Comparison between different treatment regimens of vascular targeting drug to malignant pleural effusion in patients with lung cancer: A Bayesian network meta-analysis

Background:

The presence of malignant pleural effusion in lung cancer patients often suggests a poor prognosis. We plan to investigate which regimen of vascular targeting drug is preferable to control the malignant pleural effusion in such patients.

Methods:

Two investigators dependently searched and screened for randomized controlled trials in PubMed, Embase, Web of Science and China National Knowledge Infrastructure from the database inception to August 2022. R software was applied to build a network model in Bayesian method. Objective response rate of malignant pleural effusion is the primary outcome measure. Besides, the incidence of 3 adverse events were compared, including gastrointestinal reaction, leukopenia and hypertension. Due to the disconnection of network, we analysis and discuss the short-term treatment (3–4 weeks) and long-term treatment (6–12 weeks) respectively.

Results:

31 studies with 2093 patients were identified. Four targeting drugs contain bevacizumab (Bev), anlotinib, apatinib and Endostar. Two administration routes include intracavity perfusion (icp) and intravenous injection. Based on the current evidence, for short-term treatments, compared with single-agent chemotherapy (CT), Bev_icp + CT, anlotinib + CT, Bev_icp and anlotinib + endorstar_icp present better objective response, and no statistical significance was found in objective response between Bev_icp + CT, anlotinib + CT and Bev_icp. For long-term treatments, compared with doublet or triplet chemotherapy (2CT or 3CT), Bev_icp + 2CT, apatinib + 2CT, Bev_icp + 3CT, and Bev_intravenous injection + 2CT are more effective option, but no statistical significance was found in objective response between the 4 combination regimens with chemotherapy.

Conclusion:

Our findings suggest that no statistical significance between above vascular targeting regimens. Pathological type of lung cancer may affect the effect of bevacizumab intracavity infusion plus chemotherapy. The influence of different administration routes of vascular targeting drugs on efficacy remains to be investigated. There are some concerns with the quality of the studies, and some limitations should be considered when interpreting these results, which includes limited geographical region and sample size of studies. Despite these limitations, this study may inform vascular targeting therapy choice in such a patient population.

1. Introduction

Lung cancer is the most common cause of cancer death in the world.^[1] More than 350 people will die from lung cancer each day in the US in 2022.^[2] In China, lung cancer is the most common cause of cancer death among men and women.^[3] Deadly metastatic disease often develops in lung cancer patients.^[4] VEGF is an essential mediator in lung cancer metastasis, such as brain metastasis^[5] and pleural spread.^[6,7] With better knowledge of gene mutations in lung cancer, such as epidermal growth factor receptor (EGFR) and VEGF, vascular targeted therapies have received increasing attention.^[8,9] Although the progress continues in vascular targeted therapy, the selection of a systemic lung cancer treatment programs and disease secondary to lung cancer remain a challenge.

Malignant pleural effusion (MPE) is a common complication of lung cancer, and the formation of it can generally be explained by the imbalance between effusion absorbtion and production. On the one hand, MPE can be caused by cancer cells invading the pleura. Lymphatic drainage obstruction caused by the visceral pleura involvement is an important pathophysiological mechanism in the generation of MPE.^[10,11] On the other hand, immune modulator such as tumor necrosis factor, interleukin, and regulators of vascular permeability such as VEGF, may promote excessive plasma leakage, thereby developing MPE.^[12,13] Cough and dyspnea are the most common clinical manifestations among these patients. MPE in lung cancer patients is usually at an advanced stage. A retrospective research based on SEER database shows that MPE is often a poor prognostic factor in lung cancer patients.^[14] A single center study with retrospective design reported that the median survival time of MPE patients is only 5.49 months (95% CI: 3.62–7.35).^[15] However, The treatment of malignant pleural effusions is palliative in nature.^[16] Several non-prospective studies based on small samples have reported impressive control rate for MPE patients with vascular targeting agents in combination with chemotherapy.^[17] Nevertheless, there are various treatment options available and it is difficult to reach a consensus. Currently, exploration of vascular targeting agents in lung cancer complicated with MPE is underway.

Anlotinib and Apatinib are multi-target tyrosine kinase inhibitor, featuring anti-vasculogenesis by suppressing intracellular tyrosine kinase domain phosphorylation process to interrupt the start of downstream signaling pathways.

Anlotinib is an oral multi-target tyrosine kinase inhibitor developed in China, which targets vascular endothelial growth factor receptor (VEGFR), as well as platelet-derived growth factor receptor, fibroblast growth factor receptor and stem-cell factor receptor.^[18] It is mostly used in the very third-line treatment for NSCLC in China. A post hoc analysis of phase II trial with small sample and placebo-controlled concluded that anlotinib is associated with progress-free survival in SCLC patients with with MPE.^[19]

Apatinib primarily improves MPE by inhibiting VEGFR-2, and other receptors such as stem-cell factor receptor, platelet-derived growth factor receptor-β, and c-src. It was first tested in stomach and liver cancers and is now in the exploratory stage of use in lung cancer.^[20] Small sample clinical trial preliminarily confirms the feasibility of apatinib for MPE.^[21]

Endostar, a recombinant human endothelial growth inhibitor, was approved for widespread use in advanced NSCLC in 2006.^[22] It can weaken angiogenesis by blocking VEGFR and can induce apoptosis of endothelial cell.^[23] Xunqing Ma et al^[24] explained that intracavity perfusion of endostar can suppress MPE formation through inhibiting angiogenesis, based on their mouse experience. It was concluded from a meta-analysis by Biaoxue Rong et al^[25] that endostar plus chemotherapy produced better efficacy than chemotherapy alone through intracavity perfusion in treating MPE.

Bevacizumab (Bev) is a human recombinant monoclonal antibody, which was approved in 2006 for advanced NSCLC, and remains an important part of the treatment landscape now.^[26] It can prevent VEGF-A from binding VEGFR, resulting decrease in vascular permeability and tumor cell migration, thereby reducing MPE.^[13] The published meta-analysis by Shen et al^[27] only assessed the efficacy of using platinum-based intracavity perfusion alone or combined with Bev in lung cancer complicated with MPE.

This Bayesian network meta-analysis based on PRISMA set out to compare the efficacy and safety between different regimens with vascular targeting drug for controlling the malignant pleural effusion in lung cancer patients.

2. Method

2.1. Search strategy and inclusion criteria

Two investigators (P. Huang and Z.K. Guo) dependently searched and screened for randomized controlled trials (RCTs) in PubMed, Embase, Web of Science and China National Knowledge Infrastructure databases from the database inception to August 2022. The detailed search strategy was listed in Table S1, Supplemental Digital Content, http://links.lww.com/MD/J324, including the following terms: “anlotinib,” “apatinib,” “bevacizumab,” “cediranib,” “lapatinib,” “pazopanib,” “sorafenib,” “ramucirumab” and “pleural effusion.” Patients concurrently receiving radiotherapy, thermal therapy, immune checkpoint inhibitors, or other treatment programs were excluded.

Lung cancer patients complicated with malignant pleural effusion were selected as the object population in this research. The histological types of Lung cancer included adenosquamous carcinoma, adenocarcinoma, squamous cell carcinoma, large cell carcinoma and small cell carcinoma. Studies of benign pleural effusion and studies that did not report the signature of pleural effusion were excluded.

The primary outcome measure is the objective response of MPE. Based on the World Health Organization standards,^[28] in this study, objective response was defined as complete response (CR) + partial response (PR). CR was defined as the MPE disappear completely lasting for at least 4 weeks. PR was defined as the MPE decrease by half lasting for at least 4 weeks. Objective response rate (ORR) was defined as the number of patients with objective response divided by the number of randomized patients. CR and PR were evaluated radiologically. Studies did not meet the outcome evaluation criteria mentioned above were excluded.

Only randomized controlled trials were included. Dissertations and conference presentations without available trial details were excluded.

2.2. Data extraction

We counted the number of CR and PR patients and calculated the number of objective response patients. Also, basic information for these studies (including name of the first author, publication year, performance status, gender, EGFR mutation status and clinical stage) were summarized. Specific treatment regimens, treatment period, available adverse event data, pathological types of cancer, randomization method, lost to follow-up and fund information were collected into spreadsheets (See Table S2, Supplemental Digital Content, http://links.lww.com/MD/J325).

2.3. Quality assessment

The risk of bias was assessed by consensus among the 3 researchers. The Cochrane Risk of Bias 2.0 assessment tool^[29] was used to assess risk of bias from 5 domains: Bias arising from the randomization process; Bias due to deviations from intended interventions; Bias due to missing outcome data; Bias in measurement of the outcome; Bias due to selection of the reported result. The possible risk of-bias judgements include 3 levels: “Low risk of bias,” “Some concerns” and “High risk of bias.” If more than 1 domain was considered to “High risk of bias,” the overall bias was assessed to “High risk.” If more than 1 domain was considered to “Some concerns,” the overall bias was assessed to “Some concerns.”

2.4. Network meta-analysis

A Bayesian network meta-analysis was performed with R software 4.2.2. (https://cran.r-project.org/) JAGS software 4.3.1 (https://sourceforge.net/projects/mcmc-jags/), “rjags” package 4 to 13 and “gemtc” package 1.0 to 1 were used in model fitting. Taking into account the potential heterogeneity between studies, a random-effect model was used. Because the closed loop network was not generated, and there were rare inconsistencies in study design, a consistent model was selected. For the model building process, 3 variations were considered, indluding each treatment regimen, together with the amount of its corresponding sample size and responders. Four Markov Monte Carlo chains with 25,000 Sample size per chain were used in this model. Five thousand adaption iteration and 25,000 simulation iteration were performed. Thinning factor was set to 1. Next, pooled relative risk (RR) with 95% credible intervals were calculated as the effect measure. If the 95% credible interval does not include 1, the differences are considered statistically significant. The larger RR indicates the stronger correlation between treatment and ORR. Heterogeneity of comparisons was evaluated using I². I² > 30%, >50% and > 70% were considered as moderate, substantial and considerable heterogeneity. No distinction was made between different modes of chemotherapy drugs administration. Rank probability and surface under the cumulative ranking (SUCRA) value were calculated for the secondary assessment of treatment effect. SUCRA value closed to 1 indicates better treatment effect. For studies that cannot be compared within the same network, we report the results narratively and synthetically.

Additionally, frequency of adverse events was analyzed with frequentist method using “netmeta” package 2.6-0. Nausea, emesis, abdominal pain and diarrhea were grouped together as “gastrointestinal reaction.” Besides, leukopenia and hypertension were also investigated, in which the Mantel-Haenszel method^[30] was used to handle sparse data. Incidence of adverse drug reactions are counted by the number of occurrences, and odd ratio (OR) with 95% confidence interval (CI) were calculated as the effect measure. Ninety-five percentage confidence intervals not including 1 are considered statistically significant.

2.5. Subgroup analysis and sensitivity analysis

The planned subgroup analysis could not be carried out due to insufficient data. To test the stability of network meta-analysis for objective response, sensitivity analysis was undertaken by comparing deviance information criterion (DIC) between fix-effect model and random-effect model.^[31] Also, to verify the stability of results, we excluded studies with treatment period more than 4 weeks in the first network, and excluded studies with treatment period more than 12 weeks in the second network.

2.6. Model convergence

Gelman-Rubin-Brooks diagnosis method^[32] was used to assess the convergence of 4 Markov Monte Carlo chains by estimating the Potential scale reduction factor (PSRF) and Shrink factor. Satisfactory convergence can be considered when the following conditions are met: (i) the median value of the shrink factor and 97.5% shrink factor are close to 1 and tend to be stable gradually iterative process; (ii) 1.00 < PSRF < 1.05.^[33] The results was given in Figure S2, Supplemental Digital Content, http://links.lww.com/MD/J326.

3. Result

3.1. Study selection

The complete process of study selection was reported in a flow chart (Fig. 1). PubMed, Embase, and China National Knowledge Infrastructure was searched by 1 researcher (P. Huang), and Web of Science was searched by the other 1 (Z.K. Guo). After removing 128 duplicates with reference management software, they excluded 1623 studies in total in the screening process. In the eligibility assessment, totally 121 articles were assessed by reading full-text. Forty-four trials that did not meet RCT design were excluded by P. Huang., while another 25 were excluded by Z.K. Guo. Eight RCTs duplicated with that from PubMed were excluded. Then, with the consensus of the 3 researchers, we excluded several studies that may contribute to the clinical heterogeneity: In order to evaluate the actual efficacy of vascular targeting drugs, 4 RCTs were excluded because unclear systemic chemotherapy was used;^[34–37] 8 RCTs with different outcome evaluation criteria were excluded;^[38–45] 1 RCT comprised 3 arms of different dose of Bev^[46] was excluded. Finally, a total of 2093 patients in intention-to-treat population were included from 31 parallel-group RCTs for qualitative synthesis.^[47–77] Among them, 1970 patients were finally included in quantitative analysis from 28 studies^{[47–62,64–71,73–77]} collectively spanning the years between 2013 and 2021.

Figure 1.

Flow chart of search process.

3.2. Study characteristics

As shown in Table 1, 31 2-arm studies were included in total, which all originated from China. These studies cover the following treatment regimens: Apatinib + 2CT, Anlotinib, Anlotinib + CT, Anlotinib + endostar_intracavity perfusion (icp), Bev_intravenous injection (iv), Bev_iv + 2CT, Bev_icp, Bev_icp + CT, Bev_icp + 2CT, Bev_icp + 3CT, CT, 2CT, 3CT and placebo (Bev, bevacizumab. icp, intracavity perfusion. iv, intravenous injection, CT, single-agent chemotherapy. 2CT, doublet drug chemotherapy). In 31 studies, the range of mean ages was 40 to 73 years. except for 3 studies that only reported a median age. One thousand one hundred twenty-six males and 823 females are included, apart from 2 studies^[67,75] did not report the amount of each gender. Furthermore, EGFR mutations and tumor stage are not always reported. Besides, the available pathological type is predominantly non-small cell lung cancer (93.6%), which includes adenosquamous carcinoma, adenocarcinoma, squamous cell carcinoma and large cell carcinoma, mostly adenocarcinoma.

Table 1.

Characteristics of studies.

Author	Year	Major Treatment	Sample size	Responders	Age (X or M, [range]) (Mean ± Sd)	Gender (male/female)	EGFR mutation	Pathology			Perform status	Stage
								NSCLC	SCLC	Undefined
Zhou Z.[41]	2021	Bev_icp + cisplatin_icp	43	38	62.43 ± 2.55	20/23	NA	43	0	0	NA	≥IV
Zhou Z.	2021	cisplatin_icp	43	29	61.34 ± 2.68	21/22	NA	43	0	0	NA	≥IV
Du N.[42]	2013	Bev_icp + cisplatin_icp	36	30	X = 52.5, [66–82]	19/17	15	36	0	0	KPS > 60	IV
Du N.	2013	cisplatin_icp	34	17		19/15	16	34	0	0	KPS > 60	IV
Zhu S.G.[43]	2020	anlotinib + cisplatin_icp	32	27	52.62 ± 2.84	17/15	NA	31	0	0	KPS > 70	Advanced
Zhu S.G.	2020	cisplatin_icp	32	19	50.08 ± 2.61	18/14	NA	31	0	0	KPS > 70	Advanced
Zheng W.H.[44]	2020	anlotinib + endostar_icp	44	32	M = 59, [30–75]	32/12	NA	44	0	0	ECOG ≤ 2	NA
Zheng W.H.	2020	anlotinib + cisplatin_icp	40	14	M = 57, [28–76]	34/6	NA	40	0	0	ECOG ≤ 2	NA
Yue K.[45]	2018	anlotinib + cisplatin_icp	40	34	69.15 ± 5.33 [55–78]	23/17	NA	33	0	7	KPS > 70	Advanced
Yue K.	2018	cisplatin_icp	40	25	69.11 ± 5.25 [55–78]	24/16	NA	34	0	6	KPS > 70	Advanced
Yang Y.[46]	2020	Bev_iv + pemetrexed_iv +cicplatin_iv	32	29	60.01 ± 3.99 [51–70]	20/12	NA	32	0	0	ECOG ≤ 2	≥IIIB
Yang Y.	2020	pemetrexed_iv +cisplatin_iv	33	20	59.89 ± 4.02 [50–69]	21/12	NA	33	0	0	ECOG ≤ 2	≥IIIB
Yan Y.H.[47]	2015	Bev_icp	46	27	53.6 ± 11.3	20/26	NA	46	0	0	NA	IV
Yan Y.H.	2015	cisplatin_icp	46	10	55.1 ± 10.3	18/28	NA	46	0	0	NA	IV
Xue D.F.[48]	2017	Bev_icp + cisplatin_icp	41	38	58.21 ± 3.25 [42–71]	23/18	NA	41	0	0	NA	NA
Xue D.F.	2017	cisplatin_icp	41	31	58.96 ± 3.43 [43–71]	24/17	NA	41	0	0	NA	NA
Xiao C.[49]	2021	Bev_icp + lobaplatin_icp	29	24	66.7 ± 3.7 [64–75]	16/13	NA	23	6	0	KPS ≥ 70	NA
Xiao C.	2021	lobaplatin_icp	29	16	67.2 ± 4.1 [63–75]	11/18	NA	21	8	0	KPS ≥ 70	NA
Wang X.L.[50]	2021	Bev_iv + carboplatin_iv +pemetrexed_iv	42	38	45.26 ± 2.31 [20–56]	26/16	NA	42	0	0	KPS ≥ 60	NA
Wang X.L.	2021	carboplatin_iv +pemetrexed_iv	42	31	45.32 ± 2.26 [21–57]	25/17	NA	42	0	0	KPS ≥ 60	NA
Wang K.[51]	2018	Bev_icp	29	20	63.2 ± 4.5 [35–75]	16/13	NA	23	6	0	KPS > 50	NA
Wang K.	2018	cisplatin_icp	29	12	62.9 ± 5.3 [33–75]	15/14	NA	25	4	0	KPS > 50	NA
Sun Z.J.[52]	2018	Bev_icp + gemcitabine_icp	24	20	53.5 ± 4.5	14/10	0	24	0	0	KPS ≥ 60	≥IV
Sun Z.J.	2018	gemcitabine_icp	22	11	51.2 ± 5.6	12/10	0	22	0	0	KPS ≥ 60	≥IV
Sun Y.[53]	2021	apatinib + paclitaxel_iv +carboplatin_iv	25	16	65.21 ± 2.93 [63–74]	13/12	NA	25	0	0	NA	≥IIIB
Sun Y.	2021	paclitaxel_iv +carboplatin_iv	25	9	65.79 ± 4.07 [64–78]	11/14	NA	25	0	0	NA	≥IIIB
Su Z.[54]	2019	apatinib + paclitaxel_iv +cisplatin_icp	31	24	M = 69, [42–75]	26/5	NA	31	0	0	KPS > 60	NA
Su Z.	2019	paclitaxel_iv +cisplatin_icp	30	16		23/7	NA	30	0	0	KPS > 60	NA
Shi E.H.[55]	2020	apatinib + pemetrexed_iv +cisplatin_icp	30	26	59.6 ± 5.8 [51–73]	17/13	0	30	0	0	NA	NA
Shi E.H.	2020	pemetrexed_iv +cisplatin_icp	30	18	62.9 ± 5.1 [51–73]	15/15	0	30	0	0	NA	NA
Qu B.[56]	2015	Bev_icp + cisplatin_icp	32	27	X = 72.8, [65–76]	19/13	NA	26	0	6	KPS > 70	NA
Qu B.	2015	cisplatin_icp	31	19	X = 73.9, [67–78]	17/14	NA	28	0	3	KPS > 70	NA
Nie K.K.[58]	2020	Bev_icp	21	16	62 ± 11	13/8	0	19	0	2	ECOG ≤ 3	IV
Nie K.K.	2020	Bev_iv	22	14	62 ± 12	14/8	0	19	0	3	ECOG ≤ 3	IV
Liu H.P.[57]	2016	Bev_icp + pemetrexed_iv +cisplatin_icp	42	35	62.37 ± 7.81	29/13	NA	42	0	0	KPS > 60	NA
Liu H.P.	2016	pemetrexed_iv +cisplatin_icp	42	27	61.42 ± 8.47	31/11	NA	42	0	0	KPS > 60	NA
Liang Y.[59]	2021	Bev_iv + pemetrexed_iv +carboplatin_iv	41	36	43.25 ± 2.23	17/24	NA	41	0	0	KPS > 70	NA
Liang Y.	2021	pemetrexed_iv +carboplatin_iv	41	27	43.42 ± 2.17	18/23	NA	41	0	0	KPS > 70	NA
Li Y.L.[60]	2018	Bev_iv + pemetrexed_iv +cisplatin_iv	43	32	45.24 ± 21.38 [23–63]	24/19	NA	43	0	0	NA	NA
Li Y.L.	2018	pemetrexed_iv +cisplatin_iv	43	22	43.28 ± 19.39 [26–57]	27/16	NA	43	0	0	NA	NA
Ke H.[61]	2019	Bev_iv + pemetrexed_iv +carboplatin_iv	46	33	45.3 ± 2.1 [18–55]	NA	NA	46	0	0	KPS > 70	NA
Ke H.	2019	pemetrexed_iv +carboplatin_iv	46	29	46.17 ± 2.04 [19–56]	NA	NA	46	0	0	KPS > 70	NA
Jiang Q.L.[62]	2021	Bev_iv + gemcitabine_iv +cisplatin_iv	30	28	48.14 ± 2.52 [20–75]	20/10	NA	30	0	0	KPS > 60	NA
Jiang Q.L.	2021	gemcitabine_iv +cisplatin_iv	30	17	47.52 ± 2.38 [20–75]	19/11	NA	30	0	0	KPS > 60	NA
Huang P.C.[63]	2021	Bev_iv + pemetrexed_iv +carboplatin_iv	46	20	39.28 ± 4.77 [27–65]	25/21	0	46	0	0	ECOG ≤ 2	≥III
Huang P.C.	2021	Bev_icp + pemetrexed_iv +carboplatin_iv	46	30	40.17 ± 5.03 [25–67]	27/19	0	46	0	0	ECOG ≤ 2	≥III
Huang B.[64]	2016	Bev_icp	37	30	60.28 ± 6.17 [54–72]	26/11	0	37	0	0	KPS > 70	IV
Huang B.	2016	cisplatin_icp	36	21	61.31 ± 6.05 [55–74]	27/9	0	36	0	0	KPS > 70	IV
Ge F.C.[65]	2020	anlotinib + nedaplatin_icp	37	21	67.23 ± 4.73 [49–79]	17/20	NA	31	0	6	KPS > 70	Advanced
Ge F.C.	2020	nedaplatin_icp	31	7	68.26 ± 4.68 [50–77]	16/15	NA	26	0	5	KPS > 70	Advanced
Cui X.X.[66]	2019	Bev_icp + cisplatin_icp	10	8	50.2 ± 9.8 [40–60]	6/4	NA	NA	NA	NA	NA	NA
Cui X.X.	2019	Cisplatin_icp	10	5	49.4 ± 8.6 [40–60]	5/5	NA	NA	NA	NA	NA	NA
Chen Z.B.[67]	2019	anlotinib + cisplatin_icp	25	23	60.12 ± 3.18 [60–81]	13/12	NA	NA	0	3	KPS > 70	Advanced
Chen Z.B.	2019	cisplatin_icp	25	13	71.07 ± 2.21 [71–80]	14/11	NA	NA	0		KPS > 70	Advanced
Chen Y.X.[68]	2020	Bev_icp + carboplatin_icp	21	19	X = 60.67, [45–72]	14/7	NA	21	0	0	BIS > 60	NA
Chen Y.X.	2020	Carboplatin_icp	21	13	X = 60.17, [48–73]	15/6	NA	21	0	0	BIS > 60	NA
Chen T.J.[69]	2016	Bev_icp + cisplatin_icp	24	20	54.6 ± 7.7 [41–74]	NA	NA	24	0	0	NA	NA
Chen T.J.	2016	cisplatin_icp	24	13		NA	NA	24	0	0	NA	NA
Bi L.Q.[70]	2020	anlotinib	30	20	M = 65.1, [40–75]	18/12	NA	30	0	0	KPS ≥ 60	≥III
Bi L.Q.	2020	placebo	30	13	M = 63.2, [38–75]	17/13	NA	30	0	0	KPS ≥ 60	≥III
Bai M.[71]	2017	Bev_icp + cisplatin_icp	43	39	60.14 ± 6.20	23/20	NA	43	0	0	NA	NA
Bai M.	2017	cisplatin_icp	43	31	60.23 ± 6.18	22/21	NA	43	0	0	NA	NA

Bev = bevacizumab, BIS = barthel index score, ECOG = Eastern Cooperative Oncology Group score, EGFR = epidermal growth factor receptor, icp = intracavitary perfusionm, iv = intravenous injection, KPS = Karnofsky score, M = median age, NA = not available, Sd = standard error, X = mean age.

For “overall risk of bias,” 29 studies were assessed to “some concerns,” and they didn’t provide details about the use of allocation concealment or blinding. Besides, 2 RCTs were assessed to “high risk.”^[63,72] In 28 studies, all participants randomized completed the planned course of treatment, and 2 participants were lost to follow-up in each of the other 3 studies.^[48,58,63] Complete risk of bias assessment result was included in Figure S1, Supplemental Digital Content, http://links.lww.com/MD/J327. Additionally, available general data and detailed treatment programs for each study were listed in Table S2, Supplemental Digital Content, http://links.lww.com/MD/J325.

3.3. Network meta-analysis

Due to the lack of a closed network diagram, the “node-split” method was not applicable.^[78] The studies were connected to the network through common interventions, and 2 disconnected networks generated (Fig. 2A and B). The network meta-analysis graph (Fig. 2) presents the interventions involved in quantitative analysis. Considering the clinical heterogeneity and high risk of bias, 2 studies were only reported narratively.^[63,72] Additionally, 1 study comparing Anlotinib vs placebo, was excluded from this quantitative analysis because it was disconnected to the network.^[76]

Figure 2.

Network of eligible comparisons (A) (B). The wider line presents larger number of studies, and the bigger circle presents larger sample size. The number beside the line refers to the number of trials. Bev = bevacizumab, icp = intracavity perfusion, iv = intravenous injection, CT = single-agent chemotherapy, 2CT = doublet drug chemotherapy.

In the study by Nie et al,^[63] the authors concluded that intracavity perfusion of Bev has a higher ORR than intravenous injection (ORR = 76.2% in Bev_icp group, ORR = 63.6% in Bev_iv group). But suspiciously, approximately half of the patients were treated with additional chemotherapy or thoracentesis for various reasons, failing to adhere to the protocol interventions of Bev. In 2 other studies, vascular targeting drugs showed impressive therapeutic efficacy. However, in Cui study^[72] (ORR = 80.0% in Bev_icp + CT group, ORR = 62.0% in CT group), additional pleural effusion drainage was used when the treatment was unsatisfactory. Hence, the true efficacy of Bev might be overestimated. In the study by Bi et al,^[76] ORR is 66.7% in anlotinib group while ORR is 43.3% in placebo group. No statistical significance found in the incidence of adverse events between the 2 groups.

In quantitative analysis, compared with single-agent chemotherapy, vascular targeting drugs with or without chemotherapy yielded better results for MPE objective response. As shown in Figure 3A, compared with single-agent chemotherapy, Bev_icp + CT, anlotinib + CT, Bev_icp and anlotinib + endorstar_icp presented statistical significance in objective response, with RR ranging from 1.32 to 3.28. But there was no statistical significance in objective response between Bev_icp, anlotinib + CT and Bev_icp + CT. As given in Figure 3B, Bev_iv + 2CT, apatinib + 2CT and Bev_icp + 2CT had statistically significant objective response compared with 2CT, with RR ranging from 1.31 to 1.52. No statistical significance was found in objective response between Bev_iv + 2CT, Bev_icp + 2CT, Bev_icp + 3CT and apatinib + 2CT.

Figure 3.

Network meta-analysis in random-effect model comparing different interventions for MPE objective response (A) (B). Sensitivity analysis including studies with treatment period of 3–4 weeks (C) or studies with treatment period of 6–12 weeks (D). Each cell shows the pooled RR value from the comparison between “column” and “row.” RR < 1 favors column-defining treatment. 95% Credible intervals were reported in round brackets. Significant results were printed in bold. Bev = bevacizumab, icp = intracavity perfusion, iv = intravenous injection, CT = single-agent chemotherapy, 2CT = doublet drug chemotherapy, MPE = malignant pleural effusion, RR = relative risk.

On the other hand, we calculated SUCRA values separately for the short-term and long-term treatment groups. In Figure 4A, their probabilities to be the best treatment in descending order is anlotinib + endostar_icp (SUCRA = 0.9975), Bev_icp (SUCRA = 0.6625), anlotinib + CT (SUCRA = 0.5525), Bev_icp + CT (SUCRA = 0.2875), and CT (SUCRA = 0.0000). In Figure 4B, that is Bev_icp + 2CT (SUCRA = 0.7782), apatinib + 2CT (SUCRA = 0.7508), Bev_icp + 3CT (SUCRA = 0.6768), Bev_iv + 2CT (SUCRA = 0.5372), 2CT (SUCRA = 0.1788) and 3CT (SUCRA = 0.0772).

Figure 4.

Folding line chart of Rank probabilities and SUCRA value table of each treatment (A) (B). Bev = bevacizumab, icp = intracavity perfusion, iv = intravenous injection, CT = single-agent chemotherapy, 2CT = doublet drug chemotherapy.

3.4. Sensitivity analysis

DIC and pD in random-effect model (network A: pD = 19.5, DIC = 48.3, network B: pD = 19.1, DIC = 44.7) were numerically close to the ones in fixed-effect model (network A: pD = 17.6, DIC = 46.9, network B: pD = 16.9, DIC = 43.3), which supports the reliability of the network meta-analysis for primary outcome. Besides, sensitivity analyses were performed in 2 disconnected networks for studies with treatment periods of 3 to 4 weeks (Fig. 3C) and 6 to 12 weeks (Fig. 3D), respectively. The results were consistent with those in primary analysis (Fig. 2A and B). Hence, several studies with long treatment period or studies that did not elaborate the treatment period, have little impact on the overall results of analysis.

3.5. Heterogeneity analysis

The results of the heterogeneity analysis for each pair of interventions are presented in Figure S3, Supplemental Digital Content, http://links.lww.com/MD/J328. There was moderate heterogeneity in comparison of Bev_icp versus CT (I² = 46.5%), while substantial heterogeneity in comparison of Bev_iv + 2CT versus Bev_icp + 2CT (I² = 64.2%) and in comparison of 2CT versus Bev_icp + 2CT (I² = 70.4%). However, there were not sufficient data for us to investigate the source of heterogeneity.

3.6. Adverse event analysis

Among 31 studies, adverse events were reported in 27 studies. The remaining 4 studies did not detail the number of adverse reactions and their diagnostic criteria. We attempted to contact the study author by e-mail to request any details but did not not receive a reply. A summary of adverse events was listed in Table S3, Supplemental Digital Content, http://links.lww.com/MD/J329.

Statistical significance were observed in some comparisons. The gastrointestinal reactions of Anlotinib + endostar_icp were less observed than CT (OR = 0.14, 95% CI: 0.02–0.76) (Fig. 5A and B). Anlotinib + endostar_icp produced fewer leukopenia than CT (OR = 0.20, 95% CI: 0.06–0.71), and apatinib + 2CT produced fewer leukopenia than 2CT (OR = 0.29, 95% CI: 0.10–0.83) (Fig. 5C and D). Moreover, Bev_icp + CT was more likely to lead hypertension than CT (OR = 8.82, 95% CI: 1.95–39.98) (Fig. 5E), which consistent with the existing study by Shen et al^[27]

Figure 5.

Forest plots of network meta-analysis for Gastrointestinal reaction (A) (B), leukopenia (C) (D) and Hypertension (E). Bev = bevacizumab, icp = intracavity perfusion, iv = intravenous injection, CT = single-agent chemotherapy, 2CT = doublet drug chemotherapy.

4. Discussion

In this research, a Bayesian meta-analysis was conducted to explore which vascular targeting regimen is more beneficial for lung cancer patients with MPE. In the absence of direct evidence derived from pairwise comparisons of interventions, we gave indirect evidence comparing various vascular targeting regimens from network meta-analysis. Based on the current evidence, we would draw the following conclusions cautiously. Despite some limitations, our exploratory study may provide information about individualized treatment in these patients.

For short-term treatments (3–4 weeks), our findings suggest that all regimens containing vascular targeting drugs presents objective response than single-agent chemotherapy. SUCRA value appear to support that Bev_icp was ranked over Bev_icp + CT, but no statistical significance was derived in the league table between Bev_icp, Bev_icp + CT, anlotinib + CT and anlotinib + endostar_icp. Although anlotinib + endostar_icp is most likely to have the best effect, it is based on only 1 study. Additionally, Patients treated with Bev_icp + CT may have a higher risk of hypertension than CT, while no hypertensive events were reported in the Bev_icp group. For long-term treatments (6–12 weeks), although no statistical significance was found in objective response between Bev_icp + 2CT, apatinib + 2CT, Bev_icp + 3CT and Bev_iv + 2CT, they are all more effective option than doublet or triplet chemotherapy. Based on SUCRA value, Bev_icp + 3CT is ranked top, but it is based on only 1 study. Additionally, compared with 2CT, apatinib + 2CT is less likely to cause leukopenia.

In quantitative meta-analysis, all “Bev_icp + CT” refer to Bev intracavity perfusion plus platinum, including cisplatin, carboplatin or lobaplatin. As previously mentioned, indirect comparison showed no statistical significance of objective response between Bev_icp + CT and Bev_icp (RR: 0.785, 95% Crl: 0.56–1.074). After literature searching, we found no head-to-head comparison between Bev plus platinum and sole administration of Bev. Therefore, based on the current evidence, we speculate that in terms of anti-MPE, using Bev alone may have less toxicity with equivalent objective response.

As described previously, there are varying degrees of heterogeneity between several studies. Moderate heterogeneity was found in comparison of Bev_icp + CT versus. CT. When 1 study^[53] was excluded in sensitivity analysis, the remaining 2 studies^[57,70] did not show heterogeneity, and it did not change the statistical significance of the results. Furthermore, In studies by Wang K.^[57] and Huang B.,^[70] the pathological type of participants is diverse, while only lung adenocarcinoma patients were included in Yan study.^[53] A current review article indicates that patients with squamous cancer would have less benefit from targeted therapies than adenocarcinoma. Hence it is possible that the different pathological type contributes to the heterogeneity. On the other hand, substantial heterogeneity was found between trials by Liu H.P.^[64] and Huang P.C.,^[69] but there are insufficient data to explore the probable cause of heterogeneity. A systematic review by Ababneh et al^[79] concluded that intracavity perfusion was superior to intravenous injection in the case of Bev combined chemotherapy. Still, it is difficult to speculate whether the mode of administration is a cause of heterogeneity, and more comparisons between intracavity perfusion and intravenous injection should be developed.

At present, vascular targeting drugs are not routinely used in MPE. The American Thoracic Society guideline recommends indwelling pleural catheter or pleurodesis as first-line treatment in symptomatic MPE.^[80] But it is facing many challenges, such as catheter-related infection and unsuccessful lung expansion after pleurodesis. In this case, vascular targeting drugs may be an option worth considering. A phase II single-arm study from Japan suggested that, it is an effective option to use Bev plus chemotherapy when pleurodesis failed.^[81] Therefore, we look forward to more researches to compare the efficacy and safety of vascular targeting regiments and pleural interventions.

There are several limitations with our study. First, all the included studies were conducted in China, which is likely to mean that almost all patients are Asian. Therefore, caution should be exercised when extending our conclusion to other races or ethnicities. Second, only short-term efficacy was assessed in our research, the long-term outcomes such as overall survival and progression-free survival require further investigation. Third, the use of blinding only was mentioned in 1 study, more high-quality RCTs are needed. Finally, a distinction is made only for the mode of administration of vascular targeting drugs, but not for chemotherapy drugs. Consequently, these results need to be interpreted with caution.

In terms of adverse reactions, in several studies, some patients experienced double gastrointestinal reactions, such as emesis coincided with abdominal pain, so it may have caused overestimation of adverse effects by summing the number of adverse events in statistic process. Besides, a comprehensive analysis of adverse events should not be limited to the consideration of RCTs, but also requires different types of researches such as cohort studies and single-arm trials. Therefore, as a secondary outcome, the results of adverse event analysis have limitation for generalization.

Author contributions

Conceptualization: Peng Huang, Zhi-Kai Guo, Zhan-Tu Xue.

Data curation: Peng Huang, Zhi-Kai Guo.

Formal analysis: Peng Huang.

Investigation: Peng Huang, Zhi-Kai Guo, Zhan-Tu Xue.

Methodology: Peng Huang, Zhi-Kai Guo.

Writing – original draft: Peng Huang, Zhan-Tu Xue.

Writing – review & editing: Peng Huang, Zhi-Kai Guo, Zhan-Tu Xue.

Supplementary Material

medi-102-e34386-s001.pdf ^{(199.7KB, pdf)}

medi-102-e34386-s002.pdf ^{(276.8KB, pdf)}

medi-102-e34386-s003.pdf ^{(219.1KB, pdf)}

medi-102-e34386-s004.pdf ^{(237KB, pdf)}

medi-102-e34386-s005.pdf ^{(204.7KB, pdf)}

medi-102-e34386-s006.xlsx ^{(21KB, xlsx)}