We preliminarily screened 3725 literatures from PubMed, EMBASE, Web of Science, and Cochrane Library databases. After reading, 3697 literatures did not conform to inclusion criteria. Therefore, 28 literatures[6, 11-18, 21-26, 28-40] were finally included for meta-analysis (Figure 1). All the included studies (Table 1) were retrospective and had an evidence of 3b or 4 according to the criteria of the Center for Evidence-Based Medicine in Oxford, UK[41]. All observation studies had a quality score of 5 or higher (Newcastle-Ottawa scale) and were considered to have high quality.
5-year OS rate, 5-year EFS rate, and LR rate
The pooled data of 5-year OS rate consist of 22 studies[6, 11-18, 21, 22, 24-26, 28-32, 34, 36, 38] with 6742 patients. The 5-year OS rate was 49% (95%CI 45%-53%). The pooled data of 5-year EFS rate consist of 8 studies[6, 11-17] with 1243 patients. The 5-year EFS rate was 37% (95%CI 32%-43%). The pooled data of LR rate consist of 17 studies[11, 13, 14, 17, 21-23, 28-30, 32-38] with 1660 patients. The LR rate was 37% (95%CI 28%-46%) (Table 2; Figure. 1S).
Prognostic factors
The prognostic factors with similar variables were pooled in the meta-analysis. The details of meta-analysis results are shown in Table 3 and Figure 2S.
Sex
Seven studies (including subgroups)[16, 23-25, 29] comparing the overall survival between male and female were included. Values of I2 = 55.3% and p= 0.037 were obtained after the HR values of OS were merged, indicating that heterogeneity existed. A random-effect model was used to merge the HR=1.09, 95% CI: 0.87-1.37 and p=0.466, suggesting that OS is of no significant difference in sex.
Age
Nine studies (including subgroups)[14, 16, 23-26, 29, 31] compared the overall survival between the older and the younger subgroups. Values of I2 = 80.3% and p= 0.000 were obtained after the HR values of OS were merged, indicating that heterogeneity existed. A random-effect model was used to merge the HR=1.45, 95% CI: 0.98-2.16 and p=0.065, showing no significant difference in the overall survival between the older and the younger subgroups.
NF 1 mutation
A total of 13 studies[11, 13-16, 18, 24, 28-31, 34, 36] assessed the association between the NF 1 mutation and OS. Values of I2 = 34.1% and p= 0.110 were obtained after the HR values of OS were merged, indicating that heterogeneity did not exist. The pooled result via a fixed-effect model indicated that patients with NF 1 MPNST have poorer survival than patients with non-NF 1 MPNST (HR 1.56, 95%CI 1.35-1.79, p=0.000).
Tumor size
Ten studies[13, 14, 17, 22, 24, 25, 29-31, 38] evaluated tumor size (the large vs. the small) as a risk factor for OS. Values of I2 = 44.4% and p= 0.063 were obtained after the HR values of OS were merged, indicating that heterogeneity existed. A random-effect model was used to merge the HR=2.08, 95% CI: 1.59-2.71 and p=0.000. The results showed that large tumor size correlated with a significantly higher risk for poor prognosis.
Tumor depth
Five studies[14, 15, 24, 29, 34] evaluated relation between tumor depth related to fascia and OS. Values of I2 = 0.00% and p= 0.744 were obtained after the HR values of OS were merged, indicating that heterogeneity did not exist and a fixed-effect model was applied. Collectively, deep to fascia versus superficial to fascia significantly increased the risk of poor prognosis (HR 2.09, 95%CI 1.52-2.89, p=0.000).
Tumor site
A total of 6 studies[14-16, 18, 24, 34] evaluated the relation between the tumor site and OS. Five studies[15, 17, 18, 24, 34] compared the OS between trunk and extremity, with heterogeneity existing (I2 = 74.9% and p= 0.003) and a random-effect model applied. Trunk versus extremity had an increasing risk of bad prognosis (HR 1.79, 95% CI 1.07-3.00, p=0.025). Six studies[14-16, 18, 24, 34] compared the OS between head & neck and extremity, with no heterogeneity existing (I2 = 30.3% and p= 0.208) and a fixed-effect model applied. Head & neck versus extremity had an increasing risk of bad prognosis (HR 1.38, 95%CI 1.07-1.78, p=0.014).
Tumor grade
A total of 6 studies (including subgroups)[13-16, 25] assessed the association between the tumor grade and OS. Most of the studies[14, 16, 25] used AJCC 8 Stage to define tumor grade. Six studies (including subgroups) [13-16, 25] compared the OS between grade II and I, without heterogeneity existing (I2 = 43.7% and p= 0.114) and a fixed-effect model applied. Grade II had an increasing risk of poor prognosis compared to grade I (HR 1.43, 95%CI 1.06-1.91, p=0.017). Six studies (including subgroups) [13-16, 25] compared the OS between grade III and I, with heterogeneity existing (I2 = 86.9% and p= 0.000) and a random-effect model applied. Grade III had an increasing risk of poor prognosis compared to grade I (HR 3.21, 95%CI 1.36-7.54, p=0.008).
Metastases
There were 6 studies (including subgroups)[13, 16, 25, 26, 38] exploring OS and metastases included, with heterogeneity existing (I2 = 62.3% and p= 0.021) and a random-effect model applied. Patients with metastasis had poorer OS than those without. (HR 2.30, 95%CI 1.50-3.51, p=0.000).
Surgical margin
A total of 10 studies (including subgroups)[13-15, 18, 22-24, 29, 31] estimated surgical margin as a factor influencing OS. Most studies used R0, R1, and R2 to describe margin status, while some used negative and positive. When performing meta-analysis, we regard negative and positive as R0 and R2, respectively. Six studies (including subgroups)[13, 15, 22, 24, 31] compared the OS between R1 and R0, with no heterogeneity existing (I2 =22.3% and p= 0.267) and a fixed-effect model applied. R1 had an increasing risk of poor prognosis compared to R0 (HR 1.31, 95%CI 1.04-1.64, p=0.022). Ten studies (including subgroups) [13-15, 18, 22-24, 29, 31] compared the OS between R2 and R0, with no heterogeneity existing (I2 = 0.0% and p= 0.511) and a fixed-effect model applied. R2 had an increasing risk of poor prognosis compared to R0 (HR 2.40, 95%CI 1.96-2.95, p=0.000).
Chemotherapy
Seven studies[14, 16, 18, 24, 26, 29, 34] evaluated relation between chemotherapy and OS. Values of I2 =6.3% and p= 0.379 were obtained after the HR values of OS were merged, indicating that heterogeneity did not exist and a fixed-effect model was applied. Collectively, chemotherapy for MPNST is a significantly protective factor for OS (HR 0.70, 95%CI 0.59-0.83, p=0.000).
Radiotherapy
Seven studies[14-16, 18, 22, 24, 29] evaluated relation between chemotherapy and OS. Values of I2= 50.0% and p= 0.062 were obtained after the HR values of OS were merged, indicating that heterogeneity existed and a random-effect model was applied. Collectively, radiotherapy for MPNST is a significantly protective factor for OS (HR 0.65, 95%CI 0.49-0.88, p=0.005).
Because there were rare studies discussing the risk factors of EFS and LR, so the meta-analysis associated with risk factors of EFS and LR was not performed.
Sensitivity analysis and publication bias
The sensitivity analysis was performed in these groups. The pooled HR of age became statistical significance when to exclude one of these studies including Yuan, Z. N. et al.[14], Hwang, I. K. et al.[31], Fan, Q. et al.[16], and LaFemina, J. et al.[23]. The pooled HR of tumor site (Trunk vs. Extremity) became no statistical significance when to exclude one of these studies including Stucky, C. C. et al.[34], Longhi, A. et al.[17], and Anghileri, M. et al.[18]. The pooled HR of tumor site (Head & neck vs. Extremity) became no statistical significance when to exclude Valentin, T. et al.[15]. The pooled HR of tumor grade (Grade II vs. I) became no statistical significance when to exclude one of Mowery, A. et al.[25] and Miao, R. Y. et al.[13]. The pooled HR of surgical margin (R1 vs. R0 resection) became no statistical significance when to exclude Ma, C. et al.[22]. The results of the other meta-analysis did not change after removal of any one research (Figure 3S and Figure 4S).
The Egger’s test was completed to examine the existence of publication bias. The possibilities of publication bias were found in tumor site (Trunk vs. Extremity, p=0.041; Head & neck vs. Extremity, p=0.047), tumor grade (Grade III vs. I, p=0.006), surgical margin (R2 vs. R0 resection, p=0.019), and radiotherapy (With vs. Without radiotherapy, p=0.006). The Egger’s test resulted in p≥0.05 in the other groups and indicated that the possibilities of publication bias can be excluded (Figure 5S and Figure 6S).