A systematic search of the Embase and PubMed databases from inception to June 2018 was conducted with no language restrictions. Conference proceedings from the American Society of Clinical Oncology and the European Society for Medical Oncology were also searched. The search keywords and medical subject headings used were (1) Nivolumab, Opdivo, ONO-4538, BMS-936558 and MDX1106; and (2) clinical trial. Three investigators independently performed the initial search, carefully screened the titles and abstracts for relevance, and identified trials as excluded, included and uncertain. For the uncertain studies, the full-texts were reviewed for the confirmation of eligibility. Any discrepancy was resolved by discussion.

Both the inclusion and exclusion criteria were prespecified. To be eligible, studies had to meet the following criteria: (1) population: clinical phase 2 and phase 3 prospective trials involving adult patients (> 18 years old) with solid tumors; (2) intervention: at least one arm with nivolumab monotherapy; and (3) outcomes: available information on sample size, SAEs and FAEs. Phase I trials were not included because of the small sample size of patients and various doses in these studies. Other studies on this topic, including review articles, preclinical papers, early versions of data published later, editorials, and correspondences, were not included (Fig. 1). When multiple publications of the same study occurred, only the most recent and/or most complete study was included.Fig1Fig. 1

Flow chart of the eligible trials included in this study

An FAE is defined as death caused by nivolumab treatment, and an SAE refers to any adverse event (AE) that may lead to death, hospitalization or the prolongation of hospitalization, a life-threatening condition, congenital anomaly or birth defect, disability or permanent damage, or jeopardization of the patient’s health in a way that requires treatment to prevent one of the other outcomes. The attribution of FAEs or SAEs as treatment-related or disease-related depended on the decision of the primary investigators in the eligible studies. The information on ‘treatment-related SAE’ and ‘treatment-related FAE’ was extracted from the original publications. For all of the eligible studies, we also extracted the following information: first author’s name, year of publication, phase of the trial, cancer type, number of patients enrolled, median age, gender, treatment strategy, median treatment duration and median follow-up (Table 1). In addition, the median overall survival (OS) in each treatment arm, the hazard ratio (HR) for the OS, the number of participants evaluated for safety, the number of FAEs, and the number of SAEs were listed in Table 2. All data were extracted independently by three reviewers, and any discrepancies were resolved by discussion and consensus.Tab1

Clinicopathological characteristics of trials included in this study

To calculate the incidence, the number of patients receiving nivolumab and the number of SAEs/FAEs were extracted from eligible studies. For the OR calculations, patients treated with nivolumab were compared with those assigned to a chemotherapy/placebo arm in the same trial. Four studies [13–16] were not included in the OR analysis because ipilimumab was administered in the control arms. When the trials reported no SAE/FAE in one arm, a classic half-integer continuity correction was used for the calculation.

Statistical heterogeneity across the trials was evaluated by Cochran’s Q statistic. The I ² statistic was calculated to assess the extent of inconsistency attributable to the heterogeneity across different studies [17]. The assumption of homogeneity was considered invalid for I ²  > 25% or P < 0.05. The pooled ORs and incidences were calculated using a fixed-effects model or a random-effects model, depending on the heterogeneity of the included trials. To check the impact of various clinicopathological variables on FAEs, we further conducted post hoc subgroup analysis based on the underlying malignancy and clinical phase.

All analyses were conducted using MedCalc 13.0 (MedCalc Software, Belgium) and Stata 12.0 (StataCorp, USA). Two-sided P < 0.05 were considered statistically significant.

A total of 4175 potentially relevant articles were identified from the initial search, including 2140 studies from PubMed and 2035 trials from Embase. A total of 1982 articles were excluded because of duplications. After the titles and abstracts were screened, 2086 studies did not meet the inclusion criteria. Upon further review of the whole texts of the remaining 107 potentially eligible articles, 21 trials (36 arms) with 6173 patients were enrolled for the final analysis (Fig. 1).

Within these 21 eligible studies, 9 were phase 3 RCTs, 3 were phase 2 RCTs, and 9 were single-arm phase 2 trials. Of the 36 arms, 7 arms were with chemotherapy/placebo, 3 arms were with nivolumab + ipilimumab, and 2 arms were with ipilimumab (Table 1), leaving a total of 24 arms of patients to be evaluated for the incidence of SAEs/FAEs with nivolumab. The number of patients per nivolumab arm with associated safety data ranged from 16 to 453 (average of 156), with a total of 3732 patients with SAE/FAE information available (Table 2). The most common cancer types were lung cancer (6 studies, 1623 patients) and melanoma (5 studies, 2366 patients). The dosage of nivolumab was 3 mg/kg every 14 days in all but two studies. In one trial, the applied dosage was 1 or 3 mg/kg [18], and dosages of 0.3, 2 and 10 mg/kg were applied in another trial [19]. The treatment duration with nivolumab ranged from 1.5 months to 11.9 months (average of 5.3 months).

Of the 21 trials included in our study, SAE information was unavailable in 6 trials; therefore, 15 studies (16 arms) were eligible for the analysis of SAE incidence (Table 2). A total of 1695 subjects receiving nivolumab monotherapy were included, and 196 SAEs were reported. Using a random-effects model (significant heterogeneity, I ²  = 58.2%; P = 0.01), the overall incidence of SAEs was 11.2% (95% CI, 8.7–13.8%; Fig. 2A). The incidence of SAEs varied significantly with cancer type (P < 0.001) and clinical phase (P < 0.001). The incidence rates for different cancer types were, in decreasing order, ovarian cancer (25.0%), sarcoma (19.1%), colorectal cancer (12.2%), lung cancer (11.8%), Hodgkin lymphoma (11.3%), gastric/gastro-esophageal cancer (10.9%) and melanoma (9.6%). The causes of nivolumab-related SAEs are summarized in Table 3. The most common SAEs occurred in the respiratory (n = 42, 21.4%), gastrointestinal (n = 15, 7.7%), and hepatic systems (n = 13, 6.6%). The most common SAEs were pneumonitis (n = 16, 8.2%), interstitial lung disease (n = 11, 5.6%) and colitis (n = 7, 3.6%). It was noted that pneumonitis occurred in lung cancer (n = 14) and gastric or gastro-esophageal junction cancer (n = 2); interstitial lung disease occurred in lung cancer (n = 6), gastric or gastro-esophageal junction cancer (n = 4) and Hodgkin lymphoma (n = 1); and colitis occurred in lung cancer (n = 3), gastric or gastro-esophageal junction cancer (n = 2), colorectal cancer (n = 1) and melanoma (n = 1).Fig2Fig. 2

Overall incidence of SAEs (a) and FAEs (b) associated with nivolumab treatment FAE, fatal adverse event; SAE, serious adverse event

In this study, 3386 cancer patients receiving nivolumab from 21 trials (24 arms) were included in the analysis of the incidence of FAEs. A total of 13 FAEs were reported. Using a fixed-effects model, the overall incidence of FAEs was 0.3% (95% CI, 0.1–0.5%; Fig. 2B). No significant heterogeneity was observed (heterogeneity test, I ²  = 0.0%; P = 0.96). The incidence rates for various tumor types were, in decreasing order, gastric/gastro-esophageal cancer (1.27%), head and neck cancer (0.85%), lung cancer (0.55%) and melanoma (0.09%). The causes of nivolumab-related FAEs were 4 cases of pneumonia and one case each of encephalitis, multiorgan failure, hypercalcemia, hepatitis, cardiac arrest, exertional dyspnea, ischemic stroke, neutropenia, and unknown reason (Table 3).

Of the 21 eligible studies, 9 trials were single-arm phase 2 studies, 5 trials included other immunotherapy as controls, and the remaining 7 studies were eligible for OR analysis. Among the 3397 patients (nivolumab: 1863; control: 1534) in the 7 eligible RCTs, the overall OR of FAEs induced by nivolumab was 0.61 (95% CI, 0.27–1.39, P = 0.24; incidence 0.5% versus 0.8%; Fig. 3B), indicating that the risk of nivolumab-related FAEs was not significantly different from those in the control arms. No significant heterogeneity was identified, despite clear disparities in cancer type, treatment duration and control type (Q = 2.47; I ²  = 0.0%; P = 0.87). Because no significant heterogeneity was observed, subgroup analyses were not conducted for FAEs. To account for any potential clinical heterogeneity not detected by our statistical tests, we also pooled the data using a random-effects model, and the OR and 95% CI remained unchanged.