Efficacy and safety of immune checkpoint inhibitor rechallenge in non‐small cell lung cancer: A systematic review and meta‐analysis

Abstract Background The aim of the study was to explore the efficacy and safety of immune checkpoint inhibitor (ICI) rechallenge in patients with non‐small cell lung cancer (NSCLC). Methods Studies that enrolled NSCLC patients treated with two lines of ICIs were included using four databases. The initial line (1L‐) and subsequent lines (2L‐) of ICIs were defined as 1L‐ICI and 2L‐ICI, respectively. Results A total of 17 studies involving 2100 patients were included. The pooled objective response rate (ORR), disease control rate (DCR), median progression‐free survival (mPFS), and median overall survival (mOS) for 2L‐ICIs were 10%, 50%, 3.0 months, and 13.1 months, respectively. The 2L‐ICI discontinuation rates caused by toxicities ranged from 0% to 23.5%. Original data were extracted from six studies, covering 89 patients. Patients in whom 1L‐ICIs were discontinued following clinical decision (the mPFS of 2L‐ICIs was not reach) achieved a more prolonged mPFS of 2L‐ICIs than those due to toxicity (5.2 months) and progressive disease (2.1 months) (p < 0.0001). Patients' 1L‐PFS for more than 2‐years had preferable 2L‐ORR (35.0% vs. 9.8%, p = 0.03), 2L‐DCR (85.0% vs. 49.0%, p = 0.007), and 2L‐mPFS (12.4 vs. 3.0 months, p < 0.0001) than those less than 1‐year. Patients administered the same drugs achieved a significantly prolonged mPFS compared with the remaining patients (5.4 vs. 2.3 months, p = 0.0004), and those who did not accept antitumor treatments during the intervals of two lines of ICIs achieved a prolonged mPFS compared to those patients who did accept treatments (7.6 vs. 1.9 months, p < 0.0001). Conclusions ICI rechallenge is a useful therapeutic strategy for NSCLC patients, especially suitable for those who achieve long‐term tumor remission for more than 2‐years under 1L‐ICIs.

INTRODUCTION Non-small cell lung cancer (NSCLC) accounts for about 85% of overall lung cancer patients. 1 There has been a growing number of treatment options for patients with NSCLC, 2 including the widely used immune checkpoint inhibitors (ICIs). Owing to the remarkable efficacy, immunotherapy has been the standard care for first-line treatment in NSCLC without actionable sensitive driver gene variations, irrespective of whether administered as monotherapy or in combination with platinum-based chemotherapy. 3 Once patients have failed to respond to immunotherapy, the therapeutic options are limited with barely satisfactory efficacy. For example, docetaxel as the standard second-line chemotherapy treatment only achieved a 7.1% of partial response rate in NSCLC patients. 4 For these reasons, ICI therapy is sometimes rechallenged appropriately by clinicians to maximize the benefit of immunotherapy in real-world application. 5 The rechallenge strategy dates back to the 1970s, 6 and has achieved success in certain circumstances, such as bevacizumab rechallenge in patients with progressive colorectal cancer, 7 and trastuzumab rechallenge after breast cancer progression. 8,9 However, in NSCLC, there are neither clinical guidelines to guide the specific treatment of ICI rechallenge, nor systematic data about the efficacy and safety of ICI rechallenge at present, which further lead to the complexity of the specific clinical environment of ICI rechallenge in real-world.
Reasons for discontinuation of ICIs usually include disease progression, [10][11][12][13][14] the occurrence of severe adverse events (AEs), [15][16][17] and clinician's decision after the completion of a fixed duration course of immunotherapy and the achievement of long-lasting tumor remission. 18 ICI rechallenge has been reported in several studies, but the efficacy and survival outcomes vary largely due to a very heterogeneous population. On the basis of clinical experience, patients who discontinued primary ICIs because of AEs or the completion of a fixed treatment course are more likely to benefit from ICI rechallenge. 19,20 Despite this, there is no definite evidence on which subgroups could obtain benefit from ICI rechallenge to date.
Regarding ICI rechallenge during the real-world NSCLC treatment, some data have been reported, but most studies were single-arm, with a small sample size, and did not explore factors that might affect the prognosis. 15,21,22 Therefore, in our present meta-analysis, we reviewed all the original studies about ICI rechallenge in NSCLC and comprehensively collected the detailed data, to describe the overall efficacy and safety of ICI rechallenge, and more importantly, to explore which subgroups are suitable for ICI rechallenge therapy.

Search strategy and inclusion criteria
This study was conducted and reported following the Preferred Reporting Items for Systematic reviews and metaanalysis (PRISMA) guidelines. 23 The protocol of this study was registered in PROSPERO (CRD42021284110). A literature search was performed in PubMed, Web of Science, Embase, and the Cochrane Central Register of Controlled Trials databases. The cutoff date for searching was October 15, 2021. Keywords used included "non-small cell lung cancer", "immunotherapy", "immune checkpoint inhibitors", "programmed cell death protein 1 (PD-1) inhibitors", "programmed cell death ligand 1 (PD-L1) inhibitors", "cytotoxic T lymphocyte-associated protein 4 (CTLA-4) inhibitors", "rechallenge", "retreatment", "readministration", and "restart". The detailed literature search strategy is exhibited in Table S1.
Two investigators (YF and YXT) independently conducted the initial search to identify potentially eligible articles, and determined the selected literature according to the inclusion criteria. References of the included studies and from previously published systematic reviews were manually assessed to detect any missing study. When consensus was not reached by discussing with the other two investigators (HZC and YZ), the corresponding authors were consulted for the final decision.
The inclusion criteria included: (1) pathologically or cytologically confirmed NSCLC, including but not limited to adenocarcinoma lung cancer, squamous cell lung cancer, and large cell lung cancer; (2) patients who were treated with at least two lines of immunotherapy; (3) the initial ICI discontinuation due to progressive disease (PD), toxicity, completion of a fixed treatment course, or any other reasons; (4) studies which reported any of the clinical endpoints, including objective response rate (ORR), disease response rate (DCR), progression-free survival (PFS), and overall survival (OS); (5) studies written in English. The exclusion criteria included: (1) pathologically or cytologically confirmed small cell lung cancer (SCLC) or mixed with SCLC; (2) studies that lacked sufficient information for analysis; (3) animal studies; (4) fewer than 10 patients for each study; (5) case reports; (6) reviews and meta-analyses, editorials and letters to the editors. In the case of duplicate publications, only the most recent or most informative study was included in the analyses.
Rechallenge in our study was defined as patients who received at least two lines of ICIs during their disease course, and the initial line (1L-) and subsequent line (2L-) of ICIs were defined as 1L-ICI and 2L-ICI, respectively. For studies that received three and above lines of ICIs, data of the first-and second-line of ICIs would be collected. The discontinuation reasons of the 1L-ICIs included toxicity, disease progression, completion of a fixed course of ICIs, decision of the prescriber, and other reasons that might be involved. The key inclusion criteria of patients in each study (mainly the specific conditions of two lines of ICI treatment) are seen in Table S2. proportion score [TPS]); (3) treatment information (the 1L-and 2L-ICI therapy); (4) reasons for 1L-ICI discontinuation; (5) treatment between the 1L-and 2L-ICIs; (6) clinical outcomes (ORR, DCR, PFS, OS, and safety), in which, ORR and DCR were separately extracted from the 1L-ICI and 2L-ICI therapy, PFS was extracted and calculated from the beginning of the 1L-ICI and 2L-ICI therapy separately, OS was extracted and calculated from the beginning of 2L-ICI therapy.

Data extraction and quality assessment
Additionally, deeper data mining was performed for studies that provided both 1L-ICI and 2L-ICI efficacy and survival (PFS) data of each patient, in which, extracted variables were based on a single patient. The mandatory variables included each patient's best tumor response and PFS of 1L-and 2L-ICIs, respectively. The remaining variables of each patient were extracted as much as possible according to the provisions of different studies for further analysis. Since all of the included studies were observational or retrospective cohort studies, the methodological index for nonrandomized studies (MINORS) criteria was performed to evaluate the quality of studies. 24

Statistical analysis
Categorical variables are summarized as total and percentage. Continuous variables were performed as median, mean, range, and 95% confidence interval (CI). ORR was defined as the percentage of complete response (CR) and partial response (PR) obtained as best response, DCR was defined as the percentage of CR, PR, and stable disease (SD). ORR and DCR were collectively called short-term efficacy, PFS and OS were called long-term efficacy.
We used ORR, DCR, PFS, and OS with 95% CIs from individual studies to calculate the pooled effect rate and effect time. The differences in ORR and DCR between 1L-ICIs and 2L-ICIs were compared through the form of column diagrams. Studies that provided each patient's raw survival time would be filtered further, then pooled together to plot survival curves (Kaplan-Meier method) and conduct subgroup analysis. Cox proportional hazard models were used to evaluate prognostic factors for PFS by univariate and multivariate analysis.
Cochran Q statistical test was used to evaluate betweenstudy heterogeneity, and a p-value <0.1 was considered as significant heterogeneity. Higgins I 2 statistic, in which I 2 < 25% was regarded as low, 25% ≤ I 2 < 50% as moderate, I 2 ≥ 50% as high heterogeneity, was used to calculate the proportion of total variation across studies caused by statistical heterogeneity. 25 A random-effects model was used to calculate a pooled effect size. 26 Sensitivity analyses were performed to examine statistical heterogeneity by omitting one study at each time and assessing the reliability of the pooled estimates. Publication bias was examined by using Begg's and Egger's test. 27,28 Stata software (version 17.0), IBM SPSS software (version 22.0), and GraphPad Prism (version 8.0) were used in statistical analysis. MATLAB (version R2021b) was used to extract the survival data of the swimming maps. A two-sided p-value of <0.05 was considered statistically significant, except for the heterogeneity test.
A total of 15 studies were retrospective and two were prospective. All studies were published after the year 2018, with 11 conducted in Japan, two in America, two in France, and two worldwide. A total of 12 studies used mono-ICIs, three used mono or dual ICIs as the initial immunotherapy regimens, and two studies included patients using ICIs plus chemotherapy. For the 2L-ICIs regimens, 14 studies used mono-ICIs, one used mono or dual ICIs, one used ICIs plus chemotherapy or antiangiogenic drugs, and the remaining one was not reported.
The characteristics of the included patients are shown in Table 1. A total of 436 patients from 10 studies reported ECOG PS, and 74 (17.0%) patients had a score ≥2. A total of 255 patients from seven studies reported the tumor stage details, and 154 (60.4%) were stage IV. A total of 338 patients from 11 studies reported the results of PD-L1 TPS, of whom 73 (21.6%) was 1-50%, 74 (21.9%) was ≥50%. Eight studies covering 261 patients reported the lines of prior treatment. All except one study where patients underwent at least one treatment before the 1L-ICIs, and the highest prior treatment lines were 14.
Next, we further compared the differences in shortterm efficacy between 1L-and 2L-ICIs treatment. A total of 13 studies covering 387 patients reported both 1L-and 2L-ICI efficacy data. The ORR of 1L-ICIs was higher than that of 2L-ICIs (Figure 3a). The DCR of 1L-ICIs was also more favorable than that of 2L-ICIs except for one study by Santini et al. 17 with the same 1L-ORR and 2L-ORR ( Figure 3b).

Toxicity of 2L-ICIs therapy
A total of five studies 13,16,20,30,33 covering 231 patients reported the proportion of discontinuation events caused by toxicity during the treatment of 2L-ICIs, and the rate ranged from 0% to 23.5%. Patients from one study 16 all discontinued 1L-ICIs due to toxicity, yet no patient discontinued the 2L-ICIs due to toxicities. The details are shown in Table 1.

Characteristics of 89 patients with original data extracted from six studies
We further explored the variables that may potentially affect the efficacy of 2L-ICIs. Among all the included 17 studies, 10 studies did not provide the original data on the best tumor response and PFS of each patient, and one study provided the data on time to treatment failure (TTF) instead of   the data on PFS. The specific patient's original efficacy data were extractable in six studies, 10,13,14,16,18,33 in which two were presented in the form of swimming plots, and four were presented in the form of tables, in total covering 89 patients with NSCLC. The details are shown in Table S4. Mono ICIs were used as the regimen of 1L-and 2L-ICIs in all patients. Three patients (3.4%) were treated with 1L-ICIs as first-line systemic therapy, and one received different immunotherapy (2L-ICIs) as the second-line systemic therapy. Among all the 89 NSCLC patients, information on ICI regimens, tumor response, discontinuation reasons of 1L-ICIs, and PFS of each patient were accessible. There were 48 (53.9%), 26 (29.2%), and 15 (16.9%) patients who achieved CR/PR, SD, and PD as the best tumor response during the treatment of 1L-ICIs, respectively. For the PFS of patients treated with 1L-ICIs, a total of 51 (57.3%) patients achieved PFS less than one year, 18 (20.2%) patients achieved between one and two years, and 20 (22.5%) patients achieved over two years. With regard to the discontinuation reasons for 1L-ICIs, 47 (52.8%) patients discontinued treatment due to PD, 27 (30.3%) due to toxicity, and 15 (16.9%) due to the decision of their clinician. In addition, other subgroups were defined according to the following indicators: whether there were systematic treatments between two lines of immunotherapy (yes, n = 43, 48.3%; no, n = 46, 51.7%), and whether the 2L-ICI was the same as 1L-ICI (yes, n = 36, 40.4%; no, n = 53, 59.6%). The details are shown in Table S5.

Variables affecting the PFS of 2L-ICIs based on 89 patients
We next analyzed the differences in 2L-PFS according to the above-mentioned subgroups through univariate (Table 2 and Figure 4) and multivariate (Table S6) Cox proportional hazard models.
As shown in Figure 4a, patients who discontinued 1L-ICIs because of their clinician's decision achieved a significantly prolonged median 2L-PFS than those who discontinued 1L-ICIs due to toxicity (not reached vs. 5.2 months, p = 0.002; HR: 0.22, 95% CI: 0.09-0.58) and PD (not reached vs. 2.1 months, p < 0.0001; HR: 0.15, 95% CI: 0.06-0.37). In addition, the median 2L-PFS of patients who discontinued 1L-ICIs because of toxicity was also significantly longer than those who discontinued 1L-ICIs due to PD T A B L E 2 Subgroup analysis of the short-and long-term efficacy of 2L-ICIs based on 89 patients.

Publication bias
The funnel plot revealed no significant publication bias in all the pooled analysis except the median PFS of 1L-ICIs for patients with NSCLC (Figures S11-S17). The results of Begg's and Egger's test also indicated there was no evidence of substantial publication bias for all meta-analyses (Table S7).

DISCUSSION
In the field of NSCLC, there has been no meta-analysis which has focused on the issue of ICI rechallenge and to the best of the our knowledge this is the first study to do so. Through this meta-analysis, we first performed the pooled analysis of short-term and long-term efficacy of the 1L-and 2L-ICIs individually to assess the feasibility of ICI rechallenge in NSCLC, then conducted a deeper analysis of 89 patients whose original data were extractable to explore what kind of NSCLC patients would get the best profits from ICI rechallenge. The role of ICI rechallenge therapy has not been completely investigated and established. 5,37 One previous meta-analysis 38 has made some efforts to evaluate the feasibility of ICI rechallenge in cancer patients. However, there still exist many questions that need to be addressed, including how the efficacy of 1L-ICIs influences the outcomes of 2L-ICIs, whether the different discontinuation reasons of 1L-ICIs would cause different outcomes of 2L-ICIs, and so forth.
As shown in the results, patients with NSCLC could benefit from the treatment of ICI rechallenge to a certain degree, with the pooled ORR of 10%, DCR of 50%, median PFS of 3.0 months and OS of 13.1 months. The previous meta-analysis by Inno et al. 38 investigated ICI rechallenge in multicancer types, demonstrating an overall ORR of 21.8%, median PFS of 4.9 months, and median OS of 15.6 months. The pathological type of our study was limited to NSCLC, which might be one of the important reasons contributing to our inferior results. Additionally, it should be noted that the heterogeneity of our pooled results was high (I 2 > 50%, p < 0.01), but all the sensitivity analysis results showed there was no single study that significantly influenced the heterogeneity. We found that both the short-term (ORR, 0% to 43%; DCR, 21% to 87%) 11,18,29,34,36 and the long-term efficacy (PFS, 1.6 to 7.4 months; OS, 6.5 to 20.7 months) 14,16 of the included studies varied largely. Since there was no consensus available to guide the clinical decisions of ICI rechallenge, a huge selective bias and different results with high heterogeneity were inevitable. The pooled ORR and DCR of the 1L-ICIs were 43% and 77%, respectively, also with great heterogeneity. Generally, the ORR and DCR of 2L-ICIs were worse than that of 1L-ICIs, which partly reflected the entire status quo of rechallenge strategy in cancer therapy. 39 Previous studies showed that patients enrolled in the precondition of tumor remission during the initial ICIs 18,19 achieved relatively higher ORR during the treatment of 2L-ICIs, with that of 40-60%. 18,19 However, these results are from small sample studies and could not be used as evidence to guide clinical practice. We further integrated these small sample data and conducted thorough research to explore the predictive factors for ICI rechallenge efficacy using the original data of 89 patients with NSCLC from six studies. 10,13,14,16,18,33 The findings demonstrated that patients with the tumor response of CR/PR of 1L-ICIs achieved a 2L-ORR of 18.8%, 2L-DCR of 60.4%, and 2L-PFS of 5.1 months; patients with a 1L-PFS of more than two years achieved a 2L-ORR of 35.0%, DCR of 85.0%, and 2L-PFS of 12.4 months. Actually, the clinical trial of CA180-002 proved that the ipilimumab retreatment would regain disease control in most melanoma patients (20/31) with initial clinical benefit (any response lasting ≥3 months). 40 Despite that, no widely recognized recommendations are available, these results inspired us to consider ICI reinduction in patients that showed an initial response or a lasting disease stabilization when there are few treatment options.
In clinical practice, the discontinuation reasons of ICIs mainly included drug resistance, 41 severe toxicities, 42 completion of a fixed duration of treatment, 18,43,44 tumor cured, financial reason, or death. In the respect of discontinuation reasons of the initial ICIs, the results demonstrated that patients in whom treatment with ICIs was terminated due to clinical decision achieved the most favorable efficacy of immunotherapy rechallenge compared with the reason of drug resistance (PD) and toxicity, no matter in short-or long-term efficacy.
In conventional clinical practice, once drug resistance occurs, the original drugs are no longer effective. An optional treatment strategy after immunotherapy resistance includes cessation of the original drugs, switching to chemotherapy, the addition of chemotherapy to ICI, or the addition of a novel agent to ICI. 41 In our study, patients who underwent PD to initial ICIs had dismal outcomes when they switched to other ICIs or continued to use the initial ICIs. Despite this, a small number of patients still obtained objective responses with the application of ICI rechallenge even though they had been resistant to the initial ICIs. The activation of the immune microenvironment caused by the application of other systematic antitumor treatments between the intervals of 1L-ICIs and 2L-ICIs might explain this phenomenon. 31,45 Similarly, the better outcomes for patients in whom a clinical decision had been made to discontinue initial ICIs (patients had usually finished a fixedduration of treatment or obtained long-lasting tumor remission) or toxicity may be partly explained by the fact that discontinuation of ICIs occurs before the drug resistance.
Our analysis also suggested improved prognosis for patients with the same drugs of 1L-and 2L-ICIs and for patients with no other systematic antitumor treatments during the ICI treatment intervals. As discussed above, systematic antitumor treatments between the initial and subsequent ICIs might enhance the reactiveness of the tumor microenvironment. 45 However, in our study, systematic antitumor treatments during the intervals of two lines of ICIs was not a positive indicator of better outcomes. The possible reason is that the initial ICI was discontinued in the majority of these patients due to drug resistance (PD), and that patients without treatment during the intervals mostly discontinued the initial ICI due to toxicity and clinical decision. The probable reason why patients in whom the same 1L-and 2L-ICI drugs achieved a better prognosis was also attributed to a relatively lower rate of resistance to the initial ICI. However, these conclusions need to be further verified by prospective studies.
Our analysis had several limitations. First, most of the studies included were retrospective and more susceptible to selection bias, therefore, contributing to higher heterogeneities of pooled results. To address this issue, we performed a comprehensive analysis based on the 89 patients with original data, and the positive results consequently pointed out the most important reasons for heterogeneity. Second, very few of the included studies reported the detailed and specific toxicity data of ICI rechallenge, which was the main concern of ICI rechallenge, especially for those that discontinued the 1L-ICIs due to toxicities. Third, it is well known that EOCG-PS and PD-L1 status might affect the efficacy of ICIs, but due to the absence of detailed data in many studies, we finally failed to design a proper method to perform related subgroup analysis. Therefore, there is an urgent need for well-designed prospective studies to address these issues.
In conclusion, ICI rechallenge therapy appears to be effective as an alternative regimen for NSCLC patients previously treated with ICIs, and especially suitable for those who achieve long-term tumor remission for more than 2-years under 1L-ICIs. Other factors that might affect the efficacy of ICI rechallenge, such as PD-L1 and ECOG-PS status, still need further exploration. Considering the great heterogeneity in clinical outcomes and potential severe toxicity of ICI rechallenge, it is important to identify patients who may benefit from ICI rechallenge. More investigations on exploring robust predictive factors to identify these patients are needed.

AUTHOR CONTRIBUTIONS
Yu Feng and Yunxia Tao conceived the idea of this study and conducted the literature search. Yu Feng performed the data analysis. Yu Feng, Yunxia Tao, Haizhu Chen, and Yu Zhou drafted the original manuscript. Le Tang and Chenwei Liu provided partial software technical support. Yuankai Shi and Xingsheng Hu supervised the conduct of this study. All authors reviewed the manuscript draft and revised it critically on intellectual content. All authors approved the final version of the manuscript to be published.