The flowchart summarizing the study selection process following the PRISMA statement was reported in Figure 1. A total of 1626 studies were identified in the initial search, including 108 studies from PubMed, 1131 studies from EMBASE, 148 studies from Cochrane Library, 195 studies from Web of Science, 44 studies from CNKI, and one academic meeting abstracts (Table 4). After screening based on inclusion and exclusion criteria, 372 articles were retrieved as eligible and then reviewed by two independent reviewers. Finally, fourteen studies, including 1404 patients were included in the final meta-analysis[12-25].
In terms of patient race group, all studies were performed in patients of Asian backgrounds. There were one study published in English and 13 studies in Chinese. The characteristics of the studies included in this meta-analysis were listed in Table 2 in detail.
Meta-analysis results and bias assessment results
The main results, including heterogeneity tests, effect models adopted accordingly, and the pooled SMDs with their 95% CI and the P value of this meta-analysis were presented in Table 5. The Galbr plots for the association between the use of narcotic drugs and postoperative cognitive function were shown in Figure 2, suggesting that there was no heterogeneity only among the 10 studies[12, 13, 15, 17, 19-22, 24, 25] with continuous data focusing on MMSE scores 7 days after the surgery, but not among other comparisons. Using fixed-effects model, the pooled SMD for the 10 studies was -0.422 (95 % CI: -0.549, -0.295, Z = 6.52; P < 0.001), indicating that in terms of MMSE scores 7 days postoperatively, propofol has a greater adverse effect on cognitive function in the elderly patients with lung cancer than sevoflurane. The pooled SMD in issues of preoperative MMSE scores suggested no statistical difference (SMD -0.038, 95 % CI: -0.274, 0.198, Z = 0.31; P = 0.753). Then the pooled SMD in issues of postoperative MMSE scores at different time points were calculated using the random-effects model (except the MMSE score-7d). There were significant differences in issues of MMSE 6h (11 studies; SMD -1.391, 95% CI -2.024, -0.757; p < 0.001), MMSE 24h (14 studies; SMD -1.106, 95% CI -1.588, -0.624; p < 0.001), MMSE 3d (11 studies; SMD -1.065, 95% CI -1.564, -0.566; p < 0.001), MMSE 7d (10 studies; SMD -0.422, 95% CI -0.549, -0.295; p < 0.001), and the serum S100beta concentration at 1 day after surgery (13 studies; SMD 0.746, 95% CI 0.475, 1.017; p < 0.001) (Figure 3). We assessed the risk of bias using the Cochrane risk of bias tool. Table 6 reported detailed results from the risk of a bias assessment tool.
To assess if a single study could affect the final SMDs, each study was removed one time and the data re-pooled. The analysis results demonstrated that the pooled SMDs were not affected by deleting every single study. Figure 4 showed sensitivity analysis results in issues of postoperative pain and propofol/remifentanil use.
The contour-enhanced funnel plots (this term’s explanation could be seen in Table 3) were adopted to estimate potential publication biases, showing that most of the studies had missing areas for low statistical significance (the left-hand side of the plot), indicating no publication bias in present studies (Figure 5).