Immune-Related Neurological Toxicities of PD-1/PD-L1 Inhibitors in Cancer Patients: A Systematic Review and Meta-Analysis

Background Systematic assessment of PD-1/PD-L1 inhibitor-related neurological toxicities is important for guiding anti-PD-1 and anti-PD-L1 immunotherapy. Therefore, we conducted this meta-analysis to reveal the relationship between PD-1/PD-L1 inhibitors and neurological toxicities among cancer patients. Methods Clinical trials investigating PD-1/PD-L1 inhibitors in cancer patients were identified by a systematic search of PubMed. The random-effect model was used to synthesize individual studies. Neurological toxicities, including all-grades and grades 3–5, were taken into account for the final comprehensive meta-analysis. The Newcastle Ottawa Scale (NOS) was used to assess the quality of included trials. Results Thirty-one clinical trials containing data of neurological toxicities were included. Compared with chemotherapy, the risk of all-grade neurological toxicities caused by PD-1/PD-L1 inhibitors was much lower in terms of peripheral neuropathy [OR = 0.07, 95%CI:(0.04, 0.13)], peripheral sensory neuropathy [OR = 0.07, 95%CI(0.04, 0.12)], dysgeusia [OR = 0.26, 95%CI:(0.19, 0.35)], paraesthesia [OR = 0.23, 95%CI:(0.14, 0.36)], and polyneuropathy [OR = 0.12, 95%CI:(0.01, 0.94)]. However, for grades 3–5, the statistically significant results were only seen in peripheral neuropathy [OR = 0.15, 95%CI:(0.07, 0.34)] and peripheral sensory neuropathy [OR = 0.13, 95%CI:(0.04, 0.40)]. No statistically significant difference regarding the risk of headache, dizziness, and Guillain–Barré syndrome was found between PD-1/PD-L1 inhibitors and chemotherapy. For PD-1/PD-L1 inhibitors plus chemotherapy, the risk trends of the above-mentioned neurological toxicities, especially grades 3–5 peripheral neuropathy [OR = 1.76, 95%CI:(1.10, 2.82)] was increased compared to chemotherapy alone. Conclusion Our comprehensive analysis showed that PD-1/PD-L1 inhibitors alone exhibited lower neurological toxicities than chemotherapy. However, the risk of headache, dizziness, and Guillain–Barré syndrome was similar between PD-1/PD-L1 and chemotherapy. For PD-1/PD-L1 inhibitors plus chemotherapy, the incidence trend of neurological toxicities would be increased, especially for peripheral neuropathy of grades 3–5.


INTRODUCTION
Cancer immunotherapies, developed to overcome the immune escape mechanisms of cancer progression and metastatic dissemination, are becoming familiar to oncologists (1), especially for programmed cell death protein 1 (PD-1) and its ligand (PD-L1) inhibitors. PD-1/PD-L1 inhibitors belong to immune checkpoint blocking drugs (1); they can block the binding of tumor cells to PD-1 of T cells by means of PD-L1, restore the ability to recognize tumor cells, and further restore the cell recognition and killing ability of T cells (1). Immunotherapies, including cytotoxic T lymphocyte antigen-4 (CTLA-4) and PD-1/ PD-L1 had changed the treatment landscape for plenty of solid tumors but conferred unique toxicity profiles owing to their unique mechanism of actions (1)(2)(3).

METHOD
This research was conducted and reported according to the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) (15).

Types of Enrolled Studies
Randomized, open-label, controlled clinical trials investigating the efficacy and safety of PD-1/PD-L1 inhibitors in cancer patients were included. Phase III clinical trials, limited to solid tumors, were given a priority. Then, clinical trials of other phases would be checked for eligibility and placed in an alternative location. Clinical trials investigating hematological malignancies were beyond our consideration. In order to collect as many articles as possible, the control group was not restricted to a certain therapeutic agent or intervention. For inclusion, the study must report the data of at least one type of neurological toxicities related to immunotherapy. Articles must be published in English.

Search Strategy
Keywords, including neoplasm, cancer, precancer, malignant, premalignant, tumor, PD-1, PD-L1, and clinical trial, were used for the PubMed search with reference to participants, interventions, comparisons, outcomes, and study design (PICOS) (15). The published date was limited to the last 10 years (July 9, 2010 to July 9, 2020). Of note, some data regarding peripheral neuropathy was also collected from a former systematic review and meta-analysis (16). Four authors were designated to check the eligibility of all retrieved reports. They were also responsible for the extraction of relevant data from finally included trials. In the case of duplicated clinical trials, only one was included in the final analysis step. The corresponding authors (YS and GS) were responsible for resolving all disagreements.

Evaluation of Study Quality and Publication Bias
Funnel plots, Egger's test, and the Newcastle-Ottawa scale (NOS) were used to check publication bias and risk of bias of individual trials, respectively (15,(17)(18)(19)(20). The quality assessment included the appraisal of random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, and selective outcome reporting (shown in a single figure). Harbord's test was used to check the risk of publication bias of enrolled clinical trials Heterogeneity of all enrolled clinical trials was identified by Cochrane's Q statistic test (21). The grade of heterogeneity was estimated by the DerSimonian-Laird method and I 2 values together, which was suggested by Higgins and colleagues (15,21). Heterogeneity was deemed to be low, moderate, or high according to I 2 values < 25, 25-50, and > 50%, respectively (16). All data analyses were completed by the software Review Manager 5.3. Owing to the existence of inherent heterogeneity among included trials, the random effect (RE) was used for the evaluation of odds ratio (OR) and their corresponding 95% confidence interval (CI) (58). Sometimes, the fixed effects (FE) model was used as a supplement. All reported P values are twosided, and P<0.05 was deemed to be statistically significant. Subgroup analysis was made according to tumor types, treatment regimens, and PD-1/PD-L1 inhibitors.

Risk of Bias
The results of the publication bias assessment, in the form of funnel plots, are provided in the supplement ( Supplementary  Figures 1-3, 5, 7, 9) (15,(17)(18)(19)(20). Low risk of bias was identified in all clinical trials regarding selection bias, performance bias, detection bias, attrition bias, and reporting bias ( Figure 2) . An unclear risk relating to other biases was identified in four clinical trials (36,39,40,48). None of the included trials had a high risk of bias.
The risk of dysgeusia grades 3-5 could not be analyzed in the meta-analysis due to the limited data available in the included trials (23,47).

DISCUSSION
Most of the neurological toxicities caused by PD-1/PD-L1 inhibitors might be presented as low-grade appearances, with the potential to involve any aspect of the central or peripheral nervous system (7,8). As more and more clinical trials reporting the efficacy and safety of PD-1/PD-L1 in cancer patients are being conducted, the reporting of drug-induced neurological toxicities has gradually increased (1,2,. In order to clarify the relationship between PD-1/PD-L1 inhibitors and the risk of neurological toxicities in cancer patients, this metaanalysis was designed. It was the first time that neurological toxicities were comprehensively investigated through a metaanalytic approach instead of case reports and reviews (1,(5)(6)(7)(8)(9)(10)(11)(12)(13)(14). It would be helpful in guiding anti-PD-1 and anti-PD-L1 immunotherapy. Thirty-six articles, including 31 clinical trials with available data regarding neurological toxicities, were included in our study . Among the included clinical trials, lung cancer-related clinical trials accounted for the largest proportion (N = 17) (24, 27-30, 33, 35-37, 39-41, 43, 44, 47, 49, 55-57). Of note, the majority of the included clinical trials were of high quality (low risk of bias) . Therefore, the conclusion drawn from those data would be of higher credibility.
In our meta-analysis, we noted that the risk of all-grade neurological toxicities in the PD-1/PD-L1 inhibitors group was lower compared to the chemotherapy arm. These neurological toxicities included peripheral neuropathy, peripheral sensory neuropathy, dysgeusia, paraesthesia, and polyneuropathy ( Figure 3A1, 4A1, 5A1, 6A1, S Figure 4A1, 8C). A similar observation was noted regarding peripheral neuropathy and peripheral sensory neuropathy of grades 3-5 ( Figure 3A2, 4A2) (10,(49)(50)(51). These findings highlight the need to pay more attention to the risk of neurological toxicities associated with chemotherapy in clinical practice, especially for docetaxel (26, 30-32, 34, 40, 41, 43, 44, 46). The subgroup analyses suggested that the encountered high heterogeneity in our analyses (I 2 =62%) might be related to the NSCLC subgroup (I 2 = 75%, Figure 3A1) (26,30,40). In addition, the treatment plans involved in the three NSCLC clinical trials included in the comprehensive analysis belonged to different treatment lines (first, second, or third line); this probably might be a potential contributor to the heterogeneity of the result (I 2 = 75%, Figure  3A1) (26,30,40). That being said, no obvious risk of publication bias was found from the corresponding funnel plots (Supplementary Figure 1A1, 2A1, 3A1, B1, 5A1, 9C). Interestingly, for headache, dizziness, and Guillain-Barreś yndrome, the risk was found to be of no significance (Supplementary Figure 4A, 6A, 8A) (22, 23, 25-27, 33, 34, 36, 38, 41-44, 47, 48, 51-57), which meant that the risk trend of the aforementioned three neurological toxicities caused by PD-1/ PD-L1 inhibitors was similar to that of the chemotherapy group. This finding is novel and has not been reported nor investigated by other studies in the literature.
Furthermore, Guillain-Barrésyndrome was reported in five PD-1/PD-L1 groups (all cases were reported in the PD-1/PD-L1 group), while the incidence rate of the control groups was 0 (25,27,33,42,51). No statistically significant difference was noted and this could be attributed to the small number of included trials and the sensitivity of the analysis method (25,27,33,42,51). That being said, we cannot rule out the possibility that Guillain-Barrésyndrome is a unique neurological toxicity of PD-1/PD-L1 inhibitors. Despite the fact that our analyses revealed some statistically insignificant results; however, the reported risks should not be ignored in clinical practice, and more attention should be paid to those fatal and rare reported neurological toxicities (25,27,33,42,51). These results might be of significant value in clinical practice. Once Guillain-Barreś yndrome happened, we should first consider its associations with PD-1/PD-L1 inhibitors (25,27,33,42,51).

STRENGTHS AND LIMITATIONS Strengths
This article was designed according to the PRISMA guidelines. The literature searching process was carried out in accordance with the PICOS principle. We strictly limited the selection criteria to clinical trials and checked the accuracy of the extracted data carefully. The quality of the majority of the included trials was high. Subgroup analyses were put into practice as much as possible. Therefore, our meta-analysis provided a much more reliable evaluation of the relationship between PD-1/PD-L1 inhibitors and the associated risk of neurological toxicities in cancer patients compared to available evidence in the literature.

Limitations
First, compared with the control group, all the analysis results just showed the relative risk of neurological toxicities in cancer patients. Even when the associated risk of neurological toxicity was lower than that of the control group, it did not mean that PD-1/PD-L1 would not cause neurological toxicity in the experimental group. Second, the low number of studies that reported the data of certain neurological toxicities, along with the unavailability of relevant data, made it difficult to conduct a meta-analysis in this regard. Therefore, a definite conclusion could not be reached.

CONCLUSION
Our comprehensive review showed that PD-1/PD-L1 inhibitors alone exhibited lower neurological toxicities than chemotherapy. However, in terms of headache, dizziness, and Guillain-Barreś yndrome, the risk trends were similar between both interventions. Regarding PD-1/PD-L1 inhibitors plus chemotherapy, the risk of neurological toxicities would be increased, especially for peripheral neuropathy of grades 3-5.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding authors.

AUTHOR CONTRIBUTIONS
The corresponding authors (YS and GS) had the right to deal with all the data and were responsible for the decision to submit this manuscript for publication. YT, AG, SW, SZ, and XY had the full data of the manuscript. YT, AG, SW, and SZ were responsible for checking and evaluating the quality of the data and included studies. YT was assigned to write the text of this manuscript. All authors contributed to the article and approved the submitted version.