The medical records of 199 consecutive patients with lung cancer who received at least one administration of an anti-PD1/anti-PD-L1 ICI at any treatment line at Zhongshan Hospital Fudan University between Sep 2017 and December 2019 were retrospectively reviewed. Data were collected from our electronic medical record and patient follow-up visits. We excluded 9 patients for diagnosis of resectable lung cancer. Finally, 190 evaluable patients were included.

This study was approved by the ethics committee of Zhongshan Hospital, Fudan University (No. B2020-431R). The requirement for informed consent from the patients included in this study was waived due to the retrospective nature of the study, and any personal information from the data was removed beforehand. The censored date was June 29, 2021.

The data collected for all patients included gender, age, smoking history, pack-years index, medical history, performance status (PS) on the Eastern Cooperative Oncology Group scale, tumor-node-metastasis (TNM) staging, histology, driven mutation, PD-L1 expression, treatment regimen, toxicity profile, and survival.

We evaluated the overall survival (OS) and progression-free survival (PFS). OS was measured as the period from the diagnosis of lung cancer to death. PFS1 was measured as the period from the first administration of immunotherapy to documented disease progression or death from any cause. PFS2 was measured as the period from the initiation of systemic treatment beyond disease progression after the initial immunotherapy to documented second progression or death from any cause. Treatment response was assessed using the Response Evaluation Criteria in Solid Tumors version 1.1. Adverse events were graded using the National Cancer Institute Common Terminology Criteria for Adverse Events (version 4.0).

Cytokines were detected by a solid-phase, two-site chemiluminescent immunometric assay.

Numbers and percentages were computed for categorical variables. Means and standard deviations for normally distributed continuous variables and medians and interquartile ranges for abnormally distributed continuous variables were calculated. The cumulative rates of OS or PFS were estimated with Kaplan-Meier method and survival curves were plotted. Comparisons between survival curves were performed by log-rank tests.

Factors affecting OS or PFS were analyzed using Cox proportional hazard model. Factors with statistical significance in univariate analysis were then put into multivariate regression models to further identify their effects on fatal or progressive outcomes. Hazard ratios (HR), together with its 95% confidence intervals were reported. Forest plots were drawn to visualize the results of multivariate analysis.

Comparisons between cytokine levels of TNF, IL-2R, IL-6, IL-8 and CD4/CD8 at diverse detection time and baseline levels were performed using Wilcoxon test due to abnormal distributed data. In order to build a predictive model for PFS with cytokine levels, cutoff values were determined using median values or receiver operating characteristics curves analysis. If area under curve (AUC) was smaller than 0.5, the median value was considered the cutoff value. Otherwise, cutoff values were calculated based on ROC. The results of predictive model were visualized using nomogram whose performance was evaluated by Harrell concordance index (C-index). 1000 bootstrap resamples were used for internal validation of the accuracy of the predictions. Calibration curves were drawn to assess the consistence between the predicted probability and the actual proportion (6). All statistical analyses were performed by R version 4.1.0. A two-sided P value less than 0.05 was considered as the significance level.

Detailed demographic information and baseline characteristics of the studied subjects are presented in Table 1 . Among the 190 patients enrolled in our cohort, 84 (44.2%) patients received ICI as first line therapy. 73 (38.4%) patients received ICI as monotherapy. The objective response rate (ORR) of immunotherapy in our patients was 36.3%. The Kaplan-Meier curves for all the patients are shown in Figure 1 . The median OS was 28.7 months. Median PFS1 and PFS2 were 6.9 and 4.8 months, respectively.

Univariable Cox proportional hazards regression analysis revealed that patients with EGFR/ALK/ROS1 mutation had a higher risk of ICI resistance (adjusted HR, 1.12; 95% CI, 0.58–2.15; P=0.0145). Multivariable Cox proportional hazards regression analysis revealed that second line immunotherapy (adjusted HR, 2.33; 95% CI, 1.21–4.47; P=0.011) was a significant risk factor associated with ICI resistance in lung cancer patients. PD-L1 expression of ≥50% was associated with a lower risk of ICI resistance (adjusted HR, 0.45; 95% CI, 0.21–0.97; P=0.04). ( Table 2 ; Figure 2 )

Figure 3 demonstrated the Kaplan-Meier curves for different metastatic sites. We found that PFS1 time was significantly shorter for patients with bone metastasis at baseline (adjusted HR, 1.505; 95% CI, 1.036-2.186; P=0.0195).

When disease progression after the initial immunotherapy, 32.2% (N=38) of patients experienced intrapulmonary progression and 23.7% (N=28) experienced extrapulmonary progression ( Figure 4A ). Among them, 52.5% (N=62) experienced oligo-progression ( Figure 4B ). Lung was the most common site suffering from tumor progression ( Figure 4C ).

19.8% patients received continued immunotherapy beyond ICI resistance ( Figure 5 ). Systemic therapies including chemotherapy and anti-angiogenic therapy rather than continued immunotherapy beyond ICI resistance had significant effect on PFS2 time (P=0.0256) ( Figure 6 ). 11/62 patients with oligo-progression received local treatment. Among them, 5 patients received cerebral radiotherapy, 5 patients received thoracic radiotherapy, and 1 patient received surgery for adrenal metastasis. Whether or not patients with oligo-progression received local treatment seemed to have no significant effect on PFS2 time (P=0.7267) ( Figure 7 ).

Patients with lung or lymph nodes progression when ICI resistance had poor OS time (P=0.0041 and P<0.0001) ( Figure 8 ). Patients with extrapulmonary progression and oligo-progression after initial exposure to immunotherapy had longer OS time than those with intrapulmonary progression and multi-progression ( Figures 9A, B ).

Part of our patients had plasma cytokine data at different timepoint, including baseline, disease progression, occurrence of irAE and best efficacy. Based on the available data, we evaluated the potential prognostic value of plasma cytokines for the outcomes of ICI treatment. Figures 10A-E demonstrated the cytokine levels at different timepoint. When disease progression during the initial ICI treatment, levels of TNF, IL-6 and IL-8 were significantly elevated ( Table 3 ). Figures 11A-E showed survival curves based on Kaplan-Meier for different plasma cytokines. Area under the ROC Curve of TNF is less than 0.5 (0.42), so we selected median value 7.9 pg/mL as the cutoff for TNF. Plasma TNF level higher than 7.9 pg/ml was significantly associated with longer PFS1 time ( Figure 11A ). We determined that 8 pg/mL is a clinically relevant threshold using ROC curve analysis. Plasma IL-8 level higher than 8 pg/mL was significantly associated with shorter PFS1 time ( Figure 11E ). These results suggested that TNF and IL-8 might serve as prognostic biomarkers for the PFS time of patients with advanced-stage lung cancer treated with ICI.

Next, we constructed a nomogram based on the plasma cytokines. The PFS nomogram included the following valuables: age, gender, TNF, IL-8, treatment line, PS and oligo/multi-progression ( Figure 12A ). One hundred was set as the maximum score of each variable; by adding up the total score and locating it on the bottom point scale, an estimated probability of survival could be easily determined. Calibration plot of the nomogram is displayed in Figure 12B . The C-index of the built nomogram to predict PFS (0.74; 95% CI, 0.65–0.82) demonstrated an optimal consistency between the nomogram prediction and the actual observation for lung cancer patients treated with ICI. The ROC curve indicated that the nomogram model has a good predictive ability in predicting the clinical outcomes of lung cancer patients receiving ICI treatment (AUC: 0.777 for 1‐year PFS) ( Figure 12C ).