Fill and Sign the Curriculum for Industrial Stitching Machine Operator Navttc Form

U.S. Department of Health and Human Services Food and Drug Administration Center for Drug Evaluation Research Office of Pharmacoepidemiology and Statistical Science Office of Biostatistics STATISTICAL REVIEW AND EVALUATION CLINICAL STUDIES NDA /Serial Number: 21-661/N000 Drug Name : RSR13 (efaproxiral) Injection (75 or 100 mg/kg) Applicant: Allos Therapeutics Inc. Indication(s): Brain metastases Date(s): Submission Date: December 4, 2003 PDUFA Date: June 4, 2004 ODAC Meeting Date: May 3, 2004 Review Completion Date: Review Priority: Priority Biometrics Division: Division of Biometrics I (HFD-710) Statistical Reviewer: Rajeshwari Sridhara, Ph.D. Concurring Reviewer: Kooros Mahjoob, Ph.D., Acting Director Medical Division: Oncology Drug Products (HFD-150) Clinical Team: Kevin Ridenhour, M.D. & Ramzi Dagher, M.D. Project Manager: Ms. Christy Cottrell Keywords: Active control/superiority, log-rank test, post-hoc analysis, subgroup analyses, multiple endpoints, multiple comparisons, Cox regression Table of Contents 1 Executive Summary ....................................................................................... 1 1.1 Conclusions and Recommendations....................................................... 1 1.2 Brief Overview of Clinical Studies......................................................... 1 1.3 Statistical Issues and Findings................................................................ 1 2 Introduction .................................................................................................... 4 2.1 Overview................................................................................................... 4 2.1.1 Background ......................................................................................... 4 2.1.2 Major Statistical Issues ....................................................................... 5 2.2 Data Sources............................................................................................. 5 3 Statistical Evaluation ..................................................................................... 6 3.1 Evaluation of Efficacy ............................................................................. 6 3.1.1 Study RT009 ....................................................................................... 6 3.1.1.1 Study Design ................................................................................ 6 3.1.1.2 Treatment Administration ............................................................ 8 3.1.1.3 Study Objectives........................................................................... 9 3.1.1.4 Efficacy Endpoints ....................................................................... 9 3.1.1.5 Sample Size Considerations ....................................................... 10 3.1.1.6 Interim Analysis ......................................................................... 11 3.1.1.7 Efficacy Analysis Methods ........................................................ 12 3.1.1.8 Sponsor’s Results and Statistical Reviewer’s Findings/ Comments ................................................................................................... 14 3.1.1.8.1 Baseline Characteristics......................................................... 15 3.1.1.8.2 Primary Efficacy Analyses .................................................... 18 3.1.1.8.3 Exploratory Covariate Adjusted and Subgroup Survival Analyses 21 3.1.1.8.4 Secondary Efficacy Analyses ................................................ 31 3.2 Evaluation of Safety............................................................................... 33 4 Findings in Special/Subgroup Populations ................................................ 33 4.1 Gender, Race and Age........................................................................... 33 4.2 Other Special/Subgroup Populations................................................... 35 5 Summary and Conclusions .......................................................................... 36 5.1 Statistical Issues and Collective Evidence ........................................... 36 5.2 Conclusions and Recommendations..................................................... 37 APPENDICES ..................................................................................................... 38 Appendix 1: List of Ineligible Patients ......................................................... 38 Appendix 2: List of Patients Who Were Misclassified at Randomization 39 Appendix 3: List of Patients Who Were In-evaluable (Withdrew from Study Prior to Treatment) .............................................................................. 40 Appendix 4: Survival Analysis Before Addition of Co-primary Hypothesis ........................................................................................................................... 41 Appendix 5: Interim Analysis Results .......................................................... 42 Appendix 6: Exploratory Covariate Adjusted Survival Analyses in NSCLC/Breast Primary .................................................................................. 43 Appendix 7: Exploratory Covariate Adjusted Survival Analysis in Breast Primary ............................................................................................................. 44 ii 1 1.1 Executive Summary Conclusions and Recommendations In this reviewer's opinion the registration study failed to demonstrate improved survival of RSR13 + whole brain radiation therapy (WBRT) over WBRT alone for patients with brain metastases. It is not evident that the apparent survival advantage observed in a single small subgroup of patients with primary breast cancer based on post-hoc analysis is attributable solely to the treatment effect and not due to imbalances in known and unknown prognostic factors. Therefore, the evidence submitted in this application based on results from a single trial, is not convincing and does not support the sponsor’s claim of efficacy in a subgroup of patients with breast cancer primary. 1.2 Brief Overview of Clinical Studies The sponsor has submitted results from one phase III, comparative clinical trial (registration trial Study RT009) comparing WBRT alone to RSR13 + WBRT, to demonstrate efficacy of RSR13. The sponsor has also provided supportive efficacy data from a phase II, single arm study (Study RT008). The main focus of this review is on results from Study RT009. Study RT009 was a multicenter international study conducted in patients with brain metastases. This study was initiated on February 16, 2000 and the study was completed on September 24, 2002. The data cut-off date for final efficacy analysis was January 31, 2003. Study RT009 was a phase III randomized, open-label, comparative study conducted in 538 patients from 82 international centers, who would be receiving a standard 2-week (10-day) course of WBRT for brain metastases. Patients were randomized (1:1) to receive RSR13 no longer than 30 minutes prior to daily WBRT or WBRT alone. Patients were stratified at randomization to 4 strata: (1) Radiation Therapy Oncology Group (RTOG) recursive partitioning analysis (RPA) Class I (including non-small cell lung, breast, and other primary cancers), (2) RPA Class II non-small cell lung cancer (NSCLC) primary, (3) RPA Class II breast cancer primary, and (4) RPA Class II primary tumors of various origins (hereafter referred as other primary). 1.3 Statistical Issues and Findings This NDA submission is to support administration of RSR13 as an adjunct to whole brain radiation therapy (WBRT) for patients with brain metastases from primary breast cancer. In this NDA submission, study RT009 is the only 1 randomized pivotal study conducted for the efficacy and safety of RSR13. This open-label study was designed to evaluate the efficacy and safety of combined therapy with RSR13 + WBRT versus WBRT alone in patients with brain metastases. This study enrolled a total of 538 patients with 267 patients who received WBRT alone and 271 patients who received RSR13 + WBRT. The primary efficacy endpoint of this study was survival. Statistical Issues: 1. Only one randomized open-label study conducted in patients with brain metastases, which failed to demonstrate efficacy as per the design of the study, in the intent-to-treat population (log-rank test, P-value = 0.1688) and in the co-primary subgroup of patients with NSCLC/Breast cancer primary (logrank test, P-value = 0.1217). 2. When the overall result fails to show efficacy, usually subgroup findings are not acceptable and subgroup analyses at best can be exploratory or hypothesis generating analyses (ICH E-3 guidelines, section 11.4.2.8: These analyses are not intended to "salvage" an otherwise non-supportive study but may suggest hypotheses worth examining in other studies or be helpful in refining labelling information, patient selection, dose selection etc.). When one starts to do multiple subgroups testing, one can easily make a false positive claim based on such subgroup analysis. We do not know how to interpret the P-values based on such post-hoc analysis. Furthermore, without replication of the results in a second well-controlled study, the subgroup analysis can not be ruled out for a false positive result. 3. The sponsor wishes to claim approval based on a subgroup of patients with primary breast cancer. This subgroup hypothesis corresponding to breast cancer primary patients was not stated as a hypothesis of interest to be tested in the original protocol. Any subgroup hypothesis needs to be stated in the protocol and accordingly proper allocation of α has to be specified. Otherwise, such post-hoc subgroup claim will inflate Type I error and it is difficult to interpret such P-values. 4. Some of the important issues not addressed by the sponsor are: imbalance in patients who were ineligible (per protocol) between the two treatment groups; misclassification of patients in the randomized strata; imbalance in the number of baseline brain lesions in the subgroup of patients with primary breast cancer. Findings: The protocol specified primary analysis was unadjusted log-rank test in the intentto-treat (ITT) population to compare overall survival between the two treatment 2 arms. This study failed to demonstrate survival benefit as presented in the following Table A. Table A: Primary Efficacy Survival Analysis in ITT Population Treatment WBRT RSR13 + WBRT Number of Deaths 221/267 220/271 Median Survival in Months1 (95% C.I.) 4.5 (3.7, 5.4) 5.3 (4.5, 6.2) Hazard Ratio 2 (95% C.I.) 0.877 (0.727, 1.057) P-value 3 0.1688 1 : Kaplan-Meier Estimates; 2: Hazard Ratio of RSR13 + WBRT/ WBRT; 3: unadjusted log-rank test. The sponsor amended the protocol during the course of study to include a coprimary hypothesis, to test survival difference between the two treatment arms in a subgroup of patients with NSCLC or Breast primary cancer. The results of these comparisons also failed to demonstrate survival benefit as presented in Table B below. Table B: Co-Primary Efficacy Survival Analysis in NSCLC/Breast Primary Cancer Subgroup* Treatment WBRT RSR13 + WBRT Number of Deaths 167/206 164/208 Median Survival in Months1 (95% C.I.) 4.5 (3.8, 5.4) 5.9 (4.7, 7.0) Hazard Ratio 2 (95% C.I.) 0.844 (0.680, 1.048) P-value 3 0.1217 *: Corrected for miss-classification (i.e., non-randomized subgroup); 1: Kaplan-Meier Estimates; 2 : Hazard Ratio of RSR13 + WBRT/ WBRT; 3: unadjusted log-rank test. The sponsor is seeking approval based on post-hoc analysis in a small subgroup of 115 patients with Breast cancer primary. The results of these comparisons are presented in the following Table C. Table C: Exploratory Survival Analysis in the Subgroup of Patients with Primary Breast Cancer* Treatment WBRT RSR13 + WBRT Number of Deaths 47/55 39/60 Median Survival in Months1 (95% C.I.) 4.6 (3.8, 6.2) 8.7 (6.0, 11.3) Hazard Ratio 2 (95% C.I.) 0.552 (0.359, 0.850) P-value 3 0.0061 *: Corrected for miss-classification (i.e., non-randomized subgroup); : Kaplan-Meier Estimates; 2: Hazard Ratio of RSR13 + WBRT/ WBRT; 3 : unadjusted log-rank test and not adjusted for multiple analyses. 1 3 2 Introduction 2.1 Overview It has been estimated that in the United States 80,000 to 170,000 patients develop brain metastases each year. Standard palliative treatment for symptomatic lesions consists of corticosteroids and whole brain radiation therapy (WBRT). Analysis of a large database compiled by the Radiation Therapy Oncology Group (RTOG) indicates that the overall prognosis of patients with brain metastasis is poor with median survival time of 4-7 months. 2.1.1 Background RSR13 is a synthetic allosteric modifier of hemoglobin. RSR13 emulates the function of natural allosteric modifiers such as 2,3-diphosphoglycerate (2,3DPG). RSR13 is a small molecule that reduces hemoglobin-oxygen binding affinity, described by an increase in p50 (the partial pressure of oxygen [pO2] that results in 50% hemoglobin saturation), and enhances the diffusion of oxygen from the blood to the tissues. Radiation therapy is currently the principal non-surgical therapy to achieve local control of brain metastases from solid tumors. However, the efficacy of radiation therapy (RT) is modified by the extent of tumor oxygenation. Hypoxic tumors are more resistant to cell damage by radiation and tumor hypoxia adversely affects the clinical prognosis of RT. It has been reported in literature that tumors with a low median pO 2 have a higher in-field failure rate after RT. Animal pharmacology studies have shown that RSR13 increases blood p50, increases pO2 in non-tumor and tumor tissue, and increases oxygen diffusive transport in non-tumor tissue. The effect of RSR13 on hemoglobin in the red blood cell to enhance oxygen unloading from hemoglobin, and the diffusion of that oxygen from the vascular compartment into the hypoxic tumor cells is the basis for the radioenhancement effect of RSR13. RSR13 does not need to diffuse into the brain tissue, because oxygen readily diffuses across the blood brain barrier and the cancer cell membrane to increase tumor oxygenation, thereby increasing the effectiveness of RT. The goal of adjunctive RSR13 therapy in cancer patients is to increase tumor O 2 concentration thereby maximizing the cytotoxicity of the treatment modality (RT and/or chemotherapy). The sponsor has submitted results from one phase III, randomized, controlled, open-label clinical trial (registration trial Study RT009) comparing WBRT alone to RSR13 + WBRT, to demonstrate efficacy of RSR13. The sponsor has also 4 provided supportive efficacy data from a phase II, single arm study (Study RT008). The main focus of this review will be on results from Study RT009. 2.1.2 Major Statistical Issues 1. Only one randomized open-label study conducted in patients with brain metastases, which failed to demonstrate efficacy as per the design of the study, in the intent-to-treat population (log-rank test, P-value = 0.1688) and in the co-primary subgroup of patients with NSCLC/Breast cancer primary (log-rank test, P-value = 0.1217). 2. When the overall result fails to show efficacy, usually subgroup findings are not acceptable and subgroup analyses at best can be exploratory or hypothesis generating analyses (ICH E-3 guidelines, section 11.4.2.8: These analyses are not intended to "salvage" an otherwise non-supportive study but may suggest hypotheses worth examining in other studies or be helpful in refining labelling information, patient selection, dose selection etc.). When one starts to do multiple subgroups testing, one can easily make a false positive claim based on such subgroup analysis. We do not know how to interpret the P-values based on such post-hoc analysis. Furthermore, without replication of the results in a second well-controlled study, the subgroup analysis can not be ruled out for a false positive result. 3. The sponsor wishes to claim approval based on a subgroup of patients with primary breast cancer. This subgroup hypothesis corresponding to breast cancer primary patients was not stated as a hypothesis of interest to be tested in the original protocol. Any subgroup hypothesis needs to be stated in the protocol and accordingly proper allocation of α has to be specified. Otherwise, such post-hoc subgroup claim will inflate Type I error and it is difficult to interpret such P-values. 4. Some of the important issues not addressed by the sponsor are: imbalance in patients who were ineligible (per protocol) between the two treatment groups; misclassification of patients in the randomized strata; imbalance in the number of baseline brain lesions. 2.2 Data Sources Data used for review is from the electronic submission received on 12/3/03. The network path is \\Cdsesub1\n21661\N_000\2003-12-03\crt\datasets. Specifically, datasets from Study 009 were reviewed (\\Cdsesub1\n21661\N_000\2003-1203\crt\datasets\rt009). 5 3 Statistical Evaluation 3.1 Evaluation of Efficacy The sponsor has submitted efficacy results from the following two studies: a) Study RT008: A phase II non-randomized, open-label single arm study conducted in 69 patients from 17 centers (16 US, 1 Canada), to evaluate the safety and efficacy of RSR13 with WBRT in patients with brain metastases. b) Study RT009: A phase III randomized, open-label, comparative study conducted in 538 patients from 82 centers (40 US, 15 Canada, 4 Australia, 4 Hungary, 3 Belgium, 3 France, 3 Germany, 3 Israel, 2 Italy, 2 Scotland, 2 Spain and 1 UK), to evaluate safety and efficacy of RSR13 with WBRT compared to WBRT alone in patients with brain metastases. Reviewer’s Comment: Study RT008 is a non-randomized, single arm, open-label study and as such can not evaluate efficacy based on overall survival. Therefore, this review will focus only on the randomized Study RT009 and particularly on the efficacy aspect of this study. Please refer to the clinical review of this application for the evaluation of Study RT008. 3.1.1 Study RT009 Study RT009 was a multicenter international study conducted in patients with brain metastases. This study was initiated on February 16, 2000 and the study was completed on September 24, 2002. The data cut-off date for final efficacy analysis was January 31, 2003. 3.1.1.1 Study Design Study RT009 was a phase III, randomized, open-label, comparative study in patients who would be receiving a standard 2-week (10-day) course of WBRT for brain metastases. Patients were randomized (1:1) to receive RSR13 no longer than 30 minutes prior to daily WBRT or WBRT alone. Patients were stratified at randomization to 4 strata: (1) Radiation Therapy Oncology Group (RTOG) recursive partitioning analysis (RPA) Class I (including non-small cell lung, breast, and other primary cancers), (2) RPA Class II non-small cell lung cancer (NSCLC) primary, (3) RPA Class II breast cancer primary, and (4) RPA Class II 6 primary tumors of various origins (hereafter referred as other primary). The decision tree utilized in the RPA classification is illustrated in Figure 1. Eligibility criteria included that all patients should have Karnofsky Performance Status (KPS) = 70, radiographic studies consistent with brain metastases and a histologically or cytologically confirmed primary malignancy. Patients with small cell lung cancer, extrapulmonary small cell carcinomas, germ cell tumors or lymphomas were excluded from entering the study. Patients included in the study were not to have received prior treatment for brain metastases with WBRT, sterotactic radiosurgery, chemotherapy, hormonal therapy, immunotherapy, or biological agents. All patients were to be assessed for safety from randomization until the initial follow-up visit at 1 month after completion of the radiation therapy (RT) course. Standard follow-up visits were required 3 months after the completion of RT course and every 3 months thereafter until progression, and then followed for subsequent therapies and survival. Figure 1: RPA Classification Decision Tree 7 Reviewer’s Comments: 1. It was clearly stated in the protocol that the randomized stratification was purely for balance between treatment groups and therefore the number of patients in each of the four strata would not be predetermined. 2. There were a total of 23 ineligible patients (17 patients in the WBRT alone arm and 6 patients in the RSR13 + WBRT arm) based on eligibility criteria at entry, who were entered into the study and treated (Please refer to Appendix 1 for the list of ineligible patients by treatment arm). There were greater than 2 times more patients who were ineligible in the control arm compared to the RSR13 arm. Given the open-label nature of the study there is concern for bias due to the apparent imbalance in ineligible patients between the two arms. 3. There were a total of 25 patients who were miss-classified in the strata assignment at randomization (please refer to Appendix 2 for a complete list of miss-classified patients). This miss-classification is not an issue when analyzing the overall intent-to-treat (ITT) population (all patients as randomized to the two treatment groups) using unadjusted analysis. However, this miss-classification can lead to biased results when considering adjusted analysis because the true patient strata will no longer be as randomized. In particular, there were 6 patients where there were major miss-classifications: 2 patients who were randomized as RPA II, breast cancer were later re-classified as RPA II, NSCLC; 1 patient who was randomized as NSCLC was later re-classified as RPA II, other; and 3 patients who were randomized as RPA II, other were later re-classified as RPA II, NSCLC. 3.1.1.2 Treatment Administration Daily administration of RSR13 required placement of a central venous access device (CVAD). RSR13 treatment arm patients received supplemental oxygen (4L/min via nasal cannula) beginning 5 minutes prior to initiation of infusion, during infusion and WBRT, and for at least 15 minutes after completion of daily WBRT. Patients in the control arm of the study did not receive a placebo. Starting dose and dosing adjustment thereafter of RSR13 was based on gender, weight and oxygen saturation measured by standard pulse oximetry (SpO 2). Starting dose of RSR13 in this study was 75 or 100 mg/kg. The dosing instructions were amended 2 times during the course of the study. Patients with SpO2 while breathing room air on any WBRT day < 90% were not treated with RSR13. Before the second amendment, if SpO2 while breathing room air at screening (at rest and during exercise) and on WBRT day 1 was = 93% then RSR13 100mg/kg was administered. If SpO2 while breathing room air at screening (at rest and during exercise) and on WBRT day 1 was 90-92% then 8 RSR13 75mg/kg was administered. After the second protocol amendment, if SpO2 while breathing room air at screening (at rest and during exercise) and on WBRT day 1 was = 93% then RSR13 was administered based on gender and weight as follows: (a) Males (i) if weight = 95kg then 100 mg/kg; (ii) if weight > 95kg then 75 mg/kg, and (b) Females (i) if weight = 70kg then 100 mg/kg; (ii) if weight > 70kg then 75 mg/kg. Reviewer’s comment: The dosing regimen was changed during the course of study based on weight and gender. Therefore it will be difficult to determine the optimum dose that is efficacious based on the results of this study. 3.1.1.3 Study Objectives The study objectives were: (1) to determi ne the effect of RSR13 on primary and secondary efficacy endpoints in patients with brain metastases receiving daily IV doses of RSR13 administered immediately prior to standard whole brain radiation therapy compared to patients receiving standard whole brain radiation therapy alone, and (2) to determine the safety of RSR13 in this patient population. 3.1.1.4 Efficacy Endpoints Primary Efficacy Endpoint of this study was survival. Secondary Efficacy Endpoints included time to radiographic and time to clinical tumor progression in the brain, response rate in the brain, cause of death, and quality of life. Reviewer’s Comment: In the original protocol (Jan 10, 2000), efficacy was to be established based on the primary endpoint of survival in the intent-to-treat total population. Subsequently in amendment 2 (June 5, 2001) per sponsor after enrollment of a total of 222 patients (172 patients in the NSCLC/breast primary subgroup), the protocol was amended to include a co-primary analysis in the subgroup of patients with NSCLC and breast primary cancer. This reviewer’s analysis suggested that by June 5, 2001 there were a total of 173 patients (134 patients in NSCLC/Breast primary subgroup) enrolled into the study (Appendix 4). The only reasoning given by the sponsor to include this subgroup as a co-primary was that the group 9 of patients with NSCLC/breast primary tumors comprised a large and homogenous subpopulation of patients. Given the open-label nature of this study such additions of primary hypotheses are of concern. 3.1.1.5 Sample Size Considerations In the original protocol (Jan 10, 2000), a sample size requirement of a total of 408 eligible patients was estimated based on the following assumptions: a mix of 20% RPA class I and 80% RPA class II patients, median survival in WBRT to be 4.57 months, an expected 35% increase in median survival in WBRT with RSR13 (median survival of 6.17 months), 18 months of accrual, a shape parameter of 0.20, and 80% power to detect the difference in survival at two-sided overall significance level of 0.05. It was estimated that a total of 308 deaths from both arms would be required to detect the survival difference in the overall ITT population. It was expected that there might be 5% ineligible patients and therefore a total of 408 patients were required to be entered on the study. This sample size calculation was amended in amendment 2 (June 5, 2001) as follows: The sample size was increased to a total of 501 patients in order to observe 402 deaths by increasing the power of the study to detect the survival difference (median survival 4.57 months versus 6.17 months in the overall ITT population) to 85%, increasing accrual time to 27 months, and changing the shape parameter to zero (O’Brien and Fleming). In this amendment the sponsor also added a co-primary analysis in the subgroup of patients with NSCLC/breast primary tumors. It was stated that in this subgroup a total of 308 deaths will be required to provide a power of 75% with a two-sided significance level of 0.05. Furthermore, in this amendment it was stated that the expected number of patients to be enrolled into the study would be between 501-538 patients, depending on the percentage of patients with ‘other’ as a primary cancer. If 25% of enrolled patients had ‘other’ primary, then a total of 501 patients would be enrolled; if 30% of enrolled patients had ‘other’ primary, then a total of 538 patients would be enrolled. Reviewer’s Comments: 1. Reason for changing the space parameter was not specified in the amendment. 2. There was no specific scientific reason given to include a co-primary in the subgroup of patients with NSCLC/breast primary other than that this was a large, homogenous subgroup. 3. A total of 538 patients were enrolled in this study. Only 23% of the patients had ‘other’ primary tumors. 10 3.1.1.6 Interim Analysis In the original protocol (Jan 10, 2000) it was stated that, one interim analysis of the primary study endpoint (overall survival), would be conducted. The interim analysis was planned to be performed when 50% of expected events (154 deaths) occurred. In the first amendment of the protocol (Mar 2, 2000) the section on interim analysis was revised to state that a stochastic analysis would be performed at the time of interim analysis and reported to Data Safety Monitoring Committee (DSMC). It was also stated in this amendment that if the significance level of the log-rank test between treatment arms was less than 0.0077 then the null hypothesis would be rejected. Additionally, if the stochastic analysis indicated less than 15% power to observe the alternative hypothesis, then enrollment to the study may be recommended to be stopped. In the second amendment of the protocol (June 5, 2001), this section was again revised. It stated that interim status and safely reports will be prepared for the independent DSMC every 6 months until planned study enrollment was achieved. Furthermore, with the revised increase in sample size, the interim analysis for efficacy was to be conducted when 50% events (201 deaths) had occurred in the total patient population. The results of interim analysis would be reported to the DSMC. If the significance level of the log-rank test between treatment arms was less than 0.0052 then the null hypothesis would be rejected and the DSMC might recommend stopping enrollment to the study. On the other hand if the analysis indicated less than 15% power to observe the alternative hypothesis, then also it may be recommended to stop further accrual. Reviewer’s Comment: It appears that one interim efficacy analysis was conducted (March 22, 2002). Specific results of this interim analysis or DSMC meeting minutes deliberating on interim efficacy results have not been submitted (Refer to Appendix 5 for FDA analysis). Because of this interim analysis, the significance level for testing at the time of final analysis needs to be adjusted to maintain an overall family-wise level of significance of 0.05. 11 3.1.1.7 Efficacy Analysis Methods Primary Efficacy Analysis: In the original protocol (dated Jan 10, 2000), it was specified that the primary endpoint, overall survival, measured from the time of randomization into the study, would be compared between treatment arms by unadjusted log-rank test. The median survival time would be estimated in both treatment arms. Other specific point estimates of clinical interest for each treatment were 6-month and 1year survival. It was stated that additional subgroup analyses would be performed if there were sufficient numbers of patients across subgroups. RPA class, site of primary cancer, and other important covariates such as primary tumor control, age, presence of extracranial metastases, baseline KPS, and number of metastatic lesions, would be included in a multivariate Cox model along with treatment arm to test the relative importance of these factors for survival. In the first amendment of the protocol (dated March 2, 2000), it was specified that the primary final analysis of the study would be undertaken when all patients have been potentially followed for a minimum of 6 months and the planned number of deaths (308) had been observed. The primary analysis would be conducted on an intent-to-treat patient population using log-rank statistic (unadjusted for covariates) and evaluable subgroup analyses might be performed to provide supportive evidence of efficacy. In the second (dated June 5, 2001, sample size increased, co-primary added) and third (dated October 9, 2001) amendments of the protocol, it was specified that the primary final analysis would be undertaken when the planned number of deaths in both the total study population (402) and the NSCLC/breast subpopulation (308) had been observed. The primary analysis would be conducted on an intent-to-treat basis and evaluable subgroup analyses might be performed to provide supportive evidence of efficacy. Furthermore, it was stated that a modified Bonferroni adjustment for multiple comparisons (co-primary analyses) would be made. The adjusted significance level for the final analysis after accounting for one interim analysis was set at 0.048. The statistical analysis plan (SAP) which was finalized on July 29, 2002, revised the analysis data set and specified that the analysis data set will consist of eligible patients only. The SAP specified that: (1) patients without brain metastases, patients with leptomenigeal metastases, patients with confirmed primaries of small cell lung cancer, extrapulmonary small cell carcinomas, germ cell tumors and lymphomas and (2) patients having prior treatment for brain metastases with WBRT or stereotactic radiosurgery, and patients with prior surgical resection of brain metastases with no remaining lesions would be ineligible and would be excluded from analysis. The SAP also stated that the primary analysis of the 12 overall patient population as well as the NSCLC/breast subpopulation would be performed using the unadjusted log-rank test. In the SAP the list of covariates that would be included in the Cox-model was also revised. It stated that the following covariates would be included: age (continuous as well as above and below 65 years old), baseline weight (divided by gender as per the dosing guidelines of amendment 2), number of cranial metastases (1, 2-3, 4 or more), baseline cranial tumor total area, gender, RPA class, site of primary cancer, primary tumor control, number of extracranial metastases (0, 1-2, 3 or more), presence of liver metastases, usage of subsequent treatment (systemic vs. non-systemic, any vs. none), baseline KPS, diagnosis timing (definition to follow), prior treatment for cranial metastases (yes/no; prior treatment may delay time from diagnosis to radiation therapy), worldwide location (USA vs. Canada vs. others, North America vs. Others), altitude, baseline hemoglobin, and size of center. Center size was the binary variable designating a center as large or small. Under the section on covariates, the SAP also stated that ‘While designated prospectively, supporting analyses should be considered exploratory in nature, and inferences made based on p-values should be done so with caution. Primary reasons for exploratory analyses are for estimation rather than hypothesis testing’. Secondary Efficacy Analyses: The protocol has specified that the secondary endpoints, time to radiographic tumor progression in brain and time to clinical tumor progression in the brain, would be analyzed using cumulative incidence model and that the treatment arms would be compared using the method of Pepe. It is also stated that analyses within strata, within other prognostic groups and Cox model analysis would also be performed. Secondary endpoint response rate (best maximal response) in the brain would be determined from MRI or CT scans and the frequency distribution of CR:PR:SD:PD would be compared for each treatment arm. Between treatment arms comparison would be made using Cochran-Mantel Haenzel test. Regarding secondary endpoint cause of death, frequency of neurologic/ nonneurologic/ undistinguishable deaths would be computed for each treatment arm and compared between treatment arms using Cochran-Mantel Haenzel test. Secondary endpoint of quality of life would be determined by the Spitzer Questionnaire and KPS assessment. The frequency distribution would be computed for each treatment arm by time of follow-up. 13 Reviewer’s Comments: 1. In the original protocol and in its amendments it was clearly stated that the primary analysis would be conducted in ITT population. The sample size and the power considerations were in fact based on ITT population. 2. The co-primary hypothesis testing in the NSCLC/Breast subgroup was added during the course of study. The protocol did not clearly define this subgroup, i.e., whether patients from strata 2 and 3 only to be included in this subgroup or include NSCLC/Breast primary patients in strata 1, and strata 2 and 3. 3. In using Modified Bonferroni adjustment for co-primary analysis one could consider 2 methods of adjustment after accounting for one interim analysis: (1) compare larger of the 2 p-values with 0.048, (a) if the larger p-value is < 0.048, then reject both hypotheses, or (b) if the larger p-value is > 0.048, then compare the smaller of the two p-values with 0.024 (Hochberg’s SU modified Bonferroni procedure), or (2) the second procedure (Holm’s SD modified Bonferroni) is (a) if the smaller of the two p-values is > 0.024, then both hypotheses are not significant, or (b) if the smaller of the p-value < 0.024 then the corresponding hypothesis is significant and then test if the larger of the p-value is < 0.048. 4. In the original protocol and in its amendments as well as in the SAP, it was clearly stated that the primary analysis would be based on unadjusted log-rank test. 5. The SAP was finalized after all the patients were entered into the study (last 3 patients were entered on July 29, 2002). Given the open-label nature of the study it is of concern that the analysis population was changed after all patients were entered into the study. 6. Overall there were 23 patients who were ineligible, 17 in the control arm and 6 in the RSR13 arm. With greater than 2 times more patients who were ineligible in the control arm, given the study was an open-label study, there is concern for bias. 7. There was no justification provided for the inclusion of additional several covariates for the exploratory Cox analysis in the SAP. 8. The covariate ‘diagnosis timing’ was not defined. 9. Because no apriori probability of type I error allocation has been specified, analyses of secondary efficacy endpoints can only be considered as exploratory and supportive to primary efficacy analysis. 3.1.1.8 Sponsor’s Results and Statistical Reviewer’s Findings/ Comments In the RT009 study, a total of 538 patients were randomized to receive WBRT alone (267 patients) or RSR13 followed by WBRT (271 patients). 14 3.1.1.8.1 Baseline Characteristics Table 1 lists the number of patients entered in each of the randomized strata. The baseline Characteristics of the overall population and NSCLC/Breast subgroup are presented in Tables 3 & 4. Reviewer’s Comments: 1. There were 25 patients who were miss-classified at randomization according to the sponsor / CRF. Therefore the corrected numbers in each of the strata are presented in Table 2. 2. It should be noted that the patients as listed in each strata in Table 2 are no longer as randomized. 3. This miss-classification may not affect analysis based on ITT population. However analyses based on subgroups could potentially lead to biased results. 4. The sponsor has reported that 9 patients (4 in the control (WBRT) arm and 5 in the RSR13 arm) withdrew from the study prior to WBRT day 1 (Appendix 3). 5. In the overall patient population the baseline characteristics appear to be balanced between the two treatment arms. 6. It should be noted that the NSCLC/Breast subgroup as presented in Table 3 is no longer as randomized, because: (1) NSCLC and breast cancer patients from the Strata 1, RPA class I, are included in this subgroup, and (2) patients who were miss-classified in the incorrect stratum at randomization are re-classified into corrected primary tumors based on the reported diagnosis in CRF. Table 1: Number of Patients as Randomized in Each Stratum by Treatment Arm Strata RPA Class I RPA Class II, NSCLC RPA Class II, Breast RPA Class II, Other WBRT 28 (10.5%) 132 (49.4%) 51 (19.1%) 56 (21.0%) RSR13 + WBRT 29 (10.7%) 132 (48.7%) 52 (19.2%) 58 (21.4%) Total 57 (10.6%) 264 (49.1%) 103 (19.1%) 114 (21.2%) Table 2: Number of Patients as Observed in Each Stratum by Treatment Arm Strata RPA Class I RPA Class II, NSCLC RPA Class II, Breast RPA Class II, Other WBRT 24 (9.0%) 136 (50.9%) 50 (18.7%) 57 (21.3%) RSR13 + WBRT 22 (8.1%) 133 (49.1%) 56 (20.7%) 60 (22.1%) Total 46 (8.6%) 269 (50.0%) 106 (19.7%) 117 (21.7%) 15 Table 3: Baseline Characteristics in ITT Population Characteristic Gender: Female Male Race: Caucasian Non-Caucasian Age Group: < 65 years = 65 years Age in yrs: Mean (S.D.) Median (Range) Weight in Kg: Mean (S.D.) Median (Range) KPS Group: < 90 = 90 KPS: Mean (S.D.) Bidirectional product (mm2) for baseline lesions: Mean (S.D.) Median (Range) Resting SpO2: Mean (S.D.) Primary Controlled : No Yes Extracranial metastases: 0 1 2 =3 Number of Brain Lesions: 1 2 =3 Liver Metastases: No Yes Lung Metastases: No Yes Synchronous Disease: No Yes Prior Brain Mets Treatment: No Yes Hemoglobin (g/dL): Mean (S.D.) Creatinine (mg/dL): Mean (S.D.) Albumin (g/dL): Mean (S.D.) ALT (IU/L): Mean (S.D.) Primary Site: NSCLC Breast Other WBRT 150 (56.2%) 117 (43.8%) 239 (89.5%) 28 (10.5%) 197 (73.8%) 70 (26.2%) 57.0 (11.0) 57 (23 – 81) 72.5 (17.1) 70.5 (33 – 140.9) 124 (46.4%) 143 (53.6%) 85.2 (9.7) WBRT + RSR13 153 (56.5%) 118 (43.5%) 242 (89.3%) 29 (10.7%) 196 (72.3%) 75 (27.7%) 57.1 (11.1) 57 (30 -87) 71.3 (15.0) 71.0 (39.8 – 122) 113 (41.7%) 158 (58.3%) 85.1 (9.7) 760.8 (694.8) 587.5 (4, 4200) 96.8 (1.7) 200 (74.9%) 67 (25.1%) 96 (36.0%) 69 (25.8%) 55 (20.6%) 47 (17.6%) 53 (20.2%) 81 (30.9%) 128 (48.9%) 225 (84.3%) 42 (15.7%) 183 (68.5%) 84 (31.5%) 184 (68.9%) 83 (31.1%) 238 (89.1%) 29 (10.9%) 13.5 (1.5) 0.79 (0.23) 3.70 (0.45) 40.9 (34.2) 151 (56.5%) 55 (20.6%) 61 (22.9%) 753.6 (735) 518 (16, 5080) 96.7 (1.8) 199 (73.4%) 72 (26.6%) 84 (31.0%) 72 (26.6%) 56 (20.7%) 59 (21.7%) 45 (16.9%) 82 (30.7%) 140 (52.4%) 217 (80.1%) 54 (19.9%) 179 (66.1%) 92 (33.9%) 183 (67.5%) 88 (32.5%) 250 (92.3%) 21 (7.7%) 13.3 (1.5) 0.76 (0.21) 3.65 (0.47 40.5 (49.4) 148 (54.6%) 60 (22.1%) 63 (23.3%) 16 Table 4: Baseline Characteristics in NSCLC/Breast* Subgroup Characteristic WBRT WBRT + RSR13 Gender: Female 130 (63.1%) 128 (61.5%) Male 76 (36.9%) 80 (38.5%) Race: Caucasian 184 (89.3%) 184 (88.5%) Non-Caucasian 22 (10.7%) 24 (11.5%) Age Group: < 65 years 150 (72.8%) 150 (72.1%) = 65 years 56 (27.2%) 58 (27.9%) Age in yrs: Mean (S.D.) 57.1 (11.2) 56.9 (11.0) Median (Range) 57 (26 – 81) 57 (31 – 80) Weight in Kg: Mean (S.D.) 70.7 (15.9) 71.1 (15.0) Median (Range) 68.8 (33 – 124.1) 69.6 (41.1 – 122) KPS Group: < 90 89 (43.2%) 87 (41.8%) = 90 117 (56.8%) 121 (58.2%) KPS: Mean (S.D.) 85.7 (9.5) 85.1 (9.6) 2 Bidirectional product (mm ) for baseline lesions: Mean (S.D.) 755.6 (668.2) 767.8 (767.8) Median (Range) 573.5 (17 – 3806) 459 (16 – 5080) Resting SpO2: Mean (S.D.) 96.9 (1.7) 96.8 (1.8) Primary Controlled : No 156 (75.7%) 159 (76.4%) Yes 50 (24.3%) 49 (23.6%) Extracranial metastases: 0 76 (36.9%) 75 (36.0%) 1 51 (24.8%) 54 (26.0%)) 2 42 (20.4%) 42 (20.2%) =3 37 (18.0%) 37 (17.8%) Number of Brain Lesions: 1 43 (21.1%) 37 (18.0%) 2 60 (29.4%) 66 (32.2%) =3 101 (49.5%) 102 (49.8%) Liver Metastases: No 176 (85.4%) 176 (84.6%) Yes 30 (14.6%) 32 (15.4%) Lung Metastases: No 148 (71.8%) 155 (74.5%) Yes 58 (28.2%) 53 (25.5%) Synchronous Disease: No 140 (68.0%) 133 (63.9%) Yes 66 (32.0%) 75 (36.1%) Prior Brain Mets Treatment: No 189 (91.8%) 197 (94.7%) Yes 17 (8.2%) 11 (5.3%) Hemoglobin (g/dL): Mean (S.D.) 13.4 (1.5) 13.4 (1.4) Creatinine (mg/dL): Mean (S.D.) 0.77 (0.22) 0.75 (0.20) Albumin (g/dL): Mean (S.D.) 3.7 (0.4) 3.6 (0.5) ALT (IU/L): Mean (S.D.) 39.2 (33.0) 41.9 (54.4) *: Revised group per reclassification (corrected) and including RPA Class I patients. 17 3.1.1.8.2 Primary Efficacy Analyses Primary efficacy analysis per original protocol, comparing overall survival between WBRT and RSR13 + WBRT, in the ITT population using unadjusted log-rank test is presented in Table 5 (same as reported by the sponsor). There were a total of 441/538 patients who had events (deaths) at the time of the final analysis. The Kaplan-Meier curves for the ITT population are illustrated in Figure 2. The efficacy analysis in the subgroup of NSCLC/Breast primary patients is presented in Table 6 (same as reported by the sponsor). The KaplanMeier curves for the NSCLC/Breast subgroup is presented in Figure 3. There were 331/414 deaths in this subgroup at the time of the final analysis. Table 5: Primary Efficacy Survival Analysis in ITT Population Treatment WBRT RSR13 + WBRT 1 3 Number of Deaths 221/267 220/271 Median Survival in Months1 (95% C.I.) 4.5 (3.7, 5.4) 5.3 (4.5, 6.2) Hazard Ratio 2 (95% C.I.) 0.877 (0.727, 1.057) P-value 3 0.1688 : Kaplan-Meier Estimates; 2: Hazard Ratio of RSR13 + WBRT/ WBRT; : unadjusted log-rank test. Figure 2: Kaplan-Meier Survival Curves in the ITT Population 18 Table 6: Co-Primary Efficacy Survival Analysis in NSCLC/Breast Primary Cancer Subgroup* P-value 3 Hazard Median Survival 2 1 Ratio in Months (95% C.I.) (95% C.I.) WBRT 167/206 4.5 (3.8, 5.4) 0.844 0.1217 (0.680, 1.048) RSR13 + WBRT 164/208 5.9 (4.7, 7.0) *: Corrected for miss-classification (i.e., non-randomized subgroup); 1 : Kaplan-Meier Estimates; 2: Hazard Ratio of RSR13 + WBRT/ WBRT; 3 : unadjusted log-rank test. Treatment Number of Deaths Figure 3: Kaplan-Meier Survival Curves in the Subgroup of Patients with NSCLC/Breast Primary Reviewer’s Comments: 1. RSR13 + WBRT treatment failed to demonstrate superior survival over WBRT alone in the randomized ITT population (Table 5 and Figure 2 above). The final analysis was conducted after observing the required number of deaths (required 402 deaths, observed 441 deaths) 19 specified in the protocol. The median survival in the WBRT arm was slightly less than what was expected in the protocol design. 2. RSR13 + WBRT treatment failed to demonstrate superior survival over WBRT alone in the subgroup of patients with NSCLC/Breast primary (Co-primary analysis, Table 6 and Figure 3 above). The final analysis was conducted after observing the required number of deaths (required 308 deaths, observed 331 deaths) specified in the protocol. 3. There appears to be an imbalance between the treatment arms in the number of patients who were not eligible (17 in the WBRT alone arm and 6 in the RSR13 + WBRT arm). The results of exploratory analyses in eligible patients only or per-protocol patients in the overall population and in the NSCLC/Breast primary subgroup of patients are presented respectively, in the following Tables 7 and 8. These results also fail to demonstrate superior survival of RSR13 + WBRT treatment over WBRT alone. 4. There were 30/441 early deaths within 1 month from the start of the study. Of the 30 deaths 16 were in the WBRT alone arm (2 Breast, 11 NSCLC and 3 Other primaries), and 14 were in the RSR13 + WBRT arm (4 Breast, 3 NSCLC and 7 Other primaries). It appears that there were more early deaths in the control arm compared to RSR13 arm in the NSCLC and breast primary subgroups. Given these numerical differences and open-label nature of the study it is uncertain if bias was introduced in patient selection and allocation. Table 7: Exploratory Survival Analysis in the Per-Protocol Overall Population Treatment Number Median Survival Hazard P-value 3 of Deaths in Months1 Ratio 2 (95% C.I.) (95% C.I.) WBRT 206/250 4.4 (3.7, 5.3) 0.871 0.1549 (0.719, 1.054) RSR13 + WBRT 215/265 5.4 (4.6, 6.3) 1 : Kaplan-Meier Estimates; 2: Hazard Ratio of RSR13 + WBRT/ WBRT; 3 : unadjusted log-rank test and not adjusted for multiple analyses. Table 8: Exploratory Survival Analysis in the Per-Protocol NSCLC/Breast Primary Cancer Subgroup* Treatment Number Median Survival Hazard P-value 3 of Deaths in Months1 Ratio 2 (95% C.I.) (95% C.I.) WBRT 157/194 4.4 (3.7, 5.2) 0.815 0.0693 (0.654, 1.017) RSR13 + WBRT 159/203 6.0 (4.7, 7.1) *: Corrected for miss-classification (i.e., non-randomized subgroup); 1 : Kaplan-Meier Estimates; 2: Hazard Ratio of RSR13 + WBRT/ WBRT; 3 : unadjusted log-rank test and not adjusted for multiple analyses. 20 3.1.1.8.3 Exploratory Covariate Adjusted and Subgroup Survival Analyses The sponsor had specified exploratory covariate adjusted survival analyses using Cox model. The sponsor had also specified exploratory survival analyses in each of the randomized strata. In this section the results of these exploratory analyses are presented. In the original protocol and its amendments, 7 covariates were mentioned as likely to be included in the Cox model (Refer to section 3.1.1.7, and Table 9 below). After completion of accrual this analysis was revised in the SAP to include 18 covariates (Refer to section 3.1.1.7, and Table 9 below) in various combinations of continuous and categorical variables resulting in 48 Cox models (submitted by sponsor including 17/18 covariates, all models included the same 17 covariates, not presented here). Results of Cox regression analysis including the 7 covariates specified in the protocol are presented in Table 11 (FDA analysis). One of the models including the 18 covariates in the per-protocol population as specified in the SAP is presented in Table 12 (FDA analysis). Table 9: Covariates Intended to be Included in the Cox Model Protocol Covariates RPA Class Site of Primary Cancer Primary Tumor Control Age Presence of Extracranial Metastases Baseline KPS Number of Metastatic Lesions SAP Covariates RPA Class Site of Primary Cancer Primary Tumor Control Age Baseline KPS Number of Cranial Metastases Number of Extracranial Metastases Baseline Cranial Tumor Total Area Baseline Weight (divided by gender as per the dosing guidelines) Gender Presence of Liver Metastases Usage of Subsequent Treatment* Diagnosis Timing Prior Treatment to Cranial Metastases Worldwide Location Altitude Baseline Hemoglobin Size of Center * Not included in sponsor’s adjusted Cox models 21 Table 10: Cox’s Proportional Hazard Model Adjusting for Covariates in the ITT Population (Protocol Planned Model) Covariates Hazard Ratio 95% C.I. Treatment (RSR13 + WBRT/WBRT) 0.814 0.674, 0.984 RPA Class (1 vs. 2) 0.742 0.471, 1.168 Site of Primary Cancer: Breast (Yes vs. No) 0.568 0.423, 0.764 NSCLC (Yes vs. No) 0.861 0.682, 1.085 Primary Tumor Control (Yes vs. No) 1.310 1.006, 1.707 Age 1.014 1.005, 1.023 Presence of Extracranial Metastases (No vs. Yes) 1.138 0.800, 1.618 Baseline KPS 0.968 0.958 – 0.978 Number of Metastatic Lesions 2 1.287 1.111 – 1.490 1: P-values not adjusted for multiplicity; 2: Since all patients were supposed to have brain metastases, for the purpose of this analysis ‘number of extracranial metastases’ was used as the covariate in place of ‘number of metastatic lesions’. P-value 1 0.0335 0.1973 0.0002 0.2050 0.0453 0.0022 0.4732 < 0.0001 0.0008 Table 11: Cox’s Proportional Hazard Model Adjusting for Covariates in the Overall Eligible Patient Population (SAP Planned Model)* Covariates Hazard Ratio 95% C.I. Treatment (RSR13 + WBRT/WBRT) 0.777 0.640, 0.942 RPA Class (1 vs. 2) 0.763 0.479, 1.215 Site of Primary Cancer: Breast (Yes vs. No) 0.602 0.430, 0.842 NSCLC (Yes vs. No) 0.826 0.640, 1.065 Primary Tumor Control (Yes vs. No) 1.238 0.927, 1.652 Age Group ( < 65 vs. = 65 yrs) 1.486 1.178, 1.875 Baseline KPS Group (= 90 vs. < 90) 1.564 1.283, 1.907 Number of Cranial Metastases 1.148 1.000, 1.320 Number of Extracranial Metastases 1.237 1.102, 1.389 Baseline Cranial Tumor Total Area ( 1000) Baseline Weight Group (Low vs. High) 0.971 0.765, 1.232 Gender (Female vs. Male) 1.407 1.120, 1.767 Presence of Liver Metastases (No vs. Yes) 1.249 0.941, 1.658 Usage of Subsequent Treatment (No vs. Yes) 0.910 0.695, 1.192 Diagnosis Timing (metachronous vs. synchronous) 1.122 0.870, 1.448 Prior Treatment to Cranial Metastases (No vs. Yes) 0.450 0.302, 0.671 Worldwide Location: USA (No vs. Yes) 0.921 0.665, 1.275 Canada (No vs. Yes) 0.919 0.639, 1.322 Altitude (Low vs. High) 1.096 0.805, 1.491 Baseline Hemoglobin Group (= 12 vs. < 12 g/dL) 1.336 1.027, 1.738 Size of Center (Not a big site vs. Big site) 0.965 0.762, 1.222 *: Results based on a total of 528 patients; 1 : P-values not adjusted for multiplicity; 22 P-value1 0.0103 0.2547 0.0031 0.1409 0.1481 0.0008 < 0.0001 0.0508 0.0003 0.3418 0.8096 0.0033 0.1232 0.4937 0.3737 < 0.0001 0.6199 0.6505 0.5604 0.0308 0.7679 Reviewer’s Comments: 1. Some of the covariates included in the model are likely to be highly correlated. For example, RPA Classification takes into account whether primary was controlled or not, age and whether there was metastasis in brain only or not. 2. The models specified by the sponsor both in the protocol and in SAP are ambiguous and questionable. In the protocol specified model, the meaning of the covariate, number of metastatic lesions, is unclear. One could interpret it as total number of metastases (cranial + extracranial metastases), or number of brain lesions. Furthermore both the number of cranial lesions, extracranial metastases, and baseline tumor area appear like continuous variables, but in fact are categorized as zero, 1, 2 or 3. Regarding baseline cranial tumor area, the variable included in the model was the variable submitted by the sponsor as ‘GBDPTOT’ and the explanation given for this variable is that it is bi-dimensional product for baseline lesions. It is assumed in the above analysis that these are measurements for cranial lesions only. The covariate, diagnosis timing, was not defined in the protocol. The sponsor had used this as a categorical variable: synchronous diagnosis or not. One could interpret it as the actual time in days to the diagnosis of brain metastasis. 3. Scientific basis or literature citation in choosing these covariates (either in the protocol or SAP) for the model was not provided by the sponsor. Some of the important covariates, such as, response to steroid treatment, systemic tumor activity, LDH, interval between primary tumor and development of brain metastases, reported in literature as significant prognostic factors for survival, were not included in these analyses and data needed for such evaluation were not collected. 4. Furthermore, when appropriately adjusted for multiplicity, treatment differences are unlikely to be significant. P-values from these exploratory covariate analyses can not be taken at face value. 5. Results from Cox regression analyses including only the randomized strata are presented in Tables 12-14. The treatment effect was not significant in any of these models. Please refer to Appendix 6 for exploratory Cox regression analyses in NSCLC/Breast Primary subgroup. 6. The sponsor did not include the covariate ‘usage of subsequent therapy’ (as specified in the final SAP) in their Cox models 23 Table 12: Cox’s Proportional Hazard Model Adjusting for Strata (As Randomized) in the ITT Population Covariates Hazard Ratio 95% C.I. Treatment (RT + RSR vs. RT) 0.871 0.722 – 1.050 Stratum 2** = RPA Class 2, Primary Lung 1.638 1.175 – 2.284 Stratum 3** = RPA Class 2, Primary Breast 1.388 0.954 – 2.022 Stratum 4** = RPA Class 2, Primary Other 2.142 1.488 – 3.083 * P-values not adjusted for multiplicity; ** Strata as randomized (25 patients were miss-classified) P-value* 0.1484 0.0036 0.0870 < 0.0001 Table 13: Cox’s Proportional Hazard Model Adjusting for Re-classified Strata in the ITT Population Covariates Hazard Ratio 95% C.I. Treatment (RT + RSR vs. RT) 0.879 0.729 – 1.061 Stratum 2** = RPA Class 2, Primary Lung 1.436 1.003 – 2.056 Stratum 3** = RPA Class 2, Primary Breast 1.158 0.777 – 1.726 Stratum 4** = RPA Class 2, Primary Other 1.933 1.313 – 2.846 * P-values not adjusted for multiplicity; ** Strata as observed or intended (25 patients were re-classified) P-value* 0.1799 0.0479 0.4714 0.0008 Table 14: Cox’s Proportional Hazard Model Adjusting for Primary Site (Including RPA Class I and Primary Site as Observed) in the ITT Population Covariates Treatment (RT + RSR vs. RT) Primary Lung Primary Breast * P-values not adjusted for multiplicity Hazard Ratio 0.893 0.774 0.618 95% C.I. 0.740, 1.077 0.617, 0.969 0.466, 0.821 7. Exploratory survival analyses results in each of the primary site subgroups (NSCLC, Breast, and Other) are presented in Tables 15-17 (results same as reported by the sponsor) and Kaplan-Meier Curves are illustrated in Figures 4-6. 24 P-value* 0.2362 0.0258 0.0009 Table 15: Exploratory Survival Analysis in the Subgroup of Patients with Primary NSCLC P-value 3 Hazard Median Survival 2 1 Ratio in Months (95% C.I.) (95% C.I.) WBRT 120/151 4.4 (3.5, 5.7) 0.991 0.9426 (0.771, 1.273) RSR13 + WBRT 125/148 4.9 (4.1, 6.2) 1 : Kaplan-Meier Estimates; 2: Hazard Ratio of RSR13 + WBRT/ WBRT; 3 : unadjusted log-rank test and not adjusted for multiple analyses. Treatment Number of Deaths Figure 4: Kaplan-Meier Curves in the Subgroup of Patients with Primary NSCLC 25 Table 16: Exploratory Survival Analysis in the Subgroup of Patients with Primary Breast Cancer P-value 3 Hazard Median Survival 2 1 Ratio in Months (95% C.I.) (95% C.I.) WBRT 47/55 4.6 (3.8, 6.2) 0.552 0.0061 (0.359, 0.850) RSR13 + WBRT 39/60 8.7 (6.0, 11.3) 1 : Kaplan-Meier Estimates; 2: Hazard Ratio of RSR13 + WBRT/ WBRT; 3 : unadjusted log-rank test and not adjusted for multiple analyses. Treatment Number of Deaths Figure 5: Kaplan-Meier Curves in the Subgroup of Patients with Primary Breast Cancer 26 Table 17: Exploratory Survival Analysis in the Subgroup of Patients with Primary ‘Other’ Cancer P-value 3 Hazard Median Survival 2 1 Ratio in Months (95% C.I.) (95% C.I.) WBRT 54/61 3.7 (2.5, 6.0) 1.029 0.8812 (0.708, 1.496) RSR13 + WBRT 56/63 4.0 (2.9, 5.6) 1 : Kaplan-Meier Estimates; 2: Hazard Ratio of RSR13 + WBRT/ WBRT; 3 : unadjusted log-rank test and not adjusted for multiple analyses. Treatment Number of Deaths Figure 5: Kaplan-Meier Curves in the Subgroup of Patients with Primary ‘Other’ Cancer 8. Any subgroup analysis results are relevant only if overall study (ITT) is positive. 9. Hypotheses testing in these subgroups, and a allocation for testing these subgroup hypotheses were not prespecified. Therefore, the P-values obtained in these subgroup analyses are not interpretable without a prespecified significance level. 27 10. There were no significant differences in the treatment effect in the subgroups of patients with primary NSCLC or Other Cancer. 11. The apparent treatment difference observed in the subgroup of patients with primary breast cancer can only be considered as hypothesis generating. The efficacy claim in this non-randomized, non-prespecified subgroup is based on post-hoc, data driven hypothesis testing in a very small subgroup (115 patients in total) of patients from a single study. 12. Well-controlled studies are required to approve a drug (21 CFR 314.126(a): Reports of adequate and well-controlled investigations provide the primary basis for determining whether there is "substantial evidence" to support the claims of effectiveness for new drugs). Large, well conducted, controlled, randomized study is required particularly when considering single study for consideration of approval. 13. The Guidance for Industry: Providing clinical evidence of effectiveness from human drug and biological products (May 1998) clearly states that: ‘When considering whether to rely on a single multicenter trial, it is critical that the possibility of an incorrect outcome be considered and that all the available data be examined for their potential to either support or undercut reliance on a single multicenter trial’. 14. Furthermore, of the 115 patients with breast primary, 8 patients (6 in WBRT arm and 2 in the RSR13 +WBRT arm) were not eligible patients (did not meet the inclusion criteria), 7 patients (3 in WBRT arm and 4 in RSR13 + WBRT arm) were misclassified, 1 patient in the control arm was withdrawn from the study prior to receiving any treatment, 6 patients (2 in WBRT arm, 4 in RSR13+WBRT arm) were dead within one month, and 15 patients (8 in WBRT arm, 7 in RSR13 + WBRT arm) were dead within 2 months from the start of the study. Among the patients who were terminated early in the RSR13 arm, 1 patient received only one dose, 5 patients received 2 doses only, and 2 patients received 5 doses only. 15. There appears to be imbalance in some of the baseline patient characteristics favoring the RSR13 + WBRT arm compared to WBRT alone arm in the subgroup of patients with breast primary. These baseline characteristics include, weight, performance status = 90, tumor burden, 3 or more extracranial metastases, 3 or more brain lesions, presence of lung metastases and prior brain metastases treatment (Table 18 below, bolded characteristics). Please refer to the clinical review of this application for other imbalances such as, differences in oxygen administration between the two treatment arms in the subgroup of patients with breast cancer primary. 16. Furthermore, in the breast primary subgroup majority were younger (< 65 years old) women with metachronous diagnosis compared to the other subgroups. 17. Without replication of the results in a second well-controlled study, the subgroup analysis can not be ruled out for a false positive result. 28 18. ICH E– 3, Section 11.4.2.8, clearly specifies guidelines for conducting subgroup analyses, namely, ‘These analyses are not intended to "salvage" an otherwise non-supportive study but may suggest hypotheses worth examining in other studies or be helpful in refining labelling information, patient selection, dose selection etc.’ Therefore, (a) Examining a subgroup of patients with primary breast cancer when the overall study is not-supportive, is not acceptable; (b) Because the randomized strata have been modified, the subgroup under consideration is not a randomized subgroup; (c) Apparent imbalances between the treatment groups with respect to some of the baseline characteristics may potentially be driving the difference in survival. 19. Although no drug has been approved for breast cancer patients with brain metastasis, in general the approval of drugs in advanced breast cancer are based on relatively large studies. 29 Table 18: Baseline Characteristics in Breast* Subgroup Characteristic Gender: Female Male Race: Caucasian Non-Caucasian Age Group: < 65 years = 65 years Age in yrs: Mean (S.D.) Median (Range) Weight in Kg: Mean (S.D.) Median (Range) KPS Group: < 90 = 90 KPS: Mean (S.D.) Bidirectional product (mm2) for baseline lesions: Mean (S.D.) Median (Range) Resting SpO2 : Mean (S.D.) Primary Controlled : No Yes Extracrania l metastases: 0 1 2 =3 Number of Brain Lesions: 1 2 =3 Liver Metastases: No Yes Lung Metastases: No Yes Synchronous Disease: No Yes Prior Brain Mets Treatment: No Yes Hemoglobin (g/dL): Mean (S.D.) Creatinine (mg/dL): Mean (S.D.) Albumin (g/dL): Mean (S.D.) ALT (IU/L): Mean (S.D.) WBRT 54 (98.2%) 1 (1.8%) 48 (87.3%) 7 (12.7%) 45 (81.8%) 10 (18.2%) 53.9 (11.2) 53 (30-78) 68.2 (17.5) 64 (42-124.1) 24 (43.6%) 31 (56.4%) 85.3 (9.2) WBRT + RSR13 60 (100.0%) 0 (0.0%) 50 (83.3%) 10 (16.7%) 48 (80.0%) 12 (20.0%) 52.0 (11.6) 51 (31-80) 73.2 (14.7) 72.9 (46.5-122) 24 (40.0%) 36 (60.0%) 85.5 (9.6) 882.1 (695.1) 699 (17-3588) 97.5 (1.8) 37 (67.3%) 18 (32.7%) 8 (14.6%) 8 (14.6%) 17 (30.9%) 22 (40.0%) 5 (9.3%) 9 (16.7%) 40 (74.1%) 36 (65.5%) 19 (34.5%) 23 (41.8%) 32 (58.2%) 53 (96.4%) 2 (3.6%) 51 (92.7%) 4 (7.3%) 13.0 (1.6) 0.78 (0.28) 3.9 (0.5) 36.4 (29.5) 761.9 (705.8) 578.5 (16-2936) 96.9 (1.7) 41 (68.3%) 19 (31.7%) 7 (11.7%) 14 (23.3%) 20 (33.3%) 19 (31.7%) 13 (21.7%) 13 (21.7%) 34 (56.7%) 39 (65.0%) 21 (35.0%) 31 (51.7%) 29 (48.3%) 58 (96.7%) 2 (3.3%) 58 (96.7%) 2 (3.3%) 12.7 (1.2) 0.67 (0.12) 3.7 (0.42) 40.4 (44.9) *: Revised group per reclassification (corrected) and including RPA Class I patients. 30 3.1.1.8.4 Secondary Efficacy Analyses Results submitted by the sponsor on the evaluation of secondary efficacy endpoints will be briefly summarized in this section. The protocol specified secondary endpoints were time to radiographic and time to clinical tumor progression in the brain, response rate in the brain, cause of death, and quality of life. Time to Radiographic Tumor Progression in the Brain Time to radiographic tumor progression, as determined by Central Radiology Review, was estimated for all patients using cumulative incidence analysis and Kaplan-Meier methods, and tested between treatment arms using Gray’s test. Death in this analysis was recorded as a competing risk when it occurred prior to diagnosis of radiographic progression. Per sponsor’s report, there was no statistically significant difference in the cumulative incidence of radiographic progression between the WBRT alone and RSR13 + WBRT arms (?2 = 0.458, p-value = 0.4986). The sponsor has also reported that there was no statistically significant difference in the cumulative incidence of radiographic progression between the WBRT alone and RSR13 + WBRT arms in the subset of patients with NSCLC primary (p-value = 0.8142), or Breast primary (p-value = 0.8023) or Other primary (p-value = 0.3597). Time to Clinical Progression in the Brain Time to clinical tumor progression, was estimated for all patients using cumulative incidence analysis and Kaplan-Meier methods, and tested between treatment arms using Gray’s test. Death in this analysis was recorded as a competing risk when it occurred prior to diagnosis of clinical progression. Per sponsor’s report, there was no statistically significant difference in the cumulative incidence of clinical progression between the WBRT alone and RSR13 + WBRT arms (?2 = 0.595, p-value = 0.4407). The sponsor has also reported that there was no statistically significant difference in the cumulative incidence of clinical progression between the WBRT alone and RSR13 + WBRT arms in the subset of patients with NSCLC primary (p-value = 0.8142), or Breast primary (p-value = 0.8023) or Other primary (p-value = 0.3597). Response Rate in the Brain Best response was determined from MRI o