This meta-analysis was registered in PROSPERO before its implementation (CRD42024523917). Literature was retrieved from PubMed, Embase, the Cochrane Library, Web of Science, Wanfang Data, the China National Knowledge Infrastructure, and the China Science and Technology Journal databases. The keywords “dexmedetomidine,” “precedex,” “neoplasms,” “cancer,” “cognition,” and “cognitive” were combined and searched by logical relationships of “OR and.” During this process, a pattern of integrating subject heading terms with free-text words was adopted, with search strategies adjusted according to the specific characteristics of the respective databases. The search details for the four English databases are presented in Supplementary Tables 1–4. To avoid selection bias, all literature published up until March 4, 2024 was searched, without imposing any language restrictions. Additionally, reference lists from relevant reviews and included studies were also consulted. While no language restrictions were applied, our search specifically included Chinese databases to capture regionally relevant studies.

Literature was deemed eligible for inclusion in this review if (1) the study was conducted on patients with surgically treated GIC (including gastric, colorectal, colon, and rectal cancers); (2) the study evaluated differences in the effects of DEX and saline on POCD; (3) the study was an RCT; and (4) the study reported one or more of the following outcomes: MMSE score, case number, or incidence of POCD.

Exclusion criteria were as follows: (1) non-original studies, including reviews, conference abstracts, and comments; (2) studies lacking inclusion criteria, did not report POCD criteria or measurement time, or had data errors; and (3) for duplicate publications or multiple articles sharing the same data, only the one with the most complete information was included.

Two investigators (Xiaoxia Zheng and Yang Lan) independently conducted literature screening based on the aforementioned criteria and extracted data according to a pre-specified data field. For each eligible article, information including the name of the first author, publication year, basic characteristics of the study subjects (such as sample size and age), type of cancer, type of surgery, method of anesthesia, and study outcomes were collected. Following the completion of data extraction, the fields were exchanged for review, and any inconsistencies were negotiated. The Cochrane Collaboration tool for assessing risk was employed for the quality assessment of the RCTs (27).

The MMSE scale, a widely used 30-point scale assessing cognitive domains including orientation (10 points), registration (3 points), attention/calculation (5 points), recall (3 points), and language/visual construction (9 points), was utilized to evaluate cognitive dysfunction. In most of the included studies, POCD was defined as an MMSE score <27 at any time point after surgery. The diagnostic criteria and assessment time points for POCD in the included studies are detailed in Supplementary Table 5.

Data analysis was performed using RevMan 5.3 and Stata12.0. This meta-analysis compared the differences in MMSE scores and the incidence of POCD at different follow-up times (1, 2, 3, and 7 days postoperatively). Among the differences in POCD incidence, the risk ratio (RR) and 95% confidence interval (CI) were used as effect values. The MMSE score, as a continuous variable, was assessed using the weighted mean difference (WMD) and 95% CI as effect values. Heterogeneity across the studies was monitored using Cochran’s Q and I² tests (28). The random-effects model was implemented if significant heterogeneity was defined at p < 0.05 or I² > 50% (I² statistic); otherwise, the fixed-effects model was adopted. A subgroup analysis was conducted to estimate the effect of cancer and surgery type on heterogeneity and the pooled effect. Egger’s test and the one-by-one elimination method were used to monitor publication bias across studies and the stability of pooled estimates, respectively (29, 30).

The process and results of the literature search are shown in Figure 1. A total of 1,397 records were obtained from online databases. After an initial screening by browsing the titles and abstracts, 429 duplicates and 938 articles that did not meet the inclusion criteria were excluded. Subsequently, after a thorough reading of the full text and screening of the remaining 30 articles, 12 studies were finally included in this meta-analysis (8, 24, 25, 31–39).

As shown in Table 1, all the 12 RCTs included in this meta-analysis were conducted in China. Among these, five studies were conducted on patients with gastric cancer (GC) (8, 33, 36, 37, 39), six on patients with rectal cancer (RC), including colorectal cancer (CRC) (24, 25, 31, 32, 35, 38); and one on patients with GIC (34). Regarding the type of surgery, seven publications included subjects undergoing laparoscopic RS (LRS) (8, 25, 32, 33, 35, 37, 39), whereas the remaining five reported only RS (24, 31, 34, 37, 38). The sample size of the 12 studies ranged from 60 to 120, with 881 subjects (440 in the DEX and 441 in the saline groups) enrolled in this meta-analysis. In all included studies, there were no significant differences in age, sex, or American Society of Anesthesiologists classification between the DEX and saline groups.

The results of the quality assessment are shown in Supplementary Figure 1. The included studies lacked details regarding blinding, and most of them did not report specific randomization and allocation concealment schemes. Therefore, these studies showed moderate risks of selection, performance, and detection biases. The risk of bias summary table is shown in Supplementary Table 6. The GRADE system was used to grade the quality of evidence, with assessment dimensions including study design, risk of bias, inconsistency, indirectness, imprecision, and publication bias. Due to the high risk of bias in the included studies (unclear randomization methods, inadequate blinding), significant heterogeneity (I² > 50%), and the fact that all studies were from China (with some indirectness), the quality of evidence for MMSE scores 1–7 days post-surgery was rated as ‘low.’ For the incidence of POCD, although heterogeneity was low (I² < 50%), there was still a high risk of bias (unclear blinding) and geographical limitations, and the quality of evidence was rated as ‘moderate.’

The differences in the MMSE scores between the DEX and saline groups on postoperative days 1, 2, 3, and 7 are depicted in Figure 2, representing significant heterogeneity among the included studies across all the time points. The pooled results of the random-effect model suggested that DEX significantly increased the MMSE score on postoperative days 1 (WMD [95% CI] = 2.53 [1.88, 3.19], p < 0.00001), 2 (WMD [95% CI] = 2.53 [1.07, 3.99], p = 0.0007), 3 (WMD [95% CI] = 2.24 [0.97, 3.51], p = 0.0006), and 7 (WMD [95% CI] = 1.82 [0.73, 2.92], p = 0.001).

Figure 3 shows the differences in POCD incidence between the two groups at postoperative days 1, 3, and 7. No significant heterogeneity was observed across all time points, and the combined estimates of the fixed-effect model indicated that DEX significantly decreased the risk of POCD incidence compared with the use of saline at postoperative days 1 (RR [95% CI] = 0.35 [0.21, 0.60], p = 0.0001), 3 (RR [95% CI] = 0.25 [0.09, 0.70], p = 0.008), and 7 (RR [95% CI] = 0.47 [0.22, 0.98], p = 0.04).

Considering that the studies included for most of the outcomes were ≤5, this meta-analysis only performed subgroup analyses and publication bias tests for MMSE scores on postoperative day 1. Under the subgroup of cancer types (Figure 4A), the combined results of CRC (WMD [95% CI] = 2.09 [0.85, 3.32], p = 0.0009), GC (WMD [95% CI] = 2.63 [1.91, 3.36], p < 0.00001), and GIC (WMD [95% CI] = 3.90 [3.36, 4.44], p < 0.00001) all suggested that DEX could significantly improve MMSE scores. Moreover, the combined results for three subgroup of cancer types exhibited statistically significant difference (p = 0.003). However, there was significant heterogeneity in these subgroups (CRC, p < 0.00001; I² = 88%; GC, p = 0.002, I² = 84%; and GIC, not applicable), indicating the cancer type was not a source of heterogeneity in the MMSE score.

In the subgroup of surgery types (Figure 4B), the pooled results of LRS (WMD [95% CI] = 2.56 [1.46, 3.65], p < 0.00001) and RS (WMD [95% CI] = 2.50 [1.56, 3.44], p < 0.00001) both suggested a significant improvement in MMSE scores in the DEX group compared to the use of saline. However, the pooled data exhibited significant diversity between LRS and RS (p = 0.94). Furthermore, significant heterogeneity in these subgroups was observed (LRS, p < 0.00001, I² = 91%; and RS, p < 0.00001, I² = 88%), indicating that surgery type was also not a source of heterogeneity in the MMSE score.

In addition, Egger’s test indicated no publication bias among the included studies (p = 0.447; Supplementary Figure 2).

The one-by-one elimination method revealed that for outcome indicators, except POCD at postoperative day 1, excluding any single study did not significantly alter the combined results, and the remaining studies continued to be statistically significant and consistent with the original results (Table 2 and Supplementary Figures 3, 4), which further suggested the robust outcomes and stability of this meta-analysis.

However, this study has some limitations: (1) significant heterogeneity in the combined results of MMSE scores was observed, with neither cancer nor surgery type identified as an influencing factor of heterogeneity in the subgroup analysis. The differences in DEX dosage and administration schedules included in the study may contribute to heterogeneity by influencing the drug’s neuroprotective efficacy. Most studies did not report patients’ baseline cognitive status (e.g., preoperative MMSE scores), and differences in baseline cognitive function may lead to varying responses to DEX among patients and introduce heterogeneity. The diversity of anesthetic techniques used in the studies, including differences in hypnotics, opioids, and adjunctive medications used in combination with DEX, may contribute to heterogeneity in cognitive outcomes through synergistic or antagonistic effects. Due to the lack of detailed reporting of DEX dosage and other data, as well as the absence of baseline cognitive status information in some studies, the current data were insufficient for meta-regression analysis. Future studies should standardize reporting of DEX dosing regimens, baseline cognitive assessments, and anesthetic protocols to facilitate more in-depth heterogeneity analyses. (2) Although our search strategy included international databases without language restrictions, all identified eligible studies were conducted in China. This geographic limitation should be considered when generalizing findings to other populations. In addition, differences in culture, genetics, medical practices, and anaesthesia protocols may affect the generalisability of the study results. For example, differences in perioperative management in China, such as locally adapted enhanced recovery after surgery (ERAS) protocols, genetic polymorphisms related to drug metabolism and inflammatory responses, and anesthetic drug selection, may all affect the efficacy of DEX. Therefore, the findings of this study may be most applicable to populations with similar perioperative contexts (e.g., standardized ERAS protocols, similar anesthetic regimens). The international multicenter RCTs should be conducted to validate DEX’s effects in diverse GIC cohorts, particularly those with different genetic backgrounds and healthcare systems. (3) The number of included studies was inadequate; therefore, additional high-quality RCTs with larger sample sizes are required to further validate the stability and extrapolation of the results. (4) MMSE was a limited screening tool: The MMSE was less sensitive to subtle cognitive changes (mild impairments in executive function and attention) and may struggle to capture the subtle fluctuations in cognitive function in the early post-operative period. The MMSE is significantly influenced by educational attainment and linguistic-cultural background, and since the participants in this study were all Chinese patients, the results may not be representative of assessments conducted in other linguistic-cultural contexts. Future studies should employ a combination of multidimensional cognitive assessment tools (MMSE combined with MoCA, neuropsychological battery tests) and adjust assessment tools based on the educational background and linguistic characteristics of the study population to enhance the accuracy of cognitive function assessment.