3.1. Eligible studies and their characteristics.
After initial search of the databases, we identified 59 potentially relevant reports from Pubmed and Medline (n=27) and Web of Science (n=32). After excluding repeated publications, a final 36 articles were examined. 14 articles were eliminated for analyses for the following reasons: (1) the reports were review or abstract (n=5), (2) non-English articles (n=1), (3) the reports were not related to adjuvant therapy in UCCC (n = 8). Full-text review were performed on 22 articles and 1 manually inquired articles. After carefully performing data extraction, an addition of 9 articles was removed from the analyses due to insufficient data. In the end, 12 articles were included in this meta-analysis (5-17) (Supplemental Figure 1).
3.2. Characteristics of the included studies
12 articles included consisted 3845 patients in total. The quality assessment and main characteristics of the included studies were summarized in Table 1 and Supplemental Figure 2 respectively. All studies selected were assessed to be in a good quality. The follow-up time of patients was more than 60 months to achieve the 5-year overall survival.
3.3. Comparison of OS between CT and not-CT groups
In order to compare OS between patients receiving chemotherapy (CT group) and patients receiving other therapy except chemotherapy (not-CT group), 7 trials from 6 studies (Hsu et al 2014, Chang-Halpenny et al 2016, Nguyen et al 2017, Sari et al 2017, Hong et al 2017, Armbrsuter et al 2019) were included in the pooled analysis. Data extracted from 494 patients receiving CT and 1860 patients receiving not-CT were summarized. Pooled data revealed no difference in OS between the two groups (p=0.37, I2=7%). Consequently, a fixed-effect model was applied to estimate the overall outcomes. The OR of CT versus not-CT was 0.74 (95% CI: 0.58-0.95, p= 0.02), indicating a significant higher 5-year OS in CT group when compared to the not-CT group [Figure 1]. When comparing patients receiving CT versus those without any adjuvant therapy, the OR was 1.01 (95% CI: 0.39-2.58, p=0.99), suggesting that OS was not significant between these groups [Supplement Figure 3].
3.4. Comparison of OS between RT and not-RT groups
For OS between patients receiving radiotherapy (RT group) and patients receiving other therapy except radiotherapy (not-RT group), 9 trials (Creasman et al 2004, Thomas et al 2007, Rauh-Hain et al 2009, Hsu et al 2014, Chang-Halpenny et al 2016, Nguyen et al 2017, Sari et al 2017, Shinde et al 2018, Armbruster et al 2019) were included in the pooled analysis. Extracted data from 619 patients receiving RT and 1133 patients receiving not-RT were summarized. Importantly, pooled data revealed significant difference between the OS of these two groups (p=0.02, I2=56%). So, a random-effect model was utilized to estimate the overall outcomes. Results revealed that OR of RT versus not-RT was 0.83 (95% CI: 0.47-1.46, p=0.52), indicating no significant difference of overall survival between RT and not-RT groups [Figure 2]. Comparing OS for patients receiving RT with those without any adjuvant therapy, the OR was 0.94 (95% CI: 0.33- 2.65, p=0.90) [Supplement Figure 4].
3.5. Comparison of OS between patients receiving CT plus RT and those with CT or RT alone
For OS comparison between patients receiving CT plus RT and those with CT or RT alone, 5 trials (Hsu et al 2014, Chang-Halpenny et al 2016, Sari et al 2017, Mahdi et al 2016, Armburster et al 2019) were included in the pooled analysis, in which 105 patients receiving CT plus RT and 168 receiving CT or RT alone were analyzed. Pooled data revealed significant difference in OS between these two groups (p=0.51, I2=0%). Consequently, a fixed-effect model was applied to estimate the overall outcomes. The OR of CT plus RT versus CT or RT therapy alone was 0.95 (95% CI: 0.53-1.72, p=0.88), indicating that the OS of these groups showed no significantly difference [Figure 3]. For comparison of OS between patients receiving CT plus RT and those with no adjuvant therapy, 4 trials (Hsu et al 2014, Chang-Halpenny et al 2016, Sari et al 2017, Armburster et al 2019) were included in the pooled analysis with extracted data from a total of 53 patients receiving CT plus RT and 107 receiving no adjuvant therapy. The OR of CT plus RT versus no adjuvant therapy groups was 1.07 (95% CI: 0.46-2.50, p=0.87), indicating that there was no significant difference in OS between these groups using a fixed-effect model for overall outcome estimation [Supplemental Figure 5].
3.6. Comparison of PFS between CT and not-CT groups
In order to compare PFS between patients receiving CT and not-CT, 2 trials (Hsu et al 2014, Nguyen et al 2017) were included in the pooled analysis with 47 patients receiving CT and 140 receiving not-CT. Pooled data revealed no significant difference in PFS between these two groups (p=0.30, I2=7%), and therefore, a fixed-effect model was applied to estimate the PFS outcomes. The OR of CT versus not-CT was 0.44 (95%CI: 0.17-1.13, p= 0.09), indicating that close to being statistically significant difference in PFS between CT and not-CT groups [Supplemental Figure 6].
3.7. Comparison of PFS between RT and not-RT groups
For PFS comparison between patients receiving RT and not-RT, 2 trials (Thomas et al 2007, Hsu et al 2014) were included in the pooled analysis. In particular, extracted data from 58 patients receiving RT and 79 receiving not-RT revealed statistically significant difference of PFS between these two groups (p=0.0004, I2=92%), and therefore, a random-effect model was applied to estimate the overall outcomes. The OR of RT versus not-RT was 1.18 (95%CI: 0.06-23.64, p=0.91), indicating no significant difference of PFS between patients receiving RT and not-RT groups [Supplemental Figure 7].
3.8. Meta-regression and subgroup analyses
The heterogeneity issue between all included studies was addressed. We performed meta-regression analyses to explain expected heterogeneity in the comparison of OS between patients receiving RT and not-RT, (I2=56%), by testing the effects of factors like sample size (case number >80 versus <80), nationality (Asia versus Europe and US) and publication year (before versus after 2010). Interestingly, the results indicated that difference in nationality contributed significantly to the overall heterogeneity (Z=0.04, p=0.0466) (Table 2). Therefore, subgroup analyses were performed. Data revealed that analyses on left 8 publications from Europe and US with a total number of 1660 patient indicated no heterogeneity (I2=13%). The pooled data revealed OR of RT versus not-RT was 0.61 (95% CI: 0.46-0.82) and 5-year OS in RT group was significantly higher compared to the not-RT group using a fixed-effect model (p= 0.001) (Supplemental Figure 8). Analysis on patients from Asia was not performed due to limited number of publication (n=1) (Table 3).
3.9. Publication bias
In CT and not-CT groups, funnel plots formed a very distinctive symmetrical funnel shape with log ORs of the OS. These results implied that no publication bias was detected in the present study (Begg’s test, p=0.07, Egger’s test, p=0.62) (Supplemental Figure 9). In RT and not-RT groups, no evidence of publication bias was detected as well (Begg’s test, p=0.61, Egger’s test, p=0.39) (Supplemental Figure 10).