Study selection

A flow diagram of the search and selection processes is shown in Fig. 1. In total, 2460 studies were identified from the databases and other sources. After duplicates were excluded (n = 885), 1575 studies were subjected to title and abstract screening, and 30 full-text studies were then evaluated for eligibility. Twenty-one further studies were excluded due to considering adult patients (n = 18) or lacking an EA assessment (n = 3). Nine studies were subjected to qualitative analysis. After one study with a high risk of bias was excluded, eight studies were included in this meta-analysis.

Fig. 1

Flow diagram of the literature search and study selection

Study characteristics

The study characteristics are shown in Table 1. A total of 951 participants was included in nine RCTs evaluating the preventive effects of melatonin premedication on pediatric EA [7, 9,10,11, 15, 22,23,24,25]. Eight studies were peer-reviewed publications [7, 9,10,11, 15, 22,23,24], while one study was an unpublished master’s thesis [25]. The sample size ranged from 48 to 163, and the children ranged in age from 1.5 to 9 years old. Eight studies used oral administration of 0.05 to 0.5 mg kg− 1 melatonin, and one study used a melatonin analog (ramelteon 0.1 mg kg− 1) [15]. The comparators included placebos [7, 9, 15, 23,24,25], midazolam [7, 9, 22, 24, 25], dexmedetomidine [10, 11, 24], and clonidine [11]. Anesthesia was induced by propofol in two studies [11, 25] and sevoflurane with or without N2O in the remaining studies [7, 9, 10, 15, 22,23,24]. Anesthesia was maintained under sevoflurane anesthesia with or without N2O in all studies. No study reported total intravenous anesthesia. The surgery types were minor elective surgery [7], elective ambulatory surgery [9, 22], tonsillectomy [15, 25] or adenoidectomy [25], ophthalmic surgery [10], elective infraumbilical surgery [11], and oesophageal dilatation procedures [24]. The diagnosis tools of EA included the Watcha scale [9, 11], pediatric anesthesia emergence delirium scale [10, 15, 25], Aono’s scale [15], Keegan scale [22], five-point scale [23], EA scale [24], and pain/discomfort scale [7]. One study did not provide a cut-off value for the Watcha scale [11]. The corresponding author was consulted by email, and it was determined that the patients were considered to have EA when the Watcha scale score was > 2 in this study.

Table 1 Clinical characteristics of the included studiesRisk of bias in the studies

The risk of bias is shown in Fig. 2. One study [25] was considered to have a high risk because the researcher reported in the trial registry record that it was an open-label study and the thesis did not provide any blinding information. Six studies [7, 11, 22,23,24,25] raised some concerns regarding the randomization process because the allocation concealment was not described. There were some concerns regarding the selection of the reported results in all the studies: prospectively registered protocols were missing in five studies [7, 11, 22,23,24], and the multiple time points of EA assessment were not mentioned in the registered protocols of the remaining four studies [9, 10, 15, 25].

Fig. 2

Risk bias of the included studies

Incidence of EAMelatonin or its analogs vs. placebos

In studies comparing melatonin with placebos, the incidence of EA was 21.9% in the melatonin and its analogs group and 47.1% in the placebos group. Melatonin and its analogs remarkably decreased EA incidence compared with placebos (RR 0.49, 95% CI 0.26 to 0.90, P = 0.02; TSA-adjusted CI 0.03 to 7.30; participants n = 307; studies n = 5) (Fig. 3). In the TSA, the cumulative Z-curve did not pass through the TSA boundary, with 10.7% of RIS cases (n = 2873) accrued (eFig. 1 in the Supplement). The statistical heterogeneity was substantial (I2 = 65%, P = 0.02). One study [15] using a melatonin analog (ramelteon) instead of melatonin had a major impact on the heterogeneity. Excluding this study obviously reduced the heterogeneity (I2 = 15%, P = 0.32) with no change in the meta-analysis results (RR 0.40, 95% CI 0.26 to 0.61, P < 0.01; TSA-adjusted CI 0.18 to 0.88; participants n = 259; studies n = 4) (Fig. 4A). In the TSA of melatonin premedication, the cumulative Z-curve passed through the TSA boundary before reaching the RIS (n = 755) (eFig. 2 in the Supplement). In addition, the meta-regression analysis showed no significant effect modification by dose of melatonin compared with placebo (regression coefficient 0.99, 95% CI -3.02 to 5.00, P = 0.63).

Fig. 3

A forest plot comparing the incidence of pediatric emergence agitation between melatonin or its analogs and placebo groups

Fig. 4

Forest plots comparing the incidence of pediatric emergence agitation between melatonin and control groups. A, Melatonin vs. placebos; B, Melatonin vs. midazolam; C, Melatonin vs. dexmedetomidine

Melatonin vs. midazolam

In studies comparing melatonin with midazolam, the incidence of EA was 12.9% in the melatonin group and 31.1% in the midazolam group. Melatonin significantly decreased EA incidence compared with midazolam (RR 0.48, 95% CI 0.32 to 0.73, P < 0.01; TSA-adjusted CI 0.21 to 1.12; participants n = 346; studies n = 4) (Fig. 4B). In the TSA, the cumulative Z-curve did not pass through the TSA boundary, with 29.4% of RIS cases (n = 1175) accrued (eFig. 3 in the Supplement). Heterogeneity was not detected (I2 = 5%, P = 0.37). In addition, the meta-regression analysis showed no significant effect modification by dose of melatonin compared with midazolam (regression coefficient − 3.85, 95% CI -7.84 to 0.13, P = 0.06).

Melatonin vs. dexmedetomidine

In studies comparing melatonin with dexmedetomidine, EA incidence was 20.8% in the melatonin group and 10.0% in the dexmedetomidine group, with a significant difference (RR 2.04, 95% CI 1.11 to 3.73, P = 0.02; TSA-adjusted CI 0.17 to 24.04; participants n = 240; studies n = 3) (Fig. 4C). In the TSA, the cumulative Z-curve did not pass through the TSA boundary, with 6.0% of RIS cases (n = 4011) accrued (eFig. 4 in the Supplement). No obvious heterogeneity (I2 = 0%, P = 0.72) was found. In addition, the meta-regression analysis showed no significant effect modification by dose of melatonin compared with dexmedetomidine (regression coefficient 1.50, 95% CI -3.13 to 6.12, P = 0.53).

Melatonin vs. clonidine

Only one study performed this comparison. Hence, no meta-analysis or TSA was performed. Melatonin did not attenuate the incidence of EA compared with clonidine (RR 3.0, 95% CI 0.13 to 71.22, P = 0.50). The wide 95% CI reveals the statistical imprecision.

Adverse effects

One study reported that no melatonin-relevant adverse effects were observed [24]. The corresponding authors of the other studies were contacted by email. Two authors [9, 11] responded that no adverse effects related to melatonin were found. One author [7] responded to a previous meta-analysis [16] stating that no melatonin-related adverse effects were found, but they did not respond to our contact. Two studies [10, 11] focusing on melatonin and dexmedetomidine reported a lower heart rate after dexmedetomidine premedication compared with melatonin. One of these [10] reported that no participants had symptomatic bradycardia requiring pharmacological intervention.

Reporting biases

The funnel plots of melatonin compared with placebos and midazolam are shown in eFig. 5 in the Supplement. Funnel plots of melatonin compared with dexmedetomidine or clonidine could not be generated because the paucity of studies precluded meaningful analysis. A visual inspection of the melatonin and placebo funnel plots indicated obvious asymmetry, suggesting the existence of publication bias (eFig. 5 A in the Supplement). No publication bias was observed in the melatonin and midazolam funnel plots (eFig. 5B in the Supplement).

Certainty of evidence

A summary of the findings is presented in Table 2, and the certainty of the evidence was assessed as very low or moderate in all outcomes.

Table 2 Summary of the findings