Subgroup analysis of survival by stage
Eight trials12,14,28,31–33,35,36 (2,833 patients) allowed for a quantitative aggregation analysis of DFS for stage I, among which three trials14,31,35 (550 patients) reported available information on stage IB. The combined HR of stages I and IB was 0.68 (P=0.002; Figure 3A) and 0.55 (P=0.02; Figure 3B), respectively. An evaluation of DFS for stage IA and II patients could not be conducted due to the limited data. In terms of OS, survival in the chemotherapy group was significantly improved for patients with stage I (HR: 0.74, P<0.00001; Figure 4A) and IB (HR: 0.74, P=0.0003; Figure 4B). There was no statistically significant benefit for stage IA patients (HR: 0.76, P=0.43; Figure 4C), though there was a positive trend for improved survival in the chemotherapy group. No publication bias was detected by Begg’s and Egger’s tests. In this analysis, the OS data on stage II patients was not sufficient to perform a reliable analysis.
(To view a larger version of Figure 3, click here.)
(To view a larger version of Figure 4, click here.)
Subgroup analysis of survival by chemotherapy regimens
Applicable DFS information was analyzed in nine cisplatin-based trials (1,689 patients) and three single UFT trials (1,298 patients). Cisplatin-based chemotherapy showed results consistent with the overall DFS, which favored postoperative chemotherapy (HR: 0.61, 95% CI, 0.47–0.81, P=0.0005; Figure 3C). There was evidence of publication bias by Egger’s test (P=0.029). However, single UFT chemotherapy did not show a DFS benefit with an HR of 1.04 (95% CI, 0.76–1.43, P=0.81; Figure 3D) without publication bias. OS information was available for analysis in nine cisplatin-based trials (1,854 patients) and five single UFT trials (1,802 patients). Compared with control groups, both cisplatin-based (HR: 0.75, P<0.0001; Figure 4D) and single UFT (HR: 0.72, P=0.002; Figure 4E) chemotherapy showed survival benefits in the combined OS analysis. Some evidence of publication bias was identified in DFS analysis of cisplatin-based chemotherapy by Begg’s (P=0.048) and Egger’s tests (P=0.045). ITC demonstrated that cisplatin-based chemotherapy had a longer DFS than single UFT chemotherapy (HR: 0.587, 95% CI, 0.387–0.89, P=0.04), but it failed to show difference of OS between the two regimen types (HR: 1.042, 95% CI, 0.816–1.33, P=0.90).
The toxicity was assessed in 1,692 (97%) of 1,745 patients assigned to chemotherapy group. One trial did not collect toxicity data.15 Grade 3–4 neutropenia, nausea and vomiting, thrombocytopenia, and infection were observed in 16.4%, 10.7%, 2%, and 1.9% of the patients who received cisplatin-based chemotherapy, respectively. Incidence of other adverse effects like sensory neuropathy, anemia, and diarrhea were less than 1%. Incidence of grade 3 or 4 nausea/vomiting and anorexia were observed in 0.8% and 0.7% of the patients who received single UFT chemotherapy, respectively. Only four treatment-related deaths (0.2%) occurred in chemotherapy group (data not shown). These findings indicate that chemotherapy toxicity was mild and well tolerated.
Sensitivity analysis and publication bias
Although we applied a random effects model that took variation across studies into consideration, heterogeneity was still significant in DFS analysis. Therefore, we conducted sensitivity analysis to search for the source of heterogeneity. The findings showed some trials might influence integral heterogeneity, but the results of DFS did not essentially change after excluding these individual trials, so the results reported in this study were stable and acceptable. Publication bias by Begg’s and Egger’s tests is summarized in Table 2.