Urologic Oncology: Seminars and Original Investigations 33 (2015) 167.e1–167.e6
Original article
Critical appraisal of first-generation renal tumor complexity scoring systems: Creation of a second-generation model of tumor complexity Conrad M. Tobert, M.D.a,b,1, Allen Shoemaker, Ph.D.c, Richard J. Kahnoski, M.D.a,b, Brian R. Lane, M.D., Ph.D.a,b,* a
Urology, Michigan State University College of Human Medicine, Grand Rapids, MI b Urology, Spectrum Health Hospital System, Grand Rapids, MI c Department of Biostatistics, Grand Rapids Medical Education Partners, Grand Rapids, MI Received 8 July 2014; received in revised form 19 December 2014; accepted 29 December 2014
Abstract Objective: To investigate whether a combination of variables from each nephrometry system improves performance. There are 3 firstgeneration systems that quantify tumor complexity: R.E.N.A.L. nephrometry score (RNS), preoperative aspects and dimensions used for an anatomical (PADUA) classification (PC), and centrality index (CI). Although each has been subjected to validation and comparative analysis, to our knowledge, no work has been done to combine variables from each method to optimize their performance. Patients and methods: Scores were assigned to each of 276 patients undergoing partial nephrectomy (PN) or radical nephrectomy (RN). Individual components of all 3 systems were evaluated in multivariable logistic regression analysis of surgery type (PN vs. RN) and combined into a “second-generation model.” Results: In multivariable analysis, each scoring system was a significant predictor of PN vs. RN (P o 0.0001). Of the first-generation systems, CI was most highly correlated with surgery type (area under the curve [AUC] ¼ 0.91), followed by RNS (AUC ¼ 0.90) and PC (AUC ¼ 0.88). Each individual component of these scoring systems was also a predictor of surgery type (P o 0.0001). In a multivariable model incorporating each component individually, 4 were independent predictors of surgery type (each P o 0.005): tumor size (RNS and PC), nearness to the collecting system (RNS), location along the lateral rim (PC), and centrality (CI). A novel model in which these 4 variables were rescaled outperformed each first-generation system (AUC ¼ 0.91). Conclusions: Optimization of first-generation models of renal tumor complexity results in a novel scoring system, which strongly predicts surgery type. This second-generation model should aid comprehension, but future work is still needed to establish the most clinically useful model. r 2015 Elsevier Inc. All rights reserved.
Keywords: R.E.N.A.L. nephrometry score; PADUA classification; Centrality index
1. Introduction Partial nephrectomy (PN) has emerged as the gold standard for treating small renal masses that are amenable to such an approach [1]. The decision to undergo PN is based on multiple factors but relies heavily on the complexity of the tumor and the clinical gestalt of the surgeon. Funding was provided in part by the Spectrum Health Foundation. 1
Present address: Department of Urology, University of Iowa, Iowa City,
IA. Corresponding author. Tel.: þ1-616-267-9333; fax: þ1-616-267-8040. E-mail address: [email protected] (B.R. Lane). *
http://dx.doi.org/10.1016/j.urolonc.2014.12.016 1078-1439/r 2015 Elsevier Inc. All rights reserved.
Multiple systems have been developed to provide a systematic method to quantify tumor complexity. These systems were initially purported to be useful both in the research setting and for assessing tumors in clinical practice, but the extent that they are used in clinical practice remains unclear at present. R.E.N.A.L. nephrometry score (RNS), preoperative aspects and dimensions used for an anatomical (PADUA) classification (PC), and centrality index (CI) are firstgeneration scoring systems designed to provide a quantitative assessment of renal tumor complexity [2–4]. These tools were initially designed to enable comparisons of renal masses treated at various institutions and have been shown
167.e2
C.M. Tobert et al. / Urologic Oncology: Seminars and Original Investigations 33 (2015) 167.e1–167.e6
to have significant correlation with clinical practice patterns in both academic and community settings [5,6]. Each system has been individually verified as an acceptable model for predicting a range of variables that are relevant to practicing urologists, including the type of surgery performed (PN vs. radical nephrectomy [RN] or minimally invasive PN vs. open PN) renal tumor pathology, postoperative renal function, and several other outcomes specific to PN [7–15]. To our knowledge, there have been few studies that compared the 3 individual scoring methods in an effort to identify the most significant components of each system [12,16,17]. By analyzing the individual components of each scoring system, we provide a novel perspective on the firstgeneration models of complexity. We compared all 3 scoring methods on the same group of localized renal tumors treated at our institution. Based on these findings, we integrated the components of each model into a robust predictive model for surgery type. Based on these analyses, we then generated a second-generation complexity score using the most predictive variables.
2. Patients and methods 2.1. Cohort Institutional review board approval was received for the use of data maintained within our institutional kidney tumor registry. Patients who were younger than 18 years, had locally advanced or metastatic renal cell carcinoma at presentation, had multiple tumors, had a solitary kidney, and who underwent nephrectomy for upper tract urothelial carcinoma or other reasons were excluded from analysis. All patients undergoing a partial or RN for a suspected renal cortical tumor meeting the aforementioned inclusion criteria were included. The cohort included 276 consecutive surgeries by 5 surgeons at a single institution. Surgical management included RN in 151 patients (55%) and PN in 125 patients (45%). Pathologic tumor stage was T1a (66%), T1b (13%), T2 (9%), T3a (11%), and T3b (1%). RNS, PC, and CI scores were assigned based on retrospective review of preoperative cross-sectional imaging (computed tomography or magnetic resonance imaging)
for all cases that were included in accordance with published guidelines [2–4]. Scores were assigned for all the cases by a single investigator while blinded to details of the case, including surgeon and surgery type. RNS has 4 numeric components and 1 descriptive component [2]. PC has 6 numeric components, 2 of which are shared with RNS (size group and exophytic) [3]. CI has 1 numeric component, based on 2 components, which determines the distance of the center of the mass to the center of the kidney [4]. Quantitative variables are expressed as mean and standard deviation (SD). Area under the curve (AUC) is expressed as value and 95% CI. 2.2. Statistical analysis Statistical analysis was done on the cohort to determine the efficacy of each of the systems at predicting PN vs. RN. Multivariable analysis included the individual components of the 3 scoring systems and was performed to determine the independent predictors of PN vs. RN. Using stepwise assignment of rank, based on the chi-squared values for each of these variables, a novel scoring method was created and tested against the first-generation scoring systems. The DeLong method was used to analyze 4 systems to predict surgery type. The procedure “crossfold,” a STATA macro that performs k-fold cross-validation on a specific model, was used to evaluate the optimized model's ability to fit outof-sample data. All statistical analysis was done using JMP/ SAS version 9 and SPSS version 17. 3. Results Overall, 276 consecutive patients undergoing PN or RN for a localized renal tumor were included. Of them, 151 (55%) and 125 (45%) underwent PN and RN, respectively. Mean age was 61.1 years (SD ¼ 14.2), and 61% were male. Mean glomerular filtration rate was 73.6 (SD ¼ 24.9), and mean tumor size was 4.6 cm (SD ¼ 3.0). Complexity scores (RNS, PC, and CI) were calculated for each tumor, and each individual system was a strong predictor of PN vs. RN (Table 1). Mean RNS scores were 6.03 and 9.24 for PN and RN, respectively (P o 0.0001). Mean PC scores were 8.08 and 11.22 (P o 0.0001) and
Table 1 Complexity scores and tumor size for tumors treated with PN vs. RN according to 4 scoring systems Tumor size Mean ⫾ SD
RENAL Mean ⫾ SD
PADUA Mean ⫾ SD
C index Mean ⫾ SD
Optimized model Mean ⫾ SD
PN RN P value
2.86 ⫾ 1.44 6.29 ⫾ 3.25 o0.0001
6.03 ⫾ 1.59 9.24 ⫾ 1.54 o0.0001
8.08 ⫾ 1.57 11.22 ⫾ 1.74 o0.0001
3.40 ⫾ 1.85 1.13 ⫾ 0.93 o0.0001
5.78 ⫾ 1.44 8.54 ⫾ 1.43 o0.0001
AUC of model for PN vs. RN
0.85
0.90
0.88
0.91
0.91
Each score was strongly correlated with surgery type, as evidenced by P o 0.0001. Correlation was assessed by multivariable analysis accounting for age, sex, glomerular filtration rate (GFR), surgeon, and surgery year.
C.M. Tobert et al. / Urologic Oncology: Seminars and Original Investigations 33 (2015) 167.e1–167.e6
mean CI scores were 3.40 and 1.13 for PN and RN, respectively (P o 0.0001). CI was most strongly correlated with surgery type (AUC ¼ 0.91 ⫾ 0.034), followed by RNS (AUC ¼ 0.90 ⫾ 0.036) and PC (AUC ¼ 0.89 ⫾ 0.039) (Table 1). By comparison, tumor size alone was less strongly correlated with surgery type (AUC ¼ 0.85 ⫾ 0.044). We next evaluated whether each individual variable of the first-generation scoring systems was associated with surgery type in univariable and multivariable analyses (Table 2). Associations between preoperative glomerular filtration rate, surgeon, surgery year, tumor size, and each component of RNS, PC, and CI with surgery type were statistically significant (P o 0.0001). To create a “secondgeneration” scoring system, a multivariable analysis was performed with the significant variables from the previous analysis (Table 2). Only 4 predictors of complexity remained—size group (RNS and PC), nearness to the collecting system (RNS), location of the tumor relative to the rim of the kidney (PC), and the centrality variable (CI) —which were significant predictors (P o 0.005). These 4 predictors were then recalibrated for use in a “second-generation” model of tumor complexity. Based on the strength of correlation with surgery type, each variable was assigned a point value between 0 and 3 (Supplementary Table S1). For example, scoring for location relative to the rim of the kidney was changed to assign 0 points to a tumor on the lateral margin and 2 points for a tumor on the medial margin. Nearness to the collecting system was rescaled from a 3-point scale for distances 47, 4–7, and o4 mm to a system for which 3 points are assigned for a distance o7 mm and 0 points for a distance 47 mm. Tumor radius was rescaled to 0 points for o4 cm, 1 point for tumors 44 and o7 cm, and 2 points for tumors 47 cm. CI was Table 2 Univariable and multivariable analyses of factors associated with surgery type Univariable model
Multivariable model
χ2
P value
χ2
P value
0.81 0.11 0.03 o0.0001 o0.0001 o0.0001 o0.0001 o0.0001 o0.0001 o0.0001 o0.0001 o0.0001 o0.0001 o0.0001 o0.0001 o0.0001
0.83 0.59 0.02 25.7 14.45 – 18.01 0.04 11.03 0.75 0.06 7.90 0.04 0.35 0.04 19.52
0.36 0.44 0.87 o0.0001 0.0001 – 0.0002 0.84 0.0009 0.68 0.81 0.005 0.83 0.55 0.84 0.0001
Age 0.06 Sex 2.51 GFR 9.09 Surgery year 16.31 Surgeon 20.54 Tumor size, cm 114.02 R (RNS and PADUA) 83.12 E (RNS and PADUA) 25.97 N (RNS) 149.99 A (RNS) 19.19 L (RNS) 80.73 Rim (PADUA) 33.38 Longitudinal (PADUA) 33.68 Sinus (PADUA) 72.51 Collecting system (PADUA) 109.97 C index (c) 88.93 GFR ¼ glomerular filtration rate.
167.e3
rescaled to 0 points for a score 44, 1 point for a score of o4 and 43, 2 points for a score o3 and 42, and 3 points for a score o2. Mean complexity score was 5.78 (SD ¼ 1.44) and 8.54 (SD ¼ 1.43) for PN and RN, respectively (P o 0.0001). The distribution of complexity scores of the optimization model and the 3 first-generation models is demonstrated in Fig. 1. Based on the new model, 93% of low complexity tumors (score: 0–3) underwent PN, 31% of intermediate complexity tumors (score: 4–7) underwent PN, and 6% of high complexity tumors (score: 8–10) underwent PN (P o 0.0001). Overall, the predictive ability of the second-generation complexity score exceeded or was equivalent to that of each first-generation complexity scoring systems (Table 1). Correlation with surgery type, according to the AUC, was stronger or equivalent (AUC ¼ 0.91 ⫾ 0.035) than that observed with CI (AUC ¼ 0.91 ⫾ 0.034), RNS (AUC ¼ 0.90 ⫾ 0.036), and PC (AUC ¼ 0.89 ⫾ 0.039) as demonstrated in a decision curve analysis (Fig. 2). The optimized model (pseudo-R2 of 0.7289) was validated using 5-fold cross-validation, with pseudo-R2 values of 0.718, 0.714, 0.768, 0.604, and 0.680. With this novel model of complexity, each new variable was a strong predictor of surgery type. Additionally, the association of 3 of the 4 variables with surgery type was increased by using the new scoring system (Table 3). The restructuring of the individual variables strengthened the predictive capacity of the radius, laterality, and centrality with chi-squared vales of 22.48, 7.70, and 22.76, respectively (all P o 0.005), and maintained the predictive capacity of nearness to the collecting system (χ2 ¼ 19.19, P o 0.0001) while reducing the complexity of the variable from 3 levels (47, 4–7, and o4 mm) to 2 levels (47 and o7 mm).
4. Discussion Although developed only 5 years ago, a great deal of literature has been generated regarding multiple measures of renal tumor complexity [15,18–23]. Each of the 3 firstgeneration scoring systems (RNS, PC, and CI), individually, is a strong predictor of surgery type and perioperative and postoperative outcomes of PN [7–14]. Although direct comparisons of these systems have been made, the overall conclusion from these has been that no clear winner could be identified [12,16,17]. Moreover, use of these models in clinical practice appears rather limited. Although some groups have compared the ability of these systems to predict various outcomes of PN, none have combined the components of these systems in an attempt to improve their performance, particularly with respect to arguably the most important end point, that is, type of surgery (PN vs. RN). We believe there remain 2 potential areas for improvement in models of tumor complexity: increased predictive ability for the clinician and ease of understanding for the patient.
167.e4
C.M. Tobert et al. / Urologic Oncology: Seminars and Original Investigations 33 (2015) 167.e1–167.e6
Fig. 1. Distribution of scoring systems by surgery type: (A) optimized score, (B) R.E.N.A.L. score, (C) PADUA score, and (D) C-index score. (Color version of figure is available online.)
We herein developed and tested a new model of complexity that addresses both of these concerns. The current study adds to the literature first by evaluating the contribution of the individual components of RNS, PC, and CI to prediction of operative approach. The 4 factors demonstrating independent predictive ability were maintained in the new model: radius, nearness to the collecting system, location relative to the lateral rim, and centrality. Second, we calibrated each of these variables to optimize their association with the outcome of interest. Third, we fit the scoring to a 10-point scale for better understanding by clinicians and patients alike, so that a score of 5 indicates intermediate complexity and a score of 10 (out of 10) is the most complex. By rescaling to this metric, we anticipate improvement in clinical integration, although this could not be evaluated in the present study. This second-generation model of tumor complexity outperformed or was equivalent to CI, RNS, and PC with respect to the primary outcome: prediction of surgery type (AUC was 0.91 vs. 0.91, 0.90, and 0.88, respectively). Although this model may not be superior to first-generation models, it provides a unique perspective regarding the individual variables used in each model.
Other second-generation scoring systems have recently been developed by other investigators with the goal of improving upon the first-generation systems [19,21,24–26]. The diameter-axial-polarity (DAP) system was developed to improve upon limitations of RNS and CI, including variable interobserver agreement and complicated methodology [24]. Analysis of 299 tumors treated with PN revealed better correlation of DAP with volume and functional preservation than obtained with RNS or CI. Subsequent studies validated the association of DAP with warm ischemia and postoperative renal damage in patients following PN [18,22,27,28], but none have been able to ascertain whether it is a predictor of surgery type as only patients undergoing PN were included. Even more recently, the zonal NePhRO scoring system has been developed using a series of 166 tumors treated with open PN [25]. Although NePhRO was shown to better predict complications than RNS, RNS better predicted warm ischemia in this initial report. Assessment of the contact surface area of a tumor with adjacent renal parenchyma has been found to predict perioperative and renal functional outcomes in an initial study that included 162 patients [19]. The surgical approach renal ranking is yet
C.M. Tobert et al. / Urologic Oncology: Seminars and Original Investigations 33 (2015) 167.e1–167.e6
167.e5
The limitations of our study include those commonly related to a retrospective study, including selection bias and single-institution design. In addition, there was observed variability seen in type of surgery performed across the included surgeons and years of surgery. This variability was accounted for in multivariable analysis and, thus, could be viewed as a strength of the study. Additionally, although the second-generation model we have developed outperforms RNS, PC, and CI, it is somewhat more difficult to calculate, as it requires measurement of the variables of all 3 systems (including CI) to evaluate all of these variables. If it is to be used in its current form, only 4 factors would be scored: radius, nearness, laterality, and centrality, which would be slightly more time consuming than RNS or PC. In addition, the reliance on surgeon decision making as the main outcome under analysis may be viewed as a weaker end point than other patient-specific outcomes such as complications. Nevertheless, we view this study as hypothesis generating, encouraging the field to move toward evaluation and adaption of a second-generation complexity score that performs with greater operating characteristics and may be better received by the urologic community. Testing of such systems using larger, multi-institutional data sets should accelerate this process. Fig. 2. Receiver operating characteristic (ROC) curve of 4 complexity scores. Plotted on the y-axis is the true-positive rate (sensitivity) and on the x-axis is the false-positive rate (1—specificity). (Color version of figure is available online.)
another tool developed as a more practical and intuitive scoring system, which has been evaluated in 80 patients thus far [21]. None of these newer systems has been sufficiently compared with each of the first-generation systems at present to justify their use in lieu of the prior models and, like the earlier models, the general urologic community has not adapted them at the present time. A measure of tumor complexity should be discussed with patients during the initial treatment discussion and recorded in the medical record prospectively. Such information is vital for decision making on a case-by-case basis by individual clinicians to better communicate risk and understand their personal outcomes. We feel that it is time for a critical appraisal of each of the available systems by an expert panel with recommendation toward the use of a single, optimized model of tumor complexity for the next 5 years of investigation. Table 3 Comparison of original scoring vs. optimized scoring of components that independently predict surgery type Original scoring Optimized scoring Multivariable Multivariable χ2
P value
χ2
5. Conclusion Each of the first-generation models of renal tumor complexity (RNS, PC, and CI) is based on individual factors that are significant predictors of surgery type. Combination of the independent predictors of PN from each of these models, rescaled to a 10-point scale, results in a complexity score with a strong predictive capability. With such incremental improvements, newer models of tumor complexity may perform better in research studies and be more widely used in clinical practice. Acknowledgments The corresponding author would like to thank the Betz Family Endowment for Cancer Research for their support as well as Sabrina Noyes for administrative support and technical editing. Appendix A. Supporting Information Supplementary material cited in this article is available online at http://dx.doi.org/10.1016/j.urolonc.2014.12.016.
P value
Radius—max diameter (RNS and PC) 13.10 o0.0001 22.48 o0.0001 Nearness to collecting system (RNS) 26.94 o0.0001 19.19 o0.0001 Location relative to lateral rim (PC) 6.27 0.01 7.70 0.005 C score (CI) 18.95 o0.0001 22.76 o0.0001
References [1] Campbell SC, Novick AC, Belldegrun A, et al. Guideline for management of the clinical T1 renal mass. J Urol 2009;182:1271–9.
167.e6
C.M. Tobert et al. / Urologic Oncology: Seminars and Original Investigations 33 (2015) 167.e1–167.e6
[2] Kutikov A, Uzzo RG. The R.E.N.A.L. nephrometry score: a comprehensive standardized system for quantitating renal tumor size, location and depth. J Urol 2009;182:844–53. [3] Ficarra V, Novara G, Secco S, et al. Preoperative aspects and dimensions used for an anatomical (PADUA) classification of renal tumours in patients who are candidates for nephron-sparing surgery. Eur Urol 2009;56:786–93. [4] Simmons MN, Ching CB, Samplaski MK, Park CH, Gill IS. Kidney tumor location measurement using the C index method. J Urol 2010;183:1708–13. [5] Dulabon LM, Lowrance WT, Russo P, Huang WC. Trends in renal tumor surgery delivery within the United States. Cancer 2010; 116:2316–21. [6] Miller DC, Hollingsworth JM, Hafez KS, Daignault S, Hollenbeck BK. Partial nephrectomy for small renal masses: an emerging quality of care concern? J Urol 2006;175:853–7:[discussion 858]. [7] Hew MN, Baseskioglu B, Barwari K, et al. Critical appraisal of the PADUA classification and assessment of the R.E.N.A.L. nephrometry score in patients undergoing partial nephrectomy. J Urol 2011;186:42–6. [8] Kolla SB, Spiess PE, Sexton WJ. Interobserver reliability of the RENAL nephrometry scoring system. Urology 2011;78:592–4. [9] Kutikov A, Smaldone MC, Egleston BL, et al. Anatomic features of enhancing renal masses predict malignant and high-grade pathology: a preoperative nomogram using the RENAL nephrometry score. Eur Urol 2011;60:241–8. [10] Simmons MN, Hillyer SP, Lee BH, Fergany AF, Kaouk J, Campbell SC. Nephrometry score is associated with volume loss and functional recovery after partial nephrectomy. J Urol 2012;188:39–44. [11] Samplaski MK, Hernandez A, Gill IS, Simmons MN. C-index is associated with functional outcomes after laparoscopic partial nephrectomy. J Urol 2010;184:2259–63. [12] Okhunov Z, Rais-Bahrami S, George AK, et al. The comparison of three renal tumor scoring systems: C-index, P.A.D.U.A., and R.E.N. A.L. nephrometry scores. J Endourol 2011;25:1921–4. [13] Bylund JR, Gayheart D, Fleming T, et al. Association of tumor size, location, R.E.N.A.L., PADUA and centrality index score with perioperative outcomes and postoperative renal function. J Urol 2012;188:1684–9. [14] Weight CJ, Atwell TD, Fazzio RT, et al. A multidisciplinary evaluation of inter-reviewer agreement of the nephrometry score and the prediction of long-term outcomes. J Urol 2011;186:1223–8. [15] Canter D, Kutikov A, Manley B, et al. Utility of the R.E.N.A.L. nephrometry scoring system in objectifying treatment decisionmaking of the enhancing renal mass. Urology 2011;78:1089–94. [16] Kobayashi K, Saito T, Kitamura Y, et al. The RENAL nephrometry score and the PADUA classification for the prediction of
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
perioperative outcomes in patients receiving nephron-sparing surgery: feasible tools to predict intraoperative conversion to nephrectomy. Urol Int 2013;91:261–8. Yeon JS, Son SJ, Lee YJ, et al. The nephrometry score: is it effective for predicting perioperative outcome during robot-assisted partial nephrectomy? Korean J Urol 2014;55:254–9. Altunrende F, Laydner H, Hernandez AV, et al. Correlation of the RENAL nephrometry score with warm ischemia time after robotic partial nephrectomy. World J Urol 2013;31:1165–9. Leslie S, Gill IS, de Castro Abreu AL, et al. Renal tumor contact surface area: a novel parameter for predicting complexity and outcomes of partial nephrectomy. Eur Urol 2014;S0302-2838: 00246-2. Monn MF, Gellhaus PT, Masterson TA, et al. R.E.N.A.L. nephrometry scoring: how well correlated are urologist, radiologist, and collaborator scores? J Endourol 2014;28:1006–10. Tannus M, Goldman SM, Andreoni C. Practical and intuitive surgical approach renal ranking to predict outcomes in the management of renal tumors: a novel score tool. J Endourol 2014;28:487–492. Tomaszewski JJ, Smaldone MC, Mehrazin R, et al. Anatomic complexity quantitated by nephrometry score is associated with prolonged warm ischemia time during robotic partial nephrectomy. Urology 2014;84:340–4. Miyake H, Furukawa J, Hinata N, Muramaki M, Tanaka K, Fujisawa M. Significant impact of R.E.N.A.L. nephrometry score on changes in postoperative renal function early after robot-assisted partial nephrectomy. Int J Clin Oncol 2014: [in press]. Simmons MN, Hillyer SP, Lee BH, Fergany AF, Kaouk J, Campbell SC. Diameter-axial-polar nephrometry: integration and optimization of R.E.N.A.L. and centrality index scoring systems. J Urol 2012; 188:384–90. Hakky TS, Baumgarten AS, Allen B, et al. Zonal NePhRO scoring system: a superior renal tumor complexity classification model. Clin Genitourin Cancer 2014;12:e13–8. Nisen H, Ruutu M, Glucker E, Visapaa H, Taari K. Renal tumour invasion index as a novel anatomical classification predicting urological complications after partial nephrectomy. Scand J Urol 2014;48:41–51. Wang L, Li M, Chen W, et al. Is diameter-axial-polar scoring predictive of renal functional damage in patients undergoing partial nephrectomy? An evaluation using technetium Tc 99m (99Tcm) diethylene-triamine-penta-acetic acid (DTPA) glomerular filtration rate. BJU Int 2013;111:1191–8. Maeda M, Funahashi Y, Sassa N, Fujita T, Majima T, Gotoh M. Prediction of partial nephrectomy outcomes by using the diameteraxis-polar nephrometry score. Int J Urol 2014;21:442–6.
Original article
Critical appraisal of first-generation renal tumor complexity scoring systems: Creation of a second-generation model of tumor complexity Conrad M. Tobert, M.D.a,b,1, Allen Shoemaker, Ph.D.c, Richard J. Kahnoski, M.D.a,b, Brian R. Lane, M.D., Ph.D.a,b,* a
Urology, Michigan State University College of Human Medicine, Grand Rapids, MI b Urology, Spectrum Health Hospital System, Grand Rapids, MI c Department of Biostatistics, Grand Rapids Medical Education Partners, Grand Rapids, MI Received 8 July 2014; received in revised form 19 December 2014; accepted 29 December 2014
Abstract Objective: To investigate whether a combination of variables from each nephrometry system improves performance. There are 3 firstgeneration systems that quantify tumor complexity: R.E.N.A.L. nephrometry score (RNS), preoperative aspects and dimensions used for an anatomical (PADUA) classification (PC), and centrality index (CI). Although each has been subjected to validation and comparative analysis, to our knowledge, no work has been done to combine variables from each method to optimize their performance. Patients and methods: Scores were assigned to each of 276 patients undergoing partial nephrectomy (PN) or radical nephrectomy (RN). Individual components of all 3 systems were evaluated in multivariable logistic regression analysis of surgery type (PN vs. RN) and combined into a “second-generation model.” Results: In multivariable analysis, each scoring system was a significant predictor of PN vs. RN (P o 0.0001). Of the first-generation systems, CI was most highly correlated with surgery type (area under the curve [AUC] ¼ 0.91), followed by RNS (AUC ¼ 0.90) and PC (AUC ¼ 0.88). Each individual component of these scoring systems was also a predictor of surgery type (P o 0.0001). In a multivariable model incorporating each component individually, 4 were independent predictors of surgery type (each P o 0.005): tumor size (RNS and PC), nearness to the collecting system (RNS), location along the lateral rim (PC), and centrality (CI). A novel model in which these 4 variables were rescaled outperformed each first-generation system (AUC ¼ 0.91). Conclusions: Optimization of first-generation models of renal tumor complexity results in a novel scoring system, which strongly predicts surgery type. This second-generation model should aid comprehension, but future work is still needed to establish the most clinically useful model. r 2015 Elsevier Inc. All rights reserved.
Keywords: R.E.N.A.L. nephrometry score; PADUA classification; Centrality index
1. Introduction Partial nephrectomy (PN) has emerged as the gold standard for treating small renal masses that are amenable to such an approach [1]. The decision to undergo PN is based on multiple factors but relies heavily on the complexity of the tumor and the clinical gestalt of the surgeon. Funding was provided in part by the Spectrum Health Foundation. 1
Present address: Department of Urology, University of Iowa, Iowa City,
IA. Corresponding author. Tel.: þ1-616-267-9333; fax: þ1-616-267-8040. E-mail address: [email protected] (B.R. Lane). *
http://dx.doi.org/10.1016/j.urolonc.2014.12.016 1078-1439/r 2015 Elsevier Inc. All rights reserved.
Multiple systems have been developed to provide a systematic method to quantify tumor complexity. These systems were initially purported to be useful both in the research setting and for assessing tumors in clinical practice, but the extent that they are used in clinical practice remains unclear at present. R.E.N.A.L. nephrometry score (RNS), preoperative aspects and dimensions used for an anatomical (PADUA) classification (PC), and centrality index (CI) are firstgeneration scoring systems designed to provide a quantitative assessment of renal tumor complexity [2–4]. These tools were initially designed to enable comparisons of renal masses treated at various institutions and have been shown
167.e2
C.M. Tobert et al. / Urologic Oncology: Seminars and Original Investigations 33 (2015) 167.e1–167.e6
to have significant correlation with clinical practice patterns in both academic and community settings [5,6]. Each system has been individually verified as an acceptable model for predicting a range of variables that are relevant to practicing urologists, including the type of surgery performed (PN vs. radical nephrectomy [RN] or minimally invasive PN vs. open PN) renal tumor pathology, postoperative renal function, and several other outcomes specific to PN [7–15]. To our knowledge, there have been few studies that compared the 3 individual scoring methods in an effort to identify the most significant components of each system [12,16,17]. By analyzing the individual components of each scoring system, we provide a novel perspective on the firstgeneration models of complexity. We compared all 3 scoring methods on the same group of localized renal tumors treated at our institution. Based on these findings, we integrated the components of each model into a robust predictive model for surgery type. Based on these analyses, we then generated a second-generation complexity score using the most predictive variables.
2. Patients and methods 2.1. Cohort Institutional review board approval was received for the use of data maintained within our institutional kidney tumor registry. Patients who were younger than 18 years, had locally advanced or metastatic renal cell carcinoma at presentation, had multiple tumors, had a solitary kidney, and who underwent nephrectomy for upper tract urothelial carcinoma or other reasons were excluded from analysis. All patients undergoing a partial or RN for a suspected renal cortical tumor meeting the aforementioned inclusion criteria were included. The cohort included 276 consecutive surgeries by 5 surgeons at a single institution. Surgical management included RN in 151 patients (55%) and PN in 125 patients (45%). Pathologic tumor stage was T1a (66%), T1b (13%), T2 (9%), T3a (11%), and T3b (1%). RNS, PC, and CI scores were assigned based on retrospective review of preoperative cross-sectional imaging (computed tomography or magnetic resonance imaging)
for all cases that were included in accordance with published guidelines [2–4]. Scores were assigned for all the cases by a single investigator while blinded to details of the case, including surgeon and surgery type. RNS has 4 numeric components and 1 descriptive component [2]. PC has 6 numeric components, 2 of which are shared with RNS (size group and exophytic) [3]. CI has 1 numeric component, based on 2 components, which determines the distance of the center of the mass to the center of the kidney [4]. Quantitative variables are expressed as mean and standard deviation (SD). Area under the curve (AUC) is expressed as value and 95% CI. 2.2. Statistical analysis Statistical analysis was done on the cohort to determine the efficacy of each of the systems at predicting PN vs. RN. Multivariable analysis included the individual components of the 3 scoring systems and was performed to determine the independent predictors of PN vs. RN. Using stepwise assignment of rank, based on the chi-squared values for each of these variables, a novel scoring method was created and tested against the first-generation scoring systems. The DeLong method was used to analyze 4 systems to predict surgery type. The procedure “crossfold,” a STATA macro that performs k-fold cross-validation on a specific model, was used to evaluate the optimized model's ability to fit outof-sample data. All statistical analysis was done using JMP/ SAS version 9 and SPSS version 17. 3. Results Overall, 276 consecutive patients undergoing PN or RN for a localized renal tumor were included. Of them, 151 (55%) and 125 (45%) underwent PN and RN, respectively. Mean age was 61.1 years (SD ¼ 14.2), and 61% were male. Mean glomerular filtration rate was 73.6 (SD ¼ 24.9), and mean tumor size was 4.6 cm (SD ¼ 3.0). Complexity scores (RNS, PC, and CI) were calculated for each tumor, and each individual system was a strong predictor of PN vs. RN (Table 1). Mean RNS scores were 6.03 and 9.24 for PN and RN, respectively (P o 0.0001). Mean PC scores were 8.08 and 11.22 (P o 0.0001) and
Table 1 Complexity scores and tumor size for tumors treated with PN vs. RN according to 4 scoring systems Tumor size Mean ⫾ SD
RENAL Mean ⫾ SD
PADUA Mean ⫾ SD
C index Mean ⫾ SD
Optimized model Mean ⫾ SD
PN RN P value
2.86 ⫾ 1.44 6.29 ⫾ 3.25 o0.0001
6.03 ⫾ 1.59 9.24 ⫾ 1.54 o0.0001
8.08 ⫾ 1.57 11.22 ⫾ 1.74 o0.0001
3.40 ⫾ 1.85 1.13 ⫾ 0.93 o0.0001
5.78 ⫾ 1.44 8.54 ⫾ 1.43 o0.0001
AUC of model for PN vs. RN
0.85
0.90
0.88
0.91
0.91
Each score was strongly correlated with surgery type, as evidenced by P o 0.0001. Correlation was assessed by multivariable analysis accounting for age, sex, glomerular filtration rate (GFR), surgeon, and surgery year.
C.M. Tobert et al. / Urologic Oncology: Seminars and Original Investigations 33 (2015) 167.e1–167.e6
mean CI scores were 3.40 and 1.13 for PN and RN, respectively (P o 0.0001). CI was most strongly correlated with surgery type (AUC ¼ 0.91 ⫾ 0.034), followed by RNS (AUC ¼ 0.90 ⫾ 0.036) and PC (AUC ¼ 0.89 ⫾ 0.039) (Table 1). By comparison, tumor size alone was less strongly correlated with surgery type (AUC ¼ 0.85 ⫾ 0.044). We next evaluated whether each individual variable of the first-generation scoring systems was associated with surgery type in univariable and multivariable analyses (Table 2). Associations between preoperative glomerular filtration rate, surgeon, surgery year, tumor size, and each component of RNS, PC, and CI with surgery type were statistically significant (P o 0.0001). To create a “secondgeneration” scoring system, a multivariable analysis was performed with the significant variables from the previous analysis (Table 2). Only 4 predictors of complexity remained—size group (RNS and PC), nearness to the collecting system (RNS), location of the tumor relative to the rim of the kidney (PC), and the centrality variable (CI) —which were significant predictors (P o 0.005). These 4 predictors were then recalibrated for use in a “second-generation” model of tumor complexity. Based on the strength of correlation with surgery type, each variable was assigned a point value between 0 and 3 (Supplementary Table S1). For example, scoring for location relative to the rim of the kidney was changed to assign 0 points to a tumor on the lateral margin and 2 points for a tumor on the medial margin. Nearness to the collecting system was rescaled from a 3-point scale for distances 47, 4–7, and o4 mm to a system for which 3 points are assigned for a distance o7 mm and 0 points for a distance 47 mm. Tumor radius was rescaled to 0 points for o4 cm, 1 point for tumors 44 and o7 cm, and 2 points for tumors 47 cm. CI was Table 2 Univariable and multivariable analyses of factors associated with surgery type Univariable model
Multivariable model
χ2
P value
χ2
P value
0.81 0.11 0.03 o0.0001 o0.0001 o0.0001 o0.0001 o0.0001 o0.0001 o0.0001 o0.0001 o0.0001 o0.0001 o0.0001 o0.0001 o0.0001
0.83 0.59 0.02 25.7 14.45 – 18.01 0.04 11.03 0.75 0.06 7.90 0.04 0.35 0.04 19.52
0.36 0.44 0.87 o0.0001 0.0001 – 0.0002 0.84 0.0009 0.68 0.81 0.005 0.83 0.55 0.84 0.0001
Age 0.06 Sex 2.51 GFR 9.09 Surgery year 16.31 Surgeon 20.54 Tumor size, cm 114.02 R (RNS and PADUA) 83.12 E (RNS and PADUA) 25.97 N (RNS) 149.99 A (RNS) 19.19 L (RNS) 80.73 Rim (PADUA) 33.38 Longitudinal (PADUA) 33.68 Sinus (PADUA) 72.51 Collecting system (PADUA) 109.97 C index (c) 88.93 GFR ¼ glomerular filtration rate.
167.e3
rescaled to 0 points for a score 44, 1 point for a score of o4 and 43, 2 points for a score o3 and 42, and 3 points for a score o2. Mean complexity score was 5.78 (SD ¼ 1.44) and 8.54 (SD ¼ 1.43) for PN and RN, respectively (P o 0.0001). The distribution of complexity scores of the optimization model and the 3 first-generation models is demonstrated in Fig. 1. Based on the new model, 93% of low complexity tumors (score: 0–3) underwent PN, 31% of intermediate complexity tumors (score: 4–7) underwent PN, and 6% of high complexity tumors (score: 8–10) underwent PN (P o 0.0001). Overall, the predictive ability of the second-generation complexity score exceeded or was equivalent to that of each first-generation complexity scoring systems (Table 1). Correlation with surgery type, according to the AUC, was stronger or equivalent (AUC ¼ 0.91 ⫾ 0.035) than that observed with CI (AUC ¼ 0.91 ⫾ 0.034), RNS (AUC ¼ 0.90 ⫾ 0.036), and PC (AUC ¼ 0.89 ⫾ 0.039) as demonstrated in a decision curve analysis (Fig. 2). The optimized model (pseudo-R2 of 0.7289) was validated using 5-fold cross-validation, with pseudo-R2 values of 0.718, 0.714, 0.768, 0.604, and 0.680. With this novel model of complexity, each new variable was a strong predictor of surgery type. Additionally, the association of 3 of the 4 variables with surgery type was increased by using the new scoring system (Table 3). The restructuring of the individual variables strengthened the predictive capacity of the radius, laterality, and centrality with chi-squared vales of 22.48, 7.70, and 22.76, respectively (all P o 0.005), and maintained the predictive capacity of nearness to the collecting system (χ2 ¼ 19.19, P o 0.0001) while reducing the complexity of the variable from 3 levels (47, 4–7, and o4 mm) to 2 levels (47 and o7 mm).
4. Discussion Although developed only 5 years ago, a great deal of literature has been generated regarding multiple measures of renal tumor complexity [15,18–23]. Each of the 3 firstgeneration scoring systems (RNS, PC, and CI), individually, is a strong predictor of surgery type and perioperative and postoperative outcomes of PN [7–14]. Although direct comparisons of these systems have been made, the overall conclusion from these has been that no clear winner could be identified [12,16,17]. Moreover, use of these models in clinical practice appears rather limited. Although some groups have compared the ability of these systems to predict various outcomes of PN, none have combined the components of these systems in an attempt to improve their performance, particularly with respect to arguably the most important end point, that is, type of surgery (PN vs. RN). We believe there remain 2 potential areas for improvement in models of tumor complexity: increased predictive ability for the clinician and ease of understanding for the patient.
167.e4
C.M. Tobert et al. / Urologic Oncology: Seminars and Original Investigations 33 (2015) 167.e1–167.e6
Fig. 1. Distribution of scoring systems by surgery type: (A) optimized score, (B) R.E.N.A.L. score, (C) PADUA score, and (D) C-index score. (Color version of figure is available online.)
We herein developed and tested a new model of complexity that addresses both of these concerns. The current study adds to the literature first by evaluating the contribution of the individual components of RNS, PC, and CI to prediction of operative approach. The 4 factors demonstrating independent predictive ability were maintained in the new model: radius, nearness to the collecting system, location relative to the lateral rim, and centrality. Second, we calibrated each of these variables to optimize their association with the outcome of interest. Third, we fit the scoring to a 10-point scale for better understanding by clinicians and patients alike, so that a score of 5 indicates intermediate complexity and a score of 10 (out of 10) is the most complex. By rescaling to this metric, we anticipate improvement in clinical integration, although this could not be evaluated in the present study. This second-generation model of tumor complexity outperformed or was equivalent to CI, RNS, and PC with respect to the primary outcome: prediction of surgery type (AUC was 0.91 vs. 0.91, 0.90, and 0.88, respectively). Although this model may not be superior to first-generation models, it provides a unique perspective regarding the individual variables used in each model.
Other second-generation scoring systems have recently been developed by other investigators with the goal of improving upon the first-generation systems [19,21,24–26]. The diameter-axial-polarity (DAP) system was developed to improve upon limitations of RNS and CI, including variable interobserver agreement and complicated methodology [24]. Analysis of 299 tumors treated with PN revealed better correlation of DAP with volume and functional preservation than obtained with RNS or CI. Subsequent studies validated the association of DAP with warm ischemia and postoperative renal damage in patients following PN [18,22,27,28], but none have been able to ascertain whether it is a predictor of surgery type as only patients undergoing PN were included. Even more recently, the zonal NePhRO scoring system has been developed using a series of 166 tumors treated with open PN [25]. Although NePhRO was shown to better predict complications than RNS, RNS better predicted warm ischemia in this initial report. Assessment of the contact surface area of a tumor with adjacent renal parenchyma has been found to predict perioperative and renal functional outcomes in an initial study that included 162 patients [19]. The surgical approach renal ranking is yet
C.M. Tobert et al. / Urologic Oncology: Seminars and Original Investigations 33 (2015) 167.e1–167.e6
167.e5
The limitations of our study include those commonly related to a retrospective study, including selection bias and single-institution design. In addition, there was observed variability seen in type of surgery performed across the included surgeons and years of surgery. This variability was accounted for in multivariable analysis and, thus, could be viewed as a strength of the study. Additionally, although the second-generation model we have developed outperforms RNS, PC, and CI, it is somewhat more difficult to calculate, as it requires measurement of the variables of all 3 systems (including CI) to evaluate all of these variables. If it is to be used in its current form, only 4 factors would be scored: radius, nearness, laterality, and centrality, which would be slightly more time consuming than RNS or PC. In addition, the reliance on surgeon decision making as the main outcome under analysis may be viewed as a weaker end point than other patient-specific outcomes such as complications. Nevertheless, we view this study as hypothesis generating, encouraging the field to move toward evaluation and adaption of a second-generation complexity score that performs with greater operating characteristics and may be better received by the urologic community. Testing of such systems using larger, multi-institutional data sets should accelerate this process. Fig. 2. Receiver operating characteristic (ROC) curve of 4 complexity scores. Plotted on the y-axis is the true-positive rate (sensitivity) and on the x-axis is the false-positive rate (1—specificity). (Color version of figure is available online.)
another tool developed as a more practical and intuitive scoring system, which has been evaluated in 80 patients thus far [21]. None of these newer systems has been sufficiently compared with each of the first-generation systems at present to justify their use in lieu of the prior models and, like the earlier models, the general urologic community has not adapted them at the present time. A measure of tumor complexity should be discussed with patients during the initial treatment discussion and recorded in the medical record prospectively. Such information is vital for decision making on a case-by-case basis by individual clinicians to better communicate risk and understand their personal outcomes. We feel that it is time for a critical appraisal of each of the available systems by an expert panel with recommendation toward the use of a single, optimized model of tumor complexity for the next 5 years of investigation. Table 3 Comparison of original scoring vs. optimized scoring of components that independently predict surgery type Original scoring Optimized scoring Multivariable Multivariable χ2
P value
χ2
5. Conclusion Each of the first-generation models of renal tumor complexity (RNS, PC, and CI) is based on individual factors that are significant predictors of surgery type. Combination of the independent predictors of PN from each of these models, rescaled to a 10-point scale, results in a complexity score with a strong predictive capability. With such incremental improvements, newer models of tumor complexity may perform better in research studies and be more widely used in clinical practice. Acknowledgments The corresponding author would like to thank the Betz Family Endowment for Cancer Research for their support as well as Sabrina Noyes for administrative support and technical editing. Appendix A. Supporting Information Supplementary material cited in this article is available online at http://dx.doi.org/10.1016/j.urolonc.2014.12.016.
P value
Radius—max diameter (RNS and PC) 13.10 o0.0001 22.48 o0.0001 Nearness to collecting system (RNS) 26.94 o0.0001 19.19 o0.0001 Location relative to lateral rim (PC) 6.27 0.01 7.70 0.005 C score (CI) 18.95 o0.0001 22.76 o0.0001
References [1] Campbell SC, Novick AC, Belldegrun A, et al. Guideline for management of the clinical T1 renal mass. J Urol 2009;182:1271–9.
167.e6
C.M. Tobert et al. / Urologic Oncology: Seminars and Original Investigations 33 (2015) 167.e1–167.e6
[2] Kutikov A, Uzzo RG. The R.E.N.A.L. nephrometry score: a comprehensive standardized system for quantitating renal tumor size, location and depth. J Urol 2009;182:844–53. [3] Ficarra V, Novara G, Secco S, et al. Preoperative aspects and dimensions used for an anatomical (PADUA) classification of renal tumours in patients who are candidates for nephron-sparing surgery. Eur Urol 2009;56:786–93. [4] Simmons MN, Ching CB, Samplaski MK, Park CH, Gill IS. Kidney tumor location measurement using the C index method. J Urol 2010;183:1708–13. [5] Dulabon LM, Lowrance WT, Russo P, Huang WC. Trends in renal tumor surgery delivery within the United States. Cancer 2010; 116:2316–21. [6] Miller DC, Hollingsworth JM, Hafez KS, Daignault S, Hollenbeck BK. Partial nephrectomy for small renal masses: an emerging quality of care concern? J Urol 2006;175:853–7:[discussion 858]. [7] Hew MN, Baseskioglu B, Barwari K, et al. Critical appraisal of the PADUA classification and assessment of the R.E.N.A.L. nephrometry score in patients undergoing partial nephrectomy. J Urol 2011;186:42–6. [8] Kolla SB, Spiess PE, Sexton WJ. Interobserver reliability of the RENAL nephrometry scoring system. Urology 2011;78:592–4. [9] Kutikov A, Smaldone MC, Egleston BL, et al. Anatomic features of enhancing renal masses predict malignant and high-grade pathology: a preoperative nomogram using the RENAL nephrometry score. Eur Urol 2011;60:241–8. [10] Simmons MN, Hillyer SP, Lee BH, Fergany AF, Kaouk J, Campbell SC. Nephrometry score is associated with volume loss and functional recovery after partial nephrectomy. J Urol 2012;188:39–44. [11] Samplaski MK, Hernandez A, Gill IS, Simmons MN. C-index is associated with functional outcomes after laparoscopic partial nephrectomy. J Urol 2010;184:2259–63. [12] Okhunov Z, Rais-Bahrami S, George AK, et al. The comparison of three renal tumor scoring systems: C-index, P.A.D.U.A., and R.E.N. A.L. nephrometry scores. J Endourol 2011;25:1921–4. [13] Bylund JR, Gayheart D, Fleming T, et al. Association of tumor size, location, R.E.N.A.L., PADUA and centrality index score with perioperative outcomes and postoperative renal function. J Urol 2012;188:1684–9. [14] Weight CJ, Atwell TD, Fazzio RT, et al. A multidisciplinary evaluation of inter-reviewer agreement of the nephrometry score and the prediction of long-term outcomes. J Urol 2011;186:1223–8. [15] Canter D, Kutikov A, Manley B, et al. Utility of the R.E.N.A.L. nephrometry scoring system in objectifying treatment decisionmaking of the enhancing renal mass. Urology 2011;78:1089–94. [16] Kobayashi K, Saito T, Kitamura Y, et al. The RENAL nephrometry score and the PADUA classification for the prediction of
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
perioperative outcomes in patients receiving nephron-sparing surgery: feasible tools to predict intraoperative conversion to nephrectomy. Urol Int 2013;91:261–8. Yeon JS, Son SJ, Lee YJ, et al. The nephrometry score: is it effective for predicting perioperative outcome during robot-assisted partial nephrectomy? Korean J Urol 2014;55:254–9. Altunrende F, Laydner H, Hernandez AV, et al. Correlation of the RENAL nephrometry score with warm ischemia time after robotic partial nephrectomy. World J Urol 2013;31:1165–9. Leslie S, Gill IS, de Castro Abreu AL, et al. Renal tumor contact surface area: a novel parameter for predicting complexity and outcomes of partial nephrectomy. Eur Urol 2014;S0302-2838: 00246-2. Monn MF, Gellhaus PT, Masterson TA, et al. R.E.N.A.L. nephrometry scoring: how well correlated are urologist, radiologist, and collaborator scores? J Endourol 2014;28:1006–10. Tannus M, Goldman SM, Andreoni C. Practical and intuitive surgical approach renal ranking to predict outcomes in the management of renal tumors: a novel score tool. J Endourol 2014;28:487–492. Tomaszewski JJ, Smaldone MC, Mehrazin R, et al. Anatomic complexity quantitated by nephrometry score is associated with prolonged warm ischemia time during robotic partial nephrectomy. Urology 2014;84:340–4. Miyake H, Furukawa J, Hinata N, Muramaki M, Tanaka K, Fujisawa M. Significant impact of R.E.N.A.L. nephrometry score on changes in postoperative renal function early after robot-assisted partial nephrectomy. Int J Clin Oncol 2014: [in press]. Simmons MN, Hillyer SP, Lee BH, Fergany AF, Kaouk J, Campbell SC. Diameter-axial-polar nephrometry: integration and optimization of R.E.N.A.L. and centrality index scoring systems. J Urol 2012; 188:384–90. Hakky TS, Baumgarten AS, Allen B, et al. Zonal NePhRO scoring system: a superior renal tumor complexity classification model. Clin Genitourin Cancer 2014;12:e13–8. Nisen H, Ruutu M, Glucker E, Visapaa H, Taari K. Renal tumour invasion index as a novel anatomical classification predicting urological complications after partial nephrectomy. Scand J Urol 2014;48:41–51. Wang L, Li M, Chen W, et al. Is diameter-axial-polar scoring predictive of renal functional damage in patients undergoing partial nephrectomy? An evaluation using technetium Tc 99m (99Tcm) diethylene-triamine-penta-acetic acid (DTPA) glomerular filtration rate. BJU Int 2013;111:1191–8. Maeda M, Funahashi Y, Sassa N, Fujita T, Majima T, Gotoh M. Prediction of partial nephrectomy outcomes by using the diameteraxis-polar nephrometry score. Int J Urol 2014;21:442–6.
