All subjects were patients with MPE that was diagnosed using thorax imaging, pleural fluid analysis, cytology, or pleural biopsy, without any restrictions on the tumor types. All subjects had normal heart, liver, or kidney function. The intervention used was Rh-endostatin through intrapleural administration instead of intravenous injection. Patients in the experimental group received Rh-endostatin plus chemical irritant, and the control group received chemical irritant alone, which included platinum, BLM, tetracycline, doxycycline, or silver nitrate, among others. During perfusion, all subjects did not receive hyperthermia, radiotherapy, chemotherapy, chemoradiotherapy, traditional Chinese medicine injections (TCMIs), or other biological response modifiers (BRMs). The main outcomes were clinical responses, survivals, and QOL, and the secondary outcomes were ADRs and TRAEs. The trials were randomized controlled trials (RCTs), with no restrictions on follow-up and research institutions.

Excluded studies included the duplicates; studies about non-MPE and non-Rh-endostatin; studies about Rh-endostatin plus hyperthermia, radiotherapy, chemotherapy, chemoradiotherapy, TCMIs, or BRMs; meeting abstracts and reviews without any specific data; non-RCTs as cohort studies, cross-sectional studies, case series, or case reports; unrelated SRs or meta-analyses; and studies without primary or secondary outcome data.

Based on the principle of patients (P) plus intervention (I), we applied the MeSH and free word to build the search strategies as (“Pleural Effusion” [Mesh] OR Pleural Effusion OR Pleural Effusions OR Hydrothorax OR MPEs OR MPE) AND (“Endostatins” [Mesh] OR Endostatins OR Endostatin OR Recombinant human endostatin injection OR rhES OR Rh-endostatin OR Endostar OR Sulijia OR YH-16). Two independent evaluators (C-QW and HJ) collected all the published studies of “Rh-endostatin plus chemical irritants for MPE” from Embase, PubMed, Web of Science (ISI), China Biological Medicine Database (CBM), Wanfang Database, China National Knowledge Infrastructure Database (CNKI), Chinese Scientific Journals Full-text Database (VIP), and Cochrane Central Register of Controlled Trials (CENTRAL, Issue 8 of 12, August 2020) and ongoing trials from the Chinese clinical trial registry (Chi-CTR, http://www.chictr.org.cn), WHO International Clinical Trials Registry Platform (WHO-ICTRP, http://apps.who.int/trialsearch/), and US-clinical trials (https://clinicaltrials.gov/, up to August 2020). In addition, we critically evaluated all the SRs/meta-analyses of Rh-endostatin in MPE and selected eligible trials from the references.

Two evaluators (C-QW and MH) were asked to collect the qualified trials about Rh-endostatin plus chemical irritants for MPE according to the preestablished inclusion and exclusion criteria.

Two evaluators (X-RH and QC) were asked to assess the bias risk of methodology using the Cochrane Collaboration’s risk of bias assessment tool for RCTs (33). The bias risk was assessed as a judgment (high, low, or unclear) for individual elements of five domains (selection, performance, attrition, reporting, and other).

The clinical responses were evaluated using CR, treatment failure, and progressive disease (PD). Based on previous studies (34–37), we integrated all the criteria as follows: (i) CR, (ii) PR, (iii) NR or SD; and (iv) PD ( Supplementary Material S2 ). Treatment failure was defined as NR/SD plus PD (38). Survival was defined as overall survival (OS) rate, progression-free survival (PFS) rate, or hazard ratio (HR) of the OS and PFS. Using the Karnofsky Performance Status (KPS) scale, if the KPS score increased ≥10 after perfusion, the QOL was improved.

The secondary outcomes were ADRs and TRAEs. According to the World Health Organization (WHO) (39) or Common Terminology Criteria for Adverse Events (CTCAE) standards (40), ADR was defined as neutropenia, thrombocytopenia, anemia, cardiotoxicity, hepatotoxicity, nephrotoxicity, gastrointestinal reactions, alopecia, peripheral neuritis, chest pain, and fever, among others. TRAE was defined as treatment-related mortality (TRM) and a series of clinical symptoms such as respiratory failure, pneumothorax, cutaneous emphysema, or catheter-related infection/chest infection, among others.

Two evaluators (X-TZ and T-yF ) collected all the data using a predesigned data extraction form. The data included the first author, year of publication, and demographic information of patients; baseline characteristics such as primary tumors, pleural fluid volume, KPS score, treatment history (initial treatment, retreatment, or both), anticipated survival time (AST), sample size, drainage methods [indwelling pleural catheters (IPCs) or thoracocentesis]; combinations and usages of Rh-endostatin and chemical irritants; evaluation time and follow-up protocols; and outcomes including CR, treatment failure, PD, OS, PFS, QOL, ADRs, and TRAEs. Additionally, we contacted the corresponding author to obtain the available survival data. If the authors were unavailable, we adopted the Engauge Digitizer 4.1 to transform the Kaplan–Meier survival curves into available data (41, 42).

According to the data features, the odds ratio (OR) or hazard ratio (HR) and their 95% CI were used to quantify the CR, treatment failure, PD, OS, PFS, QOL, ADRs, and TRAEs, and p < 0.05 was considered a statistical significance. Two evaluators (C-QW and X-RH) conducted a series of meta-analyses using the Review Manager 5.4.1 (as recommended by the Cochrane Collaboration). The Cochran’s χ² test and I² statistic were conducted to analyze the potential statistical heterogeneity. If p ≥ 0.1 and I² ≤ 50%, a fixed-effects model (FEM) was used to pool the OR or HR and their 95% CI. Otherwise, a random-effects model (REM) was used. If the number of trials was larger than 10, a funnel plot and Egger/Begg’s test were used to examine the potential publication bias.

When at least one item was considered a high risk, the trial was defined as poor quality. When the result was statistically different and beneficial to Rh-endostatin infusion, the trial was defined as an underestimated or overestimated trial following our experiences (38, 43, 44). According to the principle of underestimating efficacy and overestimating risk, we established a sensitivity analysis model to analyze the robustness of the results before and after eliminating the trials with poor quality, underestimation, or overestimation.

Following the guideline (45) and our previous experiences (38, 43, 44), we established a subgroup analysis model to analyze the clinical heterogeneity and the effects of variables on CR, treatment failure, and PD and to reveal their treatment thresholds and optimal usage for achieving an ideal response. The variables included patient features, drainage methods, and combinations of Rh-endostatin plus chemical irritants and their dose, treatment frequency, and times. Finally, a univariable random-effects meta-regression was conducted to reveal the relevance between each variable and CR, treatment failure, or PD and a post-hoc multiple regression analysis adjusting for their OR under all variables.

Following the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach and integrating the results of the sensitivity analysis, we developed a quality summary model to summarize the evidence quality and classify them as “high,” “moderate,” “low,” or “very low” (38, 43, 44) ( Supplementary Material S3 ). The quality was downgraded according to five domains as follows: (1) methodological bias risk; (2) statistical heterogeneity; (3) indirectness; (4) imprecision; and (5) publication bias. Two evaluators (X-FC and C-QW) used the GRADE profiler to summarize the quality and generate the absolute estimates for the CR, treatment failure, PD, OS rate, PFS rate, QOL, ADRs, and TRAEs (46).

A literature search conducted from inception to August 17, 2020, identified 959 studies. After duplicates were removed, 379 studies remained for a review of abstracts. After reviewing the abstracts, we identified 115 reports and 11 SRs/meta-analyses (16, 47–54). After reviewing full texts, we identified 73 qualified trials (29–31, 54–123). After reviewing the SRs/meta-analyses, we identified 39 trials (55–60, 62, 64, 66–69, 71–78, 81–92, 95, 97, 100, 101, 104, 109, 110). Excluding two ongoing trials without data (124, 125), we included one ongoing trial (ChiCTR-IPR-17011666) (126). Finally, we identified 75 trials (29–31, 54–123, 126, 127) for this SR/meta-analysis ( Supplementary Material S4 ; Tables S1–S4 ; Figure 1 ).

The 75 trials, published from 2010 to 2020, recruited 4,678 patients with MPE from China, which included 2,512 males and 1,738 females aged 18–89 years ( Table 1 ). Forty-seven trials recruited patients with lung cancers (30, 31, 55, 58, 59, 64–67, 70, 74, 75, 77–84, 89–95, 97, 100–107, 109–111, 113–116, 118, 120, 121, 127), three trials recruited patients with breast cancer (99, 112, 117), and the remaining 25 trials (29, 54, 56, 57, 60–63, 68, 69, 71–73, 76, 85–88, 96, 98, 108, 119, 122, 123, 126) recruited patients with malignant tumors such as lung cancer, breast cancer, malignant lymphoma, gastric cancer, hepatic carcinoma, and ovarian cancer, among others. Trials were of varied sample sizes, from 30 to 130. The patients had small to massive effusion, KPS ≥40, and AST ≥2 months. Fifteen trials reported the treatment history as initial treatment, retreatment, or both. After the drainage of hydrothorax using an IPC or thoracocentesis, 2,352 cases accepted the intrapleural administration of Rh-endostatin plus chemical irritants, while 2,326 other cases accepted the chemical irritants alone. We found six combinations of Rh-endostatin plus DDP in 60 trials (30, 31, 54–56, 58, 60, 62, 64, 66–68, 71–73, 75–78, 80–82, 84–92, 94, 95, 97–101, 103, 104, 106–123, 126, 127), nedaplatin (NDP) (69, 74, 83, 102) and BLM in four trials (57, 61, 65, 70), lobaplatin (LBP) in three trials (29, 93, 105), carboplatin (CBP) in two trials (59, 96), paclitaxel (79) or DDP/BLM in one trial (63). Rh-endostatin (30–90 mg each time) was used once or twice a week, 1–12 times by intrapleural administration. The chemical irritant was mainly DDP and used with 20–100 mg/m² each time. Three to 10 weeks after perfusion, the trials evaluated clinical responses using a Millar or Ostrowskimj criterion, ADRs using a WHO criterion, and QOL using a KPS scale. In addition, only nine trials reported the survivals (29–31, 63, 71, 75, 98, 105, 116) and TRAEs (56, 66, 76, 88, 95, 98, 99, 116, 127), and six trials (66, 88, 95, 98, 99, 116) reported the TRM.

Thirty-four trials reported the random sequence generation using a random number table (a low risk of selection bias) (30, 31, 54, 56–58, 62, 65, 67, 69, 75, 81, 82, 84, 85, 89, 90, 93, 97, 101, 102, 105, 106, 109–114, 116, 117, 120, 126, 127), and two trials reported the odd or even random (a high risk of selection bias) (60, 123). Two trials reported the allocation concealment using an envelope (a low risk of selection bias) (73, 76), and two trials reported the allocation exposure (a high risk of selection bias) (60, 123). With the exception of one open RCT (119), the remaining trials failed to clearly report the blindings (an unclear risk of performance bias). All trials reported the complete outcome data (a low risk of attrition bias). Forty-four trials selectively reported the ADRs, and one trial selectively reported the CR (a high risk of reporting bias) (66). The comparability between groups (an unclear risk of other biases) was unclear in 12 trials ( Figure 2 ).

Seventy-five trials reported the clinical responses of six combinations of Rh-endostatin with DDP (30, 31, 54–56, 58, 60, 62, 64, 66–68, 71–73, 75–78, 80–82, 84–92, 94, 95, 97–101, 103, 104, 106–123, 126, 127), NDP (69, 74, 83, 102), BLM (57, 61, 65, 70), LBP (29, 93, 105), CBP (59, 96), paclitaxel (79), or DDP/BLM (63). The statistical heterogeneity was not found using Cochran’s χ² test and I² statistic (I² = 0%). So, the data were pooled using an FEM. The ORs of fixed effects were 2.29 (95% CI 1.93–2.71, p < 0.00001), 2.50 (95% CI 1.31–4.77, p = 0.005), 2.71 (95% CI 1.37–5.35, p = 0.004), which showed that the CR of Rh-endostatin with DDP, NDP, or LBP was significantly higher than that of irritants alone ( Figure 3A ). The treatment failure of Rh-endostatin with DDP, CBP, NDP, LBP, or BLM was significantly lower than that of irritants alone. The ORs were 0.29 (95% CI 0.25–0.33, p < 0.00001), 0.28 (95% CI 0.12–0.64, p = 0.003), 0.29 (95% CI 0.16–0.51, p < 0.0001), 0.25 (95% CI 0.15–0.44, p < 0.00001), and 0.25 (95% CI 0.13–0.50, p < 0.0001), respectively ( Figure 3B ). The PD of Rh-endostatin with DDP, CBP, NDP, LBP, or BLM was significantly lower than that of irritants alone. The ORs were 0.27 (95% CI 0.22–0.34, p < 0.00001), 0.25 (95% CI 0.09–0.67, p = 0.006), 0.31 (95% CI 0.12–0.79, p = 0.01), 0.32 (95% CI 0.12–0.86, p = 0.02), and 0.31 (95% CI 0.12–0.80, p = 0.02), respectively ( Figure 3C ).

Nine trials reported the survivals (29–31, 63, 71, 75, 98, 105, 116). Only five trials reported the OS time and PFS of Rh-endostatin with DDP (30, 71, 75, 98) or LBP (29), but without the available data. Five trials reported the OS rates, and three reported the 1-year OS rate of Rh-endostatin with DDP (31, 75, 116). The statistical heterogeneity between trials was not found using Cochran’s χ² test and I² statistic (I² = 0%). So, we pooled the data using an FEM. The 1-year OS rate of Rh-endostatin with DDP was significantly higher than that of DDP alone. The OR was 3.32 (95% CI 1.63–6.75, p = 0.0009) ( Figure 4 ). The remaining OS rates were reported in only one trial, and the data were analyzed descriptively using forest plots. Statistical analysis showed that the 0.5-year OS rate of Rh-endostatin with DDP (116), 1-year OS rate of DDP/BLM (63), 2-year OS rate of DDP (75), and 3-year OS rate of LBP (105) were significantly higher than that of irritants alone. The ORs were 5.36 (95% CI 1.24–23.10, p = 0.02), 5.21 (95% CI 1.28–21.24, p = 0.02), 10.00 (95% CI 2.05–90.59, p = 0.04), and 3.60 (95% CI 1.46–8.89, p = 0.005), respectively.

Given the limited trials for Rh-endostatin with CBP, NDP, LBP, or BLM, we only evaluated the QOL of Rh-endostatin with DDP (30, 55, 56, 58, 60, 62, 64, 66, 67, 72, 73, 75, 76, 82, 84–86, 88, 90–92, 95, 98, 99, 104, 109, 111, 112, 115, 116, 118, 126) ( Figure 5 ). The statistical heterogeneity between trials was not found using Cochran’s χ² test and I² statistic (I² = 0%). So, the data were pooled using an FEM. The OR was 3.01 (95% CI 2.49–3.63, p < 0.00001), which indicated that the QOL was significantly higher than that of DDP alone.

Given the limited trials for Rh-endostatin with CBP, NDP, LBP, or BLM, we only evaluated the ADRs and TRAEs in Rh-endostatin with DDP. Fifty-eight trials observed hematotoxicity (neutropenia, thrombocytopenia, and anemia), cardiotoxicity (arrhythmia), hepatotoxicity, nephrotoxicity, gastrointestinal reaction, alopecia, neurotoxicity, rash, hypertension, hemorrhage, chest pain, and fever (30, 31, 54–56, 58, 60, 62, 64, 66–68, 71–73, 75–78, 80–82, 84–92, 94, 95, 97–100, 103, 104, 107–123, 126, 127) ( Table 2 ). Cochran’s χ² test and I² statistic showed no heterogeneity in all of the ADRs (I² = 0%). So, the data were pooled using an FEM. Rh-endostatin with DDP had a similar risk of ADRs in DDP alone. The ORs showed no significant difference between the two groups. In addition, nine trials (56, 66, 76, 88, 95, 98, 99, 116, 127) reported no risk of TRAEs, and six trials (66, 88, 95, 98, 99, 116) reported no risk of TRM.

The patient feature was defined as primary tumor, pleural fluid volume, treatment history, KPS score, and AST. First, the primary tumor was classified as lung cancer, breast cancer, or malignant tumors. In patients with lung cancer/malignant tumors, Rh-endostatin with DDP obtained a significant increase of CR and a reduction of failure and PD. In breast cancer, it only obtained a reduction of failure and PD ( Table 3A and Figures S12, S14, S16 ). The pleural fluid was classified as small to large, moderate to large, or large ( Table 3B and Figures S18, S20, S22 ); treatment history was initial treatment, retreatment, or both ( Table 3C and Figures S24, S26, S28 ); KPS score was <50, ≥50, or ≥60 ( Table 3D and Figures S30, S32, S34 ); and the AST was ≥2 months or ≥3 months ( Table 3E and Figures S36, S38, S40 ). In patients with moderate to massive effusion, initial treatment, KPS score (≥60), or AST (≥3 months), the Rh-endostatin with DDP groups obtained a significant increase of CR and a reduction of failure and PD.

The majority of patients mainly received the IPCs ( Table 3F and Figures S42, S44, S46 ). Subgroup analyses found that whether IPC is used or not had no effect on the clinical responses. Rh-endostatin was used with 30–90 mg each time, once or twice a week 1–12 times ( Tables 3G–I and Figures S48–S64 ). DDP was used with 30–40 mg/m² or 50–60 mg/m² each time ( Table 3J and Figures S66, S68, S70 ). Rh-endostatin (30–35 mg or 40–45 mg each time, once or twice a week 3–4 times) with DDP (30–40 mg/m² or 50–60 mg/m²) obtained a significant increase of response and a reduction of failure and PD in MPE. However, univariate regression analysis did not discover a positive or negative correlation between CR, treatment failure, and PD and each variable ( Table 3 and Figures S13–S71 ). Multiple meta-regression analysis also did not discover a positive or negative correlation ( Table 3 ).

In Rh-endostatin with DDP, more than 10 trials were included for the CR, treatment failure, PD, QOL, and ADRs. So, funnel plot and Egger/Begg’s tests were used to analyze their potential bias of publication. The analysis found a publication bias in CR (p < 0.001, 95% CI 0.74–1.53), treatment failure (p < 0.001, 95% CI -2.50 to -1.02), PD (p < 0.001, 95% CI -1.26 to -0.12), and QOL (p < 0.001, 95% CI 0.57–3.94) ( Figures 6A–D ). The trials overestimated the CR and QOL and underestimated the treatment failure and PD. The analysis did not find a bias in neutropenia (p = 0.10, 95% CI -1.51 to 0.14), thrombocytopenia (p = 0.82, 95% CI -1.25 to 0.99), cardiotoxicity (p = 0.34, 95% CI -59 to 1.57), hepatotoxicity (p = 0.79, 95% CI -0.73 to 0.56), nephrotoxicity (p = 0.85, 95% CI -0.66 to 0.55), gastrointestinal reactions (p = 0.97, 95% CI -0.75 to 0.72), chest pain (p = 0.28, 95% CI -0.58 to 1.82), and fever (p = 0.30, 95% CI -0.51 to 0.14) ( Figures 6E–L ). The trials objectively reported the ADRs.

In Rh-endostatin with DDP, the poor trials involved clinical response, 1-year OS rate, QOL, and ADRs. Some trials overestimated the CR, 1-year OS rate, and QOL and underestimated the treatment failure and PD. According to the underestimating efficacy and overestimating risk, we evaluated the robustness through removing the poor trials, overestimation/underestimation, and both. Before and after removing the poor trials, the results demonstrated a good robustness of all outcomes. Before and after removing the overestimation and both, the OR of 1-year OS rate was poor robust, and other indicators were robust. In addition, the OR of CR was robust in Rh-endostatin with NDP, and the OR of CR and PD was robust in Rh-endostatin with BLM ( Table 4 ).

In methodology, 46 poor trials were included for this analysis. Sensitivity analysis demonstrated that the OR of 1-year OS rate was poor robustness in Rh-endostatin with DDP, the CR and PD were poor in Rh-endostatin with NDP, and the CR was poor in Rh-endostatin with BLM. Therefore, we downgraded their quality by two grades. Other results had good robustness, and we downgraded their quality by one grade. No heterogeneity was found in all of the indicators; all indicators were not downgraded. In Rh-endostatin with DDP, the sample size of 1-year OS rate, alopecia, hypertension, and hemorrhage was lower than 300 subjects. In Rh-endostatin with NDP, CBP, LBP, or BLM, the CR, treatment failure, and PD were lower than 300. So, we downgraded their quality by one grade. In addition, the funnel plot and Egger’s test showed a publication bias of CR, treatment failure, PD, and QOL in Rh-endostatin with DDP. The sensitivity analysis results were good robust, and we did not downgrade their quality. So, we summarized a low quality for 1-year OS rate, alopecia, hypertension, and hemorrhage and a moderate quality for other results of Rh-endostatin with DDP; a low quality for CR and treatment failure in Rh-endostatin with NDP or BLM; and a very low quality for the remaining indicators ( Table 5 ).