第一部分

¹⁸F-FDG PET/CT对非小细胞肺癌新辅助免疫治疗疗效的预测价值探索

目的：探究¹⁸F-FDG PET/CT对非小细胞肺癌（non-small cell lung cancer，NSCLC）新辅助免疫治疗术后肿瘤主要病理缓解（major pathologic response，MPR）的预测价值；比较PET/CT代谢参数与生物标志物程序性细胞死亡配体1（programmed cell death ligand 1，PD-L1）表达和肿瘤突变负荷（tumor mutational burden，TMB）对MPR的预测效能。

方法：本研究纳入中国医学科学院肿瘤医院2018年4月至2022年2月接受新辅助免疫治疗NSCLC的两个前瞻性临床试验队列（注册号：ChiCTR-OIC-17013726，ChiCTR2000033588），共纳入90例NSCLC，所有患者在基线和新辅助治疗后均接受¹⁸F-FDG PET/CT检查，随后完整切除肿瘤评估术后肿瘤原发灶病理反应。测量原发肿瘤基线和新辅助免疫治疗后PET/CT的代谢参数，包括瘦体体重校正的标准摄取值（standardized uptake value lean body mass，SUL）、代谢肿瘤体积（metabolic tumor volume，MTV）和肿瘤总糖酵解（total lesion glycolysis，TLG）。应用Spearman相关分析评估PET/CT代谢参数（基线、新辅助免疫治疗后、变化百分数）、生物标志物PD-L1表达和TMB与肿瘤原发灶病理缓解程度之间的相关性。使用受试者工作特征曲线评估PET/CT代谢参数（基线、新辅助免疫治疗后、变化百分数）、实体瘤疗效PET评价标准（PET Response Criteria in Solid Tumors，PERCIST）、生物标志物PD-L1表达和TMB对MRP的预测效能。采用DeLong检验比较PET/CT代谢参数与生物标志物PD-L1表达和TMB对MPR预测效能的差异。

结果：90例NSCLC患者中40例（44.4%）达到MPR，50例（55.6%）未达到MPR。肺鳞癌的MPR率高于肺腺癌（54.9% vs. 5.6%）。MPR患者新辅助免疫治疗后原发肿瘤代谢参数明显降低（P<0.001），而非MPR患者治疗后代谢参数无明显变化（P>0.05）。基线PET/CT的SUL_max（r=0.219，P=0.038）和SUL_peak（r=0.234，P=0.026）与病理缓解程度呈正相关，基线SUL_max、SUL_peak、SUL_mean、MTV和TLG均不能预测MPR（均P>0.05）。新辅助治疗后PET/CT代谢参数SUL_max、SUL_peak、SUL_mean、MTV和TLG均与病理缓解程度呈负相关（r=-0.648～-0.457, 均P<0.001），预测MPR的AUC值为0.757～0.927（均P<0.001）。代谢参数变化百分数ΔSUL_max%、ΔSUL_peak%、ΔSUL_mean%、ΔMTV%和ΔTLG%均与病理缓解程度呈负相关（r=-0.794～-0.675，均P<0.001），预测MPR的AUC值为0.848～0.976（均P<0.001）。所有代谢参数中，ΔSUL_max%对MPR的预测效能最高，截断值为-33%，AUC值为0.976（95% CI：0.937～1.000，P<0.001），灵敏度、特异度、准确度分别为0.960、0.975、0.967。依据PERCIST评估治疗反应，1例（1.1%）完全代谢缓解，42例（46.7%）部分代谢缓解，38例（42.2%）代谢稳定，9例（10%）代谢进展。88.4%（38/43）完全和部分代谢缓解患者达到MPR，5.3%（2/38）代谢稳定患者达到MPR，无代谢进展患者达到MPR。PERCIST预测MPR的AUC值为0.931（95% CI：0.881～0.981，P<0.001）。生物标志物PD-L1表达（r=0.245，P=0.041）和TMB（r=0.298，P=0.014）与病理缓解程度呈正相关，预测MPR的AUC值分别为0.686（95% CI：0.562～0.810，P=0.008）和0.693（95% CI：0.566～0.820，P=0.006）。PET/CT代谢参数对MPR的预测效能明显优于PD-L1表达（AUC=0.757～0.976 vs. 0.686，P<0.05），也明显优于TMB（AUC=0.757～0.976 vs. 0.693，P<0.05）。

结论：新辅助治疗后PET/CT代谢参数、治疗前后代谢参数变化、PERCIST均能有效预测NSCLC新辅助免疫治疗术后MPR，其中ΔSUL_max%对MPR的预测效能最高。PET/CT代谢参数对MPR的预测效能明显优于生物标志物PD-L1表达和TMB。

第二部分

¹⁸F-FDG PET/CT对非小细胞肺癌PD-L1表达和肿瘤突变负荷的预测价值探索

目的：探讨¹⁸F-FDG PET/CT代谢参数对肺腺癌和肺鳞癌程序性细胞死亡配体1（programmed cell death ligand 1，PD-L1）表达与肿瘤突变负荷（tumor mutational burden，TMB）的预测价值；联合临床病理特征和¹⁸F-FDG PET/CT代谢特征，构建PD-L1表达与TMB的预测模型，并验证模型对新辅助免疫治疗疗效的分层价值。

方法：本研究包含两个队列，主要队列纳入2017年1月至2024年4月在中国医学科学院肿瘤医院抗肿瘤治疗前接受¹⁸F-FDG PET/CT检查、原发肿瘤PD-L1表达及TMB检测的非小细胞肺癌患者305例（肺腺癌183例，肺鳞癌122例）。首先，采用向后逐步选择的多因素Logistic回归分析主要队列中肺腺癌和肺鳞癌PD-L1表达与TMB的临床病理及PET/CT影响因素，分别对PD-L1表达与TMB进行单独分析（PD-L1阳性、PD-L1高表达、TMB高水平）和联合分析（PD-L1阳性且TMB高水平、PD-L1阴性且TMB低水平）。其次，将主要队列中肺腺癌以7:3比例随机分为训练集和验证集，构建并验证肺腺癌PD-L1表达和TMB单独和联合分析的预测模型，包括临床模型、最大瘦体体重校正的标准摄取值（maximum standardized uptake value lean body mass，SUL_max）模型和临床-SUL_max模型。应用受试者工作特征曲线评估模型的预测效能，采用校准曲线和决策曲线分析模型的观测一致性和临床有效性。最后，纳入中国医学科学院肿瘤医院2018年4月至2024年4月胸外科三个新辅助免疫治疗非小细胞肺癌队列，从中筛选出29例肺腺癌患者作为测试集，用于验证模型对新辅助免疫治疗疗效的分层价值。

结果：肺腺癌和肺鳞癌PD-L1表达水平之间无统计学差异（P=0.905），肺鳞癌TMB明显高于肺腺癌（10 vs. 5 mut/Mb，P<0.001）。对肺腺癌的分析显示：PET/CT代谢参数与PD-L1表达和TMB呈正相关（均P<0.05）。PD-L1表达与TMB单独分析的多因素Logistic回归显示：SUL_max（OR=1.157；95% CI：1.074～1.247；P<0.001）是PD-L1阳性的影响因素；肿瘤分化程度（OR=0.382；95% CI：0.157～0.925；P=0.033）、EGFR突变状态（OR=0.412；95% CI：0.145～1.170；P=0.096）和SUL_max（OR=1.077；95% CI：0.990～1.171；P=0.084）是PD-L1高表达的影响因素；年龄（OR=1.039；95% CI：1.004～1.076；P=0.030）、吸烟史（OR=1.809：95% CI：0.915～3.578：P=0.088）、EGFR突变状态（OR=0.330；95% CI：0.158～0.692；P=0.003）和SUL_max（OR=1.165；95% CI：1.073～1.265；P<0.001）是TMB高水平的影响因素。PD-L1表达与TMB联合分析的多因素Logistic回归显示：年龄（OR=1.042；95% CI：1.001～1.085；P=0.045）、EGFR突变状态（OR=0.436；95% CI：0.189～1.008；P=0.052）和SUL_max（OR=1.229；95% CI：1.123～1.344；P<0.001）是PD-L1阳性且TMB高水平的影响因素；吸烟史（OR=0.379；95% CI：0.147～0.977；P=0.045）、分叶征（OR=0.434；95% CI：0.178～1.058；P=0.066）和SUL_max（OR=0.769；95% CI：0.676～0.876；P<0.001）是PD-L1阴性且TMB低水平的影响因素。无论单独还是联合分析，SUL_max均是PD-L1表达（P<0.001）和TMB（P<0.001）的独立影响因素。在训练集和验证集中，临床-SUL_max模型的预测效能高于单独的临床模型和SUL_max模型。临床-SUL_max模型预测PD-L1阳性、PD-L1高表达和TMB高水平的效能有限（训练集AUC=0.655、0.692和0.765，验证集AUC=0.689、0.733和0.705）；预测PD-L1表达与TMB联合状态的效能更佳（预测PD-L1阳性且TMB高水平的训练集AUC=0.805，验证集AUC=0.724；预测PD-L1阴性且TMB低水平的训练集AUC=0.798，验证集AUC=0.744）。生物标志物预测模型有效分层了测试集中肺腺癌患者新辅助免疫治疗术后的病理反应（PD-L1阳性且TMB高水平模型P=0.035；PD-L1阴性且TMB低水平模型P=0.001）。对肺鳞癌的分析显示：肺鳞癌PET/CT代谢参数与PD-L1表达和TMB无相关性（均P>0.05），未筛选出影响因素构建肺鳞癌PD-L1表达和TMB的预测模型。

结论：SUL_max是肺腺癌PD-L1表达和TMB的独立预测因素。基于临床病理和PET/CT影像数据构建的临床-SUL_max模型能有效预测肺腺癌PD-L1表达和TMB状态，并进一步分层免疫治疗疗效，可协助临床医师筛选免疫治疗潜在获益的肺腺癌患者。

第三部分

免疫检查点抑制剂相关肺炎临床和胸部CT影像特征研究

目的：分析免疫检查点抑制剂相关肺炎（checkpoint inhibitor-related pneumonitis，CIP）患者临床和胸部CT影像特征，探究胸部CT影像特征对评估CIP严重程度及炎症转归的价值。

方法：回顾性分析2017年8月至2021年4月中国医学科学院肿瘤医院收治的恶性肿瘤患者中38例CIP患者的临床和胸部CT影像资料。总结CIP的临床和胸部CT特征，包括CIP发生时间、临床症状、严重程度、胸部CT炎症分布、CT征象及分类。CIP严重程度依据美国国立卫生研究院不良事件通用术语标准5.0分级。分析重度CIP的临床和胸部CT特征，采用受试者工作特征曲线评估半定量肺炎评分（临床症状评分、胸部CT征象评分、肺炎范围评分和综合评分）对重度CIP的诊断效能，综合评分采用Logistic回归分析构建。自诊断CIP之日起对患者进行随访，通过随访期间的胸部CT图像评估肺炎转归情况，比较CIP完全吸收患者与未完全吸收患者胸部CT影像特征的差异。

结果：38例CIP患者自免疫检查点抑制剂用药至发生CIP的中位时间为72.5天，范围10～228天，22例（57.9%）患者于免疫治疗后3个月内发生肺炎，34例（89.5%）患者伴有临床症状，其中咳嗽（29例，76.3%）和呼吸困难（27例，71.1%）为最多见的临床症状。胸部CT炎症分布多呈不对称分布（31例，81.6%），磨玻璃影（37例，97.4%）和实变（30例，78.9%）为最多见的CT征象，机化性肺炎型（15例，39.5%）为最多见的CT分类。1级轻度CIP 4例、2级中度CIP 20例、3-5级重度CIP 14例。重度CIP 3个月内发生的患者数多于3个月后发生的患者数（11例vs. 3例，P=0.038）；重度CIP累及肺叶数多于中度CIP（4 vs. 2个肺叶，P=0.010）；发热症状（P<0.001）与急性间质性肺炎/急性呼吸窘迫综合征型CT分类（P=0.001）是重度CIP的特征表现。临床症状评分、胸部CT征象评分、肺炎范围评分诊断重度CIP的AUC值分别为0.782（95% CI：0.622～0.942，P=0.006）、0.843（95% CI：0.702～0.984，P<0.001）和0.857（95% CI：0.716～0.998，P<0.001）。综合评分整合以上三个评分，诊断效能最佳，AUC值为0.948（95% CI：0.873～1.000，P<0.001），灵敏度、特异度、准确度分别为0.929、0.950和0.912。综合评分对重度CIP的诊断效能优于临床症状评分（P=0.028）、胸部CT征象评分（P=0.054）和肺炎范围评分（P=0.046）。在30例随访时间大于1个月的CIP患者中，完全吸收组CIP发生时间短于未完全吸收组（55.0 vs. 128.0天，P=0.022）。与未完全吸收组相比，完全吸收组实变较少（P=0.010），CT分类均为过敏性肺炎型（P=0.004）。

结论：CIP多在免疫检查点抑制剂用药后3个月内出现，发热症状和急性间质性肺炎/急性呼吸窘迫综合征型CT分类是重度CIP的特征表现。整合临床症状、胸部CT征象和肺炎范围的综合评分对重度CIP的诊断效能最佳。临床和胸部CT影像特征对评估CIP严重程度和炎症转归有重要价值。

PART I

Predictive value of ¹⁸F-FDG PET/CT on the efficacy of neoadjuvant immunotherapy in non-small cell lung cancer

Objective: To explore the predictive value of ¹⁸F-FDG PET/CT for the major pathologic response (MPR) in non-small cell lung cancer (NSCLC) receiving neoadjuvant immunotherapy. To compare the predictive performance of PET/CT metabolic parameters with biomarkers of programmed cell death ligand 1 (PD-L1) expression and tumor mutational burden (TMB) for MPR.

Methods: This study included two prospective clinical trials (registration numbers: ChiCTR-OIC-17013726, ChiCTR2000033588) in Cancer Hospital, Chinese Academy of Medical Sciences from April 2018 to February 2022, comprising 90 NSCLC patients who receiving neoadjuvant immunotherapy. All patients underwent ¹⁸F-FDG PET/CT scans at baseline and after neoadjuvant treatment, and postoperative pathological response of the primary tumor was evaluated. Metabolic parameters of primary tumor at baseline and post-neoadjuvant PET/CT were measured, including standardized uptake value lean body mass (SUL), metabolic tumor volume (MTV), and total lesion glycolysis (TLG). Spearman correlation analyses were performed to assess the correlations between PET/CT metabolic parameters (baseline, post-neoadjuvant, and percentage variation), PD-L1 expression and TMB with the degree of pathological regression. Receiver operating characteristic curve analyses were utilized to evaluate the performance of PET/CT metabolic parameters (baseline, post-neoadjuvant, and percentage variation), PET Response Criteria in Solid Tumors (PERCIST), PD-L1 expression, and TMB in predicting MPR. Additionally, DeLong test was conducted to compare the predictive performance between PET/CT metabolic parameters with PD-L1 expression and TMB.

Results: Among 90 patients with NSCLC, 40 (44.4%) achieved MPR, while 50 (55.6%) did not. Squamous cell carcinoma achieved more MPR than adenocarcinoma (54.9% vs. 5.6%). Metabolic parameters demonstrated significant reduction after neoadjuvant immunotherapy in MPR patients (P<0.001), while no changes were observed in non-MPR patients (P>0.05). At baseline PET/CT, SUL_max (r=0.219, P=0.038) and SUL_peak (r=0.234, P=0.026) showed positive correlations with the degree of pathological regression. Baseline SUL_max, SUL_peak, SUL_mean, MTV, and TLG were unable to predict MPR (all P>0.05). At post-neoadjuvant PET/CT, SUL_max, SUL_peak, SUL_mean, MTV, and TLG demonstrated negative correlations with the degree of pathological regression (r=-0.648~ -0.457, all P<0.001), with AUC values for predicting MPR ranging from 0.757 to 0.927 (all P<0.001). At the percentage variation, ΔSUL_max%, ΔSUL_peak%, ΔSUL_mean%, ΔMTV%, and ΔTLG% also showed negative correlations with the degree of pathological regression (r=-0.794~ -0.675, all P<0.001), with AUC values for predicting MPR ranging from 0.848 to 0.976 (all P<0.001). Among all metabolic parameters, ΔSUL_max% presented the highest predictive performance in threshold of -33% and AUC of 0.976 (95% CI：0.937~1.000, P<0.001), with sensitivity, specificity, and accuracy of 0.960, 0.975, and 0.967, respectively. According to PERCIST, 1 case (1.1%) had complete metabolic response, 42 cases (46.7%) had partial metabolic response, 38 cases (42.2%) had stable metabolic disease, and 9 cases (10.0%) had progressive metabolic disease. 88.4% (38/43) of patients with complete and partial metabolic response achieved MPR, while only 5.3% (2/38) of patients with stable metabolic disease achieved MPR, and none of patients with progressive metabolic disease achieved MPR. PERCIST demonstrated predictive performance for MPR with an AUC of 0.931 (95% CI: 0.881~0.981, P<0.001). Both PD-L1 expression (r=0.245, P=0.041) and TMB (r=0.298, P=0.014) were positively correlated with the degree of pathological regression, and the predictive performance for MPR presented AUC of 0.686 (95% CI: 0.562~0.810, P=0.008) for PD-L1 expression and 0.693 (95% CI: 0.566~0.820, P=0.006) for TMB. PET/CT metabolic parameters demonstrated superior predictive performance compared to PD-L1 expression (AUC=0.757~0.976 vs. 0.686, P<0.05) and TMB (AUC=0.757~0.976 vs. 0.693, P<0.05).

Conclusions: ¹⁸F-FDG PET/CT at post-neoadjuvant scan, the percentage variation after treatment, along with PERCIST can predict MPR in NSCLC receiving neoadjuvant immunotherapy, among which ΔSUL_max% demonstrated the highest performance. The predictive performance of PET/CT for MPR was significantly superior to PD-L1 expression and TMB.

PART II

Predictive value of ¹⁸F-FDG PET/CT for PD-L1 expression and tumor mutational burden in non-small cell lung cancer

Objective: To explore the predictive value of ¹⁸F-FDG PET/CT for programmed cell death ligand 1 (PD-L1) expression and tumor mutational burden (TMB) in lung adenocarcinoma and squamous cell carcinoma. We developed prediction models based on clinicopathological data and PET/CT imaging data for PD-L1 expression and TMB, and verified the usefulness of the models for stratifying neoadjuvant immunotherapy responses.

Methods: This study consisted of two cohorts. The primary cohort included 305 patients (183 cases of lung adenocarcinoma and 122 cases of lung squamous cell carcinoma) who received ¹⁸F-FDG PET/CT, PD-L1 expression and TMB test of primary tumor before anti-tumor treatment in Cancer Hospital, Chinese Academy of Medical Sciences from January 2017 to April 2024. First, multivariate logistic regression analysis with backward stepwise was employed to investigate the clinicopathological and PET/CT influencing factors of PD-L1 expression and TMB in lung adenocarcinoma and squamous cell carcinoma within the primary cohort. Separate analyses (PD-L1-Positive, PD-L1-High, TMB-High) and combined analyses (PD-L1-Positive and TMB-High, PD-L1-Negative and TMB-Low) were conducted. Second, the lung adenocarcinoma in the primary cohort were randomly divided into the training set and the validation set at a 7:3 ratio. Predictive models for PD-L1 expression and TMB in lung adenocarcinoma were constructed and validated including clinical model, the maximum standardized uptake value lean body mass (SUL_max) model, and clinical-SUL_max model. The predictive performance of the models was evaluated by the receiver operating characteristics curve. Model calibration and clinical usefulness were assessed. Finally, we included three cohorts of patients receiving neoadjuvant immunotherapy for NSCLC in the Department of Thoracic Surgery from April 2018 to April 2024 in Cancer Hospital, Chinese Academy of Medical Sciences, and selected 29 patients with lung adenocarcinoma from the three cohorts as the test set to verify the usefulness of models for stratifying neoadjuvant immunotherapy responses.

Results: PD-L1 expression did not differ between lung adenocarcinoma and squamous cell carcinoma (P=0.905); however, the TMB was higher in lung squamous cell carcinoma (10 vs. 5 mut/Mb, P<0.001). For lung adenocarcinoma, there were positive correlations between metabolic parameters and PD-L1 expression, as well as TMB (all P<0.05). When PD-L1 expression and TMB analyzed separately, the multivariate logistic regression showed: SUL_max (OR=1.157; 95% CI: 1.074~1.247; P<0.001) was an influencing factor of PD-L1-Positive; tumor differentiation (OR=0.382; 95% CI: 0.157~0.925; P=0.033), EGFR mutation status (OR=0.412; 95% CI: 0.145~1.170; P=0.096) and SUL_max (OR=1.077; 95% CI: 0.990~1.171; P=0.084) were influencing factors of PD-L1-High; age (OR=1.039; 95% CI: 1.004~1.076; P=0.030), smoking history (OR=1.809; 95% CI: 0.915~3.578; P=0.088), EGFR mutation status (OR=0.330; 95% CI: 0.158~0.692; P=0.003) and SUL_max (OR=1.165; 95% CI: 1.073~1.265; P<0.001) were influencing factors of TMB-High. When PD-L1 expression and TMB analyzed jointly, the multivariate logistic regression showed: age (OR=1.042; 95% CI: 1.001~1.085; P=0.045), EGFR mutation status (OR=0.436; 95% CI: 0.189~1.008; P=0.052), and SUL_max (OR=1.229; 95% CI: 1.123~1.344; P<0.001) were influencing factors for PD-L1-Positive and TMB-High; smoking history (OR=0.379; 95% CI: 0.147~0.977; P=0.045), lobulation sign (OR=0.434; 95% CI: 0.178~1.058; P=0.066), and SUL_max (OR=0.769; 95% CI: 0.676~0.876; P<0.001) were influencing factors for PD-L1-Negative and TMB-Low. SUL_max was an independent predictor of both PD-L1 expression (P<0.001) and TMB (P<0.001) in separate or combined analyses. The predictive performance of the clinical-SUL_max model was superior to the clinical model and the SUL_max model in both training and validation sets. The clinical-SUL_max models demonstrated limited predictive performance for PD-L1-Positive, PD-L1-High, and TMB-High individually (Training sets: AUC=0.655, 0.692, and 0.765; Validation sets: AUC=0.689, 0.733, and 0.705). The clinical-SUL_max models exhibited optimal performance in predicting PD-L1 expression combined TMB (predicting PD-L1-Positive and TMB-High: AUC=0.805 for the training set and 0.724 for the validation set; predicting PD-L1-Negative and TMB-Low: AUC=0.798 for the training set and 0.744 for the validation set). The models effectively stratified the pathological responses of lung adenocarcinoma in the test set after neoadjuvant immunotherapy (P=0.035 for the PD-L1-Positive and TMB-High model; P=0.001 for the PD-L1-Negative and TMB-Low model). For lung squamous cell carcinoma, no correlation was observed between metabolic parameters and PD-L1 expression, as well as TMB (all P>0.05), and no predictors were identified to develop predictive models for PD-L1 expression and TMB in lung squamous cell carcinoma.

Conclusions: SUL_max is an independent predictor of both PD-L1 expression and TMB in lung adenocarcinoma. The clinical-SUL_max models based on clinicopathological and PET/CT imaging features effectively predict PD-L1 expression and TMB status in lung adenocarcinoma, and further stratify clinical efficacy of immunotherapy, assisting clinicians in patient selection with potential benefits from immunotherapy.

PART III

Clinical and chest CT imaging features of immune checkpoint inhibitor-related pneumonitis

Objective: To analyze the clinical and chest computed tomography (CT) features of immune checkpoint inhibitor-related pneumonitis (CIP), and explore the value of chest CT features in assessing the disease severity and treatment outcome.

Methods: Clinical and chest CT data of 38 CIP patients with malignant tumors in the Cancer Hospital, Chinese Academy of Medical Sciences between August 2017 and April 2021 were retrospectively reviewed. We summarized the clinical and chest CT features of CIP, including occurrence time, clinical symptoms, severity grading, pneumonitis distribution on chest CT, CT findings, and radiographic pattern. CIP severity was graded according to the Common Terminology Criteria for Adverse Events 5.0. We explored the clinical and chest CT features for severe CIP, and evaluated the diagnostic performance of semi-quantitative scores (clinical symptom, CT finding, extent and combination scores) for severe CIP by receiver operating characteristics curve analysis. The combination score was developed using logistic regression analysis. Patients were followed up from the date of CIP diagnosis, with serial chest CT scans performed to evaluate the longitudinal radiographic changes and the outcomes. Compare chest CT features between patients with complete absorption and those with incomplete absorption.

Results: Among the 38 patients diagnosed with CIP, the median time from the administration of immune checkpoint inhibitors to CIP onset was 72.5 days, ranging from 10 to 228 days. Of these, 22 patients (57.9%) developed CIP within 3 months. Clinical symptoms were observed in 34 patients (89.5%), with cough (29 cases, 76.3%) and dyspnea (27 cases, 71.1%) being the frequent manifestations. The distribution of CIP on chest CT was mostly asymmetrical distribution (31 cases, 81.6%). Ground glass opacity (37 cases, 97.4%) and consolidation (30 cases, 78.9%) were the common CT findings, and organizing pneumonia pattern (15 cases, 39.5%) was the most common CT classification. There were 4 cases of mild CIP (grade 1), 20 cases of moderate CIP (grade 2), and 14 cases of severe CIP (grades 3-5). More severe CIP occurred within 3 months than after 3 months (11 vs. 3 cases, P=0.038). Severe CIP involved more lung lobes than moderate CIP (4 vs. 2 lung lobes, P=0.010). Fever (P<0.001) and the acute interstitial pneumonia/acute respiratory distress syndrome CT pattern (P=0.001) were significantly associated with severe CIP. The diagnostic performance of clinical symptom score, CT finding score and extent score for severe CIP with AUC of 0.782 (95% CI: 0.622~0.942, P=0.006), 0.843 (95% CI: 0.702~0.984, P<0.001) and 0.857 (95% CI: 0.716~0.998, P<0.001), respectively. The combination score, which integrated the three scores, demonstrated the best diagnostic performance with AUC of 0.948 (95% CI: 0.873~1.000, P<0.001), and the sensitivity, specificity, and accuracy were 0.929, 0.950, and 0.912, respectively. The combination score demonstrated superior diagnostic performance compared to clinical symptom score (P=0.028), CT finding score (P=0.054), and extent score (P=0.046). Among 30 CIP patients with followed-up longer than one month, the occurrence time in the complete absorption group was shorter than the incomplete absorption group (55.0 vs. 128.0 days, P=0.022). Compared with the incomplete absorption group, patients in the complete absorption group demonstrated less consolidation (P=0.010), and CIP were all classified as hypersensitivity pneumonitis CT pattern (P=0.004).

Conclusions: CIP mostly occurs within 3 months following immune checkpoint inhibitor initiation. Fever and the acute interstitial pneumonia/acute respiratory distress syndrome CT pattern were identified as characteristic features of severe CIP. The combination score that integrating clinical symptom, CT finding, and extent demonstrated optimal diagnostic performance for severe CIP. Both clinical and chest CT features have important value in evaluating CIP severity and treatment outcomes.