European Urology

1. Introduction

Prostate cancer is the most common nondermatologic cancer in Western men [1]. More than 90% of cases are diagnosed when the disease is organ confined and potentially curable by radical prostatectomy (RP) [2]. This was traditionally performed by open retropubic RP (ORP), although minimally invasive surgery (MIS; conventional laparoscopic and robot-assisted laparoscopic RP, LRP and RALP, respectively) has become popular recently. There is substantial evidence of lower bleeding rates for MIS [3], [4], [5], and [6] but no good evidence of an overall benefit for one modality over another, and it is uncertain whether MIS, especially robotics, justifies its increased costs and training requirements [7].

The most important outcomes to assess when comparing ORP, LRP, and RALP are cancer control, complications, urinary continence, and sexual potency. Unfortunately, biochemical recurrences, metastases, and survival statistics, as well as continence and potency, require long-term follow-up to assess and can be subject to significant reporting and interpretational biases. Positive surgical margin (PSM) and perioperative complication rates thus remain early outcome measures of importance in comparing surgical modalities. Accrual to randomized controlled studies (RCTs) has been notoriously poor [3] with only one such trial reported comparing ORP and LRP [8]. In addition, of the thousands of papers published on the surgical treatment of prostate cancer with radical prostatectomy, there have been relatively few comparative studies. A recent review found 37 comparative studies: 23 ORP and LRP, 10 ORP and RALP, and 4 LRP and RALP [9]. Meta-analyses based solely on comparative articles [9], [10], and [11], although informative, do not represent most of the literature available in this area, which are single cohort studies. Also, these reports do not adequately adjust for differences between surgical cohorts. Therefore we performed a meta-analysis of all of the available peer-reviewed observational studies, controlling for differences between ORP, LRP, and RALP cohorts with propensity score adjustments [12] and [13], a technique used by other authors [14], [15], [16], and [17].

2. Evidence acquisition

The methods used in reporting this meta-analysis follow those outlined in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [18] (see Appendix 1, available online). The study design, search strategy, data abstraction, and excluded studies (with justification) are detailed in Table 1, Figure 1[19], [20], [21], and [22], Appendix 2 (citations for excluded studies, available online), and Figure 2 (early experience analysis). PubMed and Scopus (2002–2010, English) were searched for original research articles reporting on the primary treatment of localized prostate cancer. We also reviewed the reference lists of prostatectomy review/meta-analysis articles and included those not already in our database [9], [11], [22], and [23].

Fig. 1 Flowchart outlining the literature search and article evaluation process.^aAll inclusion and exclusion criteria were applied to each search in a uniform manner.^bThe last search was completed in March 2011.^cRepresents articles that did not report on outcomes of prostate cancer treatment, articles that reported on alternative treatments for prostate cancer, and articles that did not report on margin or complication rates.^dIncludes letters to the editor, editorials, comments, news articles, review articles, and meta-analyses.^eIncludes articles in which outcomes were not stratified by surgical approach, articles with additional confounding treatments, and articles with incomplete or erroneous data (these controls were also applied at the level of outcomes). See Appendix 2 (available online) for a list of excluded studies.^fExcluded articles where some or all of the same patients were included in a different study reporting on the same parameters.^gExcluded articles where the surgery took place prior to 1990 when minimally invasive surgery was first available.^hExcluded articles that reported on the early experience of a surgeon. We defined early experience as the first 75 cases for a surgeon for each approach based on reports of the number of cases required to reach a base level of competence with the robotic approach [19], [20], [21], and [22] and confirmed with a post hoc analysis of positive margin rates as a function of the number of robot-assisted laparoscopic prostatectomy, laparoscopic radical prostatectomy, or retropubic radical prostatectomy cases performed that showed a decrease in the variability of outcomes after 75 cases (see Fig. 2).

Fig. 2 Early experience analysis. Graph showing the overall positive margin rate relative to the number of initial cases for robot-assisted laparoscopic radical prostatectomy (RALP; diamonds), laparoscopic radical prostatectomy (LRP; squares), and radical retropubic prostatectomy (Open; triangles).

All variables abstracted were chosen a priori for analysis. The main variables abstracted were surgical approach (retropubic [open], laparoscopic, robotic), type of publication (comparative series, single surgical method series) and publication year (2002–2010) (Table 2a and Table 2b), patient characteristics (age, body mass index [BMI], preoperative prostate-specific antigen [PSA], preoperative Gleason score, and pathologic stage) (Table 3), and primary (positive surgical margin rates and total intra- and perioperative complication rates) and secondary outcomes (estimated blood loss [EBL], blood transfusions, conversions to ORP, length of stay [LOS], and specific individual complication rates; Table 4, Table 5, Table 6, and Table 7). Follow-up was the perioperative period, defined as the time during surgery (intraoperative) and encompassing the first 30 d following surgery. Primary outcomes included measurement of efficacy (pT2, pT3, and overall PSM rates) and safety (total intraoperative and total perioperative complication rates). Secondary outcomes included changes in PSM rates over time, EBL, transfusion and conversion rates, and hospital LOS separated into US studies and non-US studies due to differences in discharge policies and individual complication rates (percentage of patients in a cohort experiencing the complication considered within 30 d of surgery).

2.1. Statistical methods

Patient characteristics and outcomes were summarized with descriptive statistics using weighted averages and weighted standard deviations with individual cohort size as the weight. Surgical approaches were contrasted in a pairwise manner using two sample t tests with Hochberg correction for multiple comparisons. The three pairwise comparisons were RALP versus LRP, RALP versus ORP, and LRP versus ORP. The Hochberg method was implemented as follows: All three comparisons were significant if each had a p value ≤0.05; if one of the three comparisons was not significant (p > 0.05), then the other two were significant if they each had a p value ≤0.025; if two of three comparisons were not significant (p > 0.05), then the remaining one was significant if it had a p value <0.017.

Because current literature contains almost entirely noncontrolled single institutional studies, cohorts of surgical methods to be compared often differ in terms of demographic and tumor characteristics. The statistical significance of outcomes between cohorts was adjusted for these differences using propensity score analysis. Propensity score analysis directly addresses the key problem of observational group comparisons: that of differences in patient selection criteria. First, covariates were selected that might distinguish between the surgical cohorts to be compared and have a relationship with outcome variables. Overall PSM rates were adjusted for preoperative Gleason score, preoperative PSA, and pathologic stage; pT2 and pT3 PSM rates were adjusted for preoperative Gleason score and preoperative PSA; all perioperative outcomes and complication rates were adjusted for age, BMI, preoperative Gleason score, preoperative PSA, and pathologic stage. A multivariate logistic regression modeling of the probability of being in a cohort treated with either of two surgical methods to be compared (there are three such pairwise models) yielded estimated propensity scores, where the propensity score was defined as the probability of being in either type of surgical cohort, conditioned on the average covariate values for that cohort. Second, study cohort propensity scores (from both surgical methods) were ordered from smallest to largest and adjacent scores grouped into strata. Depending on the number of cohorts available to make comparisons, up to five strata were formed, usually sufficient to remove 90% of the selection bias [12]. Within each derived stratum, surgical methods were compared using a difference in average outcome. Propensity score analysis yielded covariate distributions that are similar across cohorts to be compared within each stratum [13]. A Cochran-Mantel-Haenszel weight was then calculated for each stratum. The adjusted summary measure of difference between surgical methods was computed as the weighted average of differences of the surgical methods across strata. The weighted standard error across strata was also calculated, and pairwise comparisons were evaluated using standard normal theory. For each outcome for each surgical method, two different adjusted rates were calculated because the value of the adjusted rate depends on the pair of surgical methods being compared. The size of the adjustment was larger when there was a larger difference in preoperative characteristics included in the propensity score model. Because the adjusted rates were not unique and inferences were based on the (uniquely) adjusted difference between pairs of surgical methods, only the differences in rates were reported after propensity score adjustment.

For a given surgical type, the I² statistic was used to quantify the percentage of variability in outcomes due to differences between cohorts. This measure is based on the Q statistic, a weighted average of squared deviations of each cohort's outcome from the mean of their outcomes. The I² index quantifies the relative amount of heterogeneity between outcome values by comparing the value of the Q statistic with its expected value assuming homogeneity.

Additional analyses included a Mantel-Haenszel chi-square test to make an overall comparison and Fisher exact tests to check for pairwise differences in publications by year between surgical approaches. The Jonckheere-Terpstra test was used to check for changes in positive margin rates over time. Funnel plot analysis was performed on overall PSM and total perioperative complication rates to check for publication bias (Fig. 4). All analyses were performed using SAS v.9.2.1 (SAS Institute, Cary, NC, USA).

Fig. 4 Funnel plot analysis. Scatter plots of treatment effect (overall positive margin rate or total perioperative complication rate) by study size are shown for robot-assisted laparoscopic radical prostatectomy (RALP), laparoscopic radical prostatectomy (LRP), and radical retropubic prostatectomy (Open). Vertical dotted lines represent the weighted averages. Solid lines are visual aids for identifying symmetry versus asymmetry, with a symmetric shape indicative of a “well-behaved” data set in which publication bias is unlikely.

Summary data were abstracted from 400 studies (Appendix 3, available online) that reported on 168 ORP, 127 LRP, and 166 RALP cohorts representing 167 184 ORP, 57 303 LRP, and 62 389 RALP patients (total: 286 876 patients).

3. Evidence synthesis

There were differences in the rate of publications across the three cohorts over time (p < 0.0001). Specifically, a statistically significantly greater proportion of ORP and LRP cohorts came from early publications (2002–2004) compared with RALP (12.5% and 18.1% vs 1.2%; ORP vs LRP, p = 0.18; ORP vs RALP, p < 0.0001; LRP vs RALP, p < 0.0001) (Table 2a and Table 2b). There were differences in preoperative PSA between surgical cohorts, with the RALP cohort having a significantly lower average preoperative PSA than the ORP and LRP cohorts (p < 0.0001 for both comparisons) (Table 3). There were no significant differences in the percentage of high-grade disease (Gleason score >7) between the surgical approaches, although the ORP cohort (28.6%) and the LRP cohort (26.8%) had a significantly higher percentage of patients with pT3 disease than the RALP cohort (20.1%) (p < 0.0001 for both comparisons).

3.1. Primary outcomes (Fig. 3)

The overall PSM rates were 24.2% ORP, 20.4% LRP, and 16.2% RALP; pT2 PSM rates were 16.6% ORP, 13.0% LRP, and 10.7% RALP with only RALP versus LRP comparisons for overall and pT2 rates attaining significance after propensity score adjustment and Hochberg correction (overall PSM p = 0.002; pT2 PSM p = 0.01). The rates for pT3 cancers (42.6% ORP, 39.7% LRP, and 37.2% RALP) were not significantly different after propensity adjustment and Hochberg correction. Total intraoperative complication rates were significantly higher for ORP (1.5%) versus RALP (0.4%) (p < 0.0001) and for LRP (1.6%) versus RALP (0.4%) (p < 0.0001). For total perioperative complication rates (17.9% ORP, 11.1% LRP, and 7.8% RALP), RALP versus ORP (p < 0.0001) and versus LRP (p = 0.002) were significant. Testing for heterogeneity with Q statistics yielded p < 0.0001 in all cases, and the percentage of total variance attributable to heterogeneity of outcomes between studies as measured with the I² statistic varied from 59.7% to 98.7%. Funnel plot analysis demonstrated a lack of publication bias for overall PSM and an inconclusive finding for total perioperative complication rates (Fig. 4).

Fig. 3 Modified forest plot showing propensity-adjusted differences (X) and 95% confidence intervals (horizontal lines and numbers in brackets) for the primary outcomes. The asterisks mark values that are significant after Hochberg correction (all three comparisons are significant if each has a p value ≤0.05; if one of three comparisons is not significant (p > 0.05), then the other two are significant if they each have a p value ≤0.025; if two of three comparisons are not significant (p > 0.05), then the other one is significant if it has a p value <0.017).LRP = laparoscopic radical prostatectomy; Open = radical retropubic prostatectomy; PSM = positive surgical margin; pT2 = postoperative stage T2; pT3 = postoperative stage T3; RALP = robot-assisted laparoscopic radical prostatectomy.

4. Conclusions

This study represents the largest compilation of radical prostatectomy patients to date, and it serves as a systematic review and meta-analysis of this vast body of literature. However, due to the lack of RCTs, differences in patient characteristics between surgical cohorts might explain differences in outcomes between treatment groups. These differences cannot be fully corrected with statistical methods. Additionally, unknown differences in certain attributes of the patients and physicians themselves and/or the administered treatments could contribute to the highly heterogeneous outcomes between studies. We thus caution readers to interpret the findings of this meta-analysis within the context of the considerations just described. Furthermore, due to the large numbers of patients included in this meta-analysis, results that reach statistical significance may not necessarily be clinically meaningful.

Although there were significant differences between all three surgical approaches for the unadjusted overall and pT2 PSM rates, after propensity adjustment, the only significant differences were lower overall and pT2 PSM rates for RALP compared with LRP. This finding is not explained by differences in preoperative Gleason score or PSA because these parameters were included in the propensity adjustment. Interestingly, the crude difference in overall PSM rates for ORP and RALP of 8.0% (24.2% ORP; 16.2% RALP) was largely eliminated by the propensity adjustment. Given only slight differences between the groups in terms of preoperative Gleason score, much of this effect can be attributed to the higher preoperative PSA and pathologic stage averages of the ORP cohorts (7.5 ng/ml PSA, 28.6% pT3) compared with the RALP cohorts (6.3 ng/ml PSA, 20.1% pT3). The propensity-adjusted pT3 rates were not statistically different for all three surgical modalities after applying Hochberg corrections, which is not surprising because this is a function of the biology of the disease. A recent publication of 950 patients found higher PSM rates for RALP compared with ORP in contrast to our finding of equivalency. However, that study was a head-to-head single case series comparison of only one ORP and one robotic surgeon and not a comparison of a large number of surgeons as in this meta-analysis [24]. Other reviews have found lower PSM rates for RALP compared with LRP and/or ORP [9] and [25]. However, neither of those studies nor others included in this meta-analysis were able to account for potential inconsistencies in the pathologic processing of the specimen, such as whole mount versus standard processing; thus it is not clear how RALP's equivalence with ORP or superiority over LRP for PSM rates will translate into longer term oncologic results. However, a few recent reports have shown equivalent early (1–3 yr) [26] and [27] and midterm (5-yr) biochemical recurrence (BCR) rates for RALP, LRP, and ORP [28] and [29]. Specifically, one recent report on 2132 patients by Barocas et al. (2010) showed similar 3-yr BCR rates for ORP (83.5%) and RALP (84.0%; p = 0.19) [27], and another paper by Drouin et al. (2009) demonstrated equivalent oncologic outcomes at 5 yr between the three surgical modalities (5-yr BCR-free rates: 87.8% ORP, 88.1% LRP, and 89.6% RALP; p = 0.93) [29].

Total intraoperative complication rates and mortality rates were low for all surgical modalities, suggesting that radical prostatectomy is a safe procedure. The finding of reduced blood loss and transfusion rates in the LRP and RALP groups, with RALP causing the least bleeding, is commonly reported in the literature for MIS [3], [4], [5], and [6]. Our results also show that recovery as measured by hospital LOS was quickest for RALP, intermediate for LRP, and slowest for ORP. Along with lower readmission, reoperation, and total perioperative complication rates, this supports the notion of lower morbidity for RALP. Other studies have also confirmed a shorter overall convalescence period for RALP in terms of return to work, social activities, and activities of daily living [30] and [31]. It must be kept in mind that the complications reported in this meta-analysis were abstracted from articles that in the main (339 of 400; 85%) did not report complications using a standardized method such as the Clavien classification system [32], and therefore it is possible that differential reporting of complications affected our findings. We did attempt to reduce bias using specific abstraction measures (Table 1) and can thus be reasonably confident that RALP is at least noninferior to LRP and ORP in terms of early complication rates.

Complications are also difficult to interpret without knowledge of comorbidities. However, 349 of 400 (87%) of the included articles did not mention comorbidity status at all, and only 11 of 400 articles (2.75%) reported comorbidity using the Charlson index. Although age was clinically similar between the surgical groups, thus suggesting that differences in comorbidities between cohorts might not have been large, readers are advised to consider that these differences might have existed and thus influenced complication rates. Another factor that could have influenced the results of this meta-analysis was differences in clinical stage between the surgical groups. Unfortunately, clinical stage was only reported consistent with the TNM system in 122 of 400 articles (30%). However, the inclusion of pathologic stage in this meta-analysis compensates to some extent for the lack of availability of clinical stage information. Route of access to the prostate during surgery can also affect complications. It was not possible to compare trans- and extraperitoneal approaches for MIS because only 13 of 307 articles (4.2%) separated their data based on route. In fact, many papers did not even mention which route was used (91 of 307; 29.6%). That said, most minimally invasive radical prostatectomies are performed transperitoneally, and thus the effect of different routes on various complication differences such as hematoma and lymphocele between the minimally invasive surgical modalities is likely to be modest.

Level of experience is a factor in determining outcomes, with largely comparable results published by high-volume surgeons in the field [27], [33], and [34]. This meta-analysis does not provide evidence that one modality is superior to another among comparisons between high-volume surgeons. We were not able to do this comparison because many included studies did not state individual surgeon caseloads. Further evidence that learning curve could play a part in differential outcomes between treatment modalities is that a funnel plot analysis (Fig. 4) showed skewed complication outcomes, with smaller studies more often reporting higher complication rates, especially for LRP. It may be that included studies resulted in favorable results for RALP over LRP and ORP for some outcomes because the learning curve is shorter for RALP [20], [35], and [36], and thus more so-called expert RALP surgeons were sampled in this meta-analysis than so-called expert LRP or ORP surgeons. Hence it may be that expert surgeons of any modality achieve similar results and that less experienced surgeons fare better with RALP for certain outcomes. Another possibility for our results is that known oncologic factors controlled for in this meta-analysis do not fully represent the biology of the disease. For example, lymph node status was not included in this meta-analysis because this information was often not available. Only 22 of 400 of the papers (5.5%) reported lymph node yield, a surrogate for the extent of lymphadenectomy. An additional 127 of 400 of the papers (32%) reported on the number of patients who underwent lymphadenectomy, and the rest of the articles either gave no information at all (184 of 400; 46%), mentioned that lymphadenectomy was performed in patients who fulfilled specific criteria but did not give the number of patients (47 of 400; 12%), or just reported the number of positive lymph nodes (18 of 400; 4.5%). Therefore, readers are cautioned to interpret differences between surgical modalities in lymphatic-related complications such as lymphocele and thromboembolic disease with care.

Due to limiting the time frame of this study to 30 d postoperatively, this meta-analysis did not compare urinary continence or sexual potency outcomes. This is a limitation of the scope of this work, especially because these outcomes can be related to PSM rates, and thus these comparisons will be the subject of future work. Although Hu and colleagues [14] reported on these outcomes, interpretation of their results is difficult in the absence of validated questionnaires. A recent analysis by Alemozaffar et al. (2011) demonstrated relatively poor outcomes for radical prostatectomy in general with regard to intermediate-term (2-yr) erectile function, making comparisons between surgical modalities for this outcome yet more important [37]. Until all the preceding comparisons are made in a reliable and robust manner, it is not possible to recommend definitively one type of surgical approach over the other, but this meta-analysis is suggestive of at least equivalent PSM rates and improved perioperative morbidity profiles for RALP compared with LRP and ORP.

Author contributions: Ashutosh Tewari had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Tewari, Wiklund.

Acquisition of data: Hebert, Sooriakumaran.

Analysis and interpretation of data: Sooriakumaran, Bloch, Seshadri-Kreaden, Wiklund.

Drafting of the manuscript: Sooriakumaran, Hebert, Bloch, Seshadri-Kreaden.

Critical revision of the manuscript for important intellectual content: Tewari, Wiklund.

Statistical analysis: Bloch.

Obtaining funding: None.

Administrative, technical, or material support: Hebert.

Supervision: Tewari.

Other (specify): None.

Financial disclosures: Ashutosh Tewari certifies that all conflicts of interest, including specific financial interests and relationships and affiliations relevant to the subject matter or materials discussed in the manuscript (eg, employment/ affiliation, grants or funding, consultancies, honoraria, stock ownership or options, expert testimony, royalties, or patents filed, received, or pending), are the following: Usha Seshadri-Kreaden and April E. Hebert have support from Intuitive Surgical Inc. as paid employees. Ashutosh Tewari has received grants from Intuitive Surgical Inc. in the past. Daniel A. Bloch receives consultancy fees from Intuitive Surgical Inc. Prasanna Sooriakumaran and Peter Wiklund have nothing to disclose.

Funding/Support and role of the sponsor: None.

Acknowledgment statement: The authors acknowledge Mark Burns and Sherry Wang of Intuitive Surgical Inc. for aiding with data collection.