Initially, basing on the above mentioned search strategy, a total of 174 studies were identified (Pubmed = 24; Embase = 38; Cochrane library = 112; other = 0). Of these, 35 studies were removed due to duplication. According to the criteria mentioned above, the remaining 134 studies were excluded, and five studies were included in our final analysis [8, 19-22]. The flow diagram describing the study search is displayed in Figure 1.
The main characteristics of included studies are summarized in Table 1. Two and four studies assessed the incidence of POD [8, 20] and POCD [19-22] (a study contains both POD and POCD ), respectively. Three studies reported cardiac surgery [8, 20, 21], while the other two studies described non-cardiac surgery [19, 22]. Furthermore, the assessment methods of cognitive function were different, including neuropsychologic battery tests, Mini-mental state examination (MMSE) scores, International Study of Postoperative Cognitive Dysfunction (ISPOCD), and the Confusion Assessment Method adapted for the ICU (CAM-ICU) scale.
Risk of bias
The methodological bias of the eligible studies is presented in Figure 2. Random sequence generation was considered as low risk of bias in all included studies, while allocation concealment was described in only two RCTs [8, 19, 20]. For performance bias, three RCTs reported an unclear risk [8, 19, 22], whereas the remaining two RCTs were assigned as low risk [20, 21]. All included studies were confirmed as a low risk of detection, reporting, and other biases. Three RCTs have a high risk of attrition bias [19-21]. Some participants of studies were possible to be lost due to the long-term follow-up period (up to 30 days).
Primary outcome - incidence of POD
There are two studies that reported the incidence of POD, and the data were described as the number of patients [8, 20]. Langer et al.  performed the assessment of POD in the late afternoon with the CAM-ICU scale, while Siepe et al.  conducted it on 48 hours after surgery with MMSE scores. The study indicated a trend that patients in the LP group (89 participants) had a higher incidence of POD than those in the HP group (94 participants) (RR 3.30, 95% CI 0.80 to 13.54, P = 0.10, I2 = 15%), but the difference did not have a clinical significance (Fig. 3).
Primary outcome - incidence of POCD
For the incidence of POCD, there were four studies included [19-22]. Three studies assessed POCD at over three months postoperatively [19, 20, 22], while one study reported the values of both seven days and three months after surgery . To avoid repeated counting and ensure the accuracy of the results, we only obtained the data reported three months postoperatively. In this meta-analysis, the incidence of POCD in the LP group and the HP group was 9.5% and 7.5%, respectively, showed no significant difference (RR 1.26, 95% CI 0.76 to 2.08, P = 0.37, I2 = 0%) (Fig. 4).
Four studies reported postoperative mortality in 638 patients [8, 19-21], which no significant difference was observed between the LP group and the HP group (RR 0.86, 95% CI 0.14 to 5.37, P = 0.88, I2 = 44%). The length of hospital stay (described as days) data were available for 638 patients across four studies [8, 19-21]. It was noted that the value of the LP group was lower than the HP group, but the difference was so small that it did not have a statistical significance (MD 0.37, 95% CI -0.17 to 0.91, P = 0.18, I2 = 0%). Data on the length of ICU stay (described as hours) was extracted from three studies evaluated 537 patients [8, 19, 21], indicating that the time of the LP group was longer compared to the HP group (MD 1.82, 95% CI 0.83 to 2.82, P = 0.0003, I2 = 0%). Two studies reported MV time of the two groups [8, 21], and showed no significant difference (MD 0.40, 95% CI -1.26 to 2.06, P = 0.64, I2 = 58%). The secondary outcomes of this study are shown in Table 2. Besides, we converted data described as median and IQR to mean and SD (Table S2).
For POCD, there were two studies described cardiac surgery [19, 21] and non-cardiac surgery [20, 22], respectively. We further conducted a subgroup analysis of cardiac surgery versus non-cardiac surgery. When we excluded the results of non-cardiac surgery, no significance was found between the LP and the HP groups (RR 1.16, 95% CI 0.63 to 2.12, P = 0.64, I2 = 0%; 389 participants, Fig. 5). Also, there was no obvious difference between the subgroups (P = 0.80, Fig. 5). For POD, one RCT focuses on cardiac surgery  and another addresses non-cardiac surgery ; thus, we did not compare the incidence.
For POCD, three studies described general anesthesia [19-21], whereas one study described epidural anesthesia . The further subgroup analysis on the POCD incidence of general and epidural anesthesia indicated no obvious significance between the LP and the HP group when epidural anesthesia was excluded (RR 1.06, 95% CI 0.61 to 1.86, P = 0.84, I2 = 0%; 466 participants, Fig 6). No significant difference was observed in the subgroups (P = 0.18, Fig. 6). For POD, all patients of the included studies underwent general anesthesia [8, 20], so we did not perform a subgroup analysis.
Assessment of publication bias and sensitivity analysis
Given that the number of the eligible studies was small, we did not assess publication bias . Sensitivity analysis of the primary outcomes (the incidence of POD and POCD) by the Peto odds ratio method was yielded stable (POD: OR 3.67, 95% CI 0.86 to 15.62, P = 0.08 I2 = 12%; POCD: OR 1.30, 95% CI 0.75 to 2.25, P = 0.35, I2 = 0%).