- Email: [email protected]
Nomograms Predicting Response to Therapy and Outcomes After Bladder-Preserving Trimodality Therapy for Muscle-Invasive Bladder Cancer
International Journal of
Radiation Oncology biology
physics
www.redjournal.org
Clinical Investigation: Genitourinary Cancer
Nomograms Predicting Response to Therapy and Outcomes After Bladder-Preserving Trimodality Therapy for Muscle-Invasive Bladder Cancer John J. Coen, MD,* Jonathan J. Paly, BS,* Andrzej Niemierko, PhD,* Donald S. Kaufman, MD,y Niall M. Heney, MD,z Daphne Y. Spiegel, BS,* Jason A. Efstathiou, MD,* Anthony L. Zietman, MD,* and William U. Shipley, MD* Departments of *Radiation Oncology, yMedical Oncology, and zUrology, Massachusetts General Hospital, Boston, Massachusetts Received Aug 17, 2012, and in revised form Jan 4, 2013. Accepted for publication Jan 14, 2013
Summary Selective bladder preservation by use of trimodality therapy is an established treatment for muscleinvasive bladder cancer. Patients are often guided to this therapy based on the likelihood of a complete response and successful bladder preservation based on individual prognostic features. We have developed prognostic nomograms incorporating these features into the models. They should assist patients and clinicians making important treatment decisions.
Purpose: Selective bladder preservation by use of trimodality therapy is an established management strategy for muscle-invasive bladder cancer. Individual disease features have been associated with response to therapy, likelihood of bladder preservation, and disease-free survival. We developed prognostic nomograms to predict the complete response rate, disease-specific survival, and likelihood of remaining free of recurrent bladder cancer or cystectomy. Methods and Materials: From 1986 to 2009, 325 patients were managed with selective bladder preservation at Massachusetts General Hospital (MGH) and had complete data adequate for nomogram development. Treatment consisted of a transurethral resection of bladder tumor followed by split-course chemoradiation. Patients with a complete response at midtreatment cystoscopic assessment completed radiation, whereas those with a lesser response underwent a prompt cystectomy. Prognostic nomograms were constructed predicting complete response (CR), disease-specific survival (DSS), and bladder-intact disease-free survival (BI-DFS). BIDFS was defined as the absence of local invasive or regional recurrence, distant metastasis, bladder cancer-related death, or radical cystectomy. Results: The final nomograms included information on clinical T stage, presence of hydronephrosis, whether a visibly complete transurethral resection of bladder tumor was performed, age, sex, and tumor grade. The predictive accuracy of these nomograms was assessed. For complete response, the area under the receiving operating characteristic curve was 0.69. The Harrell concordance index was 0.61 for both DSS and BI-DFS. Conclusions: Our nomograms allow individualized estimates of complete response, DSS, and BI-DFS. They may assist patients and clinicians making important treatment decisions. Ó 2013 Elsevier Inc.
Reprint requests to: John J. Coen, MD, Department of Radiation Oncology, The Helen & Harry Gray Cancer Center, Hartford Hospital, 80 Seymour St, Hartford, CT 06102. Tel: (860) 545-2803; E-mail: jcoen@ harthosp.org Int J Radiation Oncol Biol Phys, Vol. 86, No. 2, pp. 311e316, 2013 0360-3016/$ - see front matter Ó 2013 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.ijrobp.2013.01.020
Conflict of interest: none. AcknowledgmentsdThe authors thank Dr Michael Kattan for his invaluable advice regarding nomogram construction.
312
Coen et al.
International Journal of Radiation Oncology Biology Physics
Introduction
Statistical methods
Selective bladder preservation by use of trimodality therapy is an established management strategy for muscle-invasive bladder cancer with outcomes that are comparable to radical cystectomy (1-4). This treatment paradigm incorporates a maximal transurethral resection of bladder tumor (TURBT) followed by an induction course of concurrent radiation with sensitizing chemotherapy. A midtreatment cystoscopic assessment of tumor response is performed after induction therapy, which includes a tumor site biopsy. Patients achieving a complete response (CR) continue with consolidation chemoradiation, whereas those with a lesser response are guided toward a prompt cystectomy. This treatment model has been followed at Massachusetts General Hospital (MGH) and within the Radiation Therapy Oncology Group (RTOG) where patients have been treated in a series of prospective clinical trials (2). Individual disease features have been associated with response to therapy, likelihood of bladder preservation, and disease-free survival (3). Alternate prognostic models capable of simultaneous evaluation of multiple relevant variables may result in improved predictive ability. Nomograms are statistical models designed to maximize predictive accuracy in an individual patient incorporating all relevant prognostic information. They are easier to use and have greater utility than more traditional risk group assignments (5-7). In this study we developed nomograms to predict the CR rate to induction therapy, disease-specific survival (DSS), and the likelihood of remaining free of recurrent bladder cancer while maintaining the native bladder in a series of patients managed with selective bladder preservation at MGH.
Prognostic nomograms were constructed for 3 endpoints: CR rate, 5-year DSS, and 5-year bladder-intact disease-free survival (BIDFS). All patients underwent a midtreatment cystoscopic assessment and were categorized as having either a CR or less-than-complete response. The CR rate is simply the likelihood of having a CR at this assessment. The nomogram for this endpoint was modeled by logistic regression. DSS and BI-DFS are both actuarial endpoints where patients either had an event at some interval from therapy or never had an event and were censored at the last available follow-up date. For DSS, death from bladder cancer was the only event defined. For BI-DFS, an event was scored at the earliest of the following dates: local invasive recurrence, regional recurrence, distant metastasis, bladder cancer-related death, or radical cystectomy. All other patients were censored at the last follow-up date. The nomograms for both DSS and BI-DFS were constructed by use of the Cox proportional hazards model. Clinical T stage (T2 or T3-T4a), TURBT completeness (yes or no), hydronephrosis (yes or no), age (<65 years or 65 years), sex, and grade (intermediate or high) were included as covariates in the modeling process. Backward selection was used, and only variables with a P<.3 were included in the final models. For the CR nomogram, the discriminative ability of the logistic regression model was defined as the area under the receiving operating characteristic (ROC) curve, where an area of 1 represents a test with perfect predictive ability and an area of 0.5 represents a coin toss. The discriminative ability of the DSS and BI-DFS nomograms was assessed with the Harrell concordance index (c-index). Similar to the area under the ROC curve, the cindex can range from perfect concordance (1.0) to perfect discordance (0.0), with a value of 0.5 being equivalent to a coin toss. Bootstrapping was used to account for over-fitting of the data, and corrections for optimism were made. Calibration plots were generated for all 3 nomograms. For BI-DFS and DSS, calibration of the nomogram was assessed by grouping patients with respect to their nomogram-predicted probabilities and comparing the mean of the group with the observed Kaplan-Meier estimates of 5-year BI-DFS. Bootstrapping correction was again used, where Kaplan-Meier estimates were calculated for each nomogram probability group in 200 bootstrap samples. The mean 5-year predicted KaplanMeier estimate and 95% confidence interval were calculated. These observed rates were plotted against the predicted outcomes in a nomogram calibration curve. The HosmerLemeshow goodness-of-fit test was used to assess the CR nomogram. All analyses were performed with STATA 11 (StataCorp, College Station, TX). P<.05 denoted statistical significance. All reported P values are 2 sided.
Methods and Materials From 1986 to 2009, 325 patients were managed with selective bladder preservation at MGH and had complete clinical data adequate for nomogram development. Eligible patients had clinical stage T2NXM0 to T4aNXM0, urothelial carcinoma of the bladder and were medically operable candidates for cystectomy. The majority of these patients were treated on a succession of prospective institutional and RTOG bladder-sparing protocols. Treatment began with a TURBT and pathologic confirmation of muscle-invasive bladder cancer. Induction therapy consisted of radiation to a dose of 40 Gy with concurrent cisplatin-based chemotherapy. After induction therapy, patients underwent a midtreatment cystoscopic assessment with tumor site biopsy. Those with a CR continued on to consolidation chemoradiation consisting of approximately 24 Gy with concurrent chemotherapy. Those with a less-than-complete response were advised to undergo a prompt cystectomy. These bladder-sparing protocols, including their respective eligibility criteria, chemotherapy regimens, and radiation techniques, and this cohort of patients have been previously described (3). In general, all patients were suitable for cystectomy but motivated to pursue bladder preservation. Patients with medical contraindications to the specific chemotherapy regimen used or obvious radiographic evidence of distant metastasis and/or lymph node involvement were generally excluded. Hydronephrosis has become a factor for exclusion from participation in more recent trials, but no other prognostic feature has been systematically used as an exclusion factor for this approach.
Results The characteristics of the 325 patients used to construct our nomograms are shown in Table 1. The median age was 66 years, and the median follow-up for surviving patients was 9.8 years, with 82 patients having greater than 10 years of follow-up.
Volume 86 Number 2 2013 Table 1
Nomograms predicting outcome after bladder-sparing therapy
Patient characteristics Data
Age at start of therapy [median (range)] (y) Follow-up (y) N Sex Male Female Clinical T stage 2 3a 3b 4 Hydronephrosis Yes No TURBT status Complete Incomplete Grade Intermediate High
66.3 (27.3-88.6) 4.6 (0.0-23.3) 325 240 85 176 106 20 23 53 272 215 110 45 280
Abbreviation: TURBT Z transurethral resection of bladder tumor.
CR to induction therapy The overall CR rate was 70%. A nomogram for CR was constructed using a logistic regression model. The final model included hydronephrosis, TURBT completeness, age, and sex as covariates. This nomogram is displayed in Fig. 1a. The performance of the nomogram was good, with an optimism-corrected area under the ROC curve of 0.68 (Fig. 1b). A calibration plot, shown in Fig. 1c, shows the predictive ability of the CR nomogram. The Hosmer-Lemeshow goodness-of-fit test shows that the model fits the data well (PZ.5).
Disease-specific survival The overall 5-yr DSS was 63%. The DSS nomogram was constructed by use of a Cox regression model. The final model included hydronephrosis, grade, and clinical T stage as covariates. It is displayed in Fig. 2a. The optimism-corrected Harrell concordance index was 0.60. A calibration plot for this nomogram is shown in Fig. 2b.
Bladder-intact disease-free survival The overall 5-year BI-DFS rate was 41%. The BI-DFS nomogram was also constructed using a Cox regression model. The final model included hydronephrosis, TURBT completeness, clinical T stage, and age as covariates. It is displayed in Fig. 3a. The optimism-corrected Harrell concordance index was 0.60. A calibration plot for this nomogram is shown in Fig. 3b.
Discussion Bladder-preserving trimodality therapy, when used selectively, results in rates of DFS comparable to contemporary radical
313
cystectomy with a high rate of bladder preservation (1-4). Increasingly careful selection of patients leads to higher rates of CR to induction therapy and consequently higher rates of bladder preservation. The prognostic features associated with favorable outcomes have previously been defined (3). In this singleinstitution series of well-followed patients treated with bladderpreserving trimodality therapy, we have developed a prognostic nomogram predicting rates of CR, DSS, and likelihood of remaining free of both recurrent bladder cancer and cystectomy. Nomograms are predictive models offering individualized estimates of disease-related outcomes based on multiple prognostic features. They have been shown to have greater predictive accuracy than traditional risk group assignments in a variety of clinical settings (5, 8-12). Their intuitive construction and ease of interpretation make them favored by clinicians and patients for individual decision making. In genitourinary oncology, nomograms are in widespread use for prostate cancer, but there is also a precedent in bladder cancer. The International Bladder Cancer Nomogram Consortium developed a postoperative nomogram predicting risk of recurrence after radical cystectomy (13). These investigators constructed an international multi-institutional database of over 9000 patients from which they developed a nomogram predicting risk of recurrence 5 years after cystectomy. The variables included in the nomogram were sex, age, pathologic T stage, histology, pathologic nodal status, grade of disease, and time from diagnosis to cystectomy. The predictive accuracy of the nomogram was significantly better than that of the American Joint Committee on Cancer TNM staging system or standard pathologic subgroupings. Although a postoperative nomogram for patients after radical cystectomy is certainly a useful prognostic tool, a pretreatment nomogram for patients considering either selective bladder preservation or immediate cystectomy would be of great value not just to predict outcome but also to help physicians and patients make important therapeutic decisions. Currently, patients may be steered away from bladder preservation based on risk group stratification often on the basis of a single adverse feature. Nomograms allowing more individualized estimates of CR, freedom from recurrence, and likelihood of bladder preservation may prompt patients or consultants to reconsider who is or is not a good candidate for bladder preservation rather than making this clinical decision based on a single prognostic feature. Some patients with a single poor prognostic feature may be viable candidates, whereas other patients with no one striking poor feature but with many intermediate features may be poorer candidates than previously appreciated. The potential improvement in predictive accuracy in this setting would not only improve patient selection and prognostication but also afford more rational clinical decision making. We developed nomograms predicting rates of CR to induction therapy, DSS, and BI-DFS. These represent 3 important therapeutic goals. A CR to induction therapy is required to complete the course of chemoradiation and avert an immediate cystectomy. DSS assesses the likelihood of dying of bladder cancer, an obviously critical endpoint. In addition, because the vast majority of patients with recurrent bladder cancer after radical cystectomy succumb to their cancer in fairly short order, we believe that our DSS nomogram offers a reasonable comparison to the postcystectomy nomogram in terms of disease-related outcome with bladder sparing as opposed to radical cystectomy (4). Finally, the BI-DFS endpoint assesses both goals of selective bladder preservation. The primary goal is the same as that after radical cystectomy, which is to eradicate the bladder cancer, whereas bladder
314
Coen et al.
International Journal of Radiation Oncology Biology Physics
Fig. 1. (a) Complete response (CR) nomogram. (b) Receiving operating characteristic (ROC) curve for CR model. (c) Calibration plot for CR nomogram. TURBT Z transurethral resection of bladder tumor.
preservation is a secondary goal when it can be safely accomplished. Just as the international postcystectomy nomogram was modeled to predict disease-free survival, we also modeled the same events (13). Superficial bladder recurrence was not included as an event because this is usually managed conservatively with a TURBT and intravesical therapy with no demonstrable increase in risk of bladder cancer-related death (14, 15). Any other bladder cancer recurrence was included as an event including invasive local recurrence in the bladder, pelvic recurrence, regional or nonregional lymph node recurrence, distant recurrence, or bladder cancer-related death. In addition, because selective bladder preservation has the preservation of the bladder as a secondary goal, cystectomy was also included as an event. The 5-year time point was chosen for DSS and BI-DFS because there were few events beyond 5 years, reflecting the early recurrence pattern of muscleinvasive bladder cancer (3, 4, 16). Although the database used to develop our nomogram is the largest US bladder preservation series, it is still a relatively small single-institution series as compared with the large multiinstitutional databases used to develop prognostic nomograms for men receiving various forms of therapy for prostate cancer, women with breast cancer, or men and women managed by radical
cystectomy for bladder cancer (10, 13, 17-19). Often, the variables selected for inclusion in a nomogram are preselected based on known prognostic features in the particular clinical scenario being modeled. The development process does not include subsequent elimination of nonpredictive variables. In our more limited but robust dataset, we thought it wise to only include variables that were both predictive in this dataset and clinically sensible. Despite the limited number of variables in the final models, the predictive accuracy was good and was superior to risk group stratification using any single one of these variables. Limiting the number of variables also decreases the likelihood of over-fitting the model to our dataset, making it more applicable to future patients treated with similar selective bladder-preservation regimens. Future directions include validating this nomogram both prospectively and in other pre-existing bladder-sparing datasets. Developing a multi-institutional international database may be a fruitful effort because bladder-preserving therapy is more common outside of the United States. Inclusion of molecular markers, such as MRE11, predicting response to therapy may also allow for the more precise identification of patients well or poorly suited to this approach (20). These efforts are under way both at our institution and through the RTOG. Better predictive models
Volume 86 Number 2 2013
Fig. 2.
Nomograms predicting outcome after bladder-sparing therapy
315
(a) Nomogram and (b) calibration plot for disease-specific survival (DSS) at 5 years. C Z concordance index; K-M Z Kaplan-Meier.
may allow not only improved selection of motivated patients but also wider use of this treatment paradigm through greater confidence in the ultimate outcome. In their current form, our nomograms are quite informative; however, the manner in which they are used to guide clinical decisions must be tailored to an individual patient’s disease features, overall health, bladder function, and strength of motivation toward bladder preservation. Still, some general observations can be made regarding the construction of these nomograms and some general recommendations made regarding their interpretation. The following generalizations can be inferred by inspection of the 3 nomograms. Hydronephrosis is predictive of DSS and CR. Clinical T stage is predictive of DSS but not CR. Consistent with prior publications, extent of TURBT is predictive of CR but not DSS (3). All 3 of these variables made strong contributions to the BI-DFS nomogram. Other variables had lesser but important roles. Age predicted for CR and BI-DFS, sex for CR, and grade for DSS. Application of the nomograms is more complex. Because there is no clear survival decrement to selective bladder preservation as
opposed to immediate cystectomy, the DSS nomogram cannot be used to decide for or against bladder sparing. Comparison with the surgical nomogram, before making a therapeutic decision, is not possible because it is a postcystectomy nomogram incorporating pathologic features only known after surgery (13). Although the BI-DFS nomogram is useful in giving a patient a sense of the likelihood of being disease free with an intact bladder 5 years after treatment, it is difficult to define a decisive cutoff value. Patients should be reminded that the BI-DFS rate after cystectomy is, by definition, 0%. The CR nomogram, on the other hand, may allow patients to make an informed therapeutic choice. Although patients and physicians may differ on a precise cutoff value, most define the threshold in the 60%-70% range, meaning few would advocate for patients with a chance of response of less than 60% and few, except those not generally in favor of bladder preservation, would argue that 70% was unacceptably low. Inspection of the CR nomogram shows that any patient with 3 or more favorable features would exceed this threshold. In addition, any patient without hydronephrosis who has a visibly complete resection would also exceed
Fig. 3. (a) Nomogram and (b) calibration plot for bladder-intact disease-free survival (BI-DFS) at 5 years. C Z concordance index; K-M Z Kaplan-Meier; TURBT Z transurethral resection of bladder tumor.
316
Coen et al.
this threshold even in the absence of a third favorable feature. Still, validation of these nomograms prospectively and in other preexisting datasets is in order. Caution must be exercised when using these models to inform a patient’s therapeutic decisions, especially with regard to the application of stringent cutoffs. They should be used within the context of a specific patient’s care, giving other pertinent clinical issues due respect.
Conclusion Selective bladder preservation with trimodality therapy is an established management strategy for muscle-invasive bladder cancer. Although the prognostic features for patients pursuing this therapy have been defined previously, more individualized estimates of outcome are desired. We developed nomograms that integrate predictive features to provide individualized estimates of likelihood of CR to treatment, DSS, and likelihood of both bladder preservation and freedom from recurrence (BI-DFS). These tools should prove valuable to muscle-invasive bladder cancer patients by enabling them, in conjunction with their physicians, to make more rationally important decisions about their treatment.
References 1. Rodel C, Weiss C, Sauer R. Trimodality treatment and selective organ preservation for bladder cancer. J Clin Oncol 2006;24:5536-5544. 2. Shipley WU, Kaufman DS, Tester WJ, et al. Overview of bladder cancer trials in the Radiation Therapy Oncology Group. Cancer 2003; 97:2115-2119. 3. Efstathiou JA, Spiegel DY, Shipley WU, et al. Long-term outcomes of selective bladder preservation by combined-modality therapy for invasive bladder cancer: The MGH experience. Eur Urol 2012;61:705-711. 4. Zehnder P, Studer UE, Skinner EC, et al. Super extended versus extended pelvic lymph node dissection in patients undergoing radical cystectomy for bladder cancer: A comparative study. J Urol 2011;186: 1261-1268. 5. Kattan MW, Scardino PT. Evidence for the usefulness of nomograms. Nat Clin Pract Urol 2007;4:638-639. 6. Kattan MW. Nomograms are superior to staging and risk grouping systems for identifying high-risk patients: Preoperative application in prostate cancer. Curr Opin Urol 2003;13:111-116.
International Journal of Radiation Oncology Biology Physics 7. Sternberg CN. Are nomograms better than currently available stage groupings for bladder cancer? J Clin Oncol 2006;24:3819-3820. 8. Steyerberg EW, Roobol MJ, Kattan MW, et al. Prediction of indolent prostate cancer: Validation and updating of a prognostic nomogram. J Urol 2007;177:107-112. discussion 112. 9. Stephenson AJ, Scardino PT, Eastham JA, et al. Postoperative nomogram predicting the 10-year probability of prostate cancer recurrence after radical prostatectomy. J Clin Oncol 2005;23:7005-7012. 10. Rudloff U, Jacks LM, Goldberg JI, et al. Nomogram for predicting the risk of local recurrence after breast-conserving surgery for ductal carcinoma in situ. J Clin Oncol 2010;28:3762-3769. 11. Rouzier R, Pusztai L, Delaloge S, et al. Nomograms to predict pathologic complete response and metastasis-free survival after preoperative chemotherapy for breast cancer. J Clin Oncol 2005;23:8331-8839. 12. Kattan MW, Yu C, Salomon L, et al. Development and validation of preoperative nomogram for disease recurrence within 5 years after laparoscopic radical prostatectomy for prostate cancer. Urology 2011; 77:396-401. 13. Bochner BH, Kattan MW, Vora KC. Postoperative nomogram predicting risk of recurrence after radical cystectomy for bladder cancer. J Clin Oncol 2006;24:3967-3972. 14. Weiss C, Wittlinger M, Engehausen DG, et al. Management of superficial recurrences in an irradiated bladder after combinedmodality organ-preserving therapy. Int J Radiat Oncol Biol Phys 2008;70:1502-1506. 15. Zietman AL, Grocela J, Zehr E, et al. Selective bladder conservation using transurethral resection, chemotherapy, and radiation: Management and consequences of Ta, T1, and Tis recurrence within the retained bladder. Urology 2001;58:380-385. 16. Stein JP, Lieskovsky G, Cote R, et al. Radical cystectomy in the treatment of invasive bladder cancer: Long-term results in 1,054 patients. J Clin Oncol 2001;19:666-675. 17. Eggener SE, Scardino PT, Walsh PC, et al. Predicting 15-year prostate cancer specific mortality after radical prostatectomy. J Urol 2011;185: 869-875. 18. Potters L, Roach M III, Davis BJ, et al. Postoperative nomogram predicting the 9-year probability of prostate cancer recurrence after permanent prostate brachytherapy using radiation dose as a prognostic variable. Int J Radiat Oncol Biol Phys 2010;76:1061-1065. 19. Zelefsky MJ, Kattan MW, Fearn P, et al. Pretreatment nomogram predicting ten-year biochemical outcome of three-dimensional conformal radiotherapy and intensity-modulated radiotherapy for prostate cancer. Urology 2007;70:283-287. 20. Choudhury A, Nelson LD, Teo MT, et al. MRE11 expression is predictive of cause-specific survival following radical radiotherapy for muscle-invasive bladder cancer. Cancer Res 2010;70:7017-7026.
Radiation Oncology biology
physics
www.redjournal.org
Clinical Investigation: Genitourinary Cancer
Nomograms Predicting Response to Therapy and Outcomes After Bladder-Preserving Trimodality Therapy for Muscle-Invasive Bladder Cancer John J. Coen, MD,* Jonathan J. Paly, BS,* Andrzej Niemierko, PhD,* Donald S. Kaufman, MD,y Niall M. Heney, MD,z Daphne Y. Spiegel, BS,* Jason A. Efstathiou, MD,* Anthony L. Zietman, MD,* and William U. Shipley, MD* Departments of *Radiation Oncology, yMedical Oncology, and zUrology, Massachusetts General Hospital, Boston, Massachusetts Received Aug 17, 2012, and in revised form Jan 4, 2013. Accepted for publication Jan 14, 2013
Summary Selective bladder preservation by use of trimodality therapy is an established treatment for muscleinvasive bladder cancer. Patients are often guided to this therapy based on the likelihood of a complete response and successful bladder preservation based on individual prognostic features. We have developed prognostic nomograms incorporating these features into the models. They should assist patients and clinicians making important treatment decisions.
Purpose: Selective bladder preservation by use of trimodality therapy is an established management strategy for muscle-invasive bladder cancer. Individual disease features have been associated with response to therapy, likelihood of bladder preservation, and disease-free survival. We developed prognostic nomograms to predict the complete response rate, disease-specific survival, and likelihood of remaining free of recurrent bladder cancer or cystectomy. Methods and Materials: From 1986 to 2009, 325 patients were managed with selective bladder preservation at Massachusetts General Hospital (MGH) and had complete data adequate for nomogram development. Treatment consisted of a transurethral resection of bladder tumor followed by split-course chemoradiation. Patients with a complete response at midtreatment cystoscopic assessment completed radiation, whereas those with a lesser response underwent a prompt cystectomy. Prognostic nomograms were constructed predicting complete response (CR), disease-specific survival (DSS), and bladder-intact disease-free survival (BI-DFS). BIDFS was defined as the absence of local invasive or regional recurrence, distant metastasis, bladder cancer-related death, or radical cystectomy. Results: The final nomograms included information on clinical T stage, presence of hydronephrosis, whether a visibly complete transurethral resection of bladder tumor was performed, age, sex, and tumor grade. The predictive accuracy of these nomograms was assessed. For complete response, the area under the receiving operating characteristic curve was 0.69. The Harrell concordance index was 0.61 for both DSS and BI-DFS. Conclusions: Our nomograms allow individualized estimates of complete response, DSS, and BI-DFS. They may assist patients and clinicians making important treatment decisions. Ó 2013 Elsevier Inc.
Reprint requests to: John J. Coen, MD, Department of Radiation Oncology, The Helen & Harry Gray Cancer Center, Hartford Hospital, 80 Seymour St, Hartford, CT 06102. Tel: (860) 545-2803; E-mail: jcoen@ harthosp.org Int J Radiation Oncol Biol Phys, Vol. 86, No. 2, pp. 311e316, 2013 0360-3016/$ - see front matter Ó 2013 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.ijrobp.2013.01.020
Conflict of interest: none. AcknowledgmentsdThe authors thank Dr Michael Kattan for his invaluable advice regarding nomogram construction.
312
Coen et al.
International Journal of Radiation Oncology Biology Physics
Introduction
Statistical methods
Selective bladder preservation by use of trimodality therapy is an established management strategy for muscle-invasive bladder cancer with outcomes that are comparable to radical cystectomy (1-4). This treatment paradigm incorporates a maximal transurethral resection of bladder tumor (TURBT) followed by an induction course of concurrent radiation with sensitizing chemotherapy. A midtreatment cystoscopic assessment of tumor response is performed after induction therapy, which includes a tumor site biopsy. Patients achieving a complete response (CR) continue with consolidation chemoradiation, whereas those with a lesser response are guided toward a prompt cystectomy. This treatment model has been followed at Massachusetts General Hospital (MGH) and within the Radiation Therapy Oncology Group (RTOG) where patients have been treated in a series of prospective clinical trials (2). Individual disease features have been associated with response to therapy, likelihood of bladder preservation, and disease-free survival (3). Alternate prognostic models capable of simultaneous evaluation of multiple relevant variables may result in improved predictive ability. Nomograms are statistical models designed to maximize predictive accuracy in an individual patient incorporating all relevant prognostic information. They are easier to use and have greater utility than more traditional risk group assignments (5-7). In this study we developed nomograms to predict the CR rate to induction therapy, disease-specific survival (DSS), and the likelihood of remaining free of recurrent bladder cancer while maintaining the native bladder in a series of patients managed with selective bladder preservation at MGH.
Prognostic nomograms were constructed for 3 endpoints: CR rate, 5-year DSS, and 5-year bladder-intact disease-free survival (BIDFS). All patients underwent a midtreatment cystoscopic assessment and were categorized as having either a CR or less-than-complete response. The CR rate is simply the likelihood of having a CR at this assessment. The nomogram for this endpoint was modeled by logistic regression. DSS and BI-DFS are both actuarial endpoints where patients either had an event at some interval from therapy or never had an event and were censored at the last available follow-up date. For DSS, death from bladder cancer was the only event defined. For BI-DFS, an event was scored at the earliest of the following dates: local invasive recurrence, regional recurrence, distant metastasis, bladder cancer-related death, or radical cystectomy. All other patients were censored at the last follow-up date. The nomograms for both DSS and BI-DFS were constructed by use of the Cox proportional hazards model. Clinical T stage (T2 or T3-T4a), TURBT completeness (yes or no), hydronephrosis (yes or no), age (<65 years or 65 years), sex, and grade (intermediate or high) were included as covariates in the modeling process. Backward selection was used, and only variables with a P<.3 were included in the final models. For the CR nomogram, the discriminative ability of the logistic regression model was defined as the area under the receiving operating characteristic (ROC) curve, where an area of 1 represents a test with perfect predictive ability and an area of 0.5 represents a coin toss. The discriminative ability of the DSS and BI-DFS nomograms was assessed with the Harrell concordance index (c-index). Similar to the area under the ROC curve, the cindex can range from perfect concordance (1.0) to perfect discordance (0.0), with a value of 0.5 being equivalent to a coin toss. Bootstrapping was used to account for over-fitting of the data, and corrections for optimism were made. Calibration plots were generated for all 3 nomograms. For BI-DFS and DSS, calibration of the nomogram was assessed by grouping patients with respect to their nomogram-predicted probabilities and comparing the mean of the group with the observed Kaplan-Meier estimates of 5-year BI-DFS. Bootstrapping correction was again used, where Kaplan-Meier estimates were calculated for each nomogram probability group in 200 bootstrap samples. The mean 5-year predicted KaplanMeier estimate and 95% confidence interval were calculated. These observed rates were plotted against the predicted outcomes in a nomogram calibration curve. The HosmerLemeshow goodness-of-fit test was used to assess the CR nomogram. All analyses were performed with STATA 11 (StataCorp, College Station, TX). P<.05 denoted statistical significance. All reported P values are 2 sided.
Methods and Materials From 1986 to 2009, 325 patients were managed with selective bladder preservation at MGH and had complete clinical data adequate for nomogram development. Eligible patients had clinical stage T2NXM0 to T4aNXM0, urothelial carcinoma of the bladder and were medically operable candidates for cystectomy. The majority of these patients were treated on a succession of prospective institutional and RTOG bladder-sparing protocols. Treatment began with a TURBT and pathologic confirmation of muscle-invasive bladder cancer. Induction therapy consisted of radiation to a dose of 40 Gy with concurrent cisplatin-based chemotherapy. After induction therapy, patients underwent a midtreatment cystoscopic assessment with tumor site biopsy. Those with a CR continued on to consolidation chemoradiation consisting of approximately 24 Gy with concurrent chemotherapy. Those with a less-than-complete response were advised to undergo a prompt cystectomy. These bladder-sparing protocols, including their respective eligibility criteria, chemotherapy regimens, and radiation techniques, and this cohort of patients have been previously described (3). In general, all patients were suitable for cystectomy but motivated to pursue bladder preservation. Patients with medical contraindications to the specific chemotherapy regimen used or obvious radiographic evidence of distant metastasis and/or lymph node involvement were generally excluded. Hydronephrosis has become a factor for exclusion from participation in more recent trials, but no other prognostic feature has been systematically used as an exclusion factor for this approach.
Results The characteristics of the 325 patients used to construct our nomograms are shown in Table 1. The median age was 66 years, and the median follow-up for surviving patients was 9.8 years, with 82 patients having greater than 10 years of follow-up.
Volume 86 Number 2 2013 Table 1
Nomograms predicting outcome after bladder-sparing therapy
Patient characteristics Data
Age at start of therapy [median (range)] (y) Follow-up (y) N Sex Male Female Clinical T stage 2 3a 3b 4 Hydronephrosis Yes No TURBT status Complete Incomplete Grade Intermediate High
66.3 (27.3-88.6) 4.6 (0.0-23.3) 325 240 85 176 106 20 23 53 272 215 110 45 280
Abbreviation: TURBT Z transurethral resection of bladder tumor.
CR to induction therapy The overall CR rate was 70%. A nomogram for CR was constructed using a logistic regression model. The final model included hydronephrosis, TURBT completeness, age, and sex as covariates. This nomogram is displayed in Fig. 1a. The performance of the nomogram was good, with an optimism-corrected area under the ROC curve of 0.68 (Fig. 1b). A calibration plot, shown in Fig. 1c, shows the predictive ability of the CR nomogram. The Hosmer-Lemeshow goodness-of-fit test shows that the model fits the data well (PZ.5).
Disease-specific survival The overall 5-yr DSS was 63%. The DSS nomogram was constructed by use of a Cox regression model. The final model included hydronephrosis, grade, and clinical T stage as covariates. It is displayed in Fig. 2a. The optimism-corrected Harrell concordance index was 0.60. A calibration plot for this nomogram is shown in Fig. 2b.
Bladder-intact disease-free survival The overall 5-year BI-DFS rate was 41%. The BI-DFS nomogram was also constructed using a Cox regression model. The final model included hydronephrosis, TURBT completeness, clinical T stage, and age as covariates. It is displayed in Fig. 3a. The optimism-corrected Harrell concordance index was 0.60. A calibration plot for this nomogram is shown in Fig. 3b.
Discussion Bladder-preserving trimodality therapy, when used selectively, results in rates of DFS comparable to contemporary radical
313
cystectomy with a high rate of bladder preservation (1-4). Increasingly careful selection of patients leads to higher rates of CR to induction therapy and consequently higher rates of bladder preservation. The prognostic features associated with favorable outcomes have previously been defined (3). In this singleinstitution series of well-followed patients treated with bladderpreserving trimodality therapy, we have developed a prognostic nomogram predicting rates of CR, DSS, and likelihood of remaining free of both recurrent bladder cancer and cystectomy. Nomograms are predictive models offering individualized estimates of disease-related outcomes based on multiple prognostic features. They have been shown to have greater predictive accuracy than traditional risk group assignments in a variety of clinical settings (5, 8-12). Their intuitive construction and ease of interpretation make them favored by clinicians and patients for individual decision making. In genitourinary oncology, nomograms are in widespread use for prostate cancer, but there is also a precedent in bladder cancer. The International Bladder Cancer Nomogram Consortium developed a postoperative nomogram predicting risk of recurrence after radical cystectomy (13). These investigators constructed an international multi-institutional database of over 9000 patients from which they developed a nomogram predicting risk of recurrence 5 years after cystectomy. The variables included in the nomogram were sex, age, pathologic T stage, histology, pathologic nodal status, grade of disease, and time from diagnosis to cystectomy. The predictive accuracy of the nomogram was significantly better than that of the American Joint Committee on Cancer TNM staging system or standard pathologic subgroupings. Although a postoperative nomogram for patients after radical cystectomy is certainly a useful prognostic tool, a pretreatment nomogram for patients considering either selective bladder preservation or immediate cystectomy would be of great value not just to predict outcome but also to help physicians and patients make important therapeutic decisions. Currently, patients may be steered away from bladder preservation based on risk group stratification often on the basis of a single adverse feature. Nomograms allowing more individualized estimates of CR, freedom from recurrence, and likelihood of bladder preservation may prompt patients or consultants to reconsider who is or is not a good candidate for bladder preservation rather than making this clinical decision based on a single prognostic feature. Some patients with a single poor prognostic feature may be viable candidates, whereas other patients with no one striking poor feature but with many intermediate features may be poorer candidates than previously appreciated. The potential improvement in predictive accuracy in this setting would not only improve patient selection and prognostication but also afford more rational clinical decision making. We developed nomograms predicting rates of CR to induction therapy, DSS, and BI-DFS. These represent 3 important therapeutic goals. A CR to induction therapy is required to complete the course of chemoradiation and avert an immediate cystectomy. DSS assesses the likelihood of dying of bladder cancer, an obviously critical endpoint. In addition, because the vast majority of patients with recurrent bladder cancer after radical cystectomy succumb to their cancer in fairly short order, we believe that our DSS nomogram offers a reasonable comparison to the postcystectomy nomogram in terms of disease-related outcome with bladder sparing as opposed to radical cystectomy (4). Finally, the BI-DFS endpoint assesses both goals of selective bladder preservation. The primary goal is the same as that after radical cystectomy, which is to eradicate the bladder cancer, whereas bladder
314
Coen et al.
International Journal of Radiation Oncology Biology Physics
Fig. 1. (a) Complete response (CR) nomogram. (b) Receiving operating characteristic (ROC) curve for CR model. (c) Calibration plot for CR nomogram. TURBT Z transurethral resection of bladder tumor.
preservation is a secondary goal when it can be safely accomplished. Just as the international postcystectomy nomogram was modeled to predict disease-free survival, we also modeled the same events (13). Superficial bladder recurrence was not included as an event because this is usually managed conservatively with a TURBT and intravesical therapy with no demonstrable increase in risk of bladder cancer-related death (14, 15). Any other bladder cancer recurrence was included as an event including invasive local recurrence in the bladder, pelvic recurrence, regional or nonregional lymph node recurrence, distant recurrence, or bladder cancer-related death. In addition, because selective bladder preservation has the preservation of the bladder as a secondary goal, cystectomy was also included as an event. The 5-year time point was chosen for DSS and BI-DFS because there were few events beyond 5 years, reflecting the early recurrence pattern of muscleinvasive bladder cancer (3, 4, 16). Although the database used to develop our nomogram is the largest US bladder preservation series, it is still a relatively small single-institution series as compared with the large multiinstitutional databases used to develop prognostic nomograms for men receiving various forms of therapy for prostate cancer, women with breast cancer, or men and women managed by radical
cystectomy for bladder cancer (10, 13, 17-19). Often, the variables selected for inclusion in a nomogram are preselected based on known prognostic features in the particular clinical scenario being modeled. The development process does not include subsequent elimination of nonpredictive variables. In our more limited but robust dataset, we thought it wise to only include variables that were both predictive in this dataset and clinically sensible. Despite the limited number of variables in the final models, the predictive accuracy was good and was superior to risk group stratification using any single one of these variables. Limiting the number of variables also decreases the likelihood of over-fitting the model to our dataset, making it more applicable to future patients treated with similar selective bladder-preservation regimens. Future directions include validating this nomogram both prospectively and in other pre-existing bladder-sparing datasets. Developing a multi-institutional international database may be a fruitful effort because bladder-preserving therapy is more common outside of the United States. Inclusion of molecular markers, such as MRE11, predicting response to therapy may also allow for the more precise identification of patients well or poorly suited to this approach (20). These efforts are under way both at our institution and through the RTOG. Better predictive models
Volume 86 Number 2 2013
Fig. 2.
Nomograms predicting outcome after bladder-sparing therapy
315
(a) Nomogram and (b) calibration plot for disease-specific survival (DSS) at 5 years. C Z concordance index; K-M Z Kaplan-Meier.
may allow not only improved selection of motivated patients but also wider use of this treatment paradigm through greater confidence in the ultimate outcome. In their current form, our nomograms are quite informative; however, the manner in which they are used to guide clinical decisions must be tailored to an individual patient’s disease features, overall health, bladder function, and strength of motivation toward bladder preservation. Still, some general observations can be made regarding the construction of these nomograms and some general recommendations made regarding their interpretation. The following generalizations can be inferred by inspection of the 3 nomograms. Hydronephrosis is predictive of DSS and CR. Clinical T stage is predictive of DSS but not CR. Consistent with prior publications, extent of TURBT is predictive of CR but not DSS (3). All 3 of these variables made strong contributions to the BI-DFS nomogram. Other variables had lesser but important roles. Age predicted for CR and BI-DFS, sex for CR, and grade for DSS. Application of the nomograms is more complex. Because there is no clear survival decrement to selective bladder preservation as
opposed to immediate cystectomy, the DSS nomogram cannot be used to decide for or against bladder sparing. Comparison with the surgical nomogram, before making a therapeutic decision, is not possible because it is a postcystectomy nomogram incorporating pathologic features only known after surgery (13). Although the BI-DFS nomogram is useful in giving a patient a sense of the likelihood of being disease free with an intact bladder 5 years after treatment, it is difficult to define a decisive cutoff value. Patients should be reminded that the BI-DFS rate after cystectomy is, by definition, 0%. The CR nomogram, on the other hand, may allow patients to make an informed therapeutic choice. Although patients and physicians may differ on a precise cutoff value, most define the threshold in the 60%-70% range, meaning few would advocate for patients with a chance of response of less than 60% and few, except those not generally in favor of bladder preservation, would argue that 70% was unacceptably low. Inspection of the CR nomogram shows that any patient with 3 or more favorable features would exceed this threshold. In addition, any patient without hydronephrosis who has a visibly complete resection would also exceed
Fig. 3. (a) Nomogram and (b) calibration plot for bladder-intact disease-free survival (BI-DFS) at 5 years. C Z concordance index; K-M Z Kaplan-Meier; TURBT Z transurethral resection of bladder tumor.
316
Coen et al.
this threshold even in the absence of a third favorable feature. Still, validation of these nomograms prospectively and in other preexisting datasets is in order. Caution must be exercised when using these models to inform a patient’s therapeutic decisions, especially with regard to the application of stringent cutoffs. They should be used within the context of a specific patient’s care, giving other pertinent clinical issues due respect.
Conclusion Selective bladder preservation with trimodality therapy is an established management strategy for muscle-invasive bladder cancer. Although the prognostic features for patients pursuing this therapy have been defined previously, more individualized estimates of outcome are desired. We developed nomograms that integrate predictive features to provide individualized estimates of likelihood of CR to treatment, DSS, and likelihood of both bladder preservation and freedom from recurrence (BI-DFS). These tools should prove valuable to muscle-invasive bladder cancer patients by enabling them, in conjunction with their physicians, to make more rationally important decisions about their treatment.
References 1. Rodel C, Weiss C, Sauer R. Trimodality treatment and selective organ preservation for bladder cancer. J Clin Oncol 2006;24:5536-5544. 2. Shipley WU, Kaufman DS, Tester WJ, et al. Overview of bladder cancer trials in the Radiation Therapy Oncology Group. Cancer 2003; 97:2115-2119. 3. Efstathiou JA, Spiegel DY, Shipley WU, et al. Long-term outcomes of selective bladder preservation by combined-modality therapy for invasive bladder cancer: The MGH experience. Eur Urol 2012;61:705-711. 4. Zehnder P, Studer UE, Skinner EC, et al. Super extended versus extended pelvic lymph node dissection in patients undergoing radical cystectomy for bladder cancer: A comparative study. J Urol 2011;186: 1261-1268. 5. Kattan MW, Scardino PT. Evidence for the usefulness of nomograms. Nat Clin Pract Urol 2007;4:638-639. 6. Kattan MW. Nomograms are superior to staging and risk grouping systems for identifying high-risk patients: Preoperative application in prostate cancer. Curr Opin Urol 2003;13:111-116.
International Journal of Radiation Oncology Biology Physics 7. Sternberg CN. Are nomograms better than currently available stage groupings for bladder cancer? J Clin Oncol 2006;24:3819-3820. 8. Steyerberg EW, Roobol MJ, Kattan MW, et al. Prediction of indolent prostate cancer: Validation and updating of a prognostic nomogram. J Urol 2007;177:107-112. discussion 112. 9. Stephenson AJ, Scardino PT, Eastham JA, et al. Postoperative nomogram predicting the 10-year probability of prostate cancer recurrence after radical prostatectomy. J Clin Oncol 2005;23:7005-7012. 10. Rudloff U, Jacks LM, Goldberg JI, et al. Nomogram for predicting the risk of local recurrence after breast-conserving surgery for ductal carcinoma in situ. J Clin Oncol 2010;28:3762-3769. 11. Rouzier R, Pusztai L, Delaloge S, et al. Nomograms to predict pathologic complete response and metastasis-free survival after preoperative chemotherapy for breast cancer. J Clin Oncol 2005;23:8331-8839. 12. Kattan MW, Yu C, Salomon L, et al. Development and validation of preoperative nomogram for disease recurrence within 5 years after laparoscopic radical prostatectomy for prostate cancer. Urology 2011; 77:396-401. 13. Bochner BH, Kattan MW, Vora KC. Postoperative nomogram predicting risk of recurrence after radical cystectomy for bladder cancer. J Clin Oncol 2006;24:3967-3972. 14. Weiss C, Wittlinger M, Engehausen DG, et al. Management of superficial recurrences in an irradiated bladder after combinedmodality organ-preserving therapy. Int J Radiat Oncol Biol Phys 2008;70:1502-1506. 15. Zietman AL, Grocela J, Zehr E, et al. Selective bladder conservation using transurethral resection, chemotherapy, and radiation: Management and consequences of Ta, T1, and Tis recurrence within the retained bladder. Urology 2001;58:380-385. 16. Stein JP, Lieskovsky G, Cote R, et al. Radical cystectomy in the treatment of invasive bladder cancer: Long-term results in 1,054 patients. J Clin Oncol 2001;19:666-675. 17. Eggener SE, Scardino PT, Walsh PC, et al. Predicting 15-year prostate cancer specific mortality after radical prostatectomy. J Urol 2011;185: 869-875. 18. Potters L, Roach M III, Davis BJ, et al. Postoperative nomogram predicting the 9-year probability of prostate cancer recurrence after permanent prostate brachytherapy using radiation dose as a prognostic variable. Int J Radiat Oncol Biol Phys 2010;76:1061-1065. 19. Zelefsky MJ, Kattan MW, Fearn P, et al. Pretreatment nomogram predicting ten-year biochemical outcome of three-dimensional conformal radiotherapy and intensity-modulated radiotherapy for prostate cancer. Urology 2007;70:283-287. 20. Choudhury A, Nelson LD, Teo MT, et al. MRE11 expression is predictive of cause-specific survival following radical radiotherapy for muscle-invasive bladder cancer. Cancer Res 2010;70:7017-7026.