer progression, treatment complications, or comorbidities in addition to pneumonitis itself. Radiologic extent grading relied on retrospective visual CT assessment, which may be affected by subjectivity, prior radiation changes, tumor-related abnormalities, and baseline lung disease; interobserver agreement was not formally assessed. Radiographic pattern, radiation recall pneumonitis, and pulmonary function testing were incompletely captured. Long-term pulmonary outcomes had incomplete evaluability and should be considered exploratory.

FUTURE DIRECTIONS AND CONCLUSIONS

These findings should be interpreted as hypothesis-generating. Prospective studies should integrate clinical severity, radiologic extent and pattern, pulmonary function testing, prior radiation dosimetry, baseline interstitial lung abnormalities, and steroid response into multidimensional grading frameworks. Standardized adjudication protocols distinguishing ICI pneumonitis from radiation pneumonitis, radiation recall, and infection would improve phenotypic precision. Larger multicenter cohorts with cause-specific mortality adjudication and systematic long-term follow-up are needed to validate whether integrated grading strategies improve clinical decision-making beyond symptom-based assessment alone.

In summary, clinical CTCAE grade and the highest-grade rule demonstrated stronger discrimination for 1-year all-cause mortality than radiologic extent alone in NSCLC patients with ICI-associated pneumonitis. Radiologic extent may be more relevant to chronic pulmonary morbidity than to short-term mortality. Given the study’s limitations, prospective validation of multidimensional pneumonitis grading frameworks is warranted.