Effectiveness of multi-drug regimen chemotherapy treatment in osteosarcoma patients: a network meta-analysis of randomized controlled trials

Two authors independently performed the literature search through November 2016 using electronic databases, including PubMed (MEDLINE), EmBase, and the Cochrane Library, with the keywords “osteosarcoma”, “osteogenic sarcoma”, “chemotherapy”, and “random*”, without language restriction. The bibliographies of the obtained publications and relevant reviews were also assessed to ensure that no relevant studies were inadvertently omitted. The publications included in the present study met the following criteria: (1) randomized controlled trial (RCT) design; (2) inclusion of osteosarcoma patients; (3) examination of two or more groups using different single- or multi-drug regimens; and (4) inclusion of PFS or OS as an outcome. The exclusion criteria consisted of the following: (1) non-RCT studies; (2) studies including patients with other types of sarcomas, such as Ewing sarcoma; (3) non-chemotherapy controlled studies, such as surgery or radiotherapy controlled studies; (4) studies comparing the same chemotherapeutic drug type, such as a drug dose-related study; and (5) non-desired outcome studies. Additionally, reviews, comments, case reports, basic studies, and conference reports were also excluded.

Two authors independently extracted the following information from eligible studies: first author’s name, publication year, location, research time, study register or abbreviation, sample size, average age, ratio of males, type of disease, experimental intervention, control, and follow-up. In the present analysis, the major outcome was PFS, and the secondary outcome was OS, as some patients changed the initial randomized treatment after disease progression. We assessed the methodological quality of the included trials using the Cochrane Collaboration’s tool, which assigns grades of “high risk”, “unclear risk”, or “unclear risk” of bias across the seven specified domains [17].

We initially conducted a pairwise meta-analysis using a random effect model, as this model is likely the most appropriate and conservative methodology accounting for between-trial heterogeneity within each comparison [18]. For dichotomous outcomes, odds ratios (ORs) or logarithm transformation with 95% confidence intervals (CIs) were calculated to determine the sizes of the effects. We also used a random effect network meta-analysis for mixed multiple treatment comparisons because this analysis fully preserves the within-trial randomized treatment comparisons in each trial [19]. To rank the treatments for each outcome, we used the surface under the cumulative ranking (SUCRA) probabilities [20]. Comparison-adjusted funnel plots were used to determine whether small-study effects were present in the analysis conducted in the present study [21]. All tests were two-tailed, and a p value of less than 0.05 was considered statistically significant. Data analyses were performed using STATA software (version 14.0; Stata Corporation, College Station, TX, USA).

In the present study, 747 articles were identified after the duplicates were removed. A total of 678 articles were excluded after the titles and abstracts were screened. The full texts of the remaining 69 articles were assessed, and the following types of studies were removed: non-randomized design (19); comparisons of the same type of chemotherapeutic drug (12); duplications or secondary studies (9); non-controlled studies (2); no desired outcomes (2); and other sarcoma studies (2). Eventually, 23 articles assessing a total of 5742 patients were included in the present systematic review [22‐44] (Fig. 1, Table 1).

The included studies were published from 1976 to 2016 and were researched from 1974 to 2014. The analysis contained several multicentre large-scale studies, such as The European and American Osteosarcoma Study Group-1 (EURAMOS-1), Osteosarcoma 2006 (OS2006), and the Symposium of the Cooperative Osteosarcoma Study Group (COSS-80). Many studies contained duplicate reports. Thus, we included relatively recently published studies and referred to the outcomes of the duplicate reports. All age groups of patients were included, and slightly more men than women were included. All studies included patients with osteosarcoma defined according to a pathological diagnosis. In addition, four studies included osteosarcoma patients without metastasis, two studies included metastasis patients, and one study included relapse patients. Most studies initiated chemotherapy prior to surgery. The longest median follow-up period was 9.4 years (Table 1). All included studies had an RCT design without blinding, and most randomizations were not rigorous. However, the assessed outcome was objective; thus, the overall quality of the included studies was not ideal but was acceptable (Additional file 1: Figure S1).

For chemotherapeutic drug application, we investigated all types of drugs used in the intervention arms and classified each of the drugs of the experimental arms by alphabetical order. The present study did not include a comprehensive analysis, reflecting the characteristics of applied chemotherapeutic protocols, as most application stages, durations, and dosages of drugs were different in different protocols (Table 2). Drugs showing no chemotherapeutic effect, such as granulocyte colony-stimulating factor (G-CSF) and muramyl tripeptide, were excluded. Drugs that may be included in chemotherapy, such as mistletoe, were included in the present analysis.

For the PFS analysis, we extracted all studies of 5-year PFS or the longest follow-up period for PFS. In the present study, we analysed 16 types of multi-drug regimens. Four multi-drug regimens were directly compared to a blank control, which indicated treatment without chemotherapy. In this analysis, the nodes were weighted according to the number of studies evaluated for each treatment, and the edges were weighted according to the precision of the direct estimate for each pairwise comparison (Fig. 2a). In network pairwise comparisons, the ABCDM (all protocol abbreviations are defined in Table 2) regimen was superior to the ACML (logOR, 1.38; 95% CI, 0.09–2.68) and Blank (logOR, 1.30; 95% CI, 0.19–2.41) regimens for the PFS outcome. The ABCDMP regimen was superior to the ACML (logOR, 2.14; 95% CI, 0.45–3.84), AML (logOR, 2.13; 95% CI, 0.00–4.26), Blank (logOR, 2.06; 95% CI, 0.50–3.62), and ML regimens (logOR, 2.24; 95% CI, 0.22–4.27), and the ABCDMPI regimen, combining ABCDMP with ifosfamide, was superior to the ABCDMP regimen alone (logOR, 0.84; 95% CI, 0.09–1.59). The ABCDMPI regimen was also superior to the ACML (logOR, 2.98; 95% CI, 1.13–4.84), AML (logOR, 2.97; 95% CI, 0.71–5.23), Blank (logOR, 2.90; 95% CI, 1.17–4.63), ML (logOR, 3.08; 95% CI, 0.92–5.24), and N (logOR, 2.75; 95% CI, 0.22–5.28) regimens in the network comparisons. Moreover, the ABCDMP regimen in combination with vincristine (ABCDMPL) was superior to the ACML (logOR, 1.89; 95% CI, 0.14–3.64), AMP (logOR, 0.46; 95% CI, 0.01–0.90), AMPF (logOR, 0.64; 95% CI, 0.11–1.18), and Blank (logOR, 1.80; 95% CI, 0.19–3.42) regimens. No other significant differences were found among these regimens (Additional file 2: Table S1). Based on the SUCRA rank, the ABCDMPI regimen was the most likely treatment to improve PFS in osteosarcoma patients (94.1%), followed by the ABCDMPL (81.9%) and ABCDMP (76.9%) regimens. Additionally, the comparison-adjusted funnel plot used to assess publication bias and determine the presence of small-study effects did not suggest any publication bias (Additional file 3: Figure S2a). In addition, some regimens were not included in the network meta-analysis, reflecting a disconnection, and a traditional meta-analysis showed no significant difference between interventions, except for APIZ compared to MIE (OR, 2.27; 95% CI 1.02–5.04) (Fig. 3).

For the OS analysis, we separated the regimens into two groups, reflecting the disconnection. The first group included AMP, AMPF, AMPI, AMPIE, AP, and ABCDMPL. Four regimens were directly compared to AMP, and we directly compared AP and ABCDMPL (Fig. 2b). In the network comparisons, the ABCDMPL regimen showed a significant advantage compared to the AMP (logOR, 0.47; 95% CI, 0.02–0.92), AMPF (logOR, 0.65; 95% CI, 0.01–1.29), and AP (logOR, 0.31; 95% CI, 0.04–0.57) regimens (Additional file 4: Table S2). The results showed that ABCDMPL was most likely the best regimen for improving the OS (94.7%) of osteosarcoma patients, followed by the AP (58.3%) and AMPIE (56.8%) regimens. The second group included ABCDM, ABCDMP, ACML, BCDM, ML, and N. Four regimens were directly compared to the Blank condition (Fig. 2c). In the network comparison, the ABCDM regimen was superior to the AML (logOR, 1.99; 95% CI, 0.14–3.84), Blank (logOR, 1.54; 95% CI, 0.37–2.70), and ML (logOR, 1.76; 95% CI, 0.03–3.49) regimens, and no other significant difference was found among comparisons (Additional file 5: Table S3). Regarding rank, ABCDM (89.4%) was most likely to be the best regimen, followed by N (70.1%) and BCDM (60.9%). The comparison-adjusted funnel plot showed no obvious publication bias (Additional file 3: Figure S2b and c). A comparison of regimens not included in the network meta-analysis revealed that APIR had a significant advantage over API in improving the OS (OR, 3.48; 95% CI, 1.17–10.32) of the patients (Fig. 3). However, this result was based on a single study and lacked precision and robustness.

There are several limitations to the present study. First, the present analysis was performed at a study level, not at an individual level. Second, for chemotherapy, cytoprotective agents might also improve the survival time of patients by reducing the chemotherapy-induced damage to normal tissue, but these drugs were not analysed in this study. Third, we did not perform the Grading of Recommendations Assessment, Development and Evaluation in the present analysis, as all included studies had an RCT design without blind concealment, and most of the results showed a low risk of imprecision.