Results of the search

Searches conducted for the two original reviews contributed the following studies to this combined review:

• Liu 2012 (benign gynaecological disease) included Sarlos 2010 (conference abstract only) and Paraiso 2011, and excluded three studies (not RCTs). These previously included studies comprised six citations (five conference abstracts and one full published report).

• Lu 2012 (malignant gynaecological disease) included no studies and excluded 27 studies (not RCTs).

For further details of these searches, please consult the original reviews (Liu 2012; Lu 2012). For this updated review, the combined updated searches yielded 2221 deduplicated study records. After sifting through the titles and abstracts of these records, we identified 15 records for classification (nine full articles and six conference abstracts) pertaining to 10 studies in total (Anger 2014; Campos 2013; Desille‐Gbaguidi 2013; Gocmen 2013; Green 2013; Martinez‐Maestro 2014; McNanley 2012; Palmer 2013; Paraiso 2013; Sarlos 2010). In addition, we identified one newly published paper (Lonnerfors 2014) by writing to the investigators of a trial identified by searching online trial registers. Of these 11 potentially eligible studies in total, we excluded six studies (Campos 2013; Desille‐Gbaguidi 2013; Gocmen 2013; Martinez‐Maestro 2014; McNanley 2012; Palmer 2013) and included four studies (Anger 2014; Green 2013; Lonnerfors 2014; Paraiso 2013) (Figure 1). The remaining paper was a new publication (full report) of a previously included study, which had been previously included as a conference abstract alone (Sarlos 2010).

Figure 1

---

---

Study flow diagram for updated searches (30 June 2014).

One included study (Green 2013) was a conference abstract of a study initially identified as an 'ongoing study' (see protocol citation linked to Green 2013). We requested unpublished data from these investigators and received limited data in the form of a Microsoft PowerPoint presentation. We were informed by the investigators that they had had difficulty getting the paper published because of 'too many cross‐overs in the stats'; however, we understand that they plan to make further attempts to get the study published. (See Green 2013 in Characteristics of included studies for additional details.)

We identified five ongoing RCTs (Kjolhede 2012; LAROSE 2012; Narducci 2010; Obermair 2008; RASHEC 2013) and contacted these trial investigators for up‐to‐date details about accrual and completion dates. This information is summarised in Table 1.

Included studies

This update includes six studies (four new and two previously included studies) (Anger 2014; Green 2013; Lonnerfors 2014; Paraiso 2011; Paraiso 2013; Sarlos 2010). All included studies were conducted from 2007 onwards and evaluated RAS in benign gynaecological disease. We identified no RCTs of RAS for malignant gynaecological disease.

Excluded studies

Thirty non‐randomised studies were excluded from the original reviews. For the updated review, we excluded six additional studies for the following reasons.

• Quasi‐RCT (Martinez‐Maestro 2014).

• Not an RCT (Desille‐Gbaguidi 2013; Gocmen 2013).

• Inappropriate interventions compared (Campos 2013; McNanley 2012; Palmer 2013).

Allocation

All studies were RCTs, and the method of randomisation was adequately described in all of them. We therefore considered these studies to be at low risk of selection bias, except for two RCTs that did not describe allocation concealment (Green 2013; Sarlos 2010). We considered these two studies to be at unclear risk of bias for this item.

Blinding

Three studies (Anger 2014; Paraiso 2011; Paraiso 2013) described blinding participants and assessors to group allocation; we considered these studies to be at low risk of performance and detection bias. The other three studies (Green 2013; Lonnerfors 2014; Sarlos 2010) were open‐label studies; we considered these to be at high risk of performance bias. Additionally, these studies did not report efforts to blind outcome assessment; we therefore considered them to be at unclear risk of detection bias. However, it was not clear in any of the included studies who had assessed outcomes such as extent of blood loss and length of hospital stay, which normally are determined by the surgeon who performed the procedure and therefore are at potentially high risk of bias.

Incomplete outcome data

Two studies reported no or minimal loss to follow‐up (Anger 2014; Lonnerfors 2014) (low risk of bias). In Paraiso 2011, five women were withdrawn from each study arm after randomisation (unclear risk of bias). In Paraiso 2013, nine women withdrew after randomisation (five in the CLS group and four in the RAS group), one woman allocated for CLS underwent RAS in error (protocol deviation) and one woman in the CLS group was withdrawn as the result of missing data (unclear risk of bias). Quality of life outcomes in Paraiso 2013 and Sarlos 2010 were subject to attrition of greater than 20%, so data for this outcome were considered to be at high risk of bias.

In Green 2013, of 113 women initially randomly assigned, 10 women were withdrawn because procedures were cancelled for medical or personal reasons (eight in the CLS group and two in the RAS group), and five women who had undergone alternative procedures were excluded (three in the CLS group and two in the RAS group). This left 98 participants (48 in the CLS group and 50 in the RAS group), representing attrition of 13% of the sample. In addition, 11 protocol deviations were reported. We considered this study to be at high risk of attrition bias.

Selective reporting

Most included studies reported expected and/or prespecified outcomes. We considered Green 2013 to be at high risk of reporting bias because of high attrition and protocol deviations with subsequent reporting of data per protocol. In Paraiso 2013, most expected outcomes were reported; however, no details of complications were provided despite the fact that three women required blood transfusions (unclear risk). In Lonnerfors 2014, outcomes were reported for RAS versus minimally invasive surgery (CLS and vaginal hysterectomy) together and separately. However, as separate baseline data were not reported, it was not possible to determine whether there were differences in the baseline characteristics of control women undergoing CLS or vaginal hysterectomy (unclear risk).

Other potential sources of bias

Two studies (Paraiso 2011; Sarlos 2010) were early adopters of RAS. Enrolment for these two studies occurred between 2007 and 2011, and, arguably, data from these studies may have been subject to bias caused by the learning curve, although surgeons in the latter study had performed a minimum of 30 RAS procedures before commencing the trial. Nevertheless, to assess this possibility and to avoid potential bias from early studies in the review findings, we performed sensitivity analyses for most outcomes by excluding these early studies.

In the Paraiso 2011 study report, it was unclear whether participants experienced more than one intraoperative and/or postoperative complication, and it was not possible to determine the direction of any bias as a result. This study also included relatively minor complications (e.g. urinary tract infections), unlike the other studies; this might have contributed heterogeneity to the 'complications' analyses, but not bias necessarily. Although we included these data in the analyses, we performed sensitivity analysis by excluding this study.

In Lonnerfors 2014, RAS was compared with CLS or vaginal hysterectomy, and the comparison intervention was based on the surgeon's choice. As participants were not randomly assigned to CLS and vaginal hysterectomy separately, these separate reported data were potentially subject to significant bias (e.g. a higher percentage of women who underwent CLS had a concomitant procedure (75% CLS, 59% RAS and 19% vaginal hysterectomy procedures) that would have influenced procedure time, complication rates, length of hospital stay (LOS) and costs of the separate CLS data. Therefore, we extracted combined data, not separate data, from this study. The direction and magnitude of bias resulting from inclusion of these data in this review are unclear. We performed sensitivity analyses to evaluate the impact of this study on review results.

Authors of all studies reported no potential conflicts of interest. Although most study reports declared the main study sponsor as the institution at which the study was undertaken, it was unclear whether study institutions had received financial support from the system manufacturers, directly or indirectly. Therefore, from a funding perspective, we considered most studies to be at potentially high risk of bias.

As a result of the the small number of included studies, we were unable to meaningfully evaluate publication bias as planned; however, one included study (Green 2013) with significant protocol deviations remains unpublished.

Intraoperative and postoperative complications

No statistically significant differences in complication rates were reported between RAS and CLS arms (risk ratio (RR) 0.95, 95% confidence interval (CI) 0.46 to 1.99; participants = 513; studies = 6; I² = 74%) for hysterectomy or sacrocolpopexy procedures (test for subgroup differences: Chi² = 0.34; df = 1 (P value 0.56); I² = 0%). We downgraded this evidence to 'low' because of inconsistency and imprecision. When we excluded Paraiso 2011 data, overall complication rates were not significantly different for RAS and CLS (RR 0.72, 95% CI 0.37 to 1.43). Early studies tended to favour CLS. whereas later studies favoured RAS (but were at risk of other bias). When we performed various sensitivity analyses, it became clear that the effect estimate was not robust and is likely to change with further research.

Intraoperative complications only

No statistically significant differences in intraoperative complication rates were noted between RAS and CLS arms overall (RR 1.71, 95% CI 0.83 to 3.52; participants = 337; studies = 4; I² = 0%; Analysis 1.2) or between arms for women undergoing hysterectomy (RR 1.67, 95% CI 0.75 to 3.73; participants = 269; studies = 3; I² = 0%). Only one study (Paraiso 2011) was conducted in women undergoing sacrocolpopexy (RR 1.89, 95% CI 0.37 to 9.62; participants = 68; studies = 1). Tests for subgroup differences showed no significant differences between hysterectomy and sacrocolpopexy results (Chi² = 0.02; df = 1 (P value 0.90); I² = 0%). Sensitivity analysis using a fixed‐effect (FE) model while excluding studies with limitations yielded similar results. However, most of the events in both arms of these meta‐analyses occurred in two early studies; exclusion of these studies left insufficient data for meta‐analysis. We downgraded this evidence to 'low' as the result of imprecision and study limitations (risk of bias concerns).

Intraoperative injury

For the specific complication 'intraoperative injury,' no significant differences between review arms were noted (RR 1.23, 95% CI 0.44 to 3.46; participants = 415; studies = 5; I² = 0%; 8/209 events in RAS arm versus 6/206 events in CLS arm; Analysis 1.4). Findings were similar when early studies were excluded; however, reported events were few (moderate‐quality evidence).

Postoperative complications only

With regard to postoperative complications, results of tests for subgroup differences were significant (Chi² = 7.65; df = 1 (P value 0.006); I² = 86.9%); therefore we did not pool subgroup data. For hysterectomy procedures, no statistically significant difference was noted between RAS and CLS arms (RR 0.62, 95% CI 0.30 to 1.29; participants = 315; studies = 3; I² = 27%; Analysis 1.5). Sensitivity analysis using a fixed‐effect (FE) model yielded a point estimate of 0.62 with a non‐significantly narrower CI. Early studies tended to favour CLS, whereas later studies favoured RAS but had other study limitations. Sensitivity analysis conducted by sequentially excluding studies with risk of bias concerns produced an RR point estimate of 0.63 when only Green 2013 was excluded (95% 0.18 to 2.28; participants = studies = 2; Analysis 1.6), and when both Green 2013 and Lonnerfors 2014 were excluded, only one early study (Sarlos 2010) was left, with an RR of 1.23 (95% CI 0.40 to 3.74). We downgraded this evidence to 'low' as the result of imprecision and study limitations.

For sacrocolpopexy procedures, only one small, early study contributed data (Paraiso 2011). This study found RAS to be associated with a significantly greater number of postoperative complications compared with CLS for this procedure (RR 3.54, 95% CI 1.31 to 9.56; participants = 68); however, it was unclear from the study report whether individual participants might have experienced more than one complication.

Early and late mortality

No deaths were reported in any of the included studies.

Operating time

Mean total operating time was significantly longer on average in the RAS arm than in the CLS arm overall (mean difference (MD) 41.71 minutes, 95% CI 17.08 to 66.33; P value 0.0009; participants = 294; studies = 4; I² = 82%; Analysis 1.9). Tests for subgroup differences showed no significant differences in effect between hysterectomy and sacrocolpopexy subgroups (Chi² = 0.01; df = 1 (P value 0.90); I² = 0%), and all studies included in this analysis were at low risk of bias for this outcome. When early studies were excluded from sensitivity analyses, the point estimate was similar, at an average 43 minutes longer operating time in the RAS arm.

Two studies reported this outcome as median (range). Median total operating times reported for the RAS arm versus minimally invasive arms in Lonnerfors 2014 were 76 minutes (43 to 210) versus 86 minutes (29 to 223), respectively (P value 0.54). Likewise, for Green 2013, median total operating times were 90 minutes (74 to 104) and 88 minutes ([75 to 105), respectively (P value 0.69). These individual study data, which shed a favourable light on RAS, were at high risk of bias for the reasons previously mentioned (see Risk of bias in included studies).

Results for mean operating room time were similar to those for mean total operating time (MD 42.51, 95% CI 20.96 to 64.06; participants = 294; studies = 4; I² = 70%; P value 0.0001; Analysis 1.10).

Length of hospital stay

Mean length of hospital stay in days differed significantly between hysterectomy and sacrocolpopexy procedures (tests for subgroup differences: Chi² = 5.06; df = 1 (P value 0.02); I² = 80.2%; Analysis 1.11); therefore we did not pool these subgroups.

For hysterectomy procedures, RAS resulted in significantly shorter hospital stays on average compared with CLS (MD ‐0.30 days (approximately 7.2 hours), 95% CI ‐0.54 to ‐0.06; participants = 217; studies = 2; I² = 0%). Data from one study (Lonnerfors 2014) with limitations in the context of this review had a significant impact on this evidence, which we downgraded to low quality.

For sacrocolpopexy procedures, only one small, early study contributed data on mean length of hospital stay (Paraiso 2011; MD 0.37 days, 95% CI ‐0.16 to 0.90; participants = 68; very low‐quality evidence). These data reported as median (range) were 36 hours (19 to 240) in the RAS arm compared with 29 hours (15 to 65) in the CLS arm (non‐significant).

Conversion to another approach

No significant differences between RAS and CLS approaches were observed with regard to rate of conversion (RR 1.28, 95% CI 0.40 to 4.12; participants = 337; studies = 4; I² = 0%; 6/169 versus 5/168; Analysis 1.12; moderate‐quality evidence). In the RAS arm, three conversions were required as the result of 'robot malfunction,' and one was due to an inability to ventilate the study participant. These incidents occurred in the two earlier studies (Paraiso 2011; Sarlos 2010).

Blood transfusions and blood loss

Three studies reported blood transfusions. No statistically significant difference between RAS and CLS arms was reported (RR 1.23, 95% CI 0.24 to 6.21; participants = 272; I² = 0%; Analysis 1.13). Similarly, mean estimated blood loss (millilitres) was not statistically significantly different between comparator arms (MD 3.54, 95% CI ‐20.12 to 27.21; participants = 173; studies = 2; I² = 0%; Analysis 1.14), although only two studies contributed data (tests for subgroup differences: Chi² = 0.59; df = 1 (P value 0.44); I² = 0%).

Pain

Postoperative pain was reported in four studies at different time points within the first two weeks (Anger 2014; Green 2013; Paraiso 2011; Paraiso 2013). However, only two studies provided usable data (means) for meta‐analysis: one in the hysterectomy subgroup (Paraiso 2013) and one in the sacrocolpopexy subgroup (Paraiso 2011).

With regard to the hysterectomy subgroup: Paraiso 2013 reported pain scores during normal activities at two weeks post hysterectomy with no significant differences between study arms (P value 0.79; Analysis 1.15). Green 2013 (a high risk of bias study) reported median postoperative pain scores and found no significant differences between RAS and CLS arms following hysterectomy (P value 0.73).

For the sacrocolpopexy subgroup: Anger 2014 reported mean pain scores at one week postoperatively. Differences in pain scores favoured the CLS arm but were not statistically significant (standardised mean difference (SMD) 0.41, 95% CI ‐0.03 to 0.86; participants = 78; P value 0.07; I² = 0%; Analysis 1.15). Although it did not contribute data to the meta‐analysis, the other sacrocolpopexy study (Paraiso 2011) reported that the RAS group had significantly greater pain at rest and with activity during weeks 3 to 5 and required longer use of non‐steroidal anti‐inflammatory drugs (NSAIDs) (median 20 days versus 11 days; P value < .005). This low‐quality evidence suggests that, when performed for sacrocolpopexy, RAS may be associated with greater postoperative pain than is noted with CLS.

QoL

Quality of life was reported at four to six weeks in two studies (Anger 2014; Sarlos 2010) and at six months in two studies (Paraiso 2011; Paraiso 2013). Paraiso 2011 also measured QoL at 12 months. However, data were insufficient for meta‐analysis. No significant differences in QoL were found in any of the studies reporting these data, except for Sarlos 2010. The latter study initially found a greater change in QoL at six weeks compared with before the operation in the RAS group, but this self‐reported outcome was subject to significant bias, and results did not remain significant when sensitivity analysis was conducted by the investigators. This evidence was of low quality.

Reintervention/readmission

No significant difference between RAS and CLS was noted with regard to the number of cases requiring reintervention (RR 0.35, 95% CI 0.08 to 1.54; participants = 295; studies = 3; I² = 0%; Analysis 1.18) or readmission (hysterectomy subgroup only: RR 0.46, 95% CI 0.14 to 1.48; participants = 220; studies = 2; I² = 0%; Analysis 1.19). These data were sparse, were at risk of bias and were considered to be of low quality.

Cost

Three studies (Anger 2014; Lonnerfors 2014; Paraiso 2011) reported overall costs (including equipment setup and maintenance and theatre and hospital admission costs). We did not pool these data because heterogeneity was substantial (I² > 90%); however, RAS was consistently associated with increased cost, with MDs ranging from 1936 USD to 8728 USD across the three studies (Analysis 1.20).