General Information
Conflict of Interest
Request Clinical Trials
Associated Trial(s):- NCT00638690 - A Phase 3, Randomized, Double-Blind, Placebo-Controlled Study of Abiraterone Acetate (CB7630) Plus Prednisone in Patients With Metastatic Castration-Resistant Prostate Cancer Who Have Failed Docetaxel-Based Chemotherapy
- NCT00887198 - A Phase 3, Randomized, Double-blind, Placebo-Controlled Study of Abiraterone Acetate (CB7630) Plus Prednisone in Asymptomatic or Mildly Symptomatic Patients With Metastatic Castration-Resistant Prostate Cancer
- NCT01695135 - A Phase 3, Randomized, Double-blind, Placebo-Controlled Study of Abiraterone Acetate (JNJ-212082) Plus Prednisone in Patients With Metastatic Castration-Resistant Prostate Cancer Who Have Failed Docetaxel-Based Chemotherapy
- NCT01591122 - A Phase 3, Randomized, Double-blind, Placebo-Controlled Study of Abiraterone Acetate (JNJ-212082) Plus Prednisone in Asymptomatic or Mildly Symptomatic Patients With Metastatic Castration-Resistant Prostate Cancer
- NCT01867710 - A Randomized Phase 2 Study Evaluating Abiraterone Acetate With Different Steroid Regimens for Preventing Symptoms Associated With Mineralocorticoid Excess in Asymptomatic, Chemotherapy-naïve and Metastatic Castration-resistant Prostate Cancer (mCRPC) Patients
- NCT02236637 - A Prospective Registry of Patients With a Confirmed Diagnosis of Adenocarcinoma of the Prostate Presenting With Metastatic Castrate-Resistant Prostate Cancer
Request Clinical Trials
Data Request Status
Status: OngoingResearch Proposal
Project Title: RCT-Anchored External Control Validation in mCRPC Using Trial and Registry IPD
Scientific Abstract:
Background: External control methods are increasingly discussed in oncology, but reliable use requires robust control of transportability, selection bias, and residual confounding.
Objective: To develop and validate an RCT-anchored external control framework for metastatic castration-resistant prostate cancer using de-identified individual participant-level data from randomized trials and a prospective registry.
Study Design: Retrospective methodological study of harmonized trial and registry datasets.
Participants: Men with metastatic castration-resistant prostate cancer enrolled in requested YODA studies, analyzed within clinically aligned settings rather than by naive pooling across distinct disease states.
Primary and Secondary Outcome Measure(s): Primary outcome is overall survival. Secondary outcomes include radiographic progression-free survival, time to PSA progression, PSA response, time to subsequent chemotherapy, and selected safety endpoints where harmonization is feasible.
Statistical Analysis: We will perform cohort alignment, descriptive analyses, benchmark randomized analyses, and external-control emulation analyses. Methods will include Kaplan-Meier estimation, Cox models, propensity score weighting/matching, overlap weighting, doubly robust estimation where feasible, and an RCT-anchored dynamic borrowing framework. Performance will be evaluated against randomized benchmarks using bias, RMSE, interval coverage, interval width, calibration, and conclusion concordance.
Brief Project Background and Statement of Project Significance:
External control approaches are receiving increasing attention in oncology because they may improve efficiency in settings where concurrent control enrollment is difficult, where randomized sample sizes are limited, or where historical and registry data are potentially informative. However, the practical challenge is not simply identifying a seemingly similar external dataset. Valid external borrowing requires alignment of disease state, treatment line, eligibility, baseline prognostic factors, endpoint definitions, follow-up structure, and patterns of subsequent therapy. Even when conventional balance diagnostics appear acceptable, residual transportability bias may remain and can lead to misleading treatment effect estimates or overconfident inference.
Metastatic castration-resistant prostate cancer (mCRPC) is a particularly appropriate setting in which to study this problem. Within the YODA Project, several randomized trials and a prospective registry are available in closely related mCRPC populations, including post-docetaxel and chemotherapy-naive/asymptomatic or mildly symptomatic settings. This makes it possible to evaluate external control construction in a setting where clinically relevant alignment is feasible, while also preserving the opportunity to benchmark against known randomized comparisons. In other words, these data allow not only application of external control methods, but direct validation of when such methods succeed and when they fail.
The proposed project is significant for both methodological and applied reasons. Methodologically, it will assess whether an RCT-anchored framework can improve the validity of external control construction relative to more conventional approaches such as propensity score methods or trial-only analyses. It will also quantify the consequences of limited overlap, residual covariate imbalance, and cross-dataset heterogeneity on bias, precision, and confidence interval coverage. Applied significance comes from the fact that oncology development increasingly considers externally informed comparisons, yet there remains uncertainty about how much borrowing is defensible and under what conditions.
By focusing on clinically aligned mCRPC settings and using randomized trials as reference standards, this study is designed to generate empirical evidence on the credibility, limitations, and appropriate use of external controls. The results should be relevant to trialists, biostatisticians, and regulatory scientists interested in causal inference, transportability, and the design and evaluation of externally informed oncology studies.
Specific Aims of the Project:
Aim 1: Develop an RCT-anchored external control framework for clinically aligned mCRPC settings using YODA individual participant-level data.
Aim 2: Validate the framework through trial-to-trial and trial-to-registry emulation experiments, using randomized comparisons as reference standards.
Aim 3: Compare the proposed approach with conventional methods, including trial-only analyses, propensity score methods, overlap weighting, and doubly robust estimators where feasible.
Aim 4: Quantify residual transportability challenges using prespecified diagnostics for covariate balance, overlap, effective sample size, and effect calibration.
Study Design: Methodological research
What is the purpose of the analysis being proposed? Please select all that apply.: Develop or refine statistical methods Research on clinical trial methods Research on comparison group
Software Used: Python
Data Source and Inclusion/Exclusion Criteria to be used to define the patient sample for your study:
Data will be requested only from the YODA Project: NCT00638690, NCT00887198, NCT01695135, NCT01591122, NCT01867710, and NCT02236637. No non-YODA individual participant-level datasets will be pooled in the primary analysis.
The study sample will be restricted to men with metastatic castration-resistant prostate cancer (mCRPC), age >=18 years, with histologically/cytologically confirmed adenocarcinoma of the prostate and evidence of castration-level testosterone, consistent with the parent trials/registry. To preserve clinical comparability, analyses will be conducted within aligned settings rather than by naive pooling across all studies.
For NCT00638690 and NCT01695135, we will include the intent-to-treat population meeting the post-docetaxel mCRPC setting. For NCT00887198, NCT01591122, and NCT01867710, we will include the intent-to-treat population meeting the chemotherapy-naive, asymptomatic or mildly symptomatic mCRPC setting. For NCT02236637, we will include registry patients with confirmed mCRPC who can be aligned to one of the above target settings using baseline treatment history and disease status.
Additional analytic exclusions applied across datasets will be: missing treatment assignment (if randomized), missing baseline date, missing survival follow-up information, duplicated records, and patients who cannot be harmonized to the pre-specified target trial setting. Patients from distinct disease states (eg, hormone-sensitive or non-metastatic CRPC) will not be included. No other external studies will be used for pooled IPD analyses.
Primary and Secondary Outcome Measure(s) and how they will be categorized/defined for your study:
Primary outcome: overall survival (OS), defined as time from randomization (or registry baseline/index date for NCT02236637) to death from any cause. OS is selected as the main endpoint because it is clinically important and available or derivable across the requested mCRPC datasets.
Secondary outcomes: (1) radiographic progression-free survival (rPFS), defined according to each parent study protocol and then harmonized as closely as possible across datasets; (2) time to PSA progression, using protocol-defined PSA progression criteria; (3) PSA response rate, defined as a protocol-consistent confirmed PSA decline when available; (4) time to initiation of subsequent cytotoxic chemotherapy, where available; and (5) selected safety endpoints, including grade 3/4 adverse events and treatment discontinuation due to adverse events, where harmonization is feasible.
A pre-specified endpoint hierarchy will be used. If a secondary endpoint cannot be harmonized with sufficient validity across all included datasets, it will be analyzed only in the subset with comparable definitions and clearly labeled as secondary or exploratory in any publication. No change to the primary outcome is planned.
Main Predictor/Independent Variable and how it will be categorized/defined for your study:
The main independent variable is the treatment-comparison framework, operationalized at two levels.
At the patient level, the primary exposure for effect estimation is treatment assignment within each target comparison (eg, abiraterone acetate plus prednisone/prednisolone vs protocol control among randomized trial participants, or analogous externally constructed control membership in emulation analyses).
At the design level, the principal methodological predictor is comparator type: concurrent randomized control, external trial-based control, registry-based external control, or dynamically weighted hybrid control. This variable will be categorized a priori and used to evaluate how borrowing from external data affects treatment-effect estimation, bias, precision, and interval coverage relative to the randomized benchmark.
Other Variables of Interest that will be used in your analysis and how they will be categorized/defined for your study:
Variables used for cohort characterization and multivariable adjustment will include: age; race/ethnicity where available; ECOG performance status; baseline pain/symptom burden; prior docetaxel exposure and number of prior systemic treatment lines; type of progression at baseline (PSA, radiographic, or both); baseline PSA; laboratory markers available across studies (eg, hemoglobin, alkaline phosphatase, liver function tests, albumin, creatinine); metastatic burden/site indicators (bone, liver, lymph node, visceral disease as available); prior corticosteroid use; comorbidity proxies where available; and study-level identifiers.
Additional variables of interest will include treatment start date, randomization date/index date, follow-up time, censoring indicators, radiographic assessment dates, PSA assessment dates, subsequent anticancer therapy, and adverse event summaries.
Continuous variables will generally be analyzed continuously, with clinically interpretable categories used for descriptive summaries or sensitivity analyses. Categorical variables will be harmonized to common definitions before analysis. Variables with substantial cross-study incompatibility will not be forced into pooled adjustment models and will instead be addressed through setting-specific analyses or sensitivity analyses.
Statistical Analysis Plan:
This is a retrospective methodological study using de-identified individual participant-level data from randomized trials and a prospective registry in mCRPC. The analysis will proceed in five stages.
First, we will perform documentation review, variable harmonization, and cohort alignment. We will define two clinically coherent target settings: (1) post-docetaxel mCRPC and (2) chemotherapy-naive, asymptomatic/mildly symptomatic mCRPC. Descriptive analyses will summarize baseline characteristics overall and by dataset/treatment group using means/SDs or medians/IQRs for continuous variables and counts/percentages for categorical variables.
Second, we will evaluate cross-dataset comparability using standardized mean differences, overlap diagnostics, missingness summaries, event-rate comparisons, and effective sample size after weighting. Bivariate comparisons will be descriptive and diagnostic rather than inferentially primary.
Third, we will conduct benchmark randomized analyses within the parent trials using Kaplan-Meier methods, log-rank tests, and Cox proportional hazards models for time-to-event endpoints, and generalized linear models for binary endpoints such as PSA response or adverse event outcomes. These analyses will establish the internal randomized reference for each aligned setting.
Fourth, we will emulate external-control scenarios. For each target setting, one randomized study will serve as the anchor benchmark, and comparator patients will be reconstructed from other trials and/or the registry. We will compare several approaches: (a) trial-only analysis, (b) propensity score matching, (c) inverse probability weighting, (d) overlap weighting, (e) doubly robust/AIPW estimation where feasible, and (f) an RCT-anchored dynamic borrowing framework that adaptively downweights poorly transportable external information. Survival outcomes will be analyzed primarily with weighted Cox models and restricted mean survival time analyses as sensitivity checks. Flexible machine learning methods may be used as nuisance-function learners or for representation learning in sensitivity analyses, but the primary estimands and benchmark comparisons will remain grounded in prespecified causal inference analyses.
Fifth, performance will be evaluated against the known randomized benchmark using bias, absolute error, RMSE, confidence interval coverage, interval width, calibration of estimated effects, effective sample size, and qualitative conclusion concordance. Prespecified sensitivity analyses will examine alternative covariate sets, stricter overlap restrictions, complete-case versus multiply imputed analyses when appropriate, and analyses excluding datasets/endpoints with limited harmonizability.
No non-YODA participant-level datasets will be pooled in the primary analysis. Any external non-IPD information, if referenced, will be limited to contextual interpretation and not combined with the YODA IPD for primary estimation.
Narrative Summary: There is growing interest in using data external to a randomized trial to supplement or construct control arms in oncology. However, the validity of such external controls depends on careful alignment of disease state, eligibility, endpoint definition, prognostic factors, and uncertainty quantification. We propose a methodological study in metastatic castration-resistant prostate cancer (mCRPC) using YODA individual participant-level data from multiple randomized trials and one prospective registry. The goal is to develop and validate an RCT-anchored external control framework and to compare it with conventional causal inference approaches using randomized comparisons as reference standards.
Project Timeline: Estimated project start: within 2-4 weeks after data access is granted and the Data Use Agreement is executed. Months 1-2: documentation review, data familiarization, cohort alignment, and harmonization of baseline variables and endpoints. Months 3-5: development of analysis datasets, descriptive analyses, benchmark randomized analyses, and implementation of external-control methods. Months 6-8: primary comparative analyses, sensitivity analyses, and robustness checks. Months 9-10: finalize statistical outputs, draft manuscript, and internal revision. First manuscript submission is anticipated by Month 10 or 11 after project start. Results will be reported back to the YODA Project promptly upon completion of the primary analysis, and no later than the end of the approved access period. Because the YODA access period is 12 months with possible extension, this timeline is designed to complete the primary project within the initial access window.
Dissemination Plan:
Anticipated products include at least one primary manuscript focused on RCT-anchored external control methodology in metastatic castration-resistant prostate cancer, and potentially one secondary methods paper focused on transportability diagnostics, weighting, and survival-outcome calibration. The target audiences are clinical trial methodologists, oncology statisticians, regulatory scientists, and investigators interested in external control arms and dynamic borrowing.
Potential journals include Clinical Trials, Statistics in Medicine, Pharmaceutical Statistics, Journal of Clinical Epidemiology, Trials, BMC Medical Research Methodology, and oncology-oriented methodology or translational journals as appropriate. Results may also be presented at methodological or oncology meetings. Findings will be reported in a transparent manner, including limitations related to transportability, endpoint harmonization, and residual confounding, and a summary of results will be provided back to the YODA Project.
Bibliography:
- de Bono JS, Logothetis CJ, Molina A, et al. Abiraterone and Increased Survival in Metastatic Prostate Cancer. N Engl J Med. 2011;364:1995-2005.
- Ryan CJ, Smith MR, de Bono JS, et al. Abiraterone in Metastatic Prostate Cancer without Previous Chemotherapy. N Engl J Med. 2013;368:138-148.
- Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika. 1983;70(1):41-55.
- Hernán MA, Robins JM. Using Big Data to Emulate a Target Trial When a Randomized Trial Is Not Available. Am J Epidemiol. 2016;183(8):758-764.
- Dahabreh IJ, Robertson SE, Stuart EA, Hernán MA. Generalizing causal inferences from individuals in randomized trials to all trial-eligible individuals. Biometrics. 2019;75(2):685-694.
- Dahabreh IJ, Robertson SE, Steingrimsson JA, Stuart EA, Hernán MA. Extending inferences from a randomized trial to a new target population. Stat Med. 2020;39(14):1999-2014.
- Li F, Thomas LE, Li F. Addressing Extreme Propensity Scores via the Overlap Weights. Am J Epidemiol. 2019;188(1):250-257.
- Ventz S, Lai A, Cloughesy TF, Wen PY, Trippa L, Alexander BM. Design and Evaluation of an External Control Arm Using Prior Clinical Trials and Real-World Data. Clin Cancer Res. 2019;25(16):4993-5001.
- Ventz S, Nie L, Cloughesy T, et al. The design and evaluation of hybrid controlled trials that leverage external data and randomization. Nat Commun. 2022;13:5783.
- Rahman R, Ventz S, McDunn J, et al. Leveraging external data in the design and analysis of clinical trials. Ther Innov Regul Sci. 2021.
- Kurz CF. Augmented Inverse Probability Weighting and the Double Robustness Property. Med Decis Making. 2022;42(2):156-167.