Introduction
Radical nephrectomy (RN) and partial nephrectomy (PN) are both the most effective ways to treat localized renal masses. In addition to oncological outputs, clinicians also prioritize postoperative renal function (RF), with estimated glomerular filtration rate (eGFR) being the key indicator representing total RF. PN can better preserve the eGFR than RN [
1]. However, the beneficial impact of PN does not seem to translate into survival benefits unless the preoperative eGFR is already significantly impaired [
2‐
4]. Therefore, patients with high tumor complexity, no preexisting chronic kidney disease, or proteinuria are recommended to undergo RN rather than PN if the expected eGFR is greater than 45 ml/min/1.73 m
2 after RN [
5]. In these cases, the downside of RN appears to be inconsequential [
6], and PN is associated with an increased risk of perioperative morbidity and compromised oncologic outcomes. Moreover, clinicians should consider referring patients to nephrology if the expected eGFR is not satisfactory after intervention [
5]. All of these aspects require accurate predictions of postoperative eGFR.
Unfortunately, there are few models to predict postoperative eGFR. Information about preoperative split RF is common lacking [
7‐
10], which results in limited ability in predicting eGFR after RN [
11]. Other studies estimated split RF from images using parenchymal volume analysis [
11‐
15]. These methods require volume calculating software [
13] and training, and are not suitable for routine applications. Furthermore, these studies were limited by small sample sizes. Numerous medical institutions routinely use technetium-99 m-diethylenetriaminepentaacetic acid (Tc-99 m DTPA) renal dynamic imaging (RDI) to assess preoperative split glomerular filtration rate measured by radionuclide (rGFR). However, its role in predicting postoperative eGFR is still far from clear, and its application is based on superficial clinical experience.
Thus, this study aimed to develop preoperative tools for predicting postoperative eGFR based on split rGFR. Separate models predicting short-term and long-term eGFRs were constructed to cover almost all periods after surgery.
Methods
Study design and participants
In this retrospective cohort study, data on patients with a single, unilateral renal mass who underwent RN or PN at Xijing Hospital from August 2008 to March 2022 were collected. Patients with preoperative renal failure (eGFR < 15 ml/min/1.73 m2) were excluded.
We collected these variables as potential predictors of postoperative eGFR based on previous research, including age, gender, body mass index (BMI), tumor size, preoperative eGFR, rGFR of the operative side, rGFR of the healthy side, serum cystatin C, diabetes, hypertension, surgical approach (open, laparoscopic, or robot-assisted laparoscopic surgery), and time from surgery (TFS).
Follow-up and study outcomes
The single-kidney glomerular filtration rate (GFR) was routinely calculated by Tc-99 m DTPA RDI using the Gate’s method before surgery [
16]. During hospitalization, serum creatinine was rechecked within 7 days postoperatively. The planned follow-up included serum creatinine measurements at 3 months, 6 months, and 1 year postoperatively and then annually thereafter until 5 years.
The eGFRs were calculated according to the Chronic Kidney Disease Epidemiology Collaboration formula [
17]. The long-term models predicted the eGFRs at 3 months, 6 months, and 1–5 years postoperatively. The short-term models predicted the eGFRs at 1–7 days postoperatively.
Statistical analysis
The association between preoperative characteristics and postoperative eGFRs was evaluated using linear mixed-effect models with random patient-specific intercepts. Moreover, we introduced a temporal autocorrelation structure for each model [
18].
Multivariable models were developed using backward selection of all the preoperative factors and their two-way interactions with TFS, with the significance level set to 0.01.
P-values were provided using Satterthwaite's degrees of freedom method via the R package
lmerTest. Main effects were included whenever the interaction with TFS was retained in model selection [
7].
The correlations between the preoperative variables were investigated using the Spearman correlation test. In the process of performing the Spearman correlation test and establishing multivariable models, patients without complete data for all preoperative features were excluded.
Marginal and conditional
R2 were used to summarize the predictive ability of the models [
19], with the value of 0 representing no predictive ability and 1.0 representing perfect predictive ability. The marginal
R2 helped describe the ability of preoperative characteristics to predict eGFR in future patients. The conditional
R2 also included between- and within-patient variability and can be considered as the total variation that can be explained by the models.
Statistical analyses were performed using SPSS version 26.0 (SPSS Inc., Chicago, IL, USA) for Windows and R version 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria). The measurement data with a nonnormal distribution are presented as the median (interquartile range, IQR).
Discussion
The prediction of postoperative eGFR can help make informed choices between RN and PN, especially for complex renal tumors. Tc-99 m DTPA RDI has been widely used to measure split RF [
20]. However, how to use this important information about split rGFR to predict postoperative eGFR lacks computational models. Hence, we developed a pretreatment tool for predicting eGFR based on split rGFR.
For example, a 56-year-old man with a rGFR of 24 ml/min/1.73 m
2 on the operative side, a rGFR of 36 ml/min/1.73 m
2 on the healthy side, serum cystatin C of 0.91 mg/L, and a preoperative eGFR of 67 ml/min/1.73 m
2 had a history of hypertension and diabetes. If RN was performed, the predicted eGFRs were 50, 53 and 55 ml/min/1.73 m
2 at 2 years after open, laparoscopic and robotic surgery, respectively. If PN was performed, the predicted eGFRs were 67, 73 and 73 ml/min/1.73 m
2 at 2 years after open, laparoscopic and robotic surgery, respectively. To avoid stage 3 chronic kidney disease, PN would be the preferred treatment, especially laparoscopic and robotic PN. In addition, because the predicted eGFRs were greater than 45 ml/min/1.73 m
2 after RN, RN was also feasible according to American Urological Association guidelines [
5]. Thus, if clinicians only focus on the preoperative rGFR of 36 ml/min/1.73 m
2 on the healthy side, they will not be able to make an objective judgment of postoperative eGFR.
Split rGFR is at the core of the prediction models. In our study, higher rGFR of the healthy side and lower rGFR of the operative side meant higher eGFR after RN. Moreover, patients who underwent RN with higher rGFR on the healthy side also had more recovery of eGFR over time.
Previous study suggested that patients underwent laparoscopic PN maintained slightly higher eGFR than those underwent open PN, and researchers explained this phenomenon as the transient ischemia caused by pneumoperitoneum could limit renal reperfusion injury [
21]. Additionally, another study suggested that robotic PN allows better preservation of split eGFR than laparoscopic PN by increasing the parenchymal preservation ratio [
22]. Our models further demonstrated the differences in postoperative eGFR among these three surgical methods. For long-term eGFR after PN, laparoscopic and robotic surgery were superior to open surgery, possibly because of the above reasons. Interestingly, for long-term eGFR after RN, laparoscopic and robotic surgery were also superior to open surgery. We speculate that minimally invasive technology reduced overall trauma, which indirectly benefited the recovery of eGFR. The eGFR difference between laparoscopic RN and robotic RN decreased over time, indicating that robotic RN was more conducive to early recovery of RF. To our knowledge, few studies have reported the protective effect of minimally invasive technology on eGFR after RN. Importantly, this indicates that for patients with poor rGFR on the healthy side, if RN is chosen, minimally invasive surgery should also be given priority, as it is likely to change the outcome of renal insufficiency.
Generally, the merits of this study were clear. First, we used a large cohort of 2251 patients for long-term model and 2447 patients for short-term model. Based on this large cohort, our models can predict eGFR at many time points after surgery, not just at a single time point as in most studies. Second, we utilized the information of split rGFR, which enabled our models to have high prediction ability for the eGFRs after RN. In other prediction models not based on split RF, the prediction ability for RN was significantly lower than that for PN [
11].
However, this study has several limitations. Above all, the analysis is based on retrospective data from a single center, which can be associated with information bias and selection bias. Additionally, data on tumor complexity, such as the RENAL nephrometry score [
23], 24-h proteinuria [
24], co-existing comorbidities other than diabetes and hypertension, lifestyle factors as smoking or alcohol use, or medication usage were not included. These parameters may improve the prediction ability, but they will also greatly increase the difficulty of clinical usage. Finally, the accuracy of split rGFR measurements by Tc-99 m DTPA RDI is different in patients with different degrees of RF impairment, which generally underestimates GFR [
25]. Note that we use rGFR to predict eGFRs, the deviation between rGFR and actual single-kidney GFR does not affect the application of the models.
In conclusion, our study developed novel tools to accurately predict short-term and long-term eGFRs after RN or PN based on split rGFR, and that is helpful for preoperative decision-making and communication. A minimally invasive approach seem to improve postoperative eGFRs after both PN and RN, which needs more evidence to confirm. Although awaiting external validation, the results of this study may supplement the surgical guidelines for patients with renal tumors.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.