The search yielded 125 non-duplicated articles. After excluding 75 articles based on title/abstract, 50 articles were retrieved for full text review. Of these, 34 were excluded due to different design (five not RCT), different outcome measures (27 studies not evaluating anxiety or assessing anxiety by external observers), or reporting of secondary analyses (2 studies). Other three studies could not be retrieved (one study from the 80s, one congress abstract without any further publications and one paper published in a journal no longer existing). Four additional articles were identified via hand search, thus a total of 17 RCTs [2, 4, 8-10, 12, 16-26] were included in the qualitative synthesis (Figure 1).
Study and patient characteristics
The analysis included seven crossover RCTs and 10 parallel RCTs. Characteristics of included studies are reported in Table 1. The number of enrolled participants ranged from 12 to 997 participants. The drugs that were used for conscious sedation included Passiflora incarnate, Midazolam, Dexmedetomidine, N2O, Clonidine, Remifentanil, Zaleplon, Triazolam, Chlordemethyldiazepam (CDDZ), Propofol, Fentanyl, and Methohexital. Fourteen studies investigated patient-reported anxiety using different scales: visual analogue scale (VAS), Corah dental anxiety scale (DAS), Interval scale of anxiety report (ISAR), state trait anxiety inventory (STAI); other interval scales. Two studies investigated anxiety by assessing stress hormones alteration. One study investigated anxiety using both patient-reported scale (DAS) and stress hormones alteration (salivary cortisol) (Table 1).
Risk of bias in included studies
The risk of bias is reported in Table 2. The risk of selection bias was low in five studies (adequate random sequence generation) and unclear in the others with regards to random sequence generation, while it was unclear in all studies with regards to allocation concealment (which was never specified). The risk of performance bias was low in 12 studies (blinded participants and personnel) and unclear in five. The risk of detection bias was low in four studies (blinded outcome assessor) and unclear in 13. Four studies [10, 12, 16, 25] were at high risk of attrition bias. In Smiley et al. , one patient in the dexmedetomidine-only arm withdrew from the study citing inadequate sedation. In Pereira Santos et al. , two patients did not return for the second surgery and two were excluded due to signs and symptoms of over-sedation. In Studer et al. , two patients withdrew from the study citing excessive anxiety. In Luyk et al. , two patients did not return for the second surgery. The risk of attrition bias was low in 11 studies (no dropouts or missing data) and unclear in two (no information on dropouts or missing data). The risk of reporting bias was low in all studies. Four studies [2, 10, 16, 24] were at high risk of other bias due to incorrect analysis of data from crossover designs (i.e. paired tests were performed but the period effect could not be removed because of different sequence sizes).
Narrative synthesis on patient-reported anxiety
All but two studies [17, 19] investigated patient-reported anxiety (Table 3).
Six studies found some statistically significant differences between arms. One study reported lower anxiety with triazolam vs. placebo , while one study reported lower anxiety with midazolam vs. no intervention .
One study reported lower anxiety with midazolam 2mg vs. 1mg , and one study reported lower anxiety with propofol vs. midazolam . One study reported lower anxiety with any intervention group vs. placebo, lower intraoperative anxiety with fentanyl + midazolam vs. midazolam, while subjects receiving fentanyl + midazolam + methohexital reported the lowest anxiety . One study reported lower anxiety with fentanyl + midazolam or with fentanyl + midazolam+ methohexital with respect to placebo or midazolam .
In another study, lower anxiety with midazolam vs. diazepam was shown by using VAS but not with STAI scale .
Four studies reported unclear findings on anxiety assessment / evaluation or were underpowered to draw definitive conclusions. One study comparing Passiflora incarnate vs. Midazolam  and another comparing midazolam vs. N2O/O2  reported unclear findings on anxiety, since the data suggested a period effect that was not tested and was not removed by the design due to different sequence size. Other two studies [9, 12] suggested lower anxiety with midazolam vs. placebo, and with zaleplon vs. triazolam, but were underpowered to achieve statistical significance.
Two studies [16, 18] did not find any statistically significant difference between midazolam vs. clonidine, and between remifentanil vs. placebo.
Luyk et al. found no statistically significant difference in postoperative anxiety between midazolam vs. placebo in 1992 , while they reported lower intraoperative anxiety with midazolam administered orally and intravenously vs. oral placebo + intravenous midazolam (attrition bias) in 1991 .
Narrative synthesis on anxiety by assessing stress hormones alteration
Three studies [10, 17, 19] investigated anxiety by assessing stress hormone alteration (Table 4). One study  reported lower cortisol levels with midazolam vs. placebo, while unclear findings were reported by another study comparing midazolam vs. N20/O2 . One study  found significant variation in perioperative renin levels with remifentanil vs. placebo, while aldosterone and ACTH levels were not statistically different.