The PubMed website was used to identify clinical trials involving PD-1/PD-L1 inhibitors for solid tumor patients. Relevant randomized clinical trials (RCTs), reported from inception to July 31, 2019, were collected by using search keywords and Medical Subject Headings (MeSH) terms pertinent to the intervention of interest, such as cancer, tumor, PD-1/PD-L1, nivolumab, opdivo, pembrolizumab, immune checkpoint inhibitor, Keytruda, Imfinzi, MK-3475, atezolizumab, Tecentriq, MPDL3280A, avelumab, Bavencio, durvalumab, and BMS-963558. Furthermore, some RCTs that could not be found on the PubMed website were obtained by searching and checking references of other systematic reviews, meta-analyses, and conference proceedings of the American Society of Clinical Oncology, the European Society for Medical Oncology, the American Association for Cancer Research, and the World Conference on Cancers.

According to our analysis design, the inclusion criteria were as follows: (1) RCTs would be preferred choices, (2) PD-1/PD-L1 inhibitors were used as an anti-tumor therapy, (3) The treatment regimen of the control group included anti-tumor drugs or placebo, (4) All enrolled patients were diagnosed with solid tumors rather than hematological malignancies, (5) At least one of Immune-related Digestive System Inflammation (colitis, hepatitis, and pancreatitis) was reported, (6) The results of the clinical trials are reported in English or reported in other languages and English.

Four investigators (Zewen Zhang, Xiaowei Yang, Donghua Li, Li Zhang) were designated to determine the eligibility and duplicate independently by checking titles and abstracts of enrolled studies. The data categories of enrolled studies were collected as follows: first author, the year of publication, study name and number, treatment regimen, number of evaluable cases, related drug name, phase stage, tumor type, incidence rate of immune-related digestive system inflammation.

Both all-grade and grade 3–5 immune-related digestive system inflammation were taken into account for the final comprehensive meta-analysis. The basic characteristics of all enrolled studies would be summarized in Table 1.

Risk Ratio (RR) was used to assess the risk of developing immunological gastrointestinal inflammation. 95% confidence interval (CI) were calculated by random effect (RE). P < 0.05 was regarded as statistically significance. Statistical tests were all two-sided. Subgroup analysis would be performed according to the type of tumor, treatment plan, and specific drug name. Cochrane's Q and the I² statistic, proposed by Higgins and colleagues, were used for checking the heterogeneity among enrolled trials (37, 38). I² values <25, 25–50, and >50% indicated low, medium and high heterogeneity, respectively. Newcastle-Ottawa scale, Funnel plot and Egger's test were used for assessing the bias of the analysis result (37, 39–42). Four investigators (Zewen Zhang, Xiaowei Yang, Donghua Li, Li Zhang) were appointed to evaluate the quality of all enrolled trials.

The evaluation indicators related to the quality of the included trials, named Newcastle-Ottawa scale, including random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, and selective outcome reporting, would be checked one by one and summarized in a figure (42). Review Manager 5.3, proposed by the Cochrane Collaboration, was used for the final comprehensive analysis.

All grade of colitis was evaluated first (11–20, 22–36). Clinical trials included in the study were divided into 7 groups according to treatment methods for the final comprehensive analysis (Figure 1). Compared with chemotherapy, PD-1/PD-L1 inhibitors led to an increase in the incidence risk of colitis (RR = 2.43, 95% CI: [1.23, 4.82], I² = 0%, Z = 2.54 (P = 0.01); Figure 1A), especially in clinical trials related with combined chemotherapy (RR = 2.77, 95% CI: [1.08, 7.08], I² = 0%, Z = 2.12 (P = 0.03); Figure 1A) (12, 14, 19, 20, 26–28, 30, 32–34). Similar incidence trend could also be seen when PD-1/PD-L1 inhibitors were combined with chemotherapy (RR = 2.62, 95% CI: [1.25, 5.48], I² = 0%, Z = 2.56 (P = 0.01); Figure 1B) (15–18, 36). No statistical significance could be found when Nivolumab combined with Ipilimumab were compared with Ipilimumab alone (RR = 1.20, 95% CI: [0.89, 1.63], I² = 5%, Z = 1.20 (P = 0.23); Figure 1C) (13, 24, 31). There was an obvious higher incidence risk of all grade colitis when PD-1 inhibitors were compared with placebo (RR = 5.49, 95% CI: [1.78, 16.91], I² = 0%, Z = 2.97 (P = 0.003); Figure 1D) (22, 23). Opposite incidence trend was seen in another two groups (Figures 1E,F). Whether compared with Nivolumab plus Ipilimumab or Ipilimumab alone, the incidence risk of colitis in the Nivolumab group was significantly lower than that of the control group, and the difference was statistically significant (RR = 0.15, Figure 1E; RR = 0.25, Figure 1F) (13, 24, 25, 29, 35). Meta-analysis was not performed in group G because only one group of clinical trial was enrolled (11). The corresponding funnel plots of RR were gathered in Supplemental Figure 1. Obvious heterogeneity was only found in Figure 1F (I² = 61%).

Then, we took the same method to deal with all the data for evaluating the incidence risk of grade 3–5 colitis (11–13, 15–20, 22–24, 26, 27, 29–36). However, the analysis results of statistical significance could be seen only in Figures 2E,F (13, 24, 29, 35), while no statistical significant results were seen in Figures 2A–D (12, 13, 15–20, 22–24, 26, 27, 30–34, 36). Compared with Nivolumab plus Ipilimumab, the incidence risk of colitis in the Nivolumab group was sificantly lower than that of the control group (RR = 0.11, 95% CI: [0.06, 0.22], I² = 0%, Z = 6.22 (P < 0.00001); Figure 2E), especially in the subgroup of melanoma (RR = 0.11, 95% CI: [0.05, 0.22], I² = 0%, Z = 6.08 (P < 0.00001); Figure 2E) (13, 24, 29, 35). Similar analysis results were shown in Figure 2F when Nivolumab was compared with Ipilimumab [RR = 0.11, 95% CI: [0.05, 0.23], I² = 0%, Z = 5.96 (P < 0.00001)] (13, 24, 35). The corresponding funnel plots of RR were gathered in Supplemental Figure 2. No obvious heterogeneity was found among all groups (I² = 0%).

Sixteen clinical trials with the information of all grade hepatitis were taken into account for meta-analysis (12–18, 20–25, 30, 31, 36). They were divided into seven groups according to treatment methods for the final comprehensive analysis (Figure 3). Compared with chemotherapy, the incidence risk of all grade hepatitis was higher in PD-1/PD-L1 inhibitors group (RR = 3.55, 95% CI: [1.65, 7.63], I² = 0%, Z = 3.24 (P = 0.001); Figure 3A) (12, 14, 20, 21, 30), especially in the subgroup of Pembrolizumab compared with combined chemotherapy (RR = 11.54, 95% CI: [1.51, 88.35], I² = 0%, Z = 2.35 (P = 0.02); Figure 3A) (12, 14). However, when PD-1/PD-L1 inhibitors plus chemotherapy was compared with chemotherapy, the analysis result of incidence risk was considered to be no significant (RR = 1.89, 95% CI: [0.86, 4.19], I² = 50%, Z = 1.58 (P = 0.11); Figure 3B) (15–18, 36), even if in each subgroup. Similar incidence trend could also be seen when PD-1 inhibitor was compared with placebo or Ipilimumab (Figures 3C,F) (22, 23, 25). There was an obvious higher incidence risk of all grade hepatitis when Nivolumab plus Ipilimumab were compared with Ipilimumab (RR = 8.54, 95% CI: [1.59, 45.79], I² = 0%, Z = 2.50 (P = 0.01); Figure 3D) (13, 24, 31). When the control group was Nivolumab, similar result was seen in Figure 3E [RR = 15.00, 95% CI: [1.99, 113.21], I² = 0%, Z = 2.63 (P = 0.009)] (13, 24). The corresponding funnel plots of RR were summarized in Supplemental Figure 3. Moderate heterogeneity was only found in Figure 3B (I² = 50%).

The same grouping and subgroup approach as before were taken for evaluating the incidence risk of grade 3–5 hepatitis. Fourteen clinical trials with the information of hepatitis were taken into account for the final meta-analysis (12–18, 20–24, 30, 36). Regardless of whether the experimental group was PD-1/PD-L1 plus chemotherapy or PD-1/PD-L1 inhibitor, the incidence risk of hepatitis in the PD-1/PD-L1 related experimental group was higher than that of the chemotherapy control group (RR = 5.52, 95% CI: [1.43, 21.38], I² = 0%, Z = 2.47 (P = 0.01), Figure 4A; RR = 2.21, 95% CI: [1.21, 4.06], I² = 0%, Z = 2.56 (P = 0.01), Figure 4B) (12, 14–18, 20, 21, 30, 36). Similar incidence trend happened in another two groups, which the experimental group was Nivolumab plus Ipilimumab and the control group was Nivolumab or Ipilimumab separately (RR = 6.85, 95% CI: [1.26, 37.19], I² = 0%, Z = 2.23 (P = 0.03), Figure 4D; RR = 11.00, 95% CI: [1.42, 84.95], I² = 0%, Z = 2.30 (P = 0.02), Figure 4E). No statistical significant results were seen in Figure 4C, which the control group was placebo (RR = 4.34, 95% CI: [0.75, 24.97], I² = 0%, Z = 1.64 (P = 0.10) (22, 23). The corresponding funnel plots of RR were summarized in Supplemental Figure 4. No obvious heterogeneity was found among all groups (I² = 0%).

Pancreatitis was reported in 10 enrolled clinical trials (12, 15–17, 20, 22, 27, 30, 31, 36). Among them, 8 clinical trials with the information of all grade pancreatitis were taken into account for the final meta-analysis (12, 15–17, 20, 27, 30, 36). Regardless of whether the experimental group was PD-1/PD-L1 plus chemotherapy or PD-1/PD-L1 inhibitor, the incidence risk of pancreatitis in the PD-1/PD-L1 related experimental group was of no statistical significance (RR = 2.12, 95% CI: [0.44, 10.17], I² = 0%, Z = 0.94 (P = 0.35), Figure 5A; RR = 2.54, 95% CI: [0.62, 10.40], I² = 3%, Z = 1.30 (P = 0.20), Figure 5B) (12, 15–17, 20, 27, 30, 36). Similar analysis results of grade 3–5 pancreatitis could also be seen in Figure 6 (15–17, 20, 27, 30, 36). The corresponding funnel plots of RR were summarized in Supplemental Figures 5, 6. No obvious heterogeneity was found among all groups.

The 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.