Literature search results
From the database, a total of 108 articles were included according to our retrieval strategy. First, we removed 62 duplicates. After reviewed the title and abstract, we excluded 20 articles, comprising 5 reviews, one comment, and 14 articles irrelevant to the prognosis of NKILA. In addition, we removed 17 articles because of no enough clinical data and no grouping high/low expression of NKILA. After reviewed the full-text, one provided incorrect data, so we excluded it. Ultimately, eight articles were included according to the inclusion and exclusion criteria, including 858 patients. The procession of the literature search was shown in Figure1.
Features of included studies
The general features of the enrolled studies were shown in Table 1. The publication year of the enrolled studies were from 2016 to 2020, and all came from China. The cancer types included tongue squamous cell carcinoma[12],esophageal squamous cell carcinoma[13, 14], laryngeal cancer[15], rectal cancer[16], hepatocellular carcinoma[17], colorectal cancer[18], Nasopharyngeal carcinoma[19]. All studies detected the expression level of NKILA by qRT-PCR. Among them, five studies could be directly obtained HR and 95%CI of OS, two studies provided the Kaplan-Meier survival curve, and the remaining one article only provided clinical parameters.
Association between lncRNA NKILA expression and OS
In our study, 7 studies reported the relationship between NKILA expression level and OS. The results indicated that high NKILA expression predicted a longer overall survival time(pooled HR=0.51, 95%CI: 0.35-0.74, P<0.001, random-effects model) but there was significant heterogeneity(I²=75.1%, p<0.001)(Figure 2A).Therefore, we adopted the random effect model. In order to explore the source of heterogeneity, we conducted the subgroup analysis according to cancer type and sample size. The results of the subgroup analysis for OS were shown in Table2. Our results suggested that NKILA expression level was related to well OS in digestive system tumors (HR=0.59, 95%CI: 0.39-0.90, p<0.001, random-effects model) and others tumors (HR=0.39, 95%CI: 0.26-0.60, p=0.014, random-effects model) (Figure 2C). In the subgroup analysis of sample size, the NKILA expression level obviously associated with favorable OS in sample size≤90 group(HR=0.43, 95%CI: 0.26-0.71, p=0.001, random-effects model) and sample size range from 90 to 150 (HR=0.48, 95%CI: 0.34-0.67, p<0.001, random-effects model) (Figure 2D). Nevertheless, we found that combined HRs showed significant heterogeneity in the digestive system tumor group and the sample size >150 group. Importantly, the study of Jiang et al was in both groups. We hypothesized that heterogeneity might be due to a study published by Jiang et al[18]. Finally, Heterogeneity disappeared When we removed the study of Jiang et al and a stronger relationship between NKILA expression and OS (pooled HR=0.45, 95%CI:0.35-0.59, P<0.001, fixed-effects model) in the remaining 6 articles (Figure 2B).
Association between lncRNA NKILA expression and clinical parameters
A total of 7 studies with 768 cancer patients were used to analyze the association between lncRNA NKILA and TNM stage (III/IV vs. I/II) of cancer. The result showed that patients with high expression of lncRNA NKILA associated with an earlier clinical stage (III/IV vs. I/II, OR =0.34, 95%CI: 0.25-0.46, p< 0.001, fixed-effects model) and without significant heterogeneity (I²=0.0%, p=0.658) (Figure 3A). Similarly, 4 studies with 373 cancer patients were included in LNM (positive vs. negative) analysis. The data analysis displayed that the overexpression of lncRNA NKILA was more likely to be associated with negative lymph node metastasis (OR=0.27, 95%CI:0.18-0.42, P<0.001, fixed-effects model) and no significant heterogeneity(I²=0.0%, p=0.704) (Figure 3B).
Publication bias
Despite the small number of studies included(n<10), we examined the included studies by Begg's test. No significant publication bias was found in OS(p=0.368>0.05), TNM(p=0.072>0.05) and LNM(p=1.000>0.05) group. The funnel plot of Begg's test was shown in Figure 4.
Sensitivity analysis
Sensitivity analysis checked the stability of the results by removing one study at a time. We did not find that removing one study could affect the stability of the merged HR for OS (Figure 5A). In addition, our results also confirmed that the stability of the merged OR for the TNM stage (Figure 5B) and LNM (Figure 5C) would not be affected.