Immune-Related Adverse Events Associated with Anti-PD-1/PD-L1 Treatment for Malignancies: A Meta-Analysis

1Passive immunotherapy for cancer, which involves the transfer of tumor-targeted mono-antibodies and donor T cells, has demonstrated clinical benefits against a variety of solid and hematological malignancies. In contrast to passive immunity techniques, active immunotherapy strategies aim to augment self/anti-tumor responses (Mellman et al.,; Callahan et al.,; Michot et al., ). However, cancer cells always evolve to exploit multiple pathways to resist immune attack. These immunosuppressive pathways are referred to as “immune checkpoints,” which terminate immune responses in the normal physiological state (Pardoll, ). Notably, blockade of these immune checkpoints is frequently reported to be superior to traditional treatments in improving survival of oncologic patients (Borghaei et al.,; Ribas et al.,; Fehrenbacher et al.,; Rittmeyer et al., ).

2Cytotoxic T lymphocyte-associated antigen 4 (CTLA-4) is expressed exclusively in T cells, where it inhibits their activation (Leach et al., ). Antagonistic anti-CTLA-4 antibodies, such as ipilimumab and tremelimumab with proven clinical benefits, have been developed (Pardoll,; Callahan et al., ). Ipilimumab has been approved by the FDA (Food and Drug Administration) for the treatment of advanced or recurrent melanomas. Programmed cell death protein 1 (PD-1) and its ligand PD-L1 are immune check points that physiologically limit autoimmunity during inflammatory responses (Leach et al., ). Monoclonal antibodies against PD-1 or PD-L1, such as nivolumab, pembrolizumab, atezolizumab, avelumab, and durvalumab, have been produced. These PD-1/PD-L1 inhibitors could result in a stable regression of malignancy (Borghaei et al.,; Brahmer et al.,; Motzer et al.,; Ribas et al.,; Weber et al.,; Fehrenbacher et al.,; Herbst et al.,; Rittmeyer et al., ). Currently, the FDA has approved the use of nivolumab for advanced melanoma, renal cell carcinoma (RCC), and non-small cell lung cancer (NSCLC), and pembrolizumab for advanced melanoma and NSCLC (Costa et al., ). Furthermore, combined blockade of CTLA-4 and PD-1/PD-L1 appears to achieve additional clinical benefits (Callahan et al., ).

3Treatment with immune-checkpoint inhibitors could cause immune-related adverse events (irAEs) by unbalancing the immune system (Chen et al.,; Marrone et al.,; Kumar et al., ). The definition of irAEs is different from that of AEs and treatment-related AEs. AEs that occur potentially due to immunological effects were defined as irAEs. irAEs could be serious, requiring suspension of immunotherapy and possibly leading to death (Bertrand et al.,; Chen et al.,; Eigentler et al., ). Previous studies indicated that the occurrence of irAEs induced by anti PD-1/PD-L1 agents is related to tumor types and organs and is dose-independent (Michot et al., ). Additionally, there are reports of delayed occurrence of irAEs after treatment with anti PD-1/PD-L1 agents (Nishino et al., ). Therefore, there is a need to understand the characteristics of irAEs associated with anti PD-1/PD-L1 treatments to help us manage them appropriately.

4In this study, we present a systematic review and meta-analysis and aim to assess the incidence and characteristics of irAEs in malignancies treated with anti-PD-1/PD-L1 agents.

5The protocol of this meta-analysis was registered at PROSPERO, International Prospective Register of Systematic Reviews (crd.york.ac.uk/prospero, Identifier: 42016051745).

6MEDLINE, EMBASE, and Cochrane databases were searched to determine potentially eligible studies from database inception to March 1, 2017. We put no restriction on language. The following search terms were used: “safety OR security OR side effects OR adverse events AND (anti-PD-1 OR anti-PD-L1 OR nivolumab OR pembrolizumab OR BMS-936559 OR atezolizumab OR avelumab OR durvalumab)” in MEDLINE and EMBASE, and “anti-PD-1 OR anti-PD-L1 OR nivolumab OR pembrolizumab OR BMS-936559 OR atezolizumab OR avelumab OR durvalumab” in Cochrane databases. We also screened the references of included studies and relevant reviews to find potential studies. There were too many medical records reporting the side effects of anti-PD-1/PD-L1 agents, and it was difficult to check all conference abstracts in the 2 years prior to this study.

7Eligible studies were required to meet the following criteria: (1) Prospective clinical trials that reported irAEs, which were clearly identified as “AEs of special interest,” “immune-mediated adverse events,” or “selected treatment-related adverse events of special interest,” but not “treatment-related AEs” or “drug-related AEs.” Additionally, every corresponding author of a potential irAE-related study was e-mailed and asked to provide more information about the irAEs and (2) patients were diagnosed with malignancies that were treated with anti-PD-1/PD-L1 agents. Oncologic therapy prior to anti-PD-1/PD-L1 treatment was acceptable.

8The exclusion criteria were: (1) Non-oncologic patients (e.g., hepatitis C virus-related patients) treated with anti-PD-1/PD-L1 agents; (2) Oncologic patients treated with anti-PD-1/PD-L1 agents combined with other treatments simultaneously; (3) Retrospective studies, meeting abstracts, case reports, basic research, reviews, systematic reviews and meta-analysis, letters, editorials, and expert opinions; and (4) Duplicate publications or unpublished studies.

9The titles and abstracts of all studies retrieved were independently reviewed by two authors. Next, the full texts of all potentially eligible studies were assessed. A standardized, pre-piloted form was used to extract relevant information from the included studies.

10The primary outcomes for this meta-analysis were incidence and risk ratio (RR) of irAEs and their grade (1–5; recorded according to Version 3 or 4 of the Common Terminology Criteria for Adverse Events of the National Cancer Institute). We considered that Grades ≥3 were evaluated as high grade or severe grade. The secondary outcome was incidence of death due to irAEs. Any discrepancies were solved by discussion. Missing data were requested from the principle investigator by e-mail.

11Two independent investigators assessed the risk of bias for the included studies according to the Cochrane Handbook for Systematic Reviews of Interventions (Higgins et al., ). The following components were assessed: sequence generation, allocation concealment, blinding, completeness of outcome data, incomplete outcome data, and other sources of bias. Disagreements were resolved by discussion among investigators until a consensus was reached.

12The incidence and RR of irAEs were estimated for the included studies in this meta-analysis. We pooled the incidence of irAEs in malignancies treated with anti-PD/PD-L1 agents. Heterogeneity between studies was assessed by Q test and I2 statistics. If the I2 value was less than 50%, the meta-analysis was performed using the fixed-effects model. Otherwise, the random-effects model was adopted. Potential publication bias was examined by funnel plots and Egger's test. Incidence was calculated using R software [R version 3.3.2 (2016-10-31)] with package Meta and Metaprop function. RR was calculated using Review Manager 5.3 (Nordic Cochrane Centre, Copenhagen, Denmark).

13Our search strategy identified 1,991 potential articles. Five-hundred and 51 studies were excluded owing to duplicates. The remaining articles were screened for titles and abstracts, and 1080 articles were removed based on our inclusion or exclusion criteria. Furthermore, 314 studies were dropped because they did not contain our data of interest. Finally, 46 studies were included in our meta-analysis. The study selection is shown in Figure 1.

14The detailed information of the clinical trials is presented in Table 1. Overall, 46 trials comprising 12,808 oncologic patients treated with anti-PD-1/PD-L1 agents were included in this meta-analysis (Table 1) (Brahmer et al.,,,; Topalian et al.,,; Robert et al.,,,; Borghaei et al.,; Garon et al.,; Gettinger et al.,,; Hamanishi et al.,; Larkin et al.,; McDermott et al.,,; Motzer et al.,; Patnaik et al.,; Ribas et al.,,; Rizvi et al.,; Weber et al.,,; Balar et al.,; Chatterjee et al.,; Choueiri et al.,; Fehrenbacher et al.,; Ferris et al.,; Herbst et al.,; Hua et al.,; Kaufman et al.,; Lesokhin et al.,; Massard et al.,; Muro et al.,; Nanda et al.,; Reck et al.,; Rosenberg et al.,; Seiwert et al.,; Shimizu et al.,; Younes et al.,; Bellmunt et al.,; Hui et al.,; Morris et al.,; Plimack et al.,; Sharma et al., ). Among these studies, there were 35 phase I/II clinical trials and 11 phase III clinical trials. Nine percent (4/46) of the studies were blind and 13% (6/46) were single-arm. There were 23 randomized trials and 18 multicentric studies. There were 21 studies involving nivolumab treatment (4,856 patients), 17 studies involving pembrolizumab treatment (5,581 patients), 5 trials involving atezolizumab treatment (2,015 patients), one trial involving durvalumab treatment (61 patients), and one trial involving avelumab treatment (88 patients).

15The global incidence of irAEs of any grade and severe grade in patients treated with anti-PD-1/PD-L1 agents was 26.82% (95% CI, 21.73–32.61; I2, 92.80) and 6.10% (95% CI, 4.85–7.64; I2, 52.00), respectively (Figure S1). The incidence of any grade of irAEs was 18.50% (95% CI, 15.41–22.06; I2, 77.00) with pembrolizumab and 16.67% (95% CI, 4.75–44.50; I2, 96.00) with atezolizumab, which was lower than the incidence of irAEs with nivolumab (48.00%, 95% CI, 40.13–55.98; I2, 78.50). However, there were only slight differences in the incidence of severe grade irAEs, which was 8.25% (95% CI, 5.27–12.69; I2, 66.40) with nivolumab, 5.10% (95% CI, 3.58–7.22; I2, 48.40) with pembrolizumab, and 5.28% (95% CI, 3.62–7.64; I2, 0.00) with atezolizumab (Figure 2).

16We analyzed the incidence of irAEs associated with PD-1 blockade involving the skin, endocrine system, gastrointestinal tract, liver, kidney, and lung. The most common irAEs of all grades caused by nivolumab involved the skin, followed by the gastrointestinal tract, endocrine system, liver, lung, and kidney. The severe grade irAEs caused by nivolumab mostly occurred in the gastrointestinal tract and liver. However, we found it difficult to draw robust conclusions owing to limited data on pembrolizumab. The detailed information of organ-specific irAEs is presented in Tables S3–S8.

17The most frequent irAEs during treatment with nivolumab were pruritus and rash, with an incidence of approximately 13%. However, pembrolizumab tended to have a lower incidence, less than 5% of any cutaneous irAEs of all grade. Notably, severe grade incidence of any irAEs was ~2% with nivolumab, pembrolizumab, and BMS-936559 (Table S9).

18The most common endocrine disorder associated with PD-1 blockade with nivolumab and pembrolizumab was hypothyroidism. PD-L1 blockade was associated with nearly the same incidence of irAEs for every endocrine disease. For all the drugs, the incidence of severe grade irAEs occurring in the endocrine system was ≤2% (Table S10).

19Diarrhea was the most frequent irAEs in patients treated with nivolumab or BMS-936559, and its incidence was approximately 10–13%. Pembrolizumab was likely to have a relatively low risk of diarrhea of any grade, and its incidence was less than 4%. In addition, the incidence of high-grade colitis was ~1% regardless of the treatment type. Pancreatitis was rare, with an incidence of less than 1% for either all grade or severe grade across all drug types (Table S11).

20The incidence of increased aspartate aminotransferase (AST) or alanine aminotransferase (ALT) levels ranged from 4 to 5% with nivolumab treatment. This incidence of increased AST or ALT levels was around 2% in patients treated with pembrolizumab or atezolizumab. However, the incidence of elevated AST/ALT/blood bilirubin and hepatitis was ≤2% regardless of treatment type (Table S12).

21Pneumonitis was the most common irAE related to the lung. The incidence of pneumonitis of any grade was approximately 3% and that of severe grade was approximately 1% during treatment with nivolumab or pembrolizumab. Treatment with atezolizumab was less likely to be associated with pneumonitis, with an incidence of ~1% for any grade and <1% for severe grade (Table S13).

22The incidence of any grade of nephritis was very low (less than 1%) regardless of the treatment type. There were reports of acute renal failure and renal failure only with nivolumab treatment, although the incidence was only ~2% for all grade and ~1% for severe grade (Table S14).

23We analyzed the incidence of irAEs with regard to the tumor being treated, including melanomas, NSCLCs, RCCs, and urothelial tumors. The highest incidence of any irAE associated with nivolumab and pembrolizumab occurred during the treatment of melanomas and NSCLCs, respectively. In contrast, the most frequent severe grade irAEs occurred with nivolumab treatment of melanomas.

24The global incidence of all grade irAEs was 59.06% (95% CI, 51.53–66.18; I2, 50.00) with nivolumab and 21.33% (95% CI, 12.94–33.10; I2, 90.80) with pembrolizumab. The incidence of severe grade irAEs was low, ranging from 6.94% (95% CI, 4.69–10.16; I2, 8.3) with nivolumab to 3.80% (95% CI, 2.67–5.39; I2, 0.0%) with pembrolizumab (Figure S2).

25The most frequent irAEs in patients treated with nivolumab were pruritus, followed by diarrhea and rash. Pembrolizumab treatment was mostly associated with hypothyroidism. In patients treated with nivolumab, the most frequent severe irAE was AST/ALT increase, with an incidence of ~3%, while the other irAEs were rare (<1%). The severe irAEs associated with pembrolizumab administration were maculopapular rash, colitis, increased AST, and hepatitis, occurring at 1–2% (Tables S15–S20).

26The most common irAEs associated with nivolumab were rash and diarrhea, and those associated with pembrolizumab were hypothyroidism and hyperthyroidism. The incidence of any grade of irAEs with pembrolizumab was higher than that with nivolumab. Severe grade irAEs were rare (Tables S21–S26).

27Diarrhea and colitis were the most frequent any grade irAEs and severe grade irAEs, respectively, associated with nivolumab treatment (Tables S27–S32).

28The global incidence of any grade irAEs in urothelial cancers treated with pembrolizumab and atezolizumab was 17.06% (95% CI, 13.21; 21.76; I2, 0.00) and 9.09% (95% CI, 5, 75; 14.09; I2, 50.60), respectively. The number of studies providing statistics on organ-specific irAEs in patients with urothelial cancer was relatively small. From the limited cases, we found that the incidence of all grade irAEs and severe grade irAEs in specific organs were very rare, with incidence of ≤ 1–3% (Table S33).

29The incidence of all grade irAEs in patients treated with 3 mg/kg nivolumab was 58.08% (95% CI, 34.05–78.81; I2, 84.80), while that in patients treated with 1 mg/kg nivolumab was 70.00% (95% CI, 21.76–95.14; I2, 76.50). This dose-dependent effect was also observed in severe grade irAEs, with incidence of 7.84% (95% CI, 2.62–21.22; I2, 81.30) at 3 mg/kg and 16.88% (95% CI, 2.11–65.64; I2, 88.00) at 1 mg/kg (Table S1). Owing to limited data, we could not calculate the RR for nivolumab. However, we hypothesized that dose would not affect the development of irAEs as the 95% CI overlapped greatly.

30In the treatment with pembrolizumab, the incidence of all grade irAEs was 18.93% (95% CI, 15.69–22.67; I2, 0.00) at 2 mg/kg, while that at 10 mg/kg was 15.16% (95% CI, 11.85–19.20; I2, 56.00). It was also observed that the incidence of severe grade irAEs was 1.12% (95% CI, 0.16–7.54) with 2 mg/kg pembrolizumab and 4.07% (95% CI, 1.10–13.91; I2, 69.70) with 10 mg/kg pembrolizumab (Table S2). The RR of developing any grade of global irAEs with pembrolizumab at 10 mg/kg vs. that at 2 mg/kg was not statistically different [RR, 0.91 (0.70, 1.17); P = 0.46; Figure S3].

31Deaths due to irAEs were reported in 33 studies comprising 5,090 patients. Death occurred in nine of 5,090 patients, with an incidence of 0.74 (95% CI, 00.44; 1.25; I2, 38.20). Four patients died owing to pembrolizumab treatment, with an incidence of 0.89 (95 % CI, 00.49; 1.62; I2, 0.00). Five patients died following nivolumab treatment, and four of the deaths were due to pneumonitis. The incidence of deaths following nivolumab treatment was 0.77 (95% CI, 00.32; 1.87; I2, 56.20). Drug/treatment-related deaths and studies not reporting deaths were not included in this meta-analysis.

32Risk of bias graphs and risk of bias summaries from Review Manager 5.3 were used to evaluate the methodological qualities of the studies. Blinding of participants and personnel was evaluated as a low risk item because many studies were dose-escalation and single-arm trials. The overall risk of bias was evaluated as low risk. Therefore, the quality of the studies was satisfactory (Figure S4).

33The funnel plot and Egger's test for publication bias showed a symmetric distribution of trials on either side of the funnel. The Begg's test and Egger's test (P = 0.7355) also indicated that no significant publication bias existed in this meta-analysis (Figure S5).

34In the present study, we systematically characterized the occurrence of irAEs in oncologic patients treated with anti-PD-1/PD-L1 agents according to different targets, drug types, drug dose, organ-specificity, and the tumor type treated. Although immune-check point inhibitors improved clinical outcomes in oncology, their irAEs should not be neglected (Bertrand et al.,; Michot et al.,; Costa et al.,; Tie et al., ). A previous meta-analysis indicated a high risk of irAEs with anti-CTLA-4 treatment, with incidence of 72% (95% CI, 65–79%) for all grade irAEs and 24% (95% CI, 18–30%) for high grade irAEs, and the occurrence of irAEs following anti-CTLA-4 treatment was dose-dependent (Bertrand et al., ). However, we found a much lower incidence of irAEs in patients treated with any of the anti-PD-1/PD-L1 agents, regardless of the drug type or dose. This indicates better tolerance of PD-1/PD-L1 inhibitors in cancer patients. Moreover, severe grade irAEs caused by any anti-PD-1/PD-L1 agent had a very low incidence, ranging from 5 to 8%. This helped in maximizing patient safety and treatment duration, because serious irAEs may result in the discontinuation of anti-PD-1/PD-L1 treatment in patients (Kumar et al., ).

35A wide spectrum of irAEs was induced by all anti-PD-1/PD-L1 agents, including those involving the skin, gastrointestinal tract, lung, kidney, and liver. Michot et al. suggested that irAEs of grade I-II mainly affected the skin and gastrointestinal tract, and severe grade irAEs mostly occurred within the digestive system (Michot et al., ). We also found organ-specific irAEs associated with PD-1/PD-L1 blockade; however, our results indicated that these organ-specific characteristics were related to drug types. During nivolumab treatment, the most common irAEs of all grades were skin- (e.g., pruritus, rash) and gastrointestinal tract-related (e.g., diarrhea), with an incidence of approximately 13%. In contrast, during pembrolizumab treatment, the irAE with the highest incidence (~8%) was hypothyroidism (Tables S3–14). Additionally, the incidence of irAEs caused by PD-1 blockade was tumor-specific. Pneumonitis of all grades more frequently occurred during NSCLC and RCC treatment than during melanoma treatment. We also found a higher incidence of pneumonitis during NSCLC (4.15%, 95% CI, 3.36–5.12%) and RCC (7.19%, 95% CI, 4.54–11.21%) treatments than during melanoma treatment (2.18%, 95% CI, 1.67–2.84%; Tables S19, S25, and S31). Moreover, the incidence of all grade pruritus and diarrhea was lower in NSCLC than in melanoma and RCC during nivolumab treatment (Tables S15, S17, S21, S23, S27, and S29), confirming that the irAEs were tumor-specific. Our results correspond well with those published previously (Nishino et al., ) and are arguably more convincing owing to the larger number of studies involved.

36In addition to the irAEs caused by anti-PD-1/PD-L1 agents, the response rate of this immunotherapy was another concern (Gibney et al.,; Topalian et al., ). Fortunately, genomic sequencing, RNA sequencing, and whole exome sequencing were of great importance in identifying predictive biomarkers for anti-PD-1/PD-L1 agent responses Gibney et al.,; Topalian et al.,; Hugo et al.,. Similarly, we could utilize the methods above to identify the genes associated with a higher risk of irAEs, especially with severe grade irAEs, which can sometimes be fatal. We conclude that the pattern of irAE occurrence was dependent on drug type, tumor type, irAE grades, target organ, and the particular irAE. However, the dataset in our analysis could ideally have been larger. As there are many ongoing studies on anti-PD-1/PD-L1 agents, we recommend that the factors related to irAEs be recorded in these clinical trials.

37There were some limitations in our study. First, the variability of clinical settings was high. Adopting the random-effects model was an important step to control variability. In addition, the main source of variability was significantly reduced when subgroup analysis was performed based on irAE grades, organs, drug types, and tumor types. These results provide clinical value for physicians to be wary about the occurrence of irAEs when treating specific tumors with different drugs. Second, the number of studies and patients were limited in the subgroup analysis performed based on drug type, organ-specific irAEs, particular irAEs, and tumor type. Third, some of the included studies were not random clinical trials, introducing some bias in our analysis. Last, the incidence of irAEs was not clear in patients with preexisting autoimmune diseases according to our analysis. Because all the included studies excluded patients with active autoimmune disease or a documented history of autoimmune disease or syndrome that requires systemic steroids or immunosuppressive agents, except vitiligo or resolved childhood asthma/atopy. Therefore, the safety of PD-1 inhibitors in patients with pre-existing autoimmune diseases needed to be assessed further in experimental or clinical studies.

38In conclusion, our meta-analysis provided a statistical overview of irAEs in oncologic patients treated with anti-PD-1/PD-L1 agents. The pattern of irAE occurrence was organ-specific and related to the drug and tumor types.

39Conception and design: CY and NL; Collection and assembly of data: PW and YC; Data analysis and interpretation: All authors; Manuscript writing: All authors; Final approval of manuscript: All authors; Accountable for all aspects of the work: All authors.

40The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.