1435Vol. 11 | No. 02 | March- April 2014 |U R O LO G Y J O U R N A L Postoperative Nomogram for Disease Recurrence and Cancer-Specific Death for Upper Tract Urothelial Carcinoma: Comparison to American Joint Commit- tee on Cancer Staging Classification Behfar Ehdaie,1,3 Shahrokh F. Shariat,2 Caroline Savage,3 Jonathan Coleman,1 Guido Dalbagni1 Purpose:‎We‎sought‎to‎develop‎prognostic‎models‎to‎predict‎disease‎recurrence‎and‎cancer- specific‎mortality‎in‎patients‎with‎upper‎tract‎urothelial‎carcinoma‎(UTUC)‎who‎underwent‎ radical‎nephroureterectomy‎(RNU).‎ Materials and Methods:‎Data‎on‎253‎patients‎treated‎with‎RNU‎between‎1995‎and‎2008‎at‎a‎ single‎high-volume‎tertiary‎referral‎center‎were‎analyzed.‎Statistically‎and‎clinically‎significant‎ patient‎and‎tumor‎characteristics‎were‎identified‎in‎a‎univariate‎analysis‎and‎incorporated‎into‎ a‎multivariable‎Cox‎regression‎model.‎The‎model‎was‎compared‎to‎the‎2010‎American‎Joint‎ Committee‎on‎Cancer‎(AJCC)‎staging‎classification‎using‎the‎concordance‎index‎(c-index),‎ corrected‎for‎statistical‎optimism‎using‎bootstrap‎methods. Results:‎Five-year‎recurrence-free‎survival‎(RFS)‎and‎cancer-specific‎survival‎(CSS)‎rates‎ were‎73%‎[95%‎confidence‎interval‎(CI):‎66-79%)]‎and‎78%‎(95%‎CI:‎71-84%),‎respectively.‎ On‎multivariate‎analysis,‎higher‎preoperative‎glomerular‎filtration‎rate‎(GFR)‎was‎associated‎ with‎better‎CSS‎[hazard‎ratio‎(HR)‎per‎1‎mL/min/m2‎increase‎in‎GFR‎for‎CSS:‎0.74;‎P‎=‎.002)],‎ while‎higher‎pathologic‎stage‎(HR‎for‎pT2:‎2.99‎and‎for‎≥‎pT3:‎7.34;‎P‎<‎.001)‎and‎lymph‎ node‎involvement‎(HR:‎3.75;‎P‎<‎.001)‎were‎associated‎with‎worse‎CSS;‎results‎were‎similar‎ for‎RFS.‎The‎ability‎of‎the‎final‎models,‎which‎included‎preoperative‎GFR,‎lymph‎node‎status,‎ pathologic‎grade,‎and‎stage,‎to‎predict‎RFS‎and‎CSS‎(c-index‎0.82‎and‎0.83,‎respectively)‎was‎ similar‎to‎that‎of‎the‎2010‎AJCC‎staging‎classification‎(c-index‎0.80‎and‎0.81,‎respectively).‎ Conclusion:‎Given‎the‎data-dependent‎selection‎of‎variables‎in‎this‎single‎institution‎cohort,‎ it‎is‎unlikely‎that‎the‎marginal‎improvement‎found‎with‎these‎prediction‎models‎would‎impor- tantly‎impact‎clinical‎decision-making‎or‎improve‎patient‎care.‎The‎2010‎AJCC‎staging‎clas- sification‎alone‎is‎very‎accurate‎and‎should‎continue‎to‎guide‎follow-up‎after‎RNU. Keywords:‎ nomograms;‎ prognosis;‎ survival;‎ urologic‎ neoplasms;‎ urothelium;‎ carcinoma;‎ transitional‎cell;‎retrospective‎studies. Corresponding Author: Behfar Ehdaie, MD Urology Service, Department of Surgery, Memorial Sloan- Kettering Cancer Center, 1275 York Ave., New York, NY 10065, USA. Tel: +1 646 422 4406 Fax: +1 212 988 0759. E-mail: ehdaieb@mskcc.org Received January 2013 Accepted July 2013 1Urology Service, Sidney Kim- mel Center for Prostate and Urologic Cancers, Memorial Sloan-Kettering Cancer Center, New York, NY, USA. 2 Department of Urology and Division of Medical Oncology, Weill Cornell Medical Center, New York-Presbyterian, New York, NY, USA. 3 Health Outcomes Research Group, Department of Epide- miology and Biostatistics, Me- morial Sloan-Kettering Cancer Center, New York, NY, USA. UROLOGICAL ONCOLOGY 1436 | INTRODUCTION Upper‎tract‎urothelial‎carcinoma‎(UTUC)‎is‎a‎rela-tively‎rare‎neoplasm‎and‎accounts‎for‎5-6%‎of‎all‎urothelial‎ tumors.(1)‎ Radical‎ nephroureterectomy‎ (RNU)‎with‎bladder‎cuff‎removal‎is‎the‎most‎effective‎treat- ment‎for‎invasive‎UTUC.‎Approximately‎20-40%‎of‎patients‎ initially‎ present‎ with‎ locally‎ advanced‎ disease‎ and‎ lymph‎ node‎metastases‎at‎the‎time‎of‎diagnosis.(2)‎Despite‎advances‎ in‎surgical‎technique,‎5-year‎cancer-specific‎mortality‎rates‎ are‎15%,‎45%‎and‎88%‎for‎patients‎with‎pT2N0,‎pT3N0‎and‎ pT4N0‎disease,‎respectively.(3)‎The‎rarity‎of‎the‎disease‎has‎ limited‎the‎use‎of‎prospective‎studies‎to‎evaluate‎the‎role‎of‎a‎ multimodality‎treatment‎approach,‎and‎existing‎data‎indicate‎ that‎the‎survival‎rates‎in‎these‎patients‎have‎not‎improved.(4) Accurate‎prediction‎of‎postoperative‎cancer‎recurrence‎and‎ survival‎outcomes‎could‎help‎guide‎decisions‎regarding‎ad- ministration‎of‎adjuvant‎chemotherapy‎and‎selective‎enroll- ment‎into‎clinical‎trials‎of‎novel‎therapies.‎Currently,‎use‎of‎ adjuvant‎chemotherapy‎is‎limited‎by‎the‎lack‎of‎proof‎of‎ef- ficacy‎and‎potential‎side‎effects.‎Moreover,‎physicians‎are‎ hesitant‎to‎administer‎neoadjuvant‎chemotherapy‎because‎of‎ the‎limited‎accuracy‎of‎preoperative‎staging‎based‎on‎histo- pathology.‎The‎pathological‎staging‎criteria‎defined‎by‎the‎ American‎Joint‎Committee‎on‎Cancer‎(AJCC)‎incorporate‎ tumor‎stage,‎nodal‎stage,‎and‎metastases‎information‎to‎pre- dict prognosis.(5)‎However,‎application‎of‎ the‎2010‎AJCC‎ staging‎classification‎to‎UTUC‎is‎limited‎for‎several‎reasons.‎ First,‎lymph‎node‎status‎is‎not‎consistent‎in‎UTUC‎as‎there‎ is‎no‎consensus‎defining‎an‎anatomic‎template‎for‎a‎lymph‎ node‎dissection‎and‎nearly‎60%‎of‎patients‎do‎not‎have‎ad- equate‎lymph‎node‎staging.(6)‎Second,‎tumor‎grade,‎which‎ is‎an‎important‎predictor‎of‎prognosis‎for‎UTUC,‎is‎not‎in- cluded‎in‎the‎2010‎AJCC‎classification.(7,8) Given‎these‎putative‎limitations‎of‎the‎2010‎AJCC‎staging‎ classification,‎ our‎ objective‎ was‎ to‎ develop‎ multivariable‎ models‎ to‎predict‎five-year‎recurrence-free‎survival‎(RFS)‎ and‎cancer-specific‎survival‎(CSS)‎after‎RNU‎based‎on‎pa- tient‎and‎tumor‎characteristics‎and‎to‎compare‎its‎prognostic‎ accuracy‎to‎that‎of‎the‎2010‎AJCC‎staging‎classification. MATERIALS AND METHODS Patient Cohort In‎ this‎ institutional‎ review‎ board-approved‎ study,‎ we‎ re- viewed‎all‎the‎prospectively‎collected‎data‎on‎324‎consecu- tive‎patients‎with‎upper‎tract‎tumors‎treated‎with‎RNU‎at‎Me- morial‎ Sloan-Kettering‎ Cancer‎ Center‎ (MSKCC)‎ between‎ 1995‎and‎2008.‎We‎excluded‎patients‎who‎underwent‎previ- ous‎or‎concurrent‎radical‎cystectomy‎(n‎=‎46),‎had‎prior‎con- tralateral‎UTUC‎(n‎=‎4),‎or‎received‎preoperative‎chemother- apy‎(n‎=‎21).‎The‎remaining‎253‎patients‎were‎the‎subjects‎of‎ the‎present‎analysis.‎RNU‎was‎performed‎by‎genitourinary‎ surgeons‎at‎MSKCC‎using‎a‎standardized‎approach,‎includ- ing‎the‎removal‎of‎the‎kidney‎with‎the‎entire‎length‎of‎the‎ ureter‎and‎the‎adjacent‎segment‎of‎the‎bladder‎cuff.‎The‎hi- lar‎and‎regional‎lymph‎nodes‎adjacent‎to‎the‎ipsilateral‎great‎ vessel‎generally‎were‎resected.‎All‎patients‎were‎enrolled‎in‎ standardized‎ post-operative‎ clinical‎ pathways.‎ The‎ preop- erative‎evaluation‎was‎similar‎among‎all‎patients,‎including‎ computed‎tomography‎(CT)‎scan,‎chest‎X-ray,‎cysto-ureter- oscopy,‎and‎urine‎cytology.‎No‎patient‎received‎preoperative‎ chemotherapy‎and‎post-operative‎chemotherapy‎consisted‎of‎ platinum-based‎treatment‎for‎evidence‎of‎metastatic‎disease. Outcome Evaluation To‎determine‎RFS,‎we‎defined‎disease‎recurrence‎as‎any‎ra- diographic‎documentation‎of‎disease‎or‎pathologically‎prov- en‎failure‎in‎operative‎site,‎regional‎lymph‎nodes,‎or‎distant‎ metastasis.‎Recurrences‎within‎the‎bladder‎and‎the‎contralat- eral‎collecting‎system‎were‎not‎considered‎in‎the‎analysis‎of‎ RFS‎rate.‎To‎determine‎CSS,‎cause‎of‎death‎was‎determined‎ by‎chart‎review‎corroborated‎by‎death‎certificate.‎ Statistical Analysis Our‎first‎aim‎was‎to‎identify‎postoperative‎predictors‎of‎on- cologic‎outcome‎after‎RNU.‎We‎used‎univariate‎Cox‎propor- tional-hazards‎regression‎models‎to‎predict‎RFS‎and‎CSS.‎ Predictors‎in‎our‎analyses‎included‎gender,‎age,‎American‎ Society‎of‎Anesthesiologists‎(ASA)‎classification‎score,‎pre- operative‎glomerular‎filtration‎rate‎(GFR),‎smoking‎history,‎ carcinoma‎in‎situ,‎multifocal‎disease,‎pathologic‎grade‎(≥‎2‎ or‎<‎2),‎pathologic‎stage‎(≤‎pT1,‎pT2,‎or‎≥‎pT3)‎and‎node‎sta- tus‎(NX,‎N0,‎or‎N1).‎For‎the‎outcomes‎of‎RFS‎and‎CSS,‎we‎ created‎two‎multivariable‎models‎that‎included‎carcinoma‎in‎ situ,‎grade,‎pathologic‎stage,‎and‎node‎status.‎Predictors‎for‎ the‎multivariable‎models‎were‎chosen‎from‎those‎found‎to‎ be‎significant‎on‎univariate‎analyses;‎due‎to‎the‎limited‎num- ber‎of‎events,‎we‎focused‎our‎selection‎on‎those‎that‎were‎ deemed‎the‎most‎clinically‎relevant.‎We‎evaluated‎the‎predic- tive‎accuracy‎of‎our‎multivariable‎models‎using‎concordance‎ index‎(c-index)‎with‎bootstrapping‎to‎correct‎for‎statistical‎ Urological Oncology 1437Vol. 11 | No. 02 | March- April 2014 |U R O LO G Y J O U R N A L optimism.‎All‎statistical‎analyses‎were‎conducted‎using‎STA- TA‎11.0‎(StataCorp,‎College‎Station,‎TX,‎USA).‎ RESULTS Patient‎ characteristics‎ are‎ shown‎ in‎ Table‎ 1.‎ The‎ median‎ age‎ of‎ patients‎ was‎ 72‎ years;‎ approximately‎ two-thirds (62%;‎n =‎158)‎of‎patients‎were‎male‎and‎three-quarters‎re- ported‎a‎history‎of‎smoking‎(74%).‎The‎median‎follow-up‎for‎ patients without disease recurrence or death was 4.2 years. Of‎the‎58‎patients‎who‎experienced‎recurrence,‎48‎died‎from‎ UTUC‎including‎1‎patient‎who‎died‎from‎disease‎without‎ prior‎evidence‎of‎a‎recurrence‎and‎was‎considered‎to‎have‎ experienced‎ recurrence‎ at‎ the‎ time‎ of‎ death.‎ Overall,‎ the‎ 5-year‎probability‎of‎RFS‎and‎CSS‎were‎73%‎[(95%‎confi- dence‎interval‎(CI):‎66-79%)]‎and‎78%‎(95%‎CI:‎71-84%),‎ respectively‎(Figures‎1‎and‎2).‎In‎total,‎96‎patients‎died;‎the‎ 5-year‎survival‎probability‎was‎65%‎(95%‎CI:‎58-72%).‎ Table‎2‎shows‎the‎results‎from‎the‎univariate‎Cox‎propor- tional-hazards‎(HR)‎regression‎models.‎Age,‎ASA‎score,‎and‎ preoperative‎GFR‎were‎significantly‎associated‎with‎CSS.‎ Additionally, grade, pathologic stage, and node status were significantly‎associated‎with‎RFS‎and‎CSS.‎We‎did‎not‎find‎ any‎evidence‎that‎gender‎(all‎P‎≥‎.2),‎or‎smoking‎history‎(all‎ P‎≥‎ .4)‎were‎significantly‎associated‎with‎any‎of‎ the‎out- comes.‎There‎was‎no‎evidence‎of‎non-linearity‎with‎respect‎ to‎age‎and‎preoperative‎GFR. The‎multivariable‎results‎are‎shown‎in‎Table‎3.‎Higher‎pre- operative‎GFR‎was‎associated‎with‎better‎outcomes,‎while‎ higher‎grade,‎pathologic‎stage,‎and‎lymph‎node‎involvement‎ were‎significantly‎associated‎with‎worse‎outcomes.‎ We‎were‎concerned‎that‎given‎the‎high‎rate‎of‎death‎from‎ other‎causes‎(half‎the‎deaths‎were‎due‎to‎causes‎other‎than‎ UTUC),‎ a‎ competing-risks‎ regression‎ model‎ would‎ have‎ been‎more‎appropriate.‎We‎repeated‎all‎of‎our‎analyses‎us- ing‎a‎competing‎risk‎regression‎models‎and‎found‎very‎simi- lar‎results.‎For‎example,‎the‎subhazard‎ratio‎for‎the‎effect‎of‎ GFR‎on‎recurrence‎was‎0.74‎(95%‎CI:‎0.61-0.90)‎compared‎ to‎HR‎of‎0.73‎(95%‎CI:‎0.61-0.88)‎using‎the‎Cox‎proportion- al-hazards‎model;‎both‎the‎subhazard‎ratio‎and‎HR‎estimate‎ for‎pathologic‎stage‎pT3‎and‎death‎from‎disease‎were‎7.34.‎ Given‎the‎similarity‎in‎estimates‎and‎the‎ability‎to‎compare‎ concordance‎indices,‎we‎chose‎to‎present‎our‎results‎from‎the‎ Cox‎proportional-hazards‎models.‎ The‎overall‎discriminatory‎ability‎of‎our‎two‎multivariable‎ models‎to‎predict‎RFS‎and‎CSS‎was‎high‎(bootstrap-correct- ed‎c-index:‎0.82‎and‎0.83,‎respectively).‎In‎comparison,‎the‎ discriminatory‎ability‎of‎AJCC‎stage‎alone‎to‎predict‎RFS‎ and‎CSS‎as‎measured‎by‎the‎bootstrap-corrected‎c-index‎was‎ 0.80‎and‎0.81,‎respectively.‎ DISCUSSION In‎this‎study,‎we‎combined‎patient‎and‎tumor‎characteristics‎ with‎pathologic‎stage‎and‎grade‎to‎predict‎RFS‎and‎CSS.‎ Additional patient characteristics and histology grade only Postoperative Nomogram for Upper Tract Urothelial Carcinoma | Ehdaie et al Figure 1. Recurrence-free survival in patients with upper tract urothelial carcinoma treated with radical nephroureterectomy and bladder cuff excision. Figure 2. Cancer-specific survival in patients with upper tract urothelial carcinoma treated with radical nephroureterectomy and bladder cuff excision. 1438 | marginally‎enhanced‎the‎discriminatory‎ability‎of‎the‎2010‎ AJCC‎ staging‎ classification‎ to‎ predict‎ disease‎ recurrence‎ and‎cancer-specific‎mortality‎in‎UTUC‎patients‎treated‎with‎ RNU.‎Given‎that‎the‎cohort‎was‎derived‎from‎a‎single‎insti- tution‎and‎that‎variables‎were‎selected‎for‎the‎multivariable‎ models‎in‎a‎data-dependent‎manner,‎we‎are‎cautious‎to‎rec- ommend‎the‎implementation‎of‎our‎models‎and‎replacing‎the‎ standard‎AJCC‎stage‎classification.‎As‎such,‎the‎incorpora- tion‎of‎GFR,‎pathologic‎grade,‎and‎nodal‎status‎in‎a‎formal‎ prognostic‎model‎is‎unlikely‎to‎importantly‎impact‎clinical‎ decision-making‎or‎improve‎patient‎care. Prediction‎models‎have‎been‎developed‎for‎various‎tumors‎ to‎assist‎physicians,‎provide‎patients‎with‎estimates‎of‎clini- cal‎outcomes,‎and‎aid‎in‎decision‎making.‎There‎is‎evidence‎ that‎incorporating‎multiple‎variables‎into‎a‎prediction‎model‎ provides‎more‎accurate‎risk‎prediction‎than‎classifying‎pa- tients‎ based‎ on‎ tumor‎ stage‎ alone.‎ In‎ urologic‎ oncology,‎ multivariable‎models‎have‎been‎shown‎to‎be‎more‎accurate‎ than‎clinical‎staging‎in‎predicting‎cancer-specific‎mortality‎ for‎renal‎cell‎carcinoma,‎urothelial‎carcinoma‎of‎the‎blad- der, prostate cancer, and penile cancer.(9-12)‎It‎is‎important‎to‎ evaluate‎the‎clinical‎value‎of‎a‎new‎prediction‎model‎before‎ it‎is‎implemented‎into‎clinical‎practice.‎One‎of‎the‎primary‎ steps‎is‎to‎compare‎its‎discriminatory‎ability‎with‎that‎of‎what‎ is‎currently‎used.‎Then,‎a‎direct‎comparison‎should‎be‎con- ducted‎in‎an‎external‎cohort‎using‎calibration‎and‎decision‎ curve‎analytic‎techniques‎to‎determine‎if‎the‎new‎predictive‎ outperforms‎other‎models.(13) Recently,‎Jeldres‎and‎colleagues‎developed‎a‎multivariable‎ predictive‎model‎in‎UTUC‎incorporating‎age,‎pathological‎ stage,‎and‎tumor‎grade‎using‎the‎Surveillance,‎Epidemiolo- gy,‎and‎End‎Results‎(SEER)‎database.(6)‎Their‎model’s‎accu- racy‎to‎predict‎cancer-specific‎mortality‎after‎RNU‎was‎good‎ (75.4%)‎and‎demonstrated‎statistically‎superior‎discrimina- tory‎ability‎compared‎to‎the‎AJCC‎and‎Union‎for‎Interna- tional‎Cancer‎Control‎(UICC)‎staging‎classification‎(64.8%).‎ Jeldres‎and‎colleagues‎acknowledged‎the‎limitations‎inherent‎ within‎the‎SEER‎database‎including‎lack‎of‎central‎pathol- ogy‎review,‎incomplete‎surgical‎resection‎of‎a‎bladder‎cuff‎ in‎31-35%‎of‎patients‎and‎absence‎of‎a‎lymph‎node‎dissec- tion‎in‎approximately‎60%‎of‎patients.‎The‎incompleteness‎ of‎surgical‎treatment‎defined‎by‎the‎lack‎of‎bladder‎cuff‎exci- sion‎can‎negatively‎impact‎the‎accuracy‎of‎pathological‎stag- ing‎and‎may‎limit‎the‎generalizability‎of‎the‎results.‎ Using‎a‎contemporary‎cohort‎of‎patients,‎we‎developed‎prog- nostic‎models‎to‎predict‎disease‎recurrence‎and‎cancer-specif- ic‎mortality‎incorporating‎pathologic‎stage,‎lymph‎node‎sta- tus,‎tumor‎grade,‎and‎preoperative‎GFR.‎Preoperative‎GFR‎ is‎an‎effective‎surrogate‎for‎overall‎health‎as‎multiple‎stud- Table 1. Patient characteristics. All values are median (IQR) or frequency (proportion). Variables No. = 253 Age at surgery (years) 72 (63-77) Body mass index (n = 246) (kg/m2) 27 (24-30) Male gender 158 (62%) Preoperative GFR (n = 248) (mL/min/m2) 54 (44-68) Preoperative high grade (n = 161) 110 (68%) Hematuria, no. (%) None 64 (25) Micro 28 (11) Gross 161 (64) Parenchymal invasion on CT (n = 248) 36 (15%) Hydronephrosis on CT (n = 251) 125 (50%) Cytology, no. (%) Negative 52 (21) Positive 190 (75) Not done 11 (4) Smoking history* (n = 252) 185 (73%) ASA score (n = 252), no. (%) 1 7 (3) 2 127 (50) 3 118 (47) Pathologic stage, no. (%) ≤ pT1 132 (52) pT2 56 (22) ≥ pT3 65 (26) Lymph node status, no. (%) NX 93 (37) N0 137 (54) N1 23 (9) Pathologic grade (n = 252), no. (%) High 193 (77) Low 59 (23) Previous NMIBC 86 (34) CIS 70 (28) Multifocal disease (n = 250) 65 (26%) Keys: NMIBC, non-muscle invasive bladder cancer; CIS, carcinoma in situ; GFR, glomerular filtration rate; IQR, interquartile range; CT, computed tomography; ASA, American Joint Committee on Cancer. *Smoking history reported at time of surgery (current/former vs. never). Urological Oncology 1439Vol. 11 | No. 02 | March- April 2014 |U R O LO G Y J O U R N A L ies‎have‎demonstrated‎that‎renal‎function‎impacts‎mortality‎ and‎reduced‎renal‎function‎leads‎ to‎hypertension,‎anemia,‎ malnutrition,‎cardiovascular‎disease,‎and‎reduced‎quality‎of‎ life.(14)‎In‎UTUC,‎assessment‎of‎renal‎function‎is‎important‎ for‎prognosis‎because‎RNU‎is‎the‎most‎effective‎treatment‎ for‎invasive‎disease‎while‎adjuvant‎chemotherapy‎is‎limited‎ by‎an‎increased‎risk‎of‎renal‎dysfunction.‎Furthermore,‎and‎ in‎contrast‎to‎the‎SEER‎study,‎we‎evaluated‎a‎consecutive,‎ prospectively‎collected‎cohort‎of‎patients‎in‎a‎single,‎high- volume‎tertiary‎referral‎center.‎In‎addition,‎the‎c-index‎for‎the‎ AJCC‎staging‎classification‎in‎the‎SEER‎database‎was‎low‎ (RFS:‎64.8%)‎and‎in‎our‎model,‎the‎2010‎AJCC‎staging‎clas- sification‎was‎highly‎predictive‎of‎survival‎outcomes‎(RFS:‎ 80%,‎CSS:‎81%).‎These‎results‎might‎be‎explained‎by‎the‎ heterogeneity‎in‎bladder‎cuff‎excision‎and‎lymphadenectomy‎ within‎the‎SEER‎database.‎ We‎failed‎to‎find‎evidence‎that‎the‎addition‎of‎grade,‎GFR,‎and‎ lymph‎node‎status‎importantly‎improved‎prediction‎of‎onco- logic‎outcomes‎in‎UTUC.‎Despite‎a‎marginal‎improvement‎in‎ the‎discriminatory‎ability‎of‎our‎predictive‎models‎compared‎ with‎the‎2010‎AJCC‎staging‎classification,‎we‎are‎cautious‎ with‎the‎interpretation‎of‎our‎results.‎In‎this‎single‎institution‎ study‎that‎involved‎data-dependent‎variable‎selection,‎a‎small‎ change‎in‎effect‎size‎is‎unlikely‎to‎withstand‎a‎comparative‎ analysis‎with‎the‎standard‎AJCC‎staging‎classification‎in‎an‎ external‎cohort‎using‎calibration‎and‎decision‎curve‎analytic‎ techniques‎as‎benchmarks.‎Alternatively,‎if‎we‎had‎achieved‎ these‎results‎using‎a‎large,‎multi-institutional‎database,‎then‎a‎ small‎improvement‎in‎the‎c-index‎would‎be‎more‎clinically‎ dependable‎and‎worthy‎of‎further‎investigation.‎ There‎are‎several‎important‎limitations‎of‎our‎study.‎First,‎ UTUC‎is‎a‎rare‎malignancy,‎which‎makes‎evaluation‎of‎large‎ Table 2. Univariate Cox proportional-hazards models for recurrence-free survival and upper tract urothelial carcinoma-specific survival. Variables Recurrence-Free Survival Disease-Specific Survival HR 95% CI P HR 95% CI P Male 1.11 0.65-1.89 .7 1.21 .5 Age 1.22 0.97-1534 .097 1.44 .014 ASA Score .074 .037 < 3 Ref Ref Ref ≥ 3 1.60 0.96-2.67 1.84 Preoperative GFR 0.76 0.64-0.89 < .001 0.74 .002 (per 10 mL/min/m2) Smoking 0.77 0.44-1.35 .4 0.77 .4 CIS 1.63 0.96-2.79 .072 1.66 .094 Multifocal disease 0.91 0.50-1.66 .8 0.88 .7 Grade .005 .014 Low Ref Ref Ref High 3.36 1.44-7.83 2.92 Pathologic stage < .001 < .001 ≤ pT1 Ref Ref Ref pT2 2.72 1.22-6.05 4.21 ≥ pT3 8.91 4.61-17.2 11.7 Node status < .001 < .001 N0 Ref Ref Ref N1 8.01 4.26-15.1 7.99 NX 0.78 0.42-1.46 0.76 Keys: HR, hazard ratio; CI, confidence interval; Ref, reference; NMIBC, non-muscle invasive bladder cancer; CIS, carcinoma in situ; Smoking, smoking status at time of surgery; GFR, glomerular filtration rate, ASA, American Joint Committee on Cancer. Postoperative Nomogram for Upper Tract Urothelial Carcinoma | Ehdaie et al 1440 | patient‎populations‎difficult.‎However,‎in‎spite‎of‎the‎limited‎ size‎of‎our‎cohort,‎we‎were‎able‎to‎identify‎several‎clinically‎ relevant‎predictors.‎Critically,‎despite‎identifying‎several‎var- iables‎that‎were‎independently‎and‎statistically‎significant‎on‎ multivariable‎analysis,‎incorporation‎of‎these‎factors‎into‎a‎ prediction‎model‎did‎not‎importantly‎improve‎the‎predictive‎ accuracy‎above‎that‎of‎the‎2010‎AJCC‎staging‎alone.‎Second,‎ tumor‎grade‎and‎pathology‎were‎determined‎by‎institutional‎ pathology‎report,‎as‎ there‎was‎no‎systematic‎re-review‎of‎ the‎pathologic‎material.‎Third,‎despite‎the‎standardization‎of‎ surgical‎technique‎and‎routine‎excision‎of‎a‎bladder‎cuff‎dur- ing‎RNU,‎37%‎of‎the‎patients‎in‎our‎cohort‎did‎not‎receive‎ a‎lymphadenectomy.‎Therefore,‎we‎believe‎it‎ is‎ important‎ to‎emphasize‎the‎need‎for‎a‎standardization‎of‎the‎extent‎of‎ lymph‎node‎dissection‎and‎its‎indications‎in‎UTUC‎patients‎ treated with RNU. We‎believe‎our‎study‎is‎novel‎and‎emphasizes‎the‎impor- tance‎of‎considering‎ the‎ impact‎of‎a‎prediction‎model‎on‎ clinical‎decision‎making‎rather‎ than‎focusing‎on‎marginal‎ statistical‎ significance.‎ Furthermore,‎ our‎ study‎ will‎ help‎ investigators‎develop‎an‎improved‎prediction‎model‎by‎in- corporating‎novel‎biomarkers‎for‎disease‎progression‎such‎ as‎tumor‎architecture.(15)‎In‎the‎future,‎the‎evaluation‎of‎a‎ larger‎multi-institutional‎database‎with‎additional‎patient‎and‎ tumor‎characteristics‎may‎improve‎the‎discriminatory‎ability‎ of‎a‎predictive‎model‎for‎UTUC.‎However,‎we‎demonstrated‎ that‎in‎this‎contemporary‎cohort‎of‎UTUC‎patients‎treated‎ with‎RNU‎the‎incorporation‎of‎multiple‎clinically‎significant‎ predictors‎does‎not‎provide‎more‎accurate‎prognostic‎infor- mation‎than‎pathologic‎stage. CONCLUSION The‎multivariate‎prognostic‎model‎we‎developed‎incorporat- ing‎GFR,‎tumor‎grade,‎stage‎and‎lymph‎node‎status‎did‎not‎ meaningfully‎improve‎the‎discriminatory‎ability‎of‎the‎2010‎ AJCC‎staging‎classification‎for‎UTUC.‎In‎this‎patient‎cohort,‎ the‎2010‎AJCC‎staging‎classification‎alone‎was‎very‎accu- rate.‎The‎2010‎AJCC‎staging‎classification‎should‎be‎used‎to‎ guide‎follow-up‎after‎RNU‎and‎to‎assist‎physicians‎in‎provid- ing‎estimates‎of‎cancer-specific‎outcomes‎to‎their‎patients. ACKNOWLEDGMENT This‎study‎was‎supported‎by‎the‎Sidney‎Kimmel‎Center‎for‎ Prostate‎and‎Urologic‎Cancers.‎ CONFLICT OF INTEREST None declared. Table 3. Multivariate Cox proportional-hazards models for recurrence-free survival and cancer-specific survival in patients with upper tract urothelial carcinoma treated with radical nephroureterectomy and bladder cuff excision. Variables Recurrence-Free Survival Disease-Specific Survival HR 95% CI P HR 95% CI P Preoperative GFR 0.73 0.61-0.88 < .001 0.74 0.61-0.90 .002 (per 10 mL/min/m2) Tumor grade .2 .3 Low Ref Ref Ref Ref High 1.77 0.72-4.38 1.69 0.69-4.17 Pathologic Stage < .001 < .001 ≤ pT1 Ref Ref Ref Ref pT2 1.67 0.71-3.94 2.99 1.12-7.97 ≥ pT3 5.71 2.77-1.7 7.34 2.98-18.1 Node Status < .001 .002 N0 Ref Ref Ref Ref N1 3.95 1.98-7.86 3.75 1.78-7.89 NX 1.16 0.59-2.29 1.31 0.61-2.81 Keys: HR, hazard ratio; CI, confidence interval; Ref, reference; GFR, glomerular filtration rate. Urological Oncology 1441Vol. 11 | No. 02 | March- April 2014 |U R O LO G Y J O U R N A L REFERENCES 1. Jemal A, Siegel R, Xu J, Ward E. Cancer statistics, 2010. CA Cancer J Clin. 2010;60:277-300. 2. Munoz JJ, Ellison LM. Upper tract urothelial neoplasms: incidence and survival during the last 2 decades. J Urol. 2000;164:1523-5. 3. Margulis V, Shariat SF, Matin SF, et al. Outcomes of radical nephro- ureterectomy: a series from the Upper Tract Urothelial Carcinoma Collaboration. Cancer. 2009;115:1224-33. 4. Lughezzani G, Jeldres C, Isbarn H, et al. Nephroureterectomy and segmental ureterectomy in the treatment of invasive upper tract urothelial carcinoma: a population-based study of 2299 patients. Eur J Cancer. 2009;45:3291-7. 5. Edge SB, David; Compton, Carolyn; Fritz, April; Greene, Frederick; Trotti III, Andy. AJCC Cancer Staging Manual. Springer, New York, NY. 2010; 7th edition. 6. Jeldres C, Sun M, Lughezzani G, et al. Highly predictive survival nomogram after upper urinary tract urothelial carcinoma. Cancer. 2010;116:3774-84. 7. Brown GA, Matin SF, Busby JE, et al. Ability of clinical grade to pre- dict final pathologic stage in upper urinary tract transitional cell carcinoma: implications for therapy. Urology. 2007;70:252-6. 8. Hall MC, Womack S, Sagalowsky AI, Carmody T, Erickstad MD, Roehr- born CG. Prognostic factors, recurrence, and survival in transitional cell carcinoma of the upper urinary tract: a 30-year experience in 252 patients. Urology. 1998;52:594-601. 9. Karakiewicz PI, Briganti A, Chun FK, et al. Multi-institutional valida- tion of a new renal cancer-specific survival nomogram. J Clin Oncol. 2007;25316-22. 10. Shariat SF, Karakiewicz PI, Palapattu GS, et al. Nomograms provide improved accuracy for predicting survival after radical cystectomy. Clin Cancer Res. 2006;12:6663-76. 11. Zini L, Cloutier V, Isbarn H, Perrotte P, et al. A simple and accurate model for prediction of cancer-specific mortality in patients treated with surgery for primary penile squamous cell carcinoma. Clin Can- cer Res. 2009;15:1013-8. 12. Stephenson AJ, Scardino PT, Eastham JA, et al. Postoperative nomo- gram predicting the 10-year probability of prostate cancer recur- rence after radical prostatectomy. J Clin Oncol. 2005;23:7005-12. 13. Lughezzani G, Budaus L, Isbarn H, et al. Head-to-head comparison of the three most commonly used preoperative models for pre- diction of biochemical recurrence after radical prostatectomy. Eur Urol. 2010;57:562-8. 14. Go AS, Chertow GM, Fan D, McCulloch CE, Hsu CY. Chronic kidney disease and the risks of death, cardiovascular events, and hospitali- zation. N Engl J Med. 2004;351:1296-305. 15. Fritsche HM, Novara G, Burger M, et al. Macroscopic sessile tumor architecture is a pathologic feature of biologically aggressive upper tract urothelial carcinoma. Urol Oncol. 2012;30:666-72. Postoperative Nomogram for Upper Tract Urothelial Carcinoma | Ehdaie et al