Study selection
We retrieved a total of 308 articles from Medline (n = 39), Embase (n = 121), PubMed (n = 103), and Web of Science (n = 45), and read the relevant references to obtain three studies. After excluding 154 duplicates, we reviewed the titles and abstracts of the remaining studies and excluded 113 unrelated articles. After excluding review articles and studies that analyzd patient data from the same institution in the same time period, irrelevant data, and unavailable data, 15 articles were eventually included (figure 1). A total of 4377 patients with GC underwent gastrectomy and lymphadenectomy in the included studies. Of these studies, seven were performed in China, five in Japan, two in South Korea, and one in Germany. All of the studies contained at least one risk factor for SHLNs metastasis. General characteristics and quality assessments of the included studies are listed in Table 1.
Fig 1. Flow diagram of the search and selection process.
Age
Five studies included 159 of the total 927 patients aged < 60 years with No. 10 LN+ and 179 of 896 patients aged > 60 years with No. 10 LN+ (OR = 0.90, 95% CI = 0.54–1.48, I2 = 63%, p = 0.67). There were no significant differences between the two groups (figure 2a). Because of the heterogeneity, we conducted sensitivity analysis by eliminating studies one by one, and found that the heterogeneity decreased significantly when Aoyagi (2010) was removed (I² = 18%).
Sex
Thirteen studies, with 2888 males and 1132 females, were included in the gender analysis, which revealed that were no significant differences between males and females (OR = 0.88, 95% CI = 0.72–1.07, I2 = 24%, p = 0.19) (figure 2b).
Tumor size
Five studies were included in the analysis of tumor size. Because of the difference in the statistics analyzing tumor diameter, we divided the information into two groups (> 5 cm vs. < 5 cm and > 10 cm vs. < 10 cm). There was no significant difference between the groups of > 10 cm and < 10 cm, and the studies had high heterogeneity (OR = 0.58, 95% CI = 0.21–1.58, I2 = 71%, p = 0.28). However, when the tumor diameter was demarcated by 5 cm, there was a significant difference between the two groups, and the studies had no heterogeneity (OR = 4.89, 95% CI = 2.98–8.03, I2 = 0%, p < 0.01) (figure 2c).
Tumor location
Eleven studies, containing 341 patients with tumors located in the greater curvature (Gre) and 2809 with tumors located in other locations, were included in the analysis of tumor location. There was a significant difference between tumors found in the greater curvature and those found elsewhere (OR = 3.10, 95% CI = 1.92–5.02, I2 = 54%, p < 0.01) (figure 2d). We conducted sensitivity analysis by eliminating studies one by one, and found that the heterogeneity did not change significantly following elimination of any of the studies.
Lauren’s type
Four articles containing 929 patients mentioned Lauren’s type, which included diffuse and intestinal types. After the data was combined, the results showed a significant difference between diffuse and intestinal types (OR = 2.91, 95% CI = 1.84–4.59, I2 = 0%, p < 0.01) (figure 2e).
Borrman type
Nine articles included information regarding Borrman’s type. Borrman classification is divided into four types: type I, type II, type III, and type IV. We combined types I–III and analyzed the data comparing types I–III and type IV. The heterogeneity of the two groups was small and there was a significant difference between the groups (OR = 2.49, 95% CI = 1.84–3.37, I2 = 0%, p < 0.01) (figure 2f).
Histological differentiation
Data on histological differentiation was included in 12 studies. We set the poorly differentiated and undifferentiated types as the exposure group, and the moderately differentiated and well differentiated types as the control group. Following analysis, studies were removed one by one for sensitivity analysis. There was no significant change in heterogeneity and there was a significant statistical difference betweentwo groups (OR = 2.29, 95% CI = 1.80–2.92, I2 = 25%, p < 0.01) (figure 2g).
Depth of invasion
Data regarding depth of invasion was included in 10 studies. We set T3 and T4 as the exposure group, and T1 and T2 as the control group. There was no significant heterogeneity in either group and there was a significant statistical difference between T1-2 and T3-4 (OR = 6.39, 95% CI = 4.04–10.12, I2 = 1%, p < 0.01) (figure 2h).
Lymph node metastases
Nine studies mentioned lymph node metastases, which included N1, N2, and N3. We grouped N1 and N2–3 separately. After the data was combined, the results showed a significant difference (OR = 6.96, 95% CI = 4.64–10.44, I2 = 44%, p < 0.01) (figure 2i). Because of the heterogeneity, we conducted sensitivity analysis by eliminating studies one by one, and found that the heterogeneity disappeared when Huang (2009) was removed (I2 = 0%).
Distance metastasis
Three articles containing 1272 patients included data on distant metastases. After the data were combined, there was no heterogeneity and there was a statistically significant difference (OR = 8.66, 95% CI = 5.53–13.56, I² = 0%, p < 0.01) (figure 2j).
Neurological, vascular, and lymphatic invasion
Because several studies combined data on blood vessels and lymphatics, we excluded these studies. Two studies included data regarding neurological invasion. After the data was combined, the results did not show a significant difference between the two groups (OR = 1.72, 95% CI = 0.98–3.03, I2 = 16%, p < 0.01) (figure 2k). Three studies mentioned vascular invasion. After the data was combined, the results showed a significant difference between the two groups(OR = 2.57, 95% CI = 1.21–5.47, I² = 0%, p = 0.01) (figure l). Three articles mentioned lymphatic invasion. After the data was combined, the results showed a significant difference between the two groups(OR = 3.41, 95% CI = 1.81–6.44, I² = 0%, p < 0.01) (figure 2m).
TNM stage
Eight articles mentioned TNM stage. The TNM stage was divided into four types: type I, type II, type III, and type IV. We combined types I–III and compared this group with type IV. The heterogeneity of the two groups was small and there was a significant difference between types I-II and types III-IV (OR = 22.70, 95% CI = 11.57–44.56, I² = 0%, p < 0.01) (figure 2n).
Other groups with positive lymph node metastasis
Four studies referred to other regional lymph node metastases, which were associated with SHLNs metastasis. We collected their combined values and effects; these are listed in Table 2. The results showed that other regional lymph nodes, with the exception of No. 5 LN (p = 0.14) and No. 8a LN (p = 0.10), are associated with SHLNs metastasis. We performed sensitivity analysis by changing the effect model, and found no change.
Figure 2 Forest plot analysis of risk factors.
- age, b. sex, c. tumor size, d. tumor location. e. Lauren’s type, f. Borrman type, g. histological differentiation, h. depth of invasion, i. lymph node metastases, j. distance metastasis, k. neurological invasion, l. vascular invasion, m. lymphatic invasion, n. TNM stage.
Publication bias
Publication bias was assessed only when more than 10 studies were included in the risk factor analysis. There was no obvious asymmetry in the funnel plot of histological differentiation (figure 3). Similarly, other aggregated data was not found to exhibit publication bias.
Figure 3 Funnel plot of histological differentiation.