Studies included in the meta-analysis
A total of 487 papers were identified as relevant to the search words. After screening the title and reading the abstract, 134 articles were selected to for full text review, 26 studies were excluded as review or meta-analysis, and 77 articles were removed for not involving survival data or prognostic results. In further analysis of the remaining 31 potential articles, two articles were excluded: one reported duplicate data and the other one had insufficient data. Finally, 29 publications with 105060 patients were included[4, 6–15, 23–40]. The flowchart of selection of studies and reasons for exclusion is presented in Fig. 1.
The characteristics and quality assessment results of the articles selected are summarized in Table 1. Of these, five articles only took step sectioning[15, 23–26], nine used step sectioning combined with immunohistochemical staining[8, 9, 12, 14, 29, 31, 34, 35, 37], four applied hematoxylin and eosin (H&E) staining with immunohistochemical staining[4, 6, 10, 32], while the rest only utilized immunohistochemical staining[7, 11, 13, 27, 28, 30, 33, 36, 38–40]. The breast cancer stage was described detailly in only 13 articles[4, 7, 10–12, 15, 20, 23, 24, 26, 28, 32, 33, 36], and follow-up duration ranged from 3.1 to 25.6 years. The use of adjuvant systemic therapy was not reported in 8 articles[14, 30, 31, 33, 34, 38–40], while it was applied to all or some patients in 7 articles[9, 10, 12, 13, 15, 32, 36], and none in the rest. Among the included articles, seven divided the patients with occult metastases into different subgroups[7–9, 12, 15, 29, 35]. As all subgroups analysis were regarded as separated studies, 36 studies were chosen for evaluation in this meta-analysis.
After a five-year follow-up, the results showed that occult metastases group was associated with poorer DFS (RR = 1.497; 95% CI = 1.341–1.671; I2 = 28.6%) (Fig. 2) and OS (RR = 1.440; 95% CI = 1.186–1.749; I2 = 71.7%) (Fig. 3). After a ten-year follow-up, the results also revealed poorer DFS (RR = 1.688; 95% CI = 1.256–2.268; I2 = 66.2%) (Fig. 4) and OS (RR = 1.477; 95% CI = 1.279–1.705; I2 = 58.3%) (Fig. 5) in patients with occult metastases. As obvious heterogeneity was observed in the study of 5-yr OS and 10-yr FDS, a random-effects model was utilized in these analysis. The results of the sensitivity analysis were consistent after excluding several studies[14, 33], with the confidence interval of RR not significantly decreasing (Fig. 6).
As including almost all studies, the studies of 5-yr DFS were conducted by using funnel plots and Egger’s test to assess publication bias. The funnel plot was approximately symmetrical (Fig. 7) and the result of Egger’s test (P = 0.567) revealed no obvious publication bias among the studies.