Statistical analysis

Review Manager (Version 5.2; The Cochrane Collaboration, Copenhagen, Denmark) was used for statistical analysis. Selection of fixed-effects or random-effects model was determined by statistical heterogeneity among the included trials, which was evaluated using the chi-square test and quantified using I2 statistic. The existence of homogeneity was considered unreasonable for I2>50% and P<0.10. In general, HR >1 implied that the chemotherapy group had worse prognosis than the control group. Comparisons between cisplatin-based and single UFT chemotherapy used indirect treatment comparison (ITC) software (Version 1.0; Canadian Agency for Drugs and Technologies in Health, Ottawa, Canada) proposed by Bucher et al23 which estimated the relative effects of treatment A and treatment C through the common control treatment B. Finally, publication bias was evaluated by funnel plots and further quantified by Begg’s and Egger’s tests24,25 with STATA software (Version 12.0; StataCorp LP, College Station, TX, USA). Sensitivity analysis was also conducted using STATA. Two-sided P-value <0.05 was considered statistically significant.

Continue Reading


Characteristics of included RCTs

There were 580 trials searched from databases and six from relative references lists. We identified 25 potentially eligible trials, eleven of which were excluded from our meta-analysis, including four trials that lacked survival data regarding early stage NSCLC, five trials where patients received postoperative chemotherapy plus radiotherapy, and two trials that were published previously26,27 but had been updated afterward.15,28 Finally, 14 RCTs with 3,923 patients (chemotherapy group: 1,987 patients; control group: 1,936 patients) were included in this analysis.11–15,28–36 Two chemotherapy groups were separately compared with the same control group in a study by Imaizumi et al32; so we treated this trial as two independent studies. To avoid repeat counting, the control group in these two studies was counted once in the analytical process. The screening process was briefly shown in Figure 1. The available median follow-up time and chemotherapy compliance were ranged from 51 to 120 months and 53% to 86%, respectively. The baseline characteristics of chemotherapy group and control group are balanced and summarized in Table 1.

(To view a larger version of Table 1, click here.)

All RCTs included a statement about randomization, and detailed descriptions were listed in eleven trials. Blinding of participants and personnel assessment was only described in one trial.14 However, it was not always feasible to blind in studies involving surgery. No quality difference was observed in the included RCTs except for in the study of Strauss et al35 which had a higher risk than others due to early termination (data not shown).

Overall analysis of survival

Data for 5-year disease-free survival (DFS) was available in eleven trials11,12,14,28,29,31–33,35–37 with a total of 2,937 patients. Although six trials11,12,28,35–37 showed that DFS was not significantly increased in chemotherapy group, the combined analysis displayed that survival benefit was achieved with postoperative chemotherapy (P=0.005; Figure 2A), and the recurrence risk was greater in the control group compared to the chemotherapy group (41.3% versus 32.2%). Because significant heterogeneity existed among trials (I2=71%), random-effects model was adopted and achieved a combined HR of 0.71 (95% CI, 0.57–0.90). There was evidence of publication bias by Egger’s test (P=0.029).

(To view a larger version of Figure 2, click here.)

There were 13 trials11–15,28–35 (3,656 patients) with complete OS information. The analysis showed that the survival of chemotherapy group was significantly better than control group without heterogeneity in trials (P<0.00001, I2=0%). The combined HR with fixed-effects model was 0.74 (95% CI, 0.67–0.83; Figure 2B), which represented a 26% relative reduction of death risk in chemotherapy group. No publication bias was detected by Begg’s and Egger’s tests.