Systematic review

There were 1635 references identified in the Pubmed search, 3926 in the Scopus search, and 2038 in the Embase search. Duplicates were removed, leaving 4875 unique references (Online Resource 2). Based on the aforementioned inclusion and exclusion criteria, a final cross-checked selection was made of 24 studies published between 2010 and 2020, as presented in the PRISMA flowchart in Fig. 1.

Fig. 1

Flowchart accounting for each paper found in the initial searches, from identification through screening, showing the final number of articles included in the meta-analysis and the reasons for exclusion. RCT randomized controlled trial

Overall characteristics of included studies

Of the 24 studies included in the pooled analysis [16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39], 4 were RCTs, all of which compared robotic to laparoscopic and/or vaginal cases, 5 were prospective studies (one quasi-RCT), and 15 were independent database articles. The sample size in the overall pooled population was 1,116,665, of whom 110,306 underwent robotic hysterectomy, 554,407 underwent open hysterectomy, 189,237 had a vaginal procedure, and 262,715 were operated on using a conventional laparoscopic approach. Overall, 22, 11, and 10 studies compared the robotic approach to the laparoscopic, open and vaginal approach, respectively (Table 1).

Table 1 Characteristics of included studiesRisk of bias assessment

Risk of bias (RoB) assessments for the four RCTs were low regarding randomization, missing data, and outcomes measurements (Fig. 2). However, due to some concerns regarding deviations from the intended intervention, as well as the selection of reported results, the overall ROBs of three out of the four RCTs were considered as having some concerns.

Fig. 2

Risk of bias assessments for randomized controlled trials using the Cochrane Handbook risk of bias tools for randomized trials (RoB-2). Domain 1 deals with bias arising from the randomization process, Domain 2 deals with bias due to deviations from the intended interventions, Domain 3 deals with bias due to missing outcome data, Domain 4 deals with bias in measurement of the outcome, and Domain 5 deals with bias in selection of the reported result

For the database and prospective cohort studies, the Newcastle–Ottawa scores ranged between 6 and 9 for the included cohort studies, with a lack of specifying whether patients were lost to follow-up being the most common reason for a lower score (Online Resource 3).

Baseline patient characteristics

Results for the meta-analysis of baseline patient characteristics are reported in Table 2.

Table 2 Baseline patient characteristicsRobotic versus laparoscopic approach

Patients who underwent robotic hysterectomy were significantly older than those operated via the laparoscopic approach, with a mean difference of 0.95 years (p = 0.01). The mean uterine weight was higher for laparoscopic patients, with a mean difference of 23 g (p < 0.0001) compared to robotic patients. The proportion of large uteri (i.e., uterine weight > 200–250 g) was higher in the robotic group (OR: 1.10 [1.00, 1.22]) but this difference was not significant (p = 0.06). BMI and the proportion of patients who had prior surgery was not different between groups.

Robotics versus open approach

Patients operated on by robotics were older than those operated on by the open route, with a mean difference of 0.77 years (p = 0.008). There were no differences in BMI, proportion of large uterus, or previous surgery.

Robotics versus vaginal approach

There were no statistical differences in term of age or uterus weight between the robotic and the vaginal approach. However, the proportion of large uteruses (uterus weight > 250 g/8 weeks) was higher in the robotic group (623/3315, 18.8% vs. 1305/11888, 11%, p = 0.02). There was also a significantly higher rate of previous surgery in the robotic group (518/1075, 48% vs. 1333/3801, 35%, p < 0.00001).

Peri-operative outcomesRobotic versus laparoscopic approach

Individual meta-analyses of four RCTs [21, 29, 33, 36], three prospective comparison studies [23, 31, 34] and five independent database studies [16, 25, 26, 30, 37], and the overall pooled analysis showed no difference in operative time (Fig. 3a).

Fig. 3

Forest plots for the robotic versus laparoscopic comparison showing the results of the overall pooled results and the study type subgroup analyses for a operative (OR) time in minutes (min), b estimated blood loss (EBL) in milliliters (mL), c blood transfusions, d length of hospital stay (LOS) in days, and e the number of patients requiring a hospital stay greater than 2 days. Lap laparoscopic, SD standard deviation, IV inverse variance, CI confidence interval, calc calculation, MD mean difference, hr hour, IQR interquartile range, OR odds ratio, RR risk ratio

The overall pooled estimated blood loss (EBL) was less after the robotic versus the laparoscopic approach (MD: – 52.31 mL [– 98.17, – 6.45], I2 = 95%, p = 0.03); however, the sub-analysis of RCTs [29, 36] was not significant (MD: – 14.02 mL [– 69.18, 41.15], I2 = 66%, p = 0.62 Fig. 3b). This same pattern was seen with blood transfusions; the overall pooled estimate favored robotic (RD: – 0.0043 [– 0.0059, – 0.0027], I2 = 37%, p < 0.00001), but the one RCT (RD: 0.0385 [– 0.0879, 0.1648], p = 0.55) and the prospective subgroup analysis were not significant (RD: – 0.0046 [– 0.0235, 0.0142], I2 = 0%, p = 0.63, Fig. 3c).

While the subgroup meta-analysis of four RCTs [21, 29, 33, 36] and the subgroup meta-analysis of three prospective studies [23, 31, 34] showed no difference in average LOS between the robotic and the laparoscopic approach (RCT MD: – 0.10 days [– 0.30, 0.10], I2 = 53%, p = 0.32; Prospective MD: – 0.52 [– 1.13, 0.09], I2 = 89%, p = 0.10, Fig. 3d), the meta-analysis of independent database studies (MD: – 0.12 days [– 0.20, – 0.04], I2 = 97%, p = 0.005) and the overall pooled result (MD: – 0.14 days [– 0.21, – 0.08], I2 = 95%, p < 0.0001) were both significant, but with a robotic benefit of only 3 h.

When measured as the proportion of patients requiring a hospital stay greater than 2 days, the meta-analysis (all database studies) showed that patients had a lower risk of staying more than 2 days post-operatively after a robotic procedure compared to the laparoscopic approach (OR: 0.67 [0.57, 0.79], I2 = 82%, p < 0.0001, Fig. 3e).

Conversion rates were similar between both groups in the subgroup meta-analysis of four RCTs [21, 29, 33, 36] and three prospective studies [23, 31, 34]. The subgroup meta-analysis of database studies showed a lower risk of conversion for the robotic approach (RD: – 0.06 [– 0.10, – 0.02], I2 = 100%, p = 0.003), as did the overall pooled analysis (RD: – 0.04 [– 0.06, – 0.01], I2 = 100%, p = 0.008, Fig. 4a). There were no significant differences in intraoperative or postoperative complications, mortality, or reoperations between the robotic approach and the laparoscopic approach (Fig. 4b–e). Readmissions was significant in favor of the robotic group for the overall pooled analysis (OR: 0.90 [0.83, 0.99], I2 = 20%, p = 0.03), which was driven by the database subgroup (OR: 0.91 [0.83, 0.99], I2 = 33%, p = 0.03 Fig. 4f).

Fig. 4

Forest plots for the robotic versus laparoscopic comparison showing the results of the overall pooled results and the study type subgroup analyses for a conversions to open surgery, b intraoperative complications (Intraop Comps), c mortality, d postoperative complications (Postop* Comps), e reoperations, and f readmissions. L or Lap laparoscopic, M–H Mantel–Haenszel, CI confidence interval, calc calculated, OR odds ratio, adj adjusted, NS not significant, RCT randomized controlled trial, R robotic, RR risk ratio, periop = perioperative, FU follow-up, LAVH laparoscopic-assisted vaginal hysterectomy, MIS minimally invasive surgery, TLH total laparoscopic hysterectomy. *Indicates that the postoperative complication rate within 30 days was extracted preferentially, but if the paper only reported a perioperative rate, it was included so as to not exclude that paper from the analysis

A sensitivity analysis limited to RCT and prospective studies resulted in blood loss, transfusions, conversions, and readmissions becoming non-significant (Online Resource 4). Complications, operative time, mortality, and reoperation were still comparable, and length of hospital stay was about 7 h shorter in the robotic group (7 studies, 377 robotic, 465 laparoscopic patients, MD: – 0.28 [– 0.51, – 0.05], I2 = 86%, p = 0.02).

Robotic versus open approach

Operative time was comparable in the overall pooled analysis and in all of the subgroup analyses (Fig. 5a). The overall pooled analysis and the database subgroup analysis for blood transfusions showed lower risk in the robotic group (Fig. 5b), but the subgroup analysis of prospective comparison studies was not significant. The individual meta-analyses of two prospective studies [34, 38] and four database studies [16, 24, 28, 30] reporter a shorter LOS favoring the robotic approach (MD: – 1.65 days [– 2.33, – 0.96], I2 = 95%, p < 0.00001 and MD: -1.14 days [– 1.85, – 0.43], I2 = 100%, p = 0.002, respectively, Fig. 5c) as did the overall pooled analysis (MD: – 1.31 [– 1.88, – 0.74], I2 = 100%, p < 0.00001). Consistent with this, the meta-analysis of three database studies [17, 19, 39] showed that the risk of staying more than 2 days in the hospital after the surgery was higher after an open approach (OR: 0.04 [0.02, 0.06], I2 = 99%, p < 0.00001, Fig. 5d). The meta-analyses of two prospective studies [34, 38] and the overall pooled analysis showed a lower EBL with the robotic approach versus the open approach (MD: – 81.45 mL [– 118.71, – 44.20], I2 = 50%, p < 0.0001, MD: – 123.01 [– 214.83, – 31.19], I2 = 99%, p = 0.009 respectively, Fig. 5e). The overall pooled analysis (three database studies [22, 30, 39] and one prospective study [38]) showed a lower risk of mortality with the robotic group (OR: 0.12 [0.05, 0.29], I2 = 26%, p < 0.00001, Fig. 5f). This difference persisted using a risk difference analysis (RD: – 0.0010 [– 0.0019, – 0.0002], I2 = 62%, p = 0.01). The risk of intraoperative and postoperative complications was found to be lower with the robotic approach based on database studies (OR: 0.59 [0.48, 0.71], I2 = 43%, p < 0.00001, Fig. 5g, and OR: 0.42 [0.27, 0.66], I2 = 98%, p = 0.0001, Fig. 5h, respectively). This difference persisted when using a risk difference analysis (intraoperative RD: – 0.01 [– 0.02, – 0.01], I2 = 0%, p < 0.00001; postoperative RD: – 0.09 [– 0.14, – 0.04], I2 = 98%, p = 0.0003). However, there was no difference in readmission or reoperation between the open and the robotic approach (Fig. 6).

Fig. 5

Forest plots for the robotic versus open surgery comparison showing the results of the overall pooled results and the study type subgroup analyses for a operative (OR) time in minutes (min), b blood transfusions, c length of hospital stay (LOS) in days, and d the number of patients requiring a hospital stay greater than 2 days, e estimated blood loss (EBL) in milliliters (mL), f mortality, g intraoperative complications (Intraop Comps), and h postoperative complications (Postop* Comps). SD standard deviation, IV inverse variance, CI confidence interval, BMI body mass index, RR risk ratio, O open, R robotic, FU follow-up. *Indicates that the postoperative complication rate within 30 days was extracted preferentially, but if the paper only reported a perioperative rate, it was included so as to not exclude that paper from the analysis

Fig. 6

Forest plots for the robotic versus open surgery comparison showing the results of the overall pooled results and the study type subgroup analyses for a readmissions and b reoperations. M–H Mantel–Haenszel, CI confidence interval, O open, R robotic, OR odds ratio

Robotic versus vaginal approach

Operative time was shown to be 17 min shorter with the vaginal approach compared to the robotic approach in 1 RCT [29] including 61 robotic and 25 vaginal patients (MD: 17 min [3.8, 30.2], p = 0.01, Fig. 7a). Similar results were found in one prospective study [18] and the meta-analysis of three database studies [16, 30, 37], for an overall pooled average savings of 43 min (MD: 42.87 [22.94, 62.80], I2 = 100%, p < 0.0001). EBL was lower with the robotic approach in the overall pooled result (MD: – 71.18 [– 85.15, – 57.20], I2 = 79%, p < 0.00001) and in the meta-analyses of two database studies [16, 37] (MD: – 74.49 mL [– 94.76, – 54.22], I2 = 80%, p < 0.00001, Fig. 7b). Similar results were found in one prospective study [18] (MD: – 78 mL [– 84.95, – 71.05], p < 0.00001) but one RCT [29] did not report any difference between the groups. Overall, LOS was shorter in the robotic group (MD: – 0.39 [– 0.70, – 0.08], I2 = 99%, p = 0.01, Fig. 7c), and in the one prospective study, but was comparable in the one RCT. The risk of prolonged LOS > 2 days was lower with the robotic approach (all database studies: OR: 0.56 [0.40, 0.78], I2 = 82%, p = 0.0006, Fig. 7d). There were no differences in mortality, blood transfusions, or conversions to the open approach (Fig. 7e–g). While intraoperative complications were comparable in one RCT [29] and one prospective study [18] (Fig. 7h), there was a lower risk of intraoperative complications during robotic procedures in the meta-analysis of two database studies [17, 28] (OR: 0.44 [0.28, 0.71], I2 = 0%, p = 0.0007) and in the overall pooled analysis (OR: 0.45 [0.29, 0.71], I2 = 0%, p = 0.0006), with no differences seen for postoperative complications (Fig. 7i). Overall, there were no significant differences in readmission or reoperation between the robotic and vaginal approaches (Fig. 8a, b).

Fig. 7

Forest plots for the robotic versus vaginal surgery comparison showing the results of the overall pooled results and the study type subgroup analyses for a operative (OR) time in minutes (min), b estimated blood loss (EBL) in milliliters (mL), c length of hospital stay (LOS) in days, d the number of patients requiring a hospital stay greater than 2 days, e mortality, f blood transfusions, g conversions to open surgery, h intraoperative complications (Intraop Comps), and i postoperative complications (Postop* Comps). SD standard deviation, IV inverse variance, CI confidence interval, V vaginal, R robotic, OR odds ratio, FU follow-up. *Indicates that the postoperative complication rate within 30 days w