We performed a systematic review and network meta-analysis. We searched the PubMed, Cochrane Library, Web of Science, EMBASE, China National Knowledge Infrastructure, and Baidu Wenku document download website for randomized controlled trials from the date of their inception to Feb 26, 2023, without language restrictions. To ensure the comprehensiveness of the retrieval, a combination of subject terms and free terms was used for the literature retrieval. All search results were stored using Endnote software (Thomson Corporation, Stanford, CT, USA) for further filtration. We used the following search keywords: “atrial fibrillation”, “auricular fibrillation”, “AF”, “analgesia”, “analgesic”, “sedation”, “sedative”, “radiofrequency ablation”, and “catheter ablation”, which were combined with analgesic and sedative drugs currently used in clinical practice, such as “morphine” and “fentanyl”. We only included RCTs that compared different analgesic or sedative strategies in patients with AF undergoing RFA.

The inclusion criteria were as follows: (1) patients diagnosed with AF undergoing RFA; (2) study design was an RCT to compare different analgesics or sedatives during RFA; (3) study with outcomes of “Ramsay sedation scores (RSS)”, “visual analog scale (VAS)”, “incidence of respiratory depression”, “incidence of hypotension”, or “incidence of nausea and vomiting”; (4) studies with two or multiple arms.

The exclusion criteria were as follows: (1) study types were reviews, observational studies, registry data, ongoing trials without results, case reports, systematic reviews, meta-analysis, animal experimental studies, or duplicate studies; (2) studies without definition of endpoints; (3) unrelated studies; (4) study data could not be obtained.

Our endpoints were the efficacy of an analgesic or sedative, evaluated using RSS and VAS scores. The RSS was defined: one point: the patient is awake but anxious, agitated, or restless; two points: the patient is awake but cooperative, orientated, and tranquil; three points: the patient is drowsy but responsive to commands; four points: the patient is asleep and with brisk response to glabella tap or loud auditory stimulus; five points: the patient is asleep with sluggish response to a stimulus; six points: the patient has no response to noxious stimuli. Pain was assessed by VAS, using a scale of 0 to 10. The VAS rules were as follows: higher scores represent more intense pain, a score of 0 represents no pain, and 10 points represent severe pain. The safety endpoints were the incidences of respiratory depression, hypotension, nausea, and vomiting during ablation therapy. The trial defined respiratory depression as SaO${}_2$ 90%, apnea greater than 15 seconds, respiratory rate less than 8 breaths per minute, or hypercarbia (PaCO${}_2$ $>$55 mmHg). Hypotension was defined as a systolic pressure of less than 90 mmHg or mean arterial pressure of less than 60 mmHg.

The search results were screened independently by two blinded researchers (LJ and LZ), according to inclusion and exclusion criteria. Disagreements were resolved through discussions and referral to a third coauthor (HG). Two authors independently extracted the following data from the included RCTs: the first author’s name, the publication year, the baseline characteristics, the number of patients in each group, interventions, procedure duration, Ramsay sedation scores, analgesic scores, and occurrence of adverse reactions.

Two authors independently accessed the quality and risk of bias of the included studies using the Cochrane Collaboration Systematic Evaluators manual. We used Review Manager software (Version 5.4, Cochrane, London, United Kingdom) to produce the risk of bias graph. To assess publication bias, we generated funnel graphs using STATA software (Version 17.0, StataCorp, Texas City, TX, USA).

The data were analyzed using STATA software (Version 17.0) based on a frequency model. First, we drew network evidence plots to show direct comparisons. We plotted network contributions to show the contribution of direct comparisons to indirect comparisons. We used “ifplot” in the “network” software package (Thomson Corporation, Stanford, CT, USA) to evaluate inconsistencies. The inconsistency factor (IF) and IF 95% confidence interval (CI) were used to evaluate differences between direct and indirect comparisons. If 95% CI of IF contains 0, the consistency is high. According to the inclusion and exclusion criteria, only RCTs were included in the network meta-analysis, meaning there was no loop formed regarding the efficacy endpoints and the incidence of hypotension, nausea, and vomiting. An inconsistency test was not required to evaluate the results of the direct comparison and indirect comparison. Then, interval plots were drawn. We performed league tables to show the network meta-analysis (NMA) results. To compare each strategy, we used the surface under the cumulative ranking curve (SUCRA) to estimate the probabilities (%) of each treatment, as being the best or other rankings. Finally, we developed correction funnel plots to determine the evidence of the small sample effect and publication bias.

We retrieved 708 articles from the PubMed, Cochrane Library, Web of Science, EMBASE, and China National Knowledge Infrastructure databases, and 1 article from the Baidu Wenku document download website. A total of 104 duplicate articles were excluded. After reading the titles and abstracts, 540 articles were removed, and 51 articles were further screened by reading the full text. Finally, 14 RCTs were included in the NMA, including 13 from the database and 1 from the Baidu Wenku document download website. Fig. 1 illustrates the literature screening process. The 14 studies [2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15] involved 1156 patients, 9 drugs, and 10 regimens, which were propofol, dexmedetomidine, fentanyl or its derivatives, butorphanol, oxycodone, hydromorphone, dezocine, morphine, dexmedetomidine combined with fentanyl or its derivatives, and midazolam combined with fentanyl or its derivatives. The studies and the patient characteristics are shown in Table 1 (Ref. [2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]). All included studies were RCTs. The risk of bias and quality evaluation results for the studies are shown in Fig. 2.

Fig. 1.

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flowchart for this network meta-analysis. CNKI, Chinese National Knowledge Infrastructure.

Fig. 2.

Risk of bias graph.

RSS and VAS scores were primary efficacy outcome indexes. Out of the 14 included studies, 9 studies [3, 5, 9, 10, 11, 12, 13, 14, 15] reported RSS and involved 7 regimens. Fig. 3a shows the network plot. Fig. 3b shows the contribution plot for RSS. Paired comparisons among the seven medication regimens revealed that patients receiving dexmedetomidine during RFA for AF had higher RSS than those receiving midazolam plus fentanyl or its derivatives (mean difference [MD] –0.88, 95% CI [–0.04 to –0.72], p 0.05). Dexmedetomidine plus fentanyl or its derivatives had a better sedation effect compared with fentanyl or its derivatives (MD –0.53, 95% CI [–1.06 to 0.00]), although the difference was not statistically significant (p $>$ 0.05). Hydromorphone produced better sedative effects than fentanyl or its derivatives (MD 0.48, 95% CI [0.03 to 8.59]) and morphine (MD 0.48, 95% CI [0.01 to 18.54]), although, again, the difference was not statistically significant (p $>$ 0.05). See Table 2 for details. The prediction intervals in the network meta-analysis (NMA) are shown in Fig. 3c. Fig. 3d shows the accumulated possibility plot using the area under the curve to indicate the likelihood of ranking first for RSS. The SUCRA plot shows that dexmedetomidine (SUCRA 81.9%) has the largest area under the curve and that the sedation effect is most likely to be better for RFA than by the other drugs, yet it was followed by dexmedetomidine plus fentanyl or its derivatives (SUCRA 71.7%), and butorphanol (SUCRA 66.2%). Midazolam plus fentanyl or its derivatives (SUCRA 23.2%) was associated with the lowest probability of high RSS. Fig. 3e presents a funnel plot to illustrate the publication bias. The overall publication bias showed a symmetrical distribution around the funnel plot, indicating low publication bias. The forest plot of RSS is shown in Fig. 3f.

Fig. 3.

Figures of network meta-analysis of Ramsay sedation scores. (a) Network plot of Ramsay sedation scores. Line thickness represents the number of comparisons between the two arms, while node size represents the sample size of each arm. (b) Contribution plot of Ramsay sedation scores. (c) Prediction intervals map of Ramsay sedation scores. (d) SUCRA plot of Ramsay sedation scores. (e) Funnel plot. (f) Forest plot of Ramsay sedation scores. DEX, dexmedetomidine; F, fentanyl or its derivatives; BT, butorphanol; HMOR, hydromorphone; MOR, morphine; MD, midazolam; CI, confidence interval; PrI, prediction interval; SUCRA, the surface under the cumulative ranking curve. A: midazolam plus fentanyl or its derivatives, B: dexmedetomidine, C: dexmedetomidine plus fentanyl or its derivatives, D: fentanyl or its derivatives, E: butorphanol, F: hydromorphone, G: morphine.

Five studies [5, 6, 7, 9, 15] reported VAS scores involving five regimens. Fig. 4a,b shows the network plot and contribution plot of the analgesic scores. The NMA results showed that compared to morphine, dezocine (MD 1.88, 95% CI [1.16 to 2.60], p 0.05), hydromorphone (MD 4.07, 95% CI [3.56 to 4.58], p 0.05), and butorphanol (MD 3.18, 95% CI [2.38 to 3.96], p 0.05), fentanyl or its derivatives (MD 1.57, 95% CI [1.25 to 1.89], p 0.05) had a better analgesic effect during RFA. Butorphanol (MD 2.50, 95% CI [2.11 to 2.89], p 0.05) and hydromorphone (MD 1.60, 95% CI [0.87 to 2.33], p 0.05) have better analgesic effects than fentanyl or its derivatives. While both dezocine (MD –2.19, 95% CI [–3.07 to –1.31], p 0.05) and butorphanol (MD –0.90, 95% CI [–1.73 to –0.07], p 0.05) had higher VAS scores than hydromorphone. In contrast, our results also showed that the butorphanol analgesic effect is superior to dezocine (MD –1.29, 95% CI [–2.36 to –0.22], p 0.05). The analgesic effect of dezocine may be better than fentanyl or its derivatives (MD 0.31, 95% CI [–0.48 to 1.09]), however, the difference was not statistically significant (p $>$ 0.05). Table 3 shows the results of the detailed analysis. The prediction intervals for NMA are shown in Fig. 4c. The SUCRA plot showed that hydromorphone (SUCRA 99.6%) had a larger area under the curve, and its analgesic effect was most likely to be superior to the other four regimens, followed by butorphanol (SUCRA 75.2%). The analgesic effect of morphine (SUCRA 0%) was most likely to rank last. See Fig. 4d for details. We mapped a funnel plot to illustrate publication bias (Fig. 4e). The scatter points in the study were relatively dispersed and had distribution associated with bias, thereby indicating that there may be some publication bias in the results. The forest plot of the VAS scores is shown in Fig. 4f.

Fig. 4.

Figures of network meta-analysis of VAS. (a) Network plot of VAS. Line thickness represents the number of comparisons between the two arms, while node size represents the sample size of each arm. (b) Contribution plot of VAS. (c) Prediction intervals map of VAS. (d) SUCRA plot of VAS. (e) Funnel plot. (f) forest plot of VAS. F, fentanyl or its derivatives; BT, butorphanol; HMOR, hydromorphone; MOR, morphine; DZ, dezocine; CI, confidence interval; VAS, visual analog scale; SUCRA, the surface under the cumulative ranking curve; PrI, prediction interval. A: fentanyl or its derivatives, B: butorphanol, C: morphine, D: dezocine, E: hydromorphone.

The safety outcomes reported were complication rates, mainly consisting of respiratory depression, hypotension, nausea, and vomiting. For the safety outcome, 11 studies [2, 3, 4, 5, 8, 9, 10, 11, 12, 14, 15] reported rates of respiratory depression. Fig. 5a,b shows the network plot and contribution plot of the incidence of respiratory depression. We conducted inconsistencies tests based on the loop, and the results showed that the 95% CI of IF contained 0, meaning that there was no obvious inconsistency; therefore, we used the consistency model for NMA (Fig. 5c). Dexmedetomidine combined with fentanyl or its derivatives significantly reduced the incidence of respiratory depression compared with propofol (odds ratio [OR] 16.46; 95% CI [1.54 to 175.95], p 0.05) and midazolam plus fentanyl or its derivatives (OR 7.02; 95% CI [1.33 to 36.99], p 0.05). In our study, the rate of respiratory depression was lower in the oxycodone group than in the fentanyl or its derivatives group (OR 0.10; 95% CI [0.00 to 2.80]) and the morphine group (OR 0.10; 95% CI [0.00 to 5.57]), although the differences were not statistically significant (p $>$ 0.05). See Table 4 for details. The prediction intervals for NMA are shown in Fig. 5d. The SUCRA sequencing map (Fig. 5e) indicated that by reducing the risk of respiratory depression in RFA for AF, oxycodone had the largest area under the curve and was most likely to rank as the best (SUCRA 81.4%). Butorphanol and dexmedetomidine plus fentanyl had the same SUCRA (72.2%) and were tied as being the second best. Propofol was probably the worst in terms of reducing the incidence of respiratory depression (SUCRA 10.8%). A funnel plot was drawn to illustrate the observed publication bias. Overall, publication bias showed symmetrical distribution around the funnel plot, thereby indicating low publication bias (Fig. 5f). In the funnel plot, the distribution of all the scatter points was roughly symmetric, although some research scatter points were close to the bottom of the funnel plot, thereby indicating that the results were potentially affected by publication bias and small sample effect. The forest plot is shown in Fig. 5g.

Fig. 5.

Figures of network meta-analysis of respiratory despression. (a) Network plot of respiratory depression. Line thickness represents the number of comparisons between the two arms, while node size represents the sample size of each arm. (b) Contribution plot of respiratory depression. (c) Inconsistency test plot of respiratory depression. (d) Prediction intervals map of respiratory depression. (e) SUCRA plot of respiratory depression. (f) Funnel plot. (g) Forest plot of respiratory depression. F, fentanyl or its derivatives; BT, butorphanol; HMOR, hydromorphone; MOR, morphine; OXY, oxycodone; PRO, propofol; DEX, dexmedetomidine; MD, midazolam; SUCRA, the surface under the cumulative ranking curve; CI, confidence interval; PrI, prediction interval. A: dexmedetomidine plus fentanyl or its derivatives, B: midazolam plus fentanyl or its derivatives, C: propofol, D: fentanyl or its derivatives, E: butorphanol, F: oxycodone, G: dexmedetomidine, H: hydromorphone, I: morphine.

Four studies [3, 4, 10, 14] reported the incidence of hypotension in RFA (Fig. 6a,b). Since there was no loop structure in the network plot, no inconsistency check was required. The NMA results (Table 5) illustrated that compared with dexmedetomidine, propofol (OR 27.27; 95% CI [0.05 to 15,721.14]), midazolam plus fentanyl or its derivatives (OR 7.66; 95% CI [0.07 to 845.55]), dexmedetomidine plus fentanyl or its derivatives (OR 15.67; 95% CI [0.04 to 6089.58]) probably had a higher occurrence of hypotension, although none of these results were statistically different (p $>$ 0.05). The midazolam plus fentanyl or its derivatives group had potentially less hypotension than the dexmedetomidine plus fentanyl or its derivatives group (OR 0.49; 95% CI [0.01 to 19.11], p $\mathrm{>}$0.05). Moreover, compared with propofol, midazolam plus fentanyl or its derivatives (OR 0.28; 95% CI [0.00 to 20.24], p $>$ 0.05) and dexmedetomidine plus fentanyl or its derivatives (OR 0.57; 95% CI [0.00 to 160.50], p $>$ 0.05) had a lower tendency of developing hypotension. The prediction intervals of the NMA are shown in Fig. 6c. DEX (SUCRA 88.2%) was associated with the lowest incidence of hypotension, according to the SUCRA plot, while midazolam plus fentanyl or its derivatives was the next best (SUCRA 51.9%). The propofol (SUCRA 28.8%) group had the highest incidence of hypotension according to the results of the SUCRA map (Fig. 6d). Fig. 6e shows the funnel plot. The distribution of all the scattered points in the funnel graph was symmetrical, with a small publication bias. The forest plot is shown in Fig. 6f.

Fig. 6.

Figures of network meta-analysis of incidence of hypotension. (a) Network plot of incidence of hypotension. Line thickness represents the number of comparisons between the two arms, while node size represents the sample size of each arm. (b) Contribution plot of incidence of hypotension. (c) Prediction intervals map of incidence of hypotension. (d) SUCRA plot of incidence of hypotension. (e) Funnel plot. (f) Forest plot of incidence of hypotension. PRO, propofol; F, fentanyl or its derivatives; MD, midazolam; DEX, dexmedetomidine; SUCRA, the surface under the cumulative ranking curve; CI, confidence interval; PrI, prediction interval. A: dexmedetomidine, B: propofol, C: midazolam plus fentanyl or its derivatives, D: dexmedetomidine plus fentanyl or its derivatives.

Eight articles [3, 4, 5, 9, 10, 12, 13, 15] reported the incidence of nausea and vomiting from seven strategies. Fig. 7a,b shows the network plot and contribution plot of the incidence of nausea and vomiting. Since there was no loop structure in the network plot, no inconsistency check was required. The NMA results showed that dexmedetomidine plus fentanyl or its derivatives reduced the incidence of nausea and vomiting compared with fentanyl alone (OR 4.74; 95% CI [1.01 to 22.22], p 0.05). In addition, propofol was associated with a lower incidence of nausea and vomiting than hydromorphone (OR 0.01; 95% CI [0.00 to 0.59], p 0.05) and fentanyl or its derivatives (OR 0.01; 95% CI [0.00 to 0.51], p 0.05). The pairwise comparison of the remaining therapeutic regimens showed no statistical difference (Table 6). The prediction intervals for NMA are shown in Fig. 7c. The frequency analysis results from the SUCRA plots indicated that propofol (SUCRA 93.5%) reduced the incidence of nausea and vomiting. Hydromorphone was most likely to cause nausea and vomiting (SUCRA 11.5%) (Fig. 7d). The funnel plot used to assess publication bias is shown in Fig. 7e. The distribution of all scattered points in the funnel map was approximately symmetrical, although some of the research scattered points were located at the bottom of the funnel map, thereby indicating that the results may be affected by publication bias and small sample effects. The RSS forest plot is shown in Fig. 7f.

Fig. 7.

Figures of network meta-analysis of incidence of nausea and vomiting. (a) Network plot of incidence of nausea and vomiting. Line thickness represents the number of comparisons between the two arms, while node size represents the sample size of each arm. (b) Contribution plot of incidence of nausea and vomiting. (c) Prediction intervals map of incidence of nausea and vomiting. (d) SUCRA plot of incidence of nausea and vomiting. (e) Funnel plot. (f) Forest plot of incidence of nausea and vomiting. MD, midazolam; F, fentanyl or its derivatives; DEX, dexmedetomidine; PRO, propofol; MOR, morphine; BT, butorphanol; HMOR, hydromorphone; CI, confidence interval; PrI, prediction interval; SUCRA, the surface under the cumulative ranking curve. A: midazolam plus fentanyl or its derivatives, B: dexmedetomidine plus fentanyl or its derivatives, C: propofol, D: morphine, E: butorphanol, F: fentanyl or its derivatives, G: hydromorphone.