R. M. Bukowski has received research funding from Pfizer and Wyeth; consulting fees from Pfizer, Novartis, Wyeth, Genentech, and Antigenics; and honoraria from Pfizer, Wyeth, Genentech, and Onyx Pharmaceuticals.
Robert A. Figlin MD,
Division of Medical Oncology and Therapeutics Research, City of Hope National Medical Center, Duarte, California
Editorial assistance was provided by ACUMED (Tytherington, United Kingdom) and was funded by Pfizer.
In a randomized, phase 3 trial, sunitinib demonstrated superior efficacy over interferon-alfa as first-line therapy in patients with metastatic clear-cell renal cell carcinoma (RCC). On the basis of outcome data from that trial, the authors developed a nomogram for predicting the probability of 12-month progression-free survival for patients who received sunitinib therapy.
Three-hundred seventy-five patients who received sunitinib in the phase 3 trial were the subject of the current analysis. Nomogram pretreatment predictor variables included corrected serum calcium levels, the number of metastatic sites, hemoglobin levels, prior nephrectomy, the presence of lung and liver metastases, thrombocytosis, Eastern Cooperative Oncology Group performance status, time from diagnosis to treatment, and serum levels of alkaline phosphatase and lactate dehydrogenase. Investigator-assessed progression-free survival was the predicted outcome endpoint. Internal validation of the nomogram consisted of quantification of the discrimination with the concordance index and assessment of calibration.
One-hundred seventy-four of 375 patients (46%) who received sunitinib achieved an objective response, and the median progression-free survival was 10.8 months (95% confidence interval, 10.6-12.6 months). A nomogram for predicting the probability of 12-month progression-free survival for patients who received sunitinib therapy was constructed on the basis of a Cox regression model from 11 parameters that were determined before treatment. The concordance index was 0.633.
Sunitinib malate (SUTENT; Pfizer, New York, NY) is a multitargeted tyrosine kinase inhibitor of vascular endothelial growth factor receptors and platelet-derived growth factor receptors.1 Its inhibitory profile provided the rationale for its study as an antiangiogenesis targeted therapy for clear-cell renal cell carcinoma (RCC).2, 3 Objective response rates of 40% were observed with sunitinib as second-line treatment in patients with cytokine-refractory, metastatic RCC.4, 5 The high response rate led to the conduct of a randomized phase 3 trial comparing sunitinib with interferon-alfa (IFN-α) as first-line, systemic therapy. The results of a preplanned interim analysis showed statistically significant improvements in progression-free survival and objective response rates with sunitinib compared with IFN-α.6 As a result of this trial, sunitinib is a new reference standard for the first-line treatment of metastatic RCC.
Prognostic factors are used in clinical trial design and interpretation, risk-directed treatment, and patient counseling. Pretreatment features predictive of survival in patients with metastatic RCC were examined previously, and the predictive models that were developed have been applied to the conduct of clinical trials.7–11 One model, which was developed at Memorial Sloan-Kettering Cancer Center (MSKCC), groups patients into categories (favorable, intermediate, and poor) according to the number of risk factors predictive of a short survival.7 The risk factors are: time <1 year from diagnosis to the start of systemic therapy, elevated lactate dehydrogenase and corrected serum calcium levels, anemia, and low performance status (PS).7 This model was validated independently by investigators at the Cleveland Clinic12 and has been used in phase 3 trial design and interpretation of the new, targeted agents.6, 10, 13
The MSKCC model defines risk categories, or groupings of patients, according to predicted outcome.7 It was developed with and comprised of patients who were treated with cytokines or chemotherapeutic agents, and the outcome assessed was overall survival. Because molecularly targeted therapy has replaced outpatient cytokine therapy in the first-line treatment of metastatic RCC, there is a need to reassess clinical and biologic features that are predictive of outcome. Moreover, with the identification of multiple agents now available for RCC, including temsirolimus,13 bevacizumab,14 and sorafenib,10 the development of tools that individualize patient prognostication may be helpful in directing individualized therapy.
After an update of the reported efficacy data from the randomized phase 3 trial,15 we performed an analysis of patients who were treated on the sunitinib arm of this trial to investigate the correlation between pretreatment clinical features and progression-free survival. As a result of this analysis, a predictive model, or nomogram, was developed that predicted the probability of achieving 12-month progression-free survival with sunitinib as first-line treatment for metastatic clear-cell RCC. Nomograms have been developed and validated for use in numerous malignant diseases, including prostate cancer and soft tissue sarcoma,16, 17 as well as localized RCC for use in predicting disease recurrence after nephrectomy.18 To our knowledge, to date, this is the first nomogram that has been developed specifically to predict the outcome of systemic therapy for metastatic clear-cell RCC.
MATERIALS AND METHODS
The patient population for this analysis was comprised of 375 patients with metastatic RCC who were treated with sunitinib in the randomized phase 3 trial.6 The eligibility, methods, treatment plan (1:1 randomization of sunitinib or IFN-α), and outcome for this phase 3 trial have been reported previously.6 Key eligibility criteria included metastatic RCC with a clear-cell component, measurable disease, no prior systemic therapy, adequate PS and organ function, and signed informed consent. Sunitinib was administered orally in repeated 6-week cycles at 50 mg daily for 4 weeks on treatment followed by 2 weeks off treatment (Schedule 4/2). Response and progression-free survival outcome data for this analysis were updated from the published report6 using a data cutoff of February 2007. The study was approved by the institutional review board or ethics committee at each participating center. All patients provided written informed consent.
The primary endpoint for this analysis of prognostic factors to sunitinib therapy was investigator-assessed progression-free survival. This endpoint was defined as the time from randomization to the sunitinib arm of the trial to the first documentation of disease progression or death from any cause, whichever occurred first. The time to progression-free event was estimated using the Kaplan-Meier method.19 A multivariate Cox proportional hazards regression model was the basis of the nomogram. Points for variables in the nomogram were determined by their Cox model coefficients. The proportional hazards assumption of the Cox model was verified by tests of correlation with time. The variables that we used in the nomogram were selected on the basis of knowledge of their prognostic significance from previous reports and were examined as either categorical or continuous variables.7–9, 11, 12, 20 The following baseline variables were examined: corrected serum calcium levels, the number of metastatic sites, hemoglobin levels (greater than or equal to the lower limit of normal [≥LLN] vs <LLN), prior nephrectomy, the presence of lung and liver metastases, an Eastern Cooperative Oncology Group (ECOG) PS of 0 versus 1, thrombocytosis (defined as a platelet count >400,000/μL), time from diagnosis to treatment, and the ratios of the serum levels of both alkaline phosphatase and lactate dehydrogenase to the upper limits of normal for each. No other variables were assessed. Dose reductions or interruptions were not accounted for in the construction of the nomogram.
All decisions with respect to the categorization of variables were made before modeling. Continuous variables were modeled using restricted cubic splines to accommodate potentially nonlinear effects.21 No variables were omitted for a lack of statistical significance in either univariate or multivariate analysis because of the deleterious effect this would have on predictive accuracy.21
Nomogram validation comprised 2 activities. First, discrimination was quantified with the concordance index.22 Similar to the area under the receiver-operating characteristic curve, but appropriate for censored data, the concordance index provides the probability that, in a randomly selected pair of patients in which 1 patient experiences an event (disease progression or death, whichever is first) before the other, the patient who experiences an event had the worse predicted outcome from the nomogram. We used bootstrapping with 200 resamples to obtain a relatively unbiased estimate.
Second, calibration was assessed. This was done by plotting the mean nomogram-predicted progression-free survival probabilities versus the observed Kaplan-Meier estimate of progression-free survival. Again, bootstrapping was used. All analyses were performed using S-plus 2000 Professional software (Statistical Sciences, Seattle, Wash) with the Design and Hmisc libraries added.23
Descriptive statistics for the baseline pretreatment patient features studied are described in Table 1. All patients were treated with sunitinib, and the median duration of treatment was 11 months (range, from <1 month to 31 months). In total, 101 patients (27%) were ongoing on sunitinib treatment at the time of data cutoff for this analysis. The reasons for discontinuation in the remaining patients were progressive disease (n = 199 patients; 53%), adverse events (n = 56 patients; 15%), and consent withdrawal or other (n = 19 patients; 5%).
Table 1. Descriptive Statistics for the Sunitinib Cohort (N=375)
According to an investigator assessment of Response Evaluation Criteria in Solid Tumors-defined tumor response, 174 patients (46%) achieved an objective response (169 partial responses and 5 complete responses). The best response among the remaining patients was stable disease (n = 152 patients; 41%) and progressive disease or not evaluable (n = 49 patients; 13%). The median progression-free survival was 10.8 months (95% confidence interval, 10.6-12.6 months) (Fig. 1).
In the Cox model, corrected serum calcium levels (P = .01), the number of metastatic sites (P < .01), the presence of hepatic metastases (P = .03), thrombocytosis (P < .01), serum lactate dehydrogenase levels (P < .01), and the time from diagnosis to treatment (P < .01) were associated with progression-free survival; whereas hemoglobin levels (≥LLN vs <LLN; P = .82), prior nephrectomy (P = .37), ECOG PS (0 vs 1; P = .22), the presence of lung metastases (P = .74), and serum alkaline phosphatase levels (P = .82) were not associated with progression-free survival. A nomogram for predicting the probability of 12-month progression-free survival with sunitinib therapy was constructed from the 11 parameters that were determined before treatment (Fig. 2).
Statistically insignificant variables were not omitted from the model because of the resultant bias on the remaining predictors and the subsequent deleterious effect on predictive accuracy.21 The concordance index for the model was 0.633. Figure 3 illustrates the calibration of the nomogram, which appears to be accurate.
Figure 4 provides examples of calculations using our nomogram for 2 patients in which very different prognoses are represented. For the first patient (Fig. 4a), the probability of 12-month progression-free survival on sunitinib is approximately 70%; whereas, for the second patient (Fig. 4b), the probability is approximately 10%. Both patients were treated on the phase 3 trial. Patient 1 remained on sunitinib therapy and was progression free at ≥24 months. Patient 2 experienced progressive disease after 3 months of treatment with sunitinib.
The result of the randomized phase 3 trial comparing sunitinib with IFN-α established sunitinib as a new, standard, first-line treatment for metastatic clear-cell RCC.6 By using updated outcome data from this trial, we developed and internally validated a nomogram for predicting the probability of 12-month progression-free survival for sunitinib as first-line therapy. An advantage of the nomogram is that it may provide a more individualized prognostication than models that categorize patients into risk groups.
Over the course of the past few years, there has been a shift in the medical management of patients with metastatic RCC from cytokines to agents that target the inhibition of proangiogenic growth factor activity. Favorable results of phase 3 trials with sunitinib and several other agents, including bevacizumab in combination with IFN-α (first-line therapy),14 temsirolimus (first-line therapy in poor-risk patients),13 and sorafenib (as second-line therapy),10 have resulted in a new treatment paradigm for the management of metastatic RCC.25 Moreover, with the identification of multiple targeted agents that have demonstrated benefit, now, there may be an opportunity to individualize patient treatment and to study the combination and sequencing of these new targeted agents.
In the setting of multiple treatment options, improving the accuracy of current prognostic estimates is important for making treatment recommendations in metastatic RCC. Issues, including efficacy, patient quality of life, and other ancillary effects of treatment, influence clinical decision-making, requiring that physicians more accurately assess the risk versus benefit of treatment. In response to this need, the nomogram described in this report is a tool that was developed specifically to predict patient outcome after treatment with sunitinib, a targeted agent that has demonstrated efficacy against metastatic clear-cell RCC.
Consistent with the development of nomograms for other malignancies, we included a wide variety of clinical features to develop a model that best represented the patients' baseline clinical features. The pretreatment factors that have the greatest influence on this nomogram's predictions are elevated lactate dehydrogenase levels, corrected serum calcium levels, the number of metastatic sites, and the time from diagnosis to treatment. These clinical features had previously demonstrated prognostic value in the earlier models comprised of patients who were treated with cytokines or cytotoxics.7, 12
Like any multivariate analysis in which multiple predictors are undergoing simultaneous adjustment, interpretations can be difficult or counterintuitive. From the nomogram, it may appear that a prior nephrectomy contributes to a poor outcome, in that patients who underwent prior nephrectomy appear to have a lower prediction of 12-month progression-free survival. Only 10% of patients in the study did not undergo prior nephrectomy, and, as may be expected, the progression-free survival outcome was more favorable for patients with prior nephrectomy according to the univariate analysis (median progression-free survival: 11 months vs 6 months with vs without prior nephrectomy, respectively; P = .09). The important point to remember when interpreting this axis of the nomogram is that other factors in the nomogram are being considered. In reality, it is artificial to vary a patient on 1 axis, holding all others constant (eg, sliding a patient from ‘No’ to ‘Yes’ on the Prior nephrectomy axis), because it is likely that he also will move on other axes. Thus, a patient who underwent prior nephrectomy may have other favorable characteristics, which could result in a prognosis better than that of the typical patient who has not undergone prior nephrectomy. Most important, based on the data from which it was created, this nomogram would be applied to patients at the time of evaluation for sunitinib therapy, which would occur after the completion of standard diagnostic testing and appropriate surgical management.
The next step in developing this nomogram is to assess external validation, as has been done with similar nomograms constructed for other diseases. This assessment will require the use of a similar patient cohort treated with sunitinib. One clinical trial that may be used to provide such a patient database is a randomized phase 2 trial of nearly 300 patients comparing sunitinib administered either on a continuous dosing regimen at 37.5 mg per day versus Schedule 4/2 at 50 mg per day in treatment-naive patients with metastatic clear-cell RCC.26 The nomogram was developed through analysis of a dataset comprised of patients who meet eligibility criteria for a phase 3 trial.
We recognize the limitation of predictive criteria based on pretreatment clinical features. Responses to targeted therapy for other malignancies with agents like gefitinib and imatinib mesylate have been linked to specific mutations in the tumors.27, 28 The identification of tumor-specific, predictive-specific features for sunitinib and other targeted therapies in metastatic RCC would be of great value and currently is being pursued.29
In conclusion, we have developed an internally validated nomogram for predicting the probability of 12-month progression-free survival with sunitinib. To our knowledge, it is the first such tool for use with systemic therapy in metastatic clear-cell RCC. With the identification of multiple targeted agents for metastatic RCC, the development of tools that individualize patient prognostication may be helpful for counseling patients, planning follow-up, and directing therapy. The nomogram would be applied to patients at the time of evaluation for sunitinib therapy, after the completion of standard diagnostic testing and appropriate surgical management. Independent validation of the nomogram and additional studies to identify tumor-specific prognostic factors are warranted.
We thank all of the patients and their families for their participation in this study, investigators and their staff from participating sites, and ACUMED of Tytherington, United Kingdom, for editorial assistance.