1 Introduction

Renal cell carcinoma (RCC) accounts for 2%–3% of adult malignancies and is a prevalent malignant tumor of the urinary system [1, 2]. Recently, more and more RCC are discovered in the early stages, particularly cT1 stage tumors [3]. The standard treatment for small renal masses was partial nephrectomy (PN) [3, 4]. PN has switched from an open to a minimally invasive technique due to decreased complication rates, less blood loss, shorter hospitalization time [5, 6]. The Da Vinci robotic system improves surgical efficiency and reduces difficulty, leading to its gradual adoption by urologists [7].

The retroperitoneal robotic partial nephrectomy (RRPN) and the transperitoneal robotic partial nephrectomy (TRPN) are the two surgical approaches available for robotic partial nephrectomy (RPN) [8]. Several earlier studies have compared the difficulty of operation and patients' prognosis between these two approaches [4, 7, 9-11]. Each of them has unique benefits and drawbacks as well as various application scenarios. When a surgeon is faced with an arbitrary RCC patient, how to select the best surgical approach in light of the preoperative information and the patient's condition is a conundrum that has not been satisfactorily resolved in clinical practice [12]. The majority of surgical judgments are dependent on the surgeon's personal preference and subjective experience because there is currently no quantitative tool to assist surgeons in making a rapid and accurate decision.

The current study relied on a single-institutional retrospective database of TRPN and RRPN, to perform a comparison of perioperative indicators between the two approaches. The objective is to identify the factors associated with the surgical approach and develop a predictive model that maybe potentially aid surgeons (especially young surgeons with less experience) in making clinical decisions. On the other hand, although numerous studies have compared the differences in perioperative and postoperative outcomes of different surgical approaches [4, 7, 9, 11, 13, 14], we would like to report the data from our center. The main objective of this study is to exploratively develop a preliminary prediction model based on existing data to quantify the key factors influencing the choice of surgical approach and to investigate the impact of the surgical approach on postoperative recovery. This study is intended to lay the foundation for larger prospective studies or randomized controlled trials in the future.

2 Materials and Methods

2.1 Patients

Patients who underwent RPN at Peking University Third Hospital between September 2020 and December 2024 were retrospectively collected. Eight senior and experienced urologists completed the operations. Cases with inadequate clinical or imaging data and postoperative pathological confirmation that the primary tumor was not RCC (such as urothelial carcinoma, angiomyolipoma, etc.) were excluded. A total of 455 patients were examined and divided into a training set and a validation set based on the cutoff date of February 29, 2024. After excluding patients who did not meet the criteria, 410 patients were included for subsequent analysis (Figure 1).

3 Results

3.2 Differences in Intra- and Postoperative Indicators Between the Two Approaches After PSM

Table 2 displays the differences in perioperative outcomes between the two groups after PSM. No significant difference was observed in operative time and WIT in TRPN vs. RRPN (120.00 min vs. 117.00 min, p = 0.539; 17.41 min vs. 17.92 min, p = 0.678, respectively). The mean hospitalization time was about 5 days in both groups and no statistic difference (5.07 days vs. 5.18 days, p = 0.673). The EBL (40.00 mL vs. 20.00 mL, p = 0.021) and hemoglobin change (−20.00 g/L vs. −15.00 g/L, p = 0.016) was higher in the TRPN. The postoperative creatinine did not change significantly compared with preoperative in both groups. Histopathology revealed that clear cell renal cell carcinoma (ccRCC) was the most common entity (81.3% vs. 84.4%, p = 0.709), and there were no significant differences in surgical margin, long-term creatinine (6–12 month after surgery), local recurrence rate, nuclear grade, and capsule and vascular invasion. In the postoperative course, complications were documented in 16 patients, with 8.9% in the TRPN group and 6.5% in the RRPN group (p = 0.368). During the median follow-up periods of 24 and 21 months, respectively, the OS of patients in the RRPN and TRPN groups showed no statistically significant difference (Figure 2).

TABLE 2. Intra- and postoperative results after PSM of transperitoneal and retroperitoneal approaches.

Variable	TRPN (n = 123)	RRPN (n = 77)	p
Operative time, min (median [IQR])	120.00 [95.00, 157.50]	117.00 [96.00, 154.00]	0.539
EBL, mL (median [IQR])	40.00 [20.00, 150.00]	20.00 [10.00, 50.00]	0.021
WIT, min (mean [SD])	17.41 [8.48]	17.92 [8.59]	0.678
HGB change, g/L (median [IQR])	−20.00 [−26.00, −9.00]	−15.00 [−22.00, −2.00]	0.016
CR change, μmol/L (median [IQR])	1.00 [−7.00, 11.00]	0.00 [−9.00, 10.00]	0.623
Hospitalization time, d (mean [SD])	5.07 [2.13]	5.18 [1.48]	0.673
Extubation time, d (mean [SD])	3.90 [2.80]	4.09 [2.31]	0.621
Long term creatinine, μmol/L (median [IQR])	84.00 [69.25, 95.75]	80.00 [71.00, 89.00]	0.158
Follow-up, months (median [IQR])	21.00 [13.00, 34.00]	24.00 [15.50, 36.50]	0.137
Renal artery clamping, n [%]
Yes	118 [95.9]	76 [98.7]	0.49
No	5 [4.1]	1 [1.3]
Pathological type, n [%]
ccRCC	100 [81.3]	65 [84.4]	0.709
nccRCC	23 [18.7]	12 [15.6]
WHO/ISUP grade, n [%]
1	25 [20.3]	10 [13.0]	0.341
2	83 [67.5]	59 [76.6]
3	15 [12.2]	8 [10.4]
Renal capsule invasion, n [%]
No	112 [91.1]	70 [90.9]	1
Yes	11 [8.9]	7 [9.1]
Vascular invasion, n [%]
No	119 [96.7]	75 [97.4]	1
Yes	4 [3.3]	2 [2.6]
Complication, n [%]
No	112 [91.1]	72 [93.5]	0.368
Yes	11 [8.9]	5 [6.5]
Surgical margin, n [%]
Negative	120 [97.6]	75 [97.4]	1
Positive suspiciouslya	3 [2.4]	2 [2.6]
Local recurrence, n [%]
No	118 [95.9]	74 [96.1]	1
Yes	5 [4.1]	3 [3.9]

• Abbreviations: CR, creatinine; EBL, estimated blood loss; HGB, hemoglobin; IQR, interquartile range; RRPN, retroperitoneal robotic partial nephrectomy; SD, standard deviation; TRPN, transperitoneal robotic partial nephrectomy; WIT, warm ischemic time.
• ^a Positive suspiciously: Cancer cells and suspicious but not certain cancer cells can be seen at the pathological specimen of margins.

3.3 Prediction Model Based on Logistic Regression

Logistic regression analysis was conducted using the transperitoneal approach as the reference and the retroperitoneal approach as the predictor variable. Results from both univariate and multivariate analyses are presented in Table 3. In univariate analysis, tumor diameter, anteroposterior location, longitudinal location, and accessory renal artery were statistically significant. Finally, the above four factors related to the retroperitoneal approach were included, and the logistic regression equation is $$$ \ln \left(\frac{p}{1-p}\right)=-1.71-0.04\times Size+L+M+N $$$, where p is the probability of the retroperitoneal approach. The unit of size is mm; L was 1.57 when accessory renal artery was present and 0 otherwise; M was 0, 1.32, and 2.61 when the tumor located in anterior, neither, and posterior, respectively; and N was 0, 0.79, and 1.25 when the tumor located in lower, middle, and upper poles, respectively. A nomogram was further drawn to facilitate the visualization of the model, as shown in Figure 3.

TABLE 3. Univariable and multivariable logistic regression analyses of the influence factor for the retroperitoneal approach.

Variables	Univariable logistic regression		Multivariable logistic regression
	OR (95% CI)	p	OR (95% CI)	p
Age (years)	1.00 (0.98–1.02)	0.986	—	—
Weight (kg)	1.00 (0.98–1.02)	0.939	—	—
Hight (cm)	1.00 (0.97–1.04)	0.909	—	—
HGB (g/L)	1.00 (0.98–1.01)	0.919	—	—
CR (μmol/L)	1.00 (0.98–1.01)	0.794	—	—
Tumor size (mm)	0.96 (0.94–0.99)	0.006	0.97 (0.93–0.99)	0.037
Gender
Male	1.00 (Reference)	—	—	—
Female	1.06 (0.58–1.94)	0.855	—	—
Hypertension
No	1.00 (Reference)	—	—	—
Yes	0.95 (0.53–1.70)	0.866	—	—
Diabetes
No	1.00 (Reference)	—	—	—
Yes	0.77 (0.35–1.68)	0.505	—	—
Accessory renal artery
No	1.00 (Reference)	—	1.00 (Reference)	—
Yes	3.94 (1.82–8.51)	< 0.001	4.79 (1.89–12.16)	< 0.001
Surgical history
No	1.00 (Reference)	—	—	—
Yes	1.16 (0.63–2.15)	0.633	—	—
Tumor side
Left	1.00 (Reference)	—	—	—
Right	0.79 (0.45–1.41)	0.428	—	—
Anteroposterior location
Anterior	1.00 (Reference)	—	1.00 (Reference)	—
Neither	2.78 (1.21–6.39)	0.016	3.72 (1.43–9.67)	0.007
Posterior	11.07 (5.22–23.48)	< 0.001	13.59 (5.84–31.59)	< 0.001
Longitudinal location
Lower pole	1.00 (Reference)	—	1.00 (Reference)	—
Middle pole	2.51 (1.15–5.48)	0.021	2.21 (0.87–5.61)	0.094
Upper pole	3.19 (1.48–6.88)	0.003	3.48 (1.36–8.89)	0.009
ASA grade
1	1.00 (Reference)	—	—	—
2	0.61 (0.33–1.13)	0.117	—	—
3	1.45 (0.40–5.23)	0.572	—	—
R.E.N.A.L. score
4–6	1.00 (Reference)	—	—	—
7–9	1.31 (0.72–2.40)	0.374	—	—
10–12	0.83 (0.26–2.62)	0.748	—	—
PADUA score
4–6	1.00 (Reference)	—	—	—
7–9	2.19 (0.86–5.57)	0.098	—	—
10–12	1.43 (0.51–4.04)	0.501	—	—
“3S + f” score
4–6	1.00 (Reference)	—	—	—
7–9	1.08 (0.59–1.99)	0.806	—	—
10–12	0.83 (0.23–3.02)	0.774	—	—
MAP score
0–1	1.00 (Reference)	—	—	—
2–3	1.52 (0.70–3.30)	0.284	—	—
4–5	1.44 (0.67–3.10)	0.346	—	—

• Abbreviations: CR, creatinine; HGB, hemoglobin; OR, odd ratio.

For internal validation of the regression model, the ROC curve is shown in Figure 4A, and the area under the curve (AUC) is 0.84. The nomogram model was internally validated using a 1000 bootstrap resampling method. The calibration curve (Figure 4B) exhibited a strong consistency (Hosmer–Lemeshow p = 0.224) between the actual and simulated curves. In decision curve analyses (Figure 4C), it was shown that the net benefit was high, in the range from 0.1 to 0.8, suggesting benefits within a wide probability range. The clinical impact curve (Figure 4D) indicated that when the high-risk threshold was > 80% of the predicted score probability, the nomogram determined that the approach was highly matched with the actual approach.

3.4 External Validation of the Prediction Model

A total of 92 patients were included in the validation cohort, with 59 cases (64.1%) undergoing the transperitoneal approach and 33 cases (35.9%) undergoing the retroperitoneal approach. We analyzed the distribution of baseline clinical characteristics between the training and validation cohorts (Table 4) and found that patients in the validation cohort had a higher prevalence of diabetes (p = 0.007) and a higher proportion of non-clear cell carcinoma pathology (p = 0.036) and higher ASA grade (p = 0.020). In addition, we analyzed the differences in baseline clinical characteristics between the two surgical approach groups within the validation cohort and found no statistically significant differences in any indicators (Table 1). Data from the validation cohort were used to validate the nomogram independently. In the validation cohort, the nomogram achieved an AUC of 0.77 (Figure 5A) and the calibration curve showed moderate agreement between the predicted values and the actual observations (Figure 5B). Decision curve analysis revealed that the nomogram provided greater net clinical benefits than several alternative strategies, including intervention for all and intervention for none (Figure 5C). The confusion matrix was analyzed using the cutoff value corresponding to the maximum Youden index from the training set. The results showed that the model performed well in both the training and validation cohorts (Figure 5D).

TABLE 4. Comparison of baseline characteristics between training and validation sets.

Variables	Training set (n = 200)	Validation set (n = 92)	p
Age, years (mean [SD])	53.46 [13.01]	55.38 [13.28]	0.245
Weight, kg (mean [SD])	73.60 [13.05]	72.51 [11.79]	0.496
Hight, cm (mean [SD])	169.25 [8.14]	168.18 [8.24]	0.299
Tumor size, mm (mean [SD])	31.15 [12.14]	33.12 [12.77]	0.207
HGB, g/L (median [IQR])	148.50 [136.00, 158.00]	143.00 [134.00, 154.00]	0.050
CR, μmol/L (median [IQR])	78.50 [67.00, 88.00]	76.00 [64.75, 88.00]	0.450
Gender, n [%]
Male	134 [67.00]	59 [64.13]	0.630
Female	66 [33.00]	33 [35.87]
Hypertension, n [%]
No	118 [59.00]	51 [55.43]	0.567
Yes	82 [41.00]	41 [44.57]
Diabetes, n [%]
No	167 [83.50]	64 [69.57]	0.007
Yes	33 [16.50]	28 [30.43]
Surgical history, n [%]
No	139 [69.50]	58 [63.04]	0.274
Yes	61 [30.50]	34 [36.96]
Tumor side
Left	102 [51.00]	44 [47.83]	0.614
Right	98 [49.00]	48 [52.17]
Pathological type, n [%]
ccRCC	165 [82.50]	66 [71.74]	0.036
nccRCC	35 [17.50]	26 [28.26]
ASA grade, n [%]
1	64 [32.00]	15 [16.30]	0.020
2	125 [62.50]	71 [77.17]
3	11 [5.50]	6 [6.52]

• Abbreviations: CR, creatinine; HGB, hemoglobin; IQR, interquartile range; SD, standard deviation.

3.5 Typical Cases of Patients Underwent the Two Approaches

Typical Case 1: A 59-year-old male presented with a right-sided renal mass during a routine health check-up. A contrast-enhanced CT indicated a mass of approximately 2.3 cm in diameter in the right kidney. The mass was located in the posterior and upper pole of the kidney (Figure 6A–C). According to our prediction model, the probability of RRPN was 0.85. The patient underwent RRPN (Figure 6D–G) with an arterial clamp time of 28 min, an operative time of 106 min, and an intraoperative blood loss of 5 mL. The drainage tube was removed and the patient was discharged 4 days after surgery.

Typical case 2: A 55-year-old female presented with a right-sided renal mass during a routine health checkup. Contrast-enhanced CT revealed a mass in the right kidney, with a diameter of about 4.6 cm. The mass was located at the anterior and lower pole of the kidney (Figure 7A–C). According to our prediction model, the probability of RRPN was 0.10. The patient underwent TRPN (Figure 7D–G) with an arterial clamping time of 25 min, an operation time of 110 min, and an intraoperative blood loss of 100 mL. The drainage tube was removed 2 days after surgery, and the postoperative hospitalization time was 6 days.

4 Discussion

The majority of the current research on RPN involves utilizing the transperitoneal approach [15]. This approach is favored by urologists due to its familiarity with the anatomy, adequate surgical space to avoid instrument collision [6, 16]. However, over the past decade, several reports of the retroperitoneal approach have been published, and most of them found that the perioperative outcomes were not worse than those of the transperitoneal approach [4, 7, 9-11]. Additionally, several surgeons have summarized the unique advantages of the retroperitoneal approach, such as providing better access to the posterior kidney for tumor visualization, resection, and renorrhaphy and minimizing the risk of injury to intraperitoneal organs [17, 18]. In the past, there were no clear guidelines or studies to inform surgical approach decisions. For this reason, we developed a prediction model to provide a reference for clinicians and serve as an exploratory tool for future prospective studies.

Before building a predictive model, we firstly compared the two approaches in terms of perioperative outcomes. The final results revealed no statistically significant differences for most of the variables, indicating that both approaches are comparable in terms of perioperative performance. However, we did observe significant differences in two parameters: EBL and HGB change, with the retroperitoneal approach generally showing lower blood loss and less change in hemoglobin. The findings of our study are consistent with the conclusions of a meta-analysis published by Gu et al. [19], which also highlighted similar perioperative outcomes between the two approaches, further supporting the validity of our results.

The fact that other researchers have found that RRPN can achieve shorter WIT [10] and a reduced rate of postoperative complications [11], which are inconsistent with our findings. Additionally, some surgeons have reported shorter operative times associated with RRPN, which may be attributable to a faster access to the renal artery or the simplicity of not freeing the colon [7, 13]. In addition, our center uses the video recording tool built into the robotic system to record robotic operations and track operation time. Therefore, the time spent on trocar placement and wound closure is not included in the recorded operative time, which may result in differences in operative time compared to other studies [14]. According to Archie et al. [7] and Gin et al. [20], the hospitalization time is significantly shorter for RRPN (2.5 vs. 4.6 days, 1.5 vs. 2 days, respectively). However, in our study, there is no difference between the two groups of patients in hospitalization time. Due to the retrospective nature of the study, we reviewed postoperative complications from the written medical records, which may have resulted in some complications being omitted. This may explain the discrepancies between our study and other published research [11, 14]. In the PADORA study by Dr. Mjaess et al. [14], using a large, multicenter prospective database, the differences between RRPN and TRPN were compared. The study found that the RRPN group had lower EBL and was more suitable for posterior renal tumors, consistent with our results. It is noteworthy that they also included variables such as exophytic proportion and cystic nature, though these factors did not show statistical differences, their inclusion reflects a comprehensive study design. Furthermore, Zhang et al. categorized tumors based on location (completely upper pole, completely lower pole) and complexity (R.E.N.A.L. score ≥ 7) to examine the impact of the surgical approach [11, 21-24], which inspired us to focus on tumors with specific characteristics separately in future research.

To the best of our knowledge, this is the first nomogram that predicts the surgical approach for RPN surgery. While previous studies have examined the decision of the surgical approach for laparoscopic PN, no clear preoperative determinants have been reported [23]. Liu et al. [12] researched the optimal surgical approach and concluded that the transperitoneal approach is more suitable for patients with anterior tumors and the retroperitoneal approach is more suitable for patients with posterior tumors. As can be observed, their study only took into account the tumor's anterior–posterior position. However, our model includes both demographic information and oncology characteristics.

The variables included in this model were selected using statistical methods. However, some clinically relevant variables that are often considered in actual clinical practice were not taken into account in the univariate analysis. All of the variables we included can be derived from preoperative CT scans without the need for more complex diagnostic procedures so that the nomogram can be adapted for less-resourced settings not only for use in advanced robotic surgery environments. Even more, its underlying principles can be promoted to laparoscopic surgery.

The retroperitoneal approach for posterior tumors is clear and reasonable, as it allows for direct access to the kidney's posterior surface and renal hilum while avoiding bowel manipulation and peritoneal violation [25]. Additionally, our model shows that the tumor's location in the upper pole facilitates the retroperitoneal approach, which is consistent with a previous study [11]. The transperitoneal approach for the upper pole of renal tumors will be covered by the liver and spleen, and the difficulty of separating and suturing will be greatly increased, and the risk of liver and spleen injury may be increased during the operation [11]. In regards to tumor size, our findings indicate that larger tumors decrease the probability of adopting the retroperitoneal approach. This outcome may be attributed to the bigger space in the peritoneal cavity, which allows adequate mobilization and exposure when dealing with large tumors and effectively prevents collision of the robotic arm. In addition, our study found that the presence of accessory renal artery is usually more suitable for the retroperitoneal approach, which may be related to anatomical considerations, vascular protection, and postoperative recovery. The accessory renal arteries are usually located at the renal hilum or behind the kidney [26]. The retroperitoneal approach usually provides better protection for the kidney and its blood vessels, especially when removing tumors or renal tissues. It allows the surgeon to see the course of the accessory renal artery more clearly and reduces the risks involved in the operation.

We added the R.E.N.A.L score, PADUA score, MAP score, and “3S + f” score to our data collection schedule, since we thought that the nephrometry score systems could be contributive to determining the surgical approach when this study was initially designed. However, subsequent analysis revealed that these scoring systems do not directly impact the decision of the surgical approach, although they may be able to evaluate the complexity of the tumor. Although the total score of the aforementioned scoring systems did not show a correlation with the surgical approach, certain subcategories of them were meaningful; interestingly, these subentries were parameters related to tumor size or location (Table 2). This finding indirectly supports the rationale for determining the surgical approach based on tumor characteristics.

Although PSM effectively equalized baseline characteristics between the TRPN and RRPN groups, a significant proportion of TRPN patients were not included in the matching analysis. To examine how these discordant cases might affect the results, we performed a detailed analysis of variable distribution in both groups PSM (Table 1 and Table 3). The results showed that, apart from the four variables included in the model, there were statistically significant differences in gender, preoperative creatinine, and ASA grade. These results suggest that female patients and patients with lower creatinine levels may be more suitable for the RRPN approach, while the ASA grade showed a bimodal distribution, with higher proportions of ASA grade 1 and grade 3 in the RRPN group compared to the TRPN group. To address this issue, we attempted to develop two models by directly using the training set data (without PSM): a simplified model (variables: anteroposterior location, longitudinal location, tumor size, and accessory renal artery) and a full model (simplified model + gender, creatinine + ASA grade). The results showed that the AUC of the full model improved compared to the simplified model, but only by 0.45% (Figure 1). Considering the trade-off between the increased complexity of the model and the small improvement in AUC, we thought that adding these additional variables did not provide significant benefit.

We acknowledge that there are some limitations in this study. Firstly, the data are from one single center and small sample size, so that the findings may not be generalizable to all surgeons. Additionally, this is a retrospective study, which may not be as robust as a randomized prospective trial. Only 87.5% (77/88) of the retroperitoneal cases and 53.5% (123/230) were matched in the PSM study, and we are unsure if there are any further unidentified or uncontrolled confounders, especially residual confounding factors related to the choice of the surgical approach may have a direct impact on the prognosis. Another crucial note is that we did not analyze the competence of the eight surgeons for heterogeneity or learning curve, which may have resulted in selection bias. Perhaps future studies will require a prospective design, randomized patient grouping, a larger sample size, and data from multiple centers to further validate our findings.

5 Conclusions

There are not significant differences in perioperative outcomes between the retroperitoneal and transperitoneal approaches. The retroperitoneal approach is recommended for younger patients, smaller tumors, tumors located in the upper pole, and tumors located posteriorly. Our predictive model that based on demographic information and tumor characteristics can serve as a reference for surgeon's clinical decision of the surgical approach.

Author Contributions

Fan Shu: conceptualization; methodology; investigation; writing – original draft. Zhuo Liu: investigation; writing – original draft; writing – review and editing. Peichen Duan: methodology; writing – review and editing. Yichang Hao: data curation; formal analysis. Xin Ma: data curation; formal analysis. Hongxian Zhang: data curation; formal analysis. Guoliang Wang: formal analysis; supervision. Xiaojun Tian: formal analysis; supervision. Lei Liu: formal analysis; supervision. Shudong Zhang: conceptualization; funding acquisition; writing – review and editing.

Ethics Statement

The studies involving human participants were reviewed and approved by the Institutional Ethics Committee of Peking University Third Hospital.

Consent

The patients/participants provided their written informed consent to participate in this study.

Conflicts of Interest

The authors declare no conflicts of interest.